Add new file

.gitattributes

@@ -1,6 +1,7 @@
 *.7z filter=lfs diff=lfs merge=lfs -text
 *.arrow filter=lfs diff=lfs merge=lfs -text
 *.bin filter=lfs diff=lfs merge=lfs -text
+*.csv filter=lfs diff=lfs merge=lfs -text
 *.bz2 filter=lfs diff=lfs merge=lfs -text
 *.ckpt filter=lfs diff=lfs merge=lfs -text
 *.ftz filter=lfs diff=lfs merge=lfs -text

TreeTest.py

@@ -0,0 +1,13 @@
+from ddparser import DDParser
+#2.4.0版本 的paddlepaddle
+import  nltk
+# ddp = DDParser()
+from nltk.corpus import treebank
+# print(res)
+
+text=nltk.word_tokenize("论文从存货内部控制的五要素，看YT 超市存货内部控制的整体现状，发现其内部控制各关键环节存在的问题，并对原因进行分析，结合YT 超市的具体情况，依次分析存在问题环节的关键风险点、控制目标，并提出具体改进建议。论文从存货的分类角度去发现重点环节内部控制存在的问题。论文选题恰当，结构合理，论述清晰，写作规范，合乎逻辑。论文达到硕士毕业水平。")
+t = treebank.parsed_sents('wsj_0001.mrg')[0]
+t.draw()
+print(text)
+print(nltk.pos_tag(text))
+

analyze_data.py

@@ -0,0 +1,23 @@
+import jieba
+import matplotlib.pyplot as plt
+from tqdm import tqdm
+
+from utils import parse_user_reviews
+
+
+def analyze(contents):
+    sent_lengths = []
+
+    for sentence in tqdm(contents):
+        seg_list = list(jieba.cut(sentence.strip()))
+        # Update word frequency
+        sent_lengths.append(len(seg_list))
+
+    num_bins = 100
+    n, bins, patches = plt.hist(sent_lengths, num_bins, facecolor='blue', alpha=0.5)
+    plt.show()
+
+
+if __name__ == '__main__':
+    user_reviews = parse_user_reviews('train')
+    analyze(user_reviews['content'])

bertviz_test.py

@@ -0,0 +1,23 @@
+# coding: utf-8
+from transformers import AutoTokenizer, AutoModel, utils
+from bertviz import model_view, head_view
+
+
+utils.logging.set_verbosity_error()  # Suppress standard warnings
+model_name='./chinese_wwm_pytorch'
+input_text = '论文书写规范，逻辑合理，反映作者具有一定的分析问题、解决问题的能力'
+tokenizer = AutoTokenizer.from_pretrained('./chinese_wwm_pytorch')
+tokenizer.add_special_tokens({ "additional_special_tokens": [ "[unused1]", "[unused2]", "[unused3]"] })
+model = AutoModel.from_pretrained(model_name, output_attentions=True)  # Configure model to return attention values
+inputs = tokenizer.encode(input_text, return_tensors='pt')  # Tokenize input text
+print('inputs:', inputs)
+outputs = model(inputs)  # Run model
+attention = outputs[-1]  # Retrieve attention from model outputs
+tokens = tokenizer.convert_ids_to_tokens(inputs[0])  # Convert input ids to token strings
+print('tokens:', tokens)
+# print('attention:',attention)
+# model_view(attention, tokens)  # Display model view
+html_head_view = head_view(attention, tokens, html_action='return')
+
+with open("./bert.html", 'w') as file:
+    file.write(html_head_view.data)

chinese_wwm_pytorch/bert_config.json

@@ -0,0 +1,19 @@
+{
+  "attention_probs_dropout_prob": 0.1, 
+  "directionality": "bidi", 
+  "hidden_act": "gelu", 
+  "hidden_dropout_prob": 0.1, 
+  "hidden_size": 768, 
+  "initializer_range": 0.02, 
+  "intermediate_size": 3072, 
+  "max_position_embeddings": 512, 
+  "num_attention_heads": 12, 
+  "num_hidden_layers": 12, 
+  "pooler_fc_size": 768, 
+  "pooler_num_attention_heads": 12, 
+  "pooler_num_fc_layers": 3, 
+  "pooler_size_per_head": 128, 
+  "pooler_type": "first_token_transform", 
+  "type_vocab_size": 2, 
+  "vocab_size": 21128
+}

chinese_wwm_pytorch/config.json

@@ -0,0 +1,20 @@
+{
+  "attention_probs_dropout_prob": 0.1, 
+  "directionality": "bidi", 
+  "hidden_act": "gelu", 
+  "hidden_dropout_prob": 0.1, 
+  "hidden_size": 768, 
+  "initializer_range": 0.02, 
+  "intermediate_size": 3072, 
+  "max_position_embeddings": 512, 
+  "num_attention_heads": 12, 
+  "num_hidden_layers": 12, 
+  "pooler_fc_size": 768, 
+  "pooler_num_attention_heads": 12, 
+  "pooler_num_fc_layers": 3, 
+  "pooler_size_per_head": 128, 
+  "pooler_type": "first_token_transform", 
+  "type_vocab_size": 2, 
+  "vocab_size": 21128,
+  "model_type": "bert"
+}

chinese_wwm_pytorch/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:394eb16febe46f087c2789b2f0a23179191fb6cbea4d02f673a9f65220dbf2eb
+size 411578458

config.py

@@ -0,0 +1,82 @@
+import torch
+from transformers import BertTokenizer
+
+device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+#device = torch.device("cpu")
+
+# Configure training/optimization
+learning_rate = 0.02   #0.002   5e-5
+min_word_freq = 3
+print_every = 100
+chunk_size =3   #100              这个是batchsize
+
+
+# num_labels = 20
+# num_classes = 4   # number of sentimental types
+
+# num_labels = 12
+# num_classes =3   # number of sentimental types
+
+num_labels = 14
+num_classes =3
+
+save_folder = 'models'
+
+# Configure models
+start_epoch = 0
+epochs = 150
+hidden_size = 768   #800
+encoder_n_layers = 12    #1
+dropout = 0.1
+batch_first = False
+
+filter_sizes = (5, 6, 7)
+# train_folder = 'data/ai_challenger_sentiment_analysis_trainingset_20180816'
+# valid_folder = 'data/ai_challenger_sentiment_analysis_validationset_20180816'
+# test_a_folder = 'data/ai_challenger_sentiment_analysis_testa_20180816'
+#
+# train_filename = 'sentiment_analysis_trainingset.csv'
+# valid_filename = 'sentiment_analysis_validationset.csv'
+# test_a_filename = 'sentiment_analysis_testa2.csv'
+'''自定义训练数据'''
+train_folder = 'data/train'
+valid_folder = 'data/valid'
+test_a_folder = 'data/test'
+
+train_filename = 'train_data.csv'
+valid_filename = 'val_data.csv'
+test_a_filename = 'test_data.csv'
+
+# label_names = ['location_traffic_convenience', 'location_distance_from_business_district', 'location_easy_to_find',
+#                'service_wait_time', 'service_waiters_attitude', 'service_parking_convenience', 'service_serving_speed',
+#                'price_level', 'price_cost_effective', 'price_discount', 'environment_decoration', 'environment_noise',
+#                'environment_space', 'environment_cleaness', 'dish_portion', 'dish_taste', 'dish_look',
+#                'dish_recommendation',
+#                'others_overall_experience', 'others_willing_to_consume_again']
+# label_names = ['1', '2', '3',
+#                '4', '5', '6', '7',
+#                '8', '9', '10', '11', '12']
+
+
+# label_names = ['location_traffic_convenience', 'location_distance_from_business_district', 'location_easy_to_find',
+#                'service_wait_time', 'service_waiters_attitude', 'service_parking_convenience', 'service_serving_speed',
+#                'price_level', 'price_cost_effective', 'price_discount', 'environment_decoration', 'environment_noise']
+
+label_names = ['标题', '选题角度与价值','文献综述归纳总结情况','论文工作量','是否掌握基础（专业）知识','是否具备科研能力','格式规范','行文表达','逻辑性','研究方法','研究结论','创新性及论文价值','理论深度','学术端正性'] #a 	b	c	d	e	f	g	h	u	j	k	l
+
+#assert len(label_names) == 12
+
+# Default word tokens
+PAD_token = 0  # Used for padding short sentences
+SOS_token = 1  # Start-of-sentence token
+EOS_token = 2  # End-of-sentence token
+UNK_token = 3
+
+train_path='data/train/train_data.csv'
+test_path='data/test/test_data.csv'
+dev_path='data/valid/val_data.csv'
+bert_path='./chinese_wwm_pytorch'
+pad_size = 500                   # 每段评语处理成的长度(短填长切)
+sentence_pad_size=100            #每句话处理成的长度(短填长切)
+tokenizer = BertTokenizer.from_pretrained(bert_path)
+hidden_size = 768

config2.py

@@ -0,0 +1,76 @@
+import torch
+from transformers import BertTokenizer
+
+device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+#device = torch.device("cpu")
+
+# Configure training/optimization
+learning_rate = 5e-5   #0.002
+min_word_freq = 3
+print_every = 100
+chunk_size =8   #100              这个是batchsize
+
+
+num_labels = 20
+num_classes = 4   # number of sentimental types
+
+# num_labels = 12
+# num_classes =3   # number of sentimental types
+
+
+save_folder = 'models'
+
+# Configure models
+start_epoch = 0
+epochs = 150
+hidden_size = 768   #800
+encoder_n_layers = 12    #1
+dropout = 0.1
+batch_first = False
+
+filter_sizes = (2, 3, 4)
+train_folder = 'data/ai_challenger_sentiment_analysis_trainingset_20180816'
+valid_folder = 'data/ai_challenger_sentiment_analysis_validationset_20180816'
+test_a_folder = 'data/ai_challenger_sentiment_analysis_testa_20180816'
+
+train_filename = 'sentiment_analysis_trainingset.csv'
+valid_filename = 'sentiment_analysis_validationset.csv'
+test_a_filename = 'sentiment_analysis_testa2.csv'
+'''自定义训练数据'''
+# train_folder = 'data/train'
+# valid_folder = 'data/valid'
+# test_a_folder = 'data/test'
+#
+# train_filename = 'train_data.csv'
+# valid_filename = 'val_data.csv'
+# test_a_filename = 'test_data.csv'
+
+label_names = ['location_traffic_convenience', 'location_distance_from_business_district', 'location_easy_to_find',
+               'service_wait_time', 'service_waiters_attitude', 'service_parking_convenience', 'service_serving_speed',
+               'price_level', 'price_cost_effective', 'price_discount', 'environment_decoration', 'environment_noise',
+               'environment_space', 'environment_cleaness', 'dish_portion', 'dish_taste', 'dish_look',
+               'dish_recommendation',
+               'others_overall_experience', 'others_willing_to_consume_again']
+# label_names = ['1', '2', '3',
+#                '4', '5', '6', '7',
+#                '8', '9', '10', '11', '12']
+
+
+# label_names = ['location_traffic_convenience', 'location_distance_from_business_district', 'location_easy_to_find',
+#                'service_wait_time', 'service_waiters_attitude', 'service_parking_convenience', 'service_serving_speed',
+#                'price_level', 'price_cost_effective', 'price_discount', 'environment_decoration', 'environment_noise']
+#assert len(label_names) == 12
+
+# Default word tokens
+PAD_token = 0  # Used for padding short sentences
+SOS_token = 1  # Start-of-sentence token
+EOS_token = 2  # End-of-sentence token
+UNK_token = 3
+
+train_path='data/train/train_data.csv'
+test_path='data/test/test_data.csv'
+dev_path='data/valid/val_data.csv'
+bert_path='./chinese_wwm_pytorch'
+pad_size = 500                   # 每句话处理成的长度(短填长切)
+tokenizer = BertTokenizer.from_pretrained(bert_path)
+hidden_size = 768

data/README.txt

@@ -0,0 +1,2 @@
+sentiment_analysis_testa.csv 为测试集A数据文件，共15000条评论数据
+protocol.txt 为数据集下载协议

data/WORDMAP.json

@@ -0,0 +1,2367 @@
+{
+    "\u8bba\u6587": 4,
+    "\u4ece": 5,
+    "\u5b58\u8d27": 6,
+    "\u5185\u90e8": 7,
+    "\u63a7\u5236": 8,
+    "\u7684": 9,
+    "\u4e94": 10,
+    "\uff0c": 11,
+    "\u770b": 12,
+    " ": 13,
+    "\u8d85\u5e02": 14,
+    "\u6574\u4f53": 15,
+    "\u73b0\u72b6": 16,
+    "\u53d1\u73b0": 17,
+    "\u5176": 18,
+    "\u5404": 19,
+    "\u5b58\u5728": 20,
+    "\u95ee\u9898": 21,
+    "\u5e76": 22,
+    "\u5bf9": 23,
+    "\u539f\u56e0": 24,
+    "\u8fdb\u884c": 25,
+    "\u5206\u6790": 26,
+    "\u7ed3\u5408": 27,
+    "\u73af\u8282": 28,
+    "\u5173\u952e": 29,
+    "\u98ce\u9669": 30,
+    "\u70b9": 31,
+    "\u3001": 32,
+    "\u63d0\u51fa": 33,
+    "\u5177\u4f53": 34,
+    "\u6539\u8fdb": 35,
+    "\u5efa\u8bae": 36,
+    "\u3002": 37,
+    "\n": 38,
+    "\u5206\u7c7b": 39,
+    "\u89d2\u5ea6": 40,
+    "\u53bb": 41,
+    "\u91cd\u70b9": 42,
+    "\u9009\u9898": 43,
+    "\u6070\u5f53": 44,
+    "\u7ed3\u6784\u5408\u7406": 45,
+    "\u8bba\u8ff0": 46,
+    "\u6e05\u6670": 47,
+    "\u5199\u4f5c": 48,
+    "\u89c4\u8303": 49,
+    "\u8fbe\u5230": 50,
+    "\u7855\u58eb": 51,
+    "\u6bd5\u4e1a": 52,
+    "\u6c34\u5e73": 53,
+    "\u603b\u7ed3": 54,
+    "\u4e86": 55,
+    "*": 56,
+    "ST": 57,
+    "\u7c7b": 58,
+    "\u4f01\u4e1a": 59,
+    "\u60c5\u51b5": 60,
+    "\u5168\u9762": 61,
+    "\u68b3\u7406": 62,
+    "\u88ab": 63,
+    "\u5728": 64,
+    "\u8fc7\u7a0b": 65,
+    "\u4e2d": 66,
+    "\u9010\u6e10": 67,
+    "\uff1b": 68,
+    "\u91c7\u53d6": 69,
+    "\u591a\u79cd": 70,
+    "\u8d22\u52a1": 71,
+    "\u624b\u6bb5": 72,
+    "\u4ecd": 73,
+    "\u672a": 74,
+    "\u4f7f": 75,
+    "\u540e": 76,
+    "\u91cd\u6574": 77,
+    "\u91c7\u7528": 78,
+    "\u4ee5": 79,
+    "\u5e02\u573a": 80,
+    "\u4e3a": 81,
+    "\u884c\u4e1a": 82,
+    "\u4ef7\u503c": 83,
+    "\u8d44\u4ea7": 84,
+    "\u7f6e": 85,
+    "\u51fa": 86,
+    "\u548c": 87,
+    "\u4ee5\u53ca": 88,
+    "\u503a\u8f6c\u80a1": 89,
+    "\u7b49": 90,
+    "\u6210\u529f": 91,
+    "\u6700\u540e": 92,
+    "\u5229\u76ca": 93,
+    "\u53ca": 94,
+    "\u6548\u5e94": 95,
+    "\u5f71\u54cd": 96,
+    "\u51fa\u53d1": 97,
+    "\u8bc4\u4ef7": 98,
+    "\u7ecf\u6d4e": 99,
+    "\u8fd0\u7528": 100,
+    "\u4e8b\u4ef6": 101,
+    "\u7814\u7a76\u6cd5": 102,
+    "\u524d\u540e": 103,
+    "\u53cd\u5e94": 104,
+    "\u8fd9": 105,
+    "\u4e00\u7cfb\u5217": 106,
+    "\u884c\u4e3a": 107,
+    "\u90fd": 108,
+    "\u901a\u8fc7": 109,
+    "\u7814\u7a76": 110,
+    "\u53ef\u4ee5": 111,
+    "\u8d44\u672c": 112,
+    "\u63d0\u4f9b": 113,
+    "\u501f\u9274": 114,
+    "\u7ed3\u6784": 115,
+    "\u8f83": 116,
+    "\u5408\u7406": 117,
+    "\u903b\u8f91": 118,
+    "\u4e25\u5bc6": 119,
+    "\u5e7f\u6cdb": 120,
+    "\u9605\u8bfb": 121,
+    "\u76f8\u5173": 122,
+    "\u6587\u732e": 123,
+    "\u5de5\u4f5c\u91cf": 124,
+    "\u9971\u6ee1": 125,
+    "\u7b26\u5408": 126,
+    "\u7855\u58eb\u8bba\u6587": 127,
+    "\u8981\u6c42": 128,
+    "\u8be5": 129,
+    "\u4e3b\u8981": 130,
+    "\u5de5\u4f5c": 131,
+    "\u5982\u4e0b": 132,
+    "\uff1a": 133,
+    "1": 134,
+    "\u9488\u5bf9": 135,
+    "\u578b": 136,
+    "\u7f51": 137,
+    "\u9006\u53d8\u5668": 138,
+    "\u8c10\u632f": 139,
+    "\u5904": 140,
+    "\u7535\u5bb9": 141,
+    "\u7535\u6d41": 142,
+    "\u53cd\u9988": 143,
+    "\u4e0e": 144,
+    "\u53cc": 145,
+    "\u6709\u6e90": 146,
+    "\u963b\u5c3c": 147,
+    "\u6cd5": 148,
+    "\u6291\u5236": 149,
+    "\u524d": 150,
+    "\u7ea7": 151,
+    "\u7535\u538b": 152,
+    "\u6ce2\u52a8": 153,
+    "\u52a0\u5165": 154,
+    "PI": 155,
+    "\u57fa\u4e8e": 156,
+    "\u4ee3\u8868": 157,
+    "\u4f20\u7edf": 158,
+    "\u63a7\u5236\u7b56\u7565": 159,
+    "2": 160,
+    "\u7cfb\u7edf": 161,
+    "\u975e\u7ebf\u6027": 162,
+    "\u81ea\u6297\u6270": 163,
+    "\u7b97\u6cd5": 164,
+    "\u5f85": 165,
+    "\u53c2\u6570": 166,
+    "\u591a": 167,
+    "\u7ebf\u6027": 168,
+    "\u63a7\u5236\u7b97\u6cd5": 169,
+    "\uff08": 170,
+    "\uff09": 171,
+    "\u63a8\u5bfc": 172,
+    "3": 173,
+    "\u5c06": 174,
+    "\u81ea": 175,
+    "\u63a7\u5236\u6280\u672f": 176,
+    "\u8bbe\u8ba1": 177,
+    "\u63a7\u5236\u5668": 178,
+    "\u6765\u6e90\u4e8e": 179,
+    "\u5b9e\u8df5": 180,
+    "\u6709": 181,
+    "\u4e00\u5b9a": 182,
+    "\u5b9e\u9645": 183,
+    "\u4f7f\u7528": 184,
+    "\u5f52\u7eb3": 185,
+    "\u6280\u672f": 186,
+    "\u65b9\u9762": 187,
+    "\u5177\u6709": 188,
+    "\u65b0\u9896\u6027": 189,
+    "\u6240": 190,
+    "\u65b9\u6cd5": 191,
+    "\u5b9e\u7528\u6027": 192,
+    "\u56fe\u8868": 193,
+    "\u5c42\u6b21\u5206\u660e": 194,
+    "\u6587\u5b57": 195,
+    "\u901a\u987a": 196,
+    "\u8bf4\u660e": 197,
+    "\u8be5\u751f": 198,
+    "\u624e\u5b9e": 199,
+    "\u7406\u8bba": 200,
+    "\u57fa\u7840": 201,
+    "\u5f3a": 202,
+    "\u79d1\u7814": 203,
+    "\u80fd\u529b": 204,
+    "\u4e13\u4e1a": 205,
+    "\u6c34\u51c6": 206,
+    "\u540c\u610f": 207,
+    "\u7b54\u8fa9": 208,
+    "\u6211\u56fd": 209,
+    "\u7535\u7f51": 210,
+    "\u5206\u5e03": 211,
+    "\u8f93\u7535": 212,
+    "\u7ebf\u8def": 213,
+    "\u5bb9\u6613": 214,
+    "\u906d\u53d7": 215,
+    "\u5371\u5bb3": 216,
+    "\u914d\u7535": 217,
+    "\u4f5c\u7528": 218,
+    "\u7531\u4e8e": 219,
+    "\u9020\u6210": 220,
+    "\u4e8b\u6545": 221,
+    "\u9ad8": 222,
+    "\u5de6\u53f3": 223,
+    "\u4e25\u91cd": 224,
+    "\u4f9b\u7535": 225,
+    "\u53ef\u9760\u6027": 226,
+    "\u6240\u4ee5": 227,
+    "\u91cd\u8981": 228,
+    "\u610f\u4e49": 229,
+    "\u5efa\u7acb": 230,
+    "\u6570\u5b66\u6a21\u578b": 231,
+    "\u4e24": 232,
+    "\u5f97\u51fa": 233,
+    "\u63aa\u65bd": 234,
+    "\u73b0\u5b9e\u610f\u4e49": 235,
+    "\u7814\u7a76\u6210\u679c": 236,
+    "\u5e94\u7528": 237,
+    "\u7ed3\u679c\u8868\u660e": 238,
+    "\u4f5c\u8005": 239,
+    "\u638c\u63e1": 240,
+    "\u4e13\u4e1a\u77e5\u8bc6": 241,
+    "\u4ece\u4e8b": 242,
+    "\u79d1\u7814\u5de5\u4f5c": 243,
+    "\u7b26\u5408\u89c4\u8303": 244,
+    "\u8bed\u53e5": 245,
+    "\u57fa\u672c": 246,
+    "\u8ba4\u771f": 247,
+    "\u4fee\u6539": 248,
+    "\u53c2\u52a0": 249,
+    "\u914d\u7535\u7f51": 250,
+    "\u968f\u7740": 251,
+    "\u8d1f\u8377": 252,
+    "\u4e0d\u65ad": 253,
+    "\u589e\u5927": 254,
+    "\u5f00\u5173\u67dc": 255,
+    "\u8bbe\u5907": 256,
+    "\u6570\u91cf": 257,
+    "\u589e\u52a0": 258,
+    "\u662f": 259,
+    "\u7edd\u7f18": 260,
+    "\u56e0\u6b64": 261,
+    "\u8d85\u58f0\u6ce2": 262,
+    "\u5730": 263,
+    "\u68c0\u6d4b": 264,
+    "\u5c40\u90e8": 265,
+    "\u653e\u7535": 266,
+    "\u539f\u7406": 267,
+    "\u53ca\u5176": 268,
+    "\u5b9e\u73b0": 269,
+    "\u4ecb\u7ecd": 270,
+    "\u5982\u4f55": 271,
+    "\u5229\u7528": 272,
+    "\u67d0": 273,
+    "\u4e8c\u7ef4": 274,
+    "\u6709\u9650\u5143": 275,
+    "\u7535\u573a": 276,
+    "\u6a21\u578b": 277,
+    "\u627e\u51fa": 278,
+    "\u89c2\u6d4b": 279,
+    "\u51fa\u73b0": 280,
+    "\u8fc7": 281,
+    "\u6848\u4f8b": 282,
+    "\u5224\u65ad": 283,
+    "\u6545\u969c": 284,
+    "\u7c7b\u578b": 285,
+    "\u8d77\u5230": 286,
+    "\u53cd\u6620": 287,
+    "\u7814\u7a76\u751f": 288,
+    "\u57fa\u7840\u77e5\u8bc6": 289,
+    "\u690d\u4fdd": 290,
+    "\u673a\u68b0": 291,
+    "\u55b7\u836f": 292,
+    "\u6548\u7387": 293,
+    "\u4f4e": 294,
+    "\u6548\u679c": 295,
+    "\u5dee": 296,
+    "\u673a": 297,
+    "\u836f\u6db2": 298,
+    "\u6d41\u91cf": 299,
+    "\u76d1\u63a7": 300,
+    "\u7cbe\u51c6": 301,
+    "\u63a7\u5236\u7cfb\u7edf": 302,
+    "\u9700\u6c42": 303,
+    "\u65b9\u6848": 304,
+    "\u9009\u7528": 305,
+    "-": 306,
+    "\u4f5c\u4e3a": 307,
+    "\u5d4c\u5165\u5f0f": 308,
+    "\u8f6f\u4ef6": 309,
+    "\u5f00\u53d1": 310,
+    "\u73af\u5883": 311,
+    "\u5b8c\u6210": 312,
+    "\u901a\u4fe1": 313,
+    "\u6a21\u7cca": 314,
+    "PID": 315,
+    "\u529f\u80fd\u6a21\u5757": 316,
+    "\u7f16\u5199": 317,
+    "\u4e0b": 318,
+    "\u6d4b\u8bd5": 319,
+    "\u529f\u80fd": 320,
+    "\u6b63\u5e38": 321,
+    "\u4e0d\u540c": 322,
+    "\u5de5\u51b5": 323,
+    "\u6027\u80fd": 324,
+    "\u597d": 325,
+    "\u6761\u7406": 326,
+    "\u6bd4\u8f83\u6e05\u695a": 327,
+    "\u6b63\u786e": 328,
+    "\u53d6\u5f97": 329,
+    ",": 330,
+    "\u8868\u660e": 331,
+    "\u57fa\u7840\u7406\u8bba": 332,
+    "\u7f16\u7a0b": 333,
+    "\u8f83\u5f3a": 334,
+    "\u8bba\u6587\u7b54\u8fa9": 335,
+    "\u9009\u62e9": 336,
+    "\u6c22": 337,
+    "\u975e\u6676": 338,
+    "\u5408\u91d1": 339,
+    "\u529b\u5b66\u6027\u80fd": 340,
+    "\u5bf9\u8c61": 341,
+    "\u6df1\u5165": 342,
+    "\u4e86\u89e3": 343,
+    "\u5fae\u89c2": 344,
+    "Zr": 345,
+    "\u57fa": 346,
+    "\u5757\u4f53": 347,
+    "\u7b49\u79bb\u5b50": 348,
+    "\u7535\u5f27": 349,
+    "\u5236\u5907": 350,
+    "\u542b\u91cf": 351,
+    "\u4e09\u79cd": 352,
+    "\u5178\u578b": 353,
+    "\u7eb3\u7c73": 354,
+    "\u53d8\u5f62": 355,
+    "\u5ba4\u6e29": 356,
+    "\u9ad8\u6e29": 357,
+    "\u538b\u7f29": 358,
+    "\u89c4\u5f8b": 359,
+    "\u4ea7\u751f": 360,
+    "\u5f97\u5230": 361,
+    "\u4e00\u4e9b": 362,
+    "\u7ed3\u8bba": 363,
+    "\u672c\u5b66\u79d1": 364,
+    "\u51c6\u786e": 365,
+    "\u9002\u5f53": 366,
+    "\u8868\u5f81": 367,
+    "\u6dfb\u52a0": 368,
+    "\u8868\u73b0": 369,
+    "\u6db2\u76f8": 370,
+    "\u533a": 371,
+    "\u5bbd\u5ea6": 372,
+    "\u51e0\u4e2a": 373,
+    "\u6027\u8d28": 374,
+    "\u521b\u65b0\u6027": 375,
+    "\u6d41\u7545": 376,
+    "\u8868\u8ff0": 377,
+    "\u4e25\u8c28": 378,
+    "\u672c": 379,
+    "Mg": 380,
+    "(": 381,
+    "+": 382,
+    "0.5": 383,
+    "x": 384,
+    ")": 385,
+    "Y": 386,
+    "\u7cfb": 387,
+    "\u70ed\u88c2": 388,
+    "\u654f\u611f\u6027": 389,
+    "\u9884\u6d4b": 390,
+    "\u6536\u96c6": 391,
+    "\u4e24\u79cd": 392,
+    "\u51dd\u56fa": 393,
+    "\u5e76\u4e14": 394,
+    "\u8def\u5f84": 395,
+    "\u56fa\u76f8": 396,
+    "\u5206\u6570": 397,
+    "\u6e29\u5ea6": 398,
+    "\u533a\u57df": 399,
+    "\u8303\u56f4": 400,
+    "\u7279\u5f81": 401,
+    "\u540c\u65f6": 402,
+    "\u76f8": 403,
+    "\u56e0\u7d20": 404,
+    "\u5e94\u529b": 405,
+    "\u8f83\u4e3a": 406,
+    "\u903b\u8f91\u6027": 407,
+    "\u80fd\u591f": 408,
+    "\u89e3\u51b3\u95ee\u9898": 409,
+    "\u4e00\u7bc7": 410,
+    "\u4f18\u79c0": 411,
+    "\u7855\u58eb\u5b66\u4f4d": 412,
+    "\u94f6\u884c": 413,
+    "\u5f53\u524d": 414,
+    "\u4e00\u4e2a": 415,
+    "\u6bd4\u8f83": 416,
+    "\u53d1\u5c55": 417,
+    "\u5065\u5eb7": 418,
+    "\u5b9e\u8bc1": 419,
+    "\u9700\u8981": 420,
+    "\u6587\u7ae0": 421,
+    "\u4e5f": 422,
+    "\u7a0b\u5ea6": 423,
+    "\u4e0a": 424,
+    "\u4f53\u73b0": 425,
+    "\u5bf9\u672c": 426,
+    "\u9886\u57df": 427,
+    "\u5b66\u79d1": 428,
+    "\u57fa\u672c\u77e5\u8bc6": 429,
+    "\u73b0\u5b9e": 430,
+    "\u6570\u636e": 431,
+    "\u5904\u7406": 432,
+    "\u4f46\u662f": 433,
+    "\u89c4\u8303\u6027": 434,
+    "\u5b66\u672f": 435,
+    "\u4e25\u8c28\u6027": 436,
+    "\u8fd8": 437,
+    "\u8fdb\u4e00\u6b65": 438,
+    "\u8be6\u7ec6": 439,
+    "\u610f\u89c1": 440,
+    "\u4e0d\u8db3": 441,
+    "\u65b0": 442,
+    "\u300a": 443,
+    "\u300b": 444,
+    "\u56fd\u5bb6": 445,
+    "\u5c42\u9762": 446,
+    "\u8d8a\u6765\u8d8a": 447,
+    "\u91cd\u89c6": 448,
+    "\u7ee9\u6548\u8bc4\u4ef7": 449,
+    "\u591a\u4e2a": 450,
+    "\u5173\u4e8e": 451,
+    "\u73af\u5883\u4fdd\u62a4": 452,
+    "\u5f00\u59cb": 453,
+    "\u666e\u904d": 454,
+    "\u5173\u6ce8": 455,
+    "\u4e3a\u4e86": 456,
+    "\u63d0\u9ad8": 457,
+    "\u8d28\u91cf": 458,
+    "\u5927\u91cf": 459,
+    "\u4e13\u9879\u8d44\u91d1": 460,
+    "\u5236\u5b9a": 461,
+    "\u4e00\u5957": 462,
+    "\u6709\u6548": 463,
+    "\u663e\u5f97": 464,
+    "\u201c": 465,
+    "MD": 466,
+    "\u68ee\u5de5": 467,
+    "\u516c\u53f8": 468,
+    "\u4fdd\u62a4": 469,
+    "\u5de5\u7a0b": 470,
+    "\u201d": 471,
+    "\u8c03\u67e5": 472,
+    "\u56fd\u5185\u5916": 473,
+    "\u6982\u5ff5": 474,
+    "\u754c\u5b9a": 475,
+    "\u6307\u6807\u4f53\u7cfb": 476,
+    "\u4f53\u7cfb": 477,
+    "\u4fdd\u969c": 478,
+    "\u575a\u5b9e": 479,
+    "\u5b66\u79d1\u77e5\u8bc6": 480,
+    "\u5177\u5907": 481,
+    "\u72ec\u7acb": 482,
+    "\u79d1\u5b66\u7814\u7a76": 483,
+    "\u673a\u6784": 484,
+    "\u6218\u7565": 485,
+    "\u4e00\u6587": 486,
+    "\u4e4b\u95f4": 487,
+    "\u5173\u7cfb": 488,
+    "2017": 489,
+    "\u5e74": 490,
+    "\u9009\u53d6": 491,
+    "\u7535\u5b50": 492,
+    "\u4e09\u4e2a": 493,
+    "\u5bf9\u4e8e": 494,
+    "\u663e\u793a": 495,
+    "\u8bba\u8bc1": 496,
+    "\u6c38\u78c1": 497,
+    "\u540c\u6b65\u7535\u673a": 498,
+    "\u975e\u5e38": 499,
+    "\u672c\u6587": 500,
+    "\u516d\u76f8": 501,
+    "PMSM": 502,
+    "\u8f6c\u6362": 503,
+    "\u8026": 504,
+    "\u5750\u6807\u7cfb": 505,
+    "\u78c1\u573a": 506,
+    "\u5b9a\u5411": 507,
+    "MATLAB": 508,
+    "/": 509,
+    "Simulink": 510,
+    "\u642d\u5efa": 511,
+    "\u4eff\u771f": 512,
+    "\u9a8c\u8bc1": 513,
+    "\u5176\u6b21": 514,
+    "\u5b9a\u5b50": 515,
+    "\u7ed5\u7ec4": 516,
+    "\u65f6": 517,
+    "\u975e": 518,
+    "\u6b63\u4ea4": 519,
+    "\u79cd": 520,
+    "\u5207\u6362": 521,
+    "\u5230": 522,
+    "\u8865\u507f": 523,
+    "\u5b9e\u9a8c": 524,
+    "\u518d\u6b21": 525,
+    "\u6ed1": 526,
+    "\u5f15\u5165": 527,
+    "\u65b0\u578b": 528,
+    "\u6ed1\u6a21": 529,
+    "\u901f\u5ea6": 530,
+    "\u9a71\u52a8": 531,
+    "\u6a21\u5757": 532,
+    "PWM": 533,
+    "\u4f5c\u51fa": 534,
+    "\u8868\u8fbe": 535,
+    "\u6846\u67b6": 536,
+    "\u4f46": 537,
+    "\u4e0d": 538,
+    "\u7a81\u51fa": 539,
+    "\u5c0f": 540,
+    "\u6ce2\u5305": 541,
+    "\u53d8\u6362": 542,
+    "\u9707\u52a8": 543,
+    "\u7279\u6027": 544,
+    "\u7a0b": 545,
+    "\u7a7a\u95f4": 546,
+    "\u76f8\u5173\u6027": 547,
+    "\u6a21\u62df": 548,
+    "\u573a": 549,
+    "\u5c55\u5f00": 550,
+    "\u7406\u8bba\u610f\u4e49": 551,
+    "\u65f6\u53d8": 552,
+    "\u529f\u7387": 553,
+    "\u5b9e\u6d4b": 554,
+    "\u8bb0\u5f55": 555,
+    "\u5e73\u53f0": 556,
+    "\u6027": 557,
+    "\u91cd\u6784": 558,
+    "\u4e00\u79cd": 559,
+    "\u76ee\u6807": 560,
+    "\u52a0\u901f\u5ea6": 561,
+    "\u7b97\u4f8b": 562,
+    "\u6b63\u786e\u6027": 563,
+    "\u8f93\u5165": 564,
+    "\u5386\u53f2": 565,
+    "\u5206\u89e3": 566,
+    "\u5206": 567,
+    "\u4fe1\u53f7": 568,
+    "\u8c03\u6574": 569,
+    "\u518d": 570,
+    "\u6210\u679c": 571,
+    "\u5b9e\u4f8b": 572,
+    "\u8ba1\u7b97": 573,
+    "\u77e5\u8bc6": 574,
+    "4": 575,
+    "\u4e00\u822c": 576,
+    "\u6d77\u6d0b": 577,
+    "\u6cb9\u6c14": 578,
+    "\u8d44\u6e90": 579,
+    "\u5171\u540c\u5f00\u53d1": 580,
+    "\u5165\u624b": 581,
+    "\u9610\u8ff0": 582,
+    "\u53d9\u8ff0": 583,
+    "\u56fd\u9645": 584,
+    "\u793e\u4f1a": 585,
+    "\u7acb\u6cd5": 586,
+    "\u73b0\u6709": 587,
+    "\u56fd\u5916": 588,
+    "\u5305\u62ec": 589,
+    "\u77f3\u6cb9": 590,
+    "\u5236\u5ea6": 591,
+    "\u7ecf\u9a8c": 592,
+    "\u6cd5\u5f8b": 593,
+    "\u5b8c\u5584": 594,
+    "\u5168\u6587": 595,
+    "\u6e05\u695a": 596,
+    "\u8d44\u6599": 597,
+    "\u4e30\u5bcc": 598,
+    "\u4e0d\u5c11": 599,
+    "\u5408\u683c": 600,
+    "\u6bd5\u4e1a\u8bba\u6587": 601,
+    "\u6fc0\u5149": 602,
+    "\u4fe1\u9053": 603,
+    "\u7f16\u7801": 604,
+    "\u8c03\u5236": 605,
+    "\u77e9\u9635": 606,
+    "\u6784\u9020": 607,
+    "\u8f6f": 608,
+    "\u5b9e\u7269": 609,
+    "\u524d\u666f": 610,
+    "\u80e1\u4e54\u6728": 611,
+    "\u4e0d\u4ec5": 612,
+    "\u9a6c\u514b\u601d\u4e3b\u4e49": 613,
+    "\u601d\u60f3": 614,
+    "\u6587\u5316": 615,
+    "\u5ba3\u4f20": 616,
+    "\u76ee\u524d": 617,
+    "\u5f88\u591a": 618,
+    "\u96c6\u4e2d": 619,
+    "\u6216": 620,
+    "\u6bdb\u6cfd\u4e1c\u601d\u60f3": 621,
+    "\u5ef6\u5b89": 622,
+    "\u65f6\u671f": 623,
+    "\u7279\u5b9a": 624,
+    "\u4e0d\u662f": 625,
+    "\u5386\u53f2\u80cc\u666f": 626,
+    "\u601d\u8def": 627,
+    "\u6761\u4ef6": 628,
+    "\u8d21\u732e": 629,
+    "\u9996\u5148": 630,
+    "\u80cc\u666f": 631,
+    "\u5185\u5bb9": 632,
+    "\u5b9e\u8df5\u6027": 633,
+    "\u89c2\u70b9": 634,
+    "\u7ed3\u6784\u8bbe\u8ba1": 635,
+    "\u5c42\u6b21": 636,
+    "\u5f15\u6587": 637,
+    "\u5145\u5206": 638,
+    "\u5e94": 639,
+    "\u63d0\u4ea4": 640,
+    "\u70ed\u7535": 641,
+    "\u8054\u5408": 642,
+    "\u7279\u70b9": 643,
+    "\u8fd0\u884c": 644,
+    "\u5355\u5143": 645,
+    "\u5efa\u6a21": 646,
+    "\u4e0d\u786e\u5b9a\u6027": 647,
+    "\u65f6\u95f4\u5c3a\u5ea6": 648,
+    "\u534f\u8c03": 649,
+    "\u4f18\u5316": 650,
+    "\u80fd\u91cf": 651,
+    "\u7ba1\u7406": 652,
+    "\u5f00\u5c55": 653,
+    "\u673a\u7ec4": 654,
+    "\u53d1\u7535": 655,
+    "\u5305\u542b": 656,
+    "\u7ed9": 657,
+    "\u5236": 658,
+    "\u968f": 659,
+    "\u8d1f\u8f7d": 660,
+    "\u7387": 661,
+    "\u53d8\u5316": 662,
+    "\u533a\u95f4": 663,
+    "\u5bf9\u7cfb\u7edf": 664,
+    "\u4e4b\u524d": 665,
+    "\u8003\u8651": 666,
+    "\u53ef": 667,
+    "\u518d\u751f\u80fd\u6e90": 668,
+    "\u51fa\u529b": 669,
+    "\u786e\u5b9a": 670,
+    "\u7075\u654f\u5ea6": 671,
+    "\u5316": 672,
+    "\u8c03\u5ea6": 673,
+    "\u8f6c\u5316": 674,
+    "\u4e24\u4e2a": 675,
+    "\u6df7\u5408": 676,
+    "\u6c42\u89e3": 677,
+    "\u5b9e\u65f6": 678,
+    "\u9636\u6bb5": 679,
+    "\u6839\u636e": 680,
+    "\u5185": 681,
+    "\u9152\u74f6\u76d6": 682,
+    "\u5370\u5237": 683,
+    "\u677f": 684,
+    "\u7f3a\u9677": 685,
+    "\u89c6\u89c9": 686,
+    "\u7f3a\u5931": 687,
+    "\u673a\u5668": 688,
+    "\u65b9\u5f0f": 689,
+    "\u4ee5\u4e0b": 690,
+    "\u51e0": 691,
+    ".": 692,
+    "\u6210\u50cf": 693,
+    "\u56fe\u50cf\u5904\u7406": 694,
+    "\u8f6f\u4ef6\u8bbe\u8ba1": 695,
+    "\u4ece\u800c": 696,
+    "\u8868\u9762": 697,
+    "\u5e38\u89c1": 698,
+    "\u65e2\u6709": 699,
+    "\u53c8": 700,
+    "\u64b0\u5199": 701,
+    "\u6761\u4f8b": 702,
+    "\u79d1\u6280": 703,
+    "\u4e8e": 704,
+    "\u65b0\u80fd\u6e90": 705,
+    "\u98ce\u673a": 706,
+    "\u6545\u969c\u8bca\u65ad": 707,
+    "\u57f9\u517b": 708,
+    "\u65b9\u5411": 709,
+    "\u7eea\u8bba": 710,
+    "\u63cf\u8ff0": 711,
+    "\u7cbe\u786e": 712,
+    "\u5b50": 713,
+    "\u8fa8\u8bc6": 714,
+    "K": 715,
+    "L": 716,
+    "\u98ce\u7535": 717,
+    "\u4e88\u4ee5": 718,
+    "\u72b6\u6001": 719,
+    "\u8bc4\u4f30": 720,
+    "\u6df1\u523b": 721,
+    "\u8ba4\u8bc6": 722,
+    "\u5b8c\u6574": 723,
+    "\u8bed\u8a00": 724,
+    "\u78f7\u9178": 725,
+    "\u8f9b\u916f": 726,
+    "\u5408\u6210": 727,
+    "\u5de5\u4e1a": 728,
+    "\u3000": 729,
+    "\u4e09": 730,
+    "\u539f\u6599": 731,
+    "\u50ac\u5316\u5242": 732,
+    "\u7ecf\u8fc7": 733,
+    "\u5148": 734,
+    "\u90e8\u5206": 735,
+    "\u916f\u5316": 736,
+    "\u4ea7\u7269": 737,
+    "\u5de5\u827a": 738,
+    "\u7b80\u5355": 739,
+    "\u53ef\u63a7": 740,
+    "\u76f8\u5bf9": 741,
+    "\u8bba\u636e": 742,
+    "\u53ef\u9760": 743,
+    "\u53ef\u4fe1": 744,
+    "\u6761\u7406\u6e05\u6670": 745,
+    "\u7ed3\u679c": 746,
+    "\u53c2\u8003\u4ef7\u503c": 747,
+    "\u5df2\u7ecf": 748,
+    "\u7533\u8bf7": 749,
+    "\u7ec4\u7ec7": 750,
+    "\u7535\u8def": 751,
+    "\u78c1": 752,
+    "\u6574\u6d41": 753,
+    "\u5f62\u6210": 754,
+    "\u6700\u7ec8": 755,
+    "\u5bbd": 756,
+    "\u8f6c\u901f": 757,
+    "\u8f93\u51fa": 758,
+    "\u8f83\u5927": 759,
+    "\u5b9e\u7528\u4ef7\u503c": 760,
+    "\u52a8\u6001": 761,
+    "\u660e\u786e": 762,
+    "\u7535\u673a": 763,
+    "\u505a": 764,
+    "\u6761\u7406\u6e05\u695a": 765,
+    "\u8868\u8fbe\u80fd\u529b": 766,
+    "\u4ece\u6587\u4e2d": 767,
+    "\u770b\u51fa": 768,
+    "\u7b14\u8005": 769,
+    "\u8bfe\u9898": 770,
+    "\u7406\u8bba\u77e5\u8bc6": 771,
+    "\u7ed3\u5408\u5b9e\u9645": 772,
+    "\u67e5\u9605": 773,
+    "\u6587\u732e\u8d44\u6599": 774,
+    "\u9759\u6001": 775,
+    "\u8fd0\u52a8": 776,
+    "\u63d0\u53d6": 777,
+    "\u8ddf\u8e2a": 778,
+    "\u8def\u7ebf": 779,
+    "\u8bba\u6587\u683c\u5f0f": 780,
+    "\u79bb": 781,
+    "\u8f74\u5f0f": 782,
+    "\u5207\u5272": 783,
+    "\u8bd5\u9a8c": 784,
+    "\u7efc\u5408": 785,
+    "\u8981": 786,
+    "\u76ee\u7684": 787,
+    "\u7efc\u8ff0": 788,
+    "\u7f3a\u4e4f": 789,
+    "\u9002\u7528": 790,
+    "\u6d41\u4f53": 791,
+    "\u52a8\u529b\u5b66": 792,
+    "\u8f74": 793,
+    "\u8fd9\u4e9b": 794,
+    "\u5bf9\u6bd4": 795,
+    "\u5355": 796,
+    "\u5206\u522b": 797,
+    "\u8f85\u52a9": 798,
+    "\u96be\u6613": 799,
+    "\u6d89\u53ca": 800,
+    "\u52b1\u78c1": 801,
+    "\u76f4\u7ebf": 802,
+    "\u60ac\u6d6e": 803,
+    "\u540c\u6b65": 804,
+    "\u7535\u52a8\u673a": 805,
+    "\u673a\u7406": 806,
+    "\u78c1\u60ac\u6d6e": 807,
+    "\u795e\u7ecf\u7f51\u7edc": 808,
+    "\u7f51\u7edc": 809,
+    "\u89c4\u6a21": 810,
+    "\u5f62\u5f0f": 811,
+    "\u4e09\u5c42": 812,
+    "BP": 813,
+    "\u5b66\u4e60": 814,
+    "\u6539\u5584": 815,
+    "\u4e3b\u52a8": 816,
+    "\u8f74\u627f": 817,
+    "\u8f6c\u5b50": 818,
+    "\u632f\u52a8": 819,
+    "\u6b64\u57fa\u7840": 820,
+    "\u91cd\u590d": 821,
+    "\u76f8\u7ed3\u5408": 822,
+    "\u590d\u5408": 823,
+    "\u63d0": 824,
+    "\u7b56\u7565": 825,
+    "\u76f8\u5e94": 826,
+    "\u6838\u5fc3": 827,
+    "\u6570\u5b57": 828,
+    "\u826f\u597d": 829,
+    "\u7535\u6c14\u5de5\u7a0b": 830,
+    "\u8c03\u7814": 831,
+    "\u6d3b\u6027": 832,
+    "\u521d\u6b65": 833,
+    "\u4e00": 834,
+    "\u671f\u95f4": 835,
+    "\u53d1\u8868": 836,
+    "8": 837,
+    "\u7bc7": 838,
+    "\u4e00\u9879": 839,
+    "\u56de\u5f52": 840,
+    "\u65f6\u6548": 841,
+    "7075": 842,
+    "\u70ed\u5904\u7406": 843,
+    "\u5e38\u6e29": 844,
+    "\u4f4e\u6e29": 845,
+    "\u51b2\u51fb": 846,
+    "\u7528": 847,
+    "\u5148\u8fdb": 848,
+    "\u65ad\u53e3": 849,
+    "\u89c2\u5bdf": 850,
+    "\u6d4b\u91cf": 851,
+    "\u6240\u5f97": 852,
+    "\u540c\u7c7b": 853,
+    "\u53c2\u8003": 854,
+    "\u6240\u5b66": 855,
+    "\u89e3\u91ca": 856,
+    "\u79d1\u5b66": 857,
+    "\u73b0\u8c61": 858,
+    "\u9519\u8bef": 859,
+    "\u603b\u4f53": 860,
+    "\u66f4\u52a0": 861,
+    "\u591a\u6e90": 862,
+    "\u5fae\u7f51": 863,
+    "\u80fd\u6e90": 864,
+    "\u6784\u5efa": 865,
+    "\u4f20\u8f93": 866,
+    "\u8def\u7531": 867,
+    "\u62d3\u6251": 868,
+    "\u5206\u914d": 869,
+    "\u667a\u80fd": 870,
+    "\u4f53": 871,
+    "\u7fa4": 872,
+    "\u6709\u6548\u6027": 873,
+    "\u7406\u8bba\u4f9d\u636e": 874,
+    "\u53ef\u884c": 875,
+    "\u4eba\u4f53": 876,
+    "\u4e0a\u80a2": 877,
+    "\u808c\u8089": 878,
+    "\u529b": 879,
+    "\u8bad\u7ec3": 880,
+    "\u9488\u5bf9\u6027": 881,
+    "\u5b9e\u9645\u610f\u4e49": 882,
+    "\u51c6\u786e\u6027": 883,
+    "\u89e3\u51b3": 884,
+    "\u5f88": 885,
+    "\u9c81\u68d2\u6027": 886,
+    "\u80fd": 887,
+    "\u53cd\u63a8": 888,
+    "\u6765": 889,
+    "PMLSM": 890,
+    "\u77e2\u91cf": 891,
+    "\u4f4d\u7f6e": 892,
+    "\u660e\u663e\u63d0\u9ad8": 893,
+    "\u9002\u5e94": 894,
+    "\u5f84\u5411": 895,
+    "\u4ea7\u54c1": 896,
+    "\u88c5\u914d": 897,
+    "\u4f5c\u4e1a": 898,
+    "\u5236\u9020": 899,
+    "\u6210\u672c": 900,
+    "\u5360": 901,
+    "\u6bd4": 902,
+    "\u751f\u4ea7": 903,
+    "\u5468\u671f": 904,
+    "\u5fc5\u8981": 905,
+    "\u4eba": 906,
+    "\u56e0": 907,
+    "\u52a0\u5de5": 908,
+    "\u4e2d\u5fc3": 909,
+    "\u5916": 910,
+    "\u8212\u9002\u5ea6": 911,
+    "\u5f88\u5f3a": 912,
+    "\u7c7b\u4f3c": 913,
+    "\u6280\u672f\u624b\u6bb5": 914,
+    "\u73b0\u573a": 915,
+    "\u96f6\u90e8\u4ef6": 916,
+    "\u64cd\u4f5c": 917,
+    "\u4eba\u5458": 918,
+    "\u52a8\u4f5c": 919,
+    "\u521b\u5efa": 920,
+    "\u8ddd\u79bb": 921,
+    "\u9ad8\u5ea6": 922,
+    "\u5206\u7ea7": 923,
+    "\u91c7\u96c6": 924,
+    "\u6d41\u7a0b": 925,
+    "\u7535\u4fe1\u53f7": 926,
+    "\u7b49\u7ea7": 927,
+    "\u5212\u5206": 928,
+    "\u96be\u5ea6": 929,
+    "\u5927": 930,
+    "\u6307\u6570": 931,
+    "\u91cd\u65b0": 932,
+    "\u75b2\u52b3": 933,
+    "\u6837\u672c": 934,
+    "\u71b5": 935,
+    "\u505a\u51fa": 936,
+    "\u62c5\u8d1f": 937,
+    "\u4e13\u95e8": 938,
+    "\u683c\u5f0f": 939,
+    "\u5b66\u4f4d": 940,
+    "\u89c4\u5b9a": 941,
+    "\u5b66\u8005": 942,
+    "\u9884\u8b66": 943,
+    "\u6982\u8ff0": 944,
+    "BC": 945,
+    "\u623f\u5730\u4ea7": 946,
+    "\u72b6\u51b5": 947,
+    "\u7cfb\u7edf\u6027": 948,
+    "\u5173\u8054": 949,
+    "\u6cd5\u5bf9": 950,
+    "\u8d22\u52a1\u6307\u6807": 951,
+    "\u5171\u540c": 952,
+    "\u6784\u6210": 953,
+    "\u5206\u6790\u6cd5": 954,
+    "\u6307\u6807": 955,
+    "\u6743\u91cd": 956,
+    "\u7cfb\u6570": 957,
+    "\u5747": 958,
+    "\u5bf9\u7b56": 959,
+    "\u76d1\u7ba1": 960,
+    "\u52a0\u5f3a": 961,
+    "\u6295\u8d44": 962,
+    "\u9879\u76ee": 963,
+    "\u8425\u8fd0": 964,
+    "\u6d3b\u52a8": 965,
+    "\u57ce\u5e02": 966,
+    "\u4f9d\u6258": 967,
+    "\u8f7d\u4f53": 968,
+    "\u592f": 969,
+    "\u6269\u6869": 970,
+    "\u6869": 971,
+    "\u4e09\u7ef4": 972,
+    "\u8377\u8f7d": 973,
+    "\u7ec4\u5408": 974,
+    "\u6570\u503c": 975,
+    "\u4fa7": 976,
+    "\u7aef": 977,
+    "\u603b": 978,
+    "\u627f\u8f7d\u80fd\u529b": 979,
+    "\u6269\u6563": 980,
+    "\u6270\u52a8": 981,
+    "\u6700\u4f18": 982,
+    "\u89e3\u6790": 983,
+    "Zn": 984,
+    "\u8150\u8680\u6027": 985,
+    "\u91cf": 986,
+    "Al": 987,
+    "\u5143\u7d20": 988,
+    "\u94f8\u9020": 989,
+    "\u56db\u79cd": 990,
+    "\u6210\u5206": 991,
+    "\u663e\u5fae": 992,
+    "\u62c9\u4f38": 993,
+    "\u9541\u5408\u91d1": 994,
+    "\u7535\u5316\u5b66": 995,
+    "\u8150\u8680": 996,
+    "\u9884\u671f": 997,
+    "\u4e66\u5199": 998,
+    "\u6807\u51c6": 999,
+    "\u521b\u65b0": 1000,
+    "\u89c1\u89e3": 1001,
+    "\u6709\u5173": 1002,
+    "\u98ce\u529b": 1003,
+    "\u53d1\u7535\u673a\u7ec4": 1004,
+    "\u65e0\u529f": 1005,
+    "\u53d8\u6d41\u5668": 1006,
+    "\u56db": 1007,
+    "\u76f4\u63a5": 1008,
+    "\u6709\u529f": 1009,
+    "\u964d\u4f4e": 1010,
+    "\u8dcc\u843d": 1011,
+    "\u6700\u5927": 1012,
+    "\u4fdd\u8bc1": 1013,
+    "\u5b89\u5168": 1014,
+    "\u7a33\u5b9a": 1015,
+    "\u63d0\u5347": 1016,
+    "\u5269\u4f59": 1017,
+    "\u5bb9\u91cf": 1018,
+    "\u76f4\u6d41\u7535": 1019,
+    "\u76f4\u6d41": 1020,
+    "\u503c": 1021,
+    "\u538b": 1022,
+    "\u5e45\u503c": 1023,
+    "\u7a33\u5b9a\u6027": 1024,
+    "\u98ce\u7535\u573a": 1025,
+    "\u96c6\u7fa4": 1026,
+    "\u534f\u540c": 1027,
+    "\u8c03\u9891": 1028,
+    "\u8c03\u8282": 1029,
+    "\u5c42": 1030,
+    "\u50a8\u80fd": 1031,
+    "\u53c2\u4e0e": 1032,
+    "\u9891\u7387": 1033,
+    "\u6587\u4e2d": 1034,
+    "\u57fa\u672c\u4e0a": 1035,
+    "\u53d8\u538b\u5668": 1036,
+    "\u7ebf\u5708": 1037,
+    "\u4f20\u9012": 1038,
+    "\u8f74\u5411": 1039,
+    "\u6db2\u538b": 1040,
+    "\u5408\u95f8": 1041,
+    "\u95f4": 1042,
+    "\u63a5\u89e6": 1043,
+    "\u6539\u53d8": 1044,
+    "\u8f7d\u8377": 1045,
+    "\u5927\u5c0f": 1046,
+    "\u5faa\u73af": 1047,
+    "\u6b21\u6570": 1048,
+    "\u6709\u9650\u5143\u6cd5": 1049,
+    "\u53d7\u5230": 1050,
+    "\u53d1\u751f": 1051,
+    "\u53d1\u7535\u673a": 1052,
+    "\u65e0": 1053,
+    "\u4f20\u611f\u5668": 1054,
+    "\u6709\u76ca": 1055,
+    "\u63a2\u7d22": 1056,
+    "\u5df2": 1057,
+    "\u6ee1\u8db3": 1058,
+    "\u4f30\u8ba1": 1059,
+    "\u6240\u63d0": 1060,
+    "10": 1061,
+    "\u8bf4\u670d\u529b": 1062,
+    "\u6b20\u7f3a": 1063,
+    "\u6c7d\u8f66": 1064,
+    "\u6709\u9650\u516c\u53f8": 1065,
+    "\u7814\u5236": 1066,
+    "\u8f66\u6865": 1067,
+    "\u7535\u52a8\u6c7d\u8f66": 1068,
+    "\u7b80\u5316": 1069,
+    "\u5bff\u547d": 1070,
+    "\u53d7\u529b": 1071,
+    "\u4fee\u6b63": 1072,
+    "\u6750\u6599": 1073,
+    "S": 1074,
+    "\u66f2\u7ebf": 1075,
+    "\u5df2\u6709": 1076,
+    "\u6210\u719f": 1077,
+    "\u4f7f\u7528\u5bff\u547d": 1078,
+    "\u5960\u5b9a": 1079,
+    "\u8be6\u5c3d": 1080,
+    "\u53d8\u91cf": 1081,
+    "\u6b64\u5916": 1082,
+    "\u8bc1\u660e": 1083,
+    "\u5408\u7406\u6027": 1084,
+    "\u7efc\u4e0a\u6240\u8ff0": 1085,
+    "\u8ba4\u4e3a": 1086,
+    "\u5b89\u6392": 1087,
+    "\u7136\u540e": 1088,
+    "\u8bc4\u5ba1": 1089,
+    "\u8865\u5145": 1090,
+    "\u5149\u7ea4": 1091,
+    "\u6df1\u5165\u7814\u7a76": 1092,
+    "\u6bcf": 1093,
+    "\u4e0d\u591f": 1094,
+    "\u52a8\u624b": 1095,
+    "\u6545": 1096,
+    "\u8be5\u6587": 1097,
+    "\u89c6\u89d2": 1098,
+    "\u65b0\u9896": 1099,
+    "\u5f97\u5f53": 1100,
+    "\u5de5\u5177": 1101,
+    "\u7fd4\u5b9e": 1102,
+    "\u6db2": 1103,
+    "\u751f\u7269": 1104,
+    "\u7535": 1105,
+    "\u963b\u6297": 1106,
+    "\u5206\u6790\u65b9\u6cd5": 1107,
+    "\u7acb\u9898": 1108,
+    "\u5b9e\u9a8c\u8bbe\u8ba1": 1109,
+    "\u4ea4\u53c9": 1110,
+    "\u9274\u4e8e": 1111,
+    "\u8fd1\u5e74\u6765": 1112,
+    "\u8bba\u70b9": 1113,
+    "\u81ea\u5df1": 1114,
+    "\u663e\u8457": 1115,
+    "\u8bc1\u636e": 1116,
+    "\u8ba4\u5b9a": 1117,
+    "\u53e6\u5916": 1118,
+    "\u6ce8\u91ca": 1119,
+    "\u53c2\u8003\u6587\u732e": 1120,
+    "\u78c1\u901a": 1121,
+    "\u70ed": 1122,
+    "\u78c1\u6781": 1123,
+    "\u6c38\u78c1\u4f53": 1124,
+    "\u635f\u8017": 1125,
+    "\u5f62\u72b6": 1126,
+    "\u6da1\u6d41\u635f\u8017": 1127,
+    "\u901a\u98ce": 1128,
+    "\u6563\u70ed": 1129,
+    "\u68c0\u5bdf\u673a\u5173": 1130,
+    "\u63d0\u8d77": 1131,
+    "\u884c\u653f": 1132,
+    "\u516c\u76ca": 1133,
+    "\u8bc9\u8bbc": 1134,
+    "\u800c": 1135,
+    "\u4fc3\u8fdb": 1136,
+    "\u63a2\u8ba8": 1137,
+    "\u5982": 1138,
+    "\u7ae0": 1139,
+    "\u4e3b\u4f53": 1140,
+    "\u6269\u5927": 1141,
+    "\u5916\u90e8": 1142,
+    "\u673a\u5236": 1143,
+    "\u529f\u5e95": 1144,
+    "\u719f\u7ec3": 1145,
+    "\u5f15\u8bc1": 1146,
+    "\u4e14": 1147,
+    "\u5199\u4f5c\u80fd\u529b": 1148,
+    "\u603b\u4f53\u800c\u8a00": 1149,
+    "\u6b64": 1150,
+    "\u6446\u7ebf": 1151,
+    "\u9f7f": 1152,
+    "\u9525\u9f7f\u8f6e": 1153,
+    "\u7d27\u5bc6": 1154,
+    "\u4f20\u52a8": 1155,
+    "\u66f4": 1156,
+    "\u5927\u578b": 1157,
+    "\u7cbe\u5ea6": 1158,
+    "\u7528\u4e8e": 1159,
+    "Q": 1160,
+    "\u6570\u63a7": 1161,
+    "\u87ba\u65cb": 1162,
+    "\u9f7f\u8f6e": 1163,
+    "\u5207\u524a": 1164,
+    "\u83b7\u5f97": 1165,
+    "\u6709\u5229\u4e8e": 1166,
+    "\u5206\u660e": 1167,
+    "BLDCM": 1168,
+    "\u8f6c\u77e9": 1169,
+    "\u8109\u52a8": 1170,
+    "\u51fd\u6570": 1171,
+    "DTC": 1172,
+    "TSF": 1173,
+    "\u66f4\u597d": 1174,
+    "\u6570\u5b66": 1175,
+    "\u6570\u636e\u6316\u6398": 1176,
+    "\u8fde\u9501": 1177,
+    "\u4e2a": 1178,
+    "\u4e0d\u8db3\u4e4b\u5904": 1179,
+    "\u53ef\u884c\u6027": 1180,
+    "\u6574\u7406": 1181,
+    "\u5404\u4e2a": 1182,
+    "\u6570\u636e\u5e93": 1183,
+    "\u6a21\u5f0f": 1184,
+    "\u6316\u6398": 1185,
+    "\u53ea\u662f": 1186,
+    "\u6df1\u5ea6": 1187,
+    "\u5355\u6676": 1188,
+    "\u670d\u5f79": 1189,
+    "\u590d\u6742": 1190,
+    "\u957f\u671f": 1191,
+    "\u8815\u53d8": 1192,
+    "\u5f62\u8c8c": 1193,
+    "\u6f14\u5316": 1194,
+    "\u754c\u9762": 1195,
+    "\u53d8\u5316\u89c4\u5f8b": 1196,
+    "\u7c89\u672b": 1197,
+    "\u6210\u578b": 1198,
+    "\u771f\u7a7a": 1199,
+    "\u70e7\u7ed3": 1200,
+    "\u63a2\u7a76": 1201,
+    "\u65f6\u95f4": 1202,
+    "Si": 1203,
+    "\u5c01\u88c5": 1204,
+    "\u590d\u5408\u6750\u6599": 1205,
+    "\u8ba8\u8bba": 1206,
+    "\u5c31": 1207,
+    "\u800c\u8a00": 1208,
+    "\u9898\u76ee": 1209,
+    "\u8fd8\u6709": 1210,
+    "\u6765\u8bf4": 1211,
+    "\u6ca1\u6709": 1212,
+    "\u7279\u522b": 1213,
+    "\u6c88\u9633": 1214,
+    "\u5927\u5b66": 1215,
+    "\u6388\u4e88": 1216,
+    "\u7ecf": 1217,
+    "\u5c11\u91cf": 1218,
+    "\u91cf\u5b50": 1219,
+    "\u8bc6\u522b": 1220,
+    "n": 1221,
+    "\u56fe\u50cf": 1222,
+    "\u5b58\u50a8": 1223,
+    "\u5377\u79ef": 1224,
+    "\u96c6": 1225,
+    ".%": 1226,
+    "\u6676\u7c92": 1227,
+    "\u5c3a\u5bf8": 1228,
+    "\u7ed9\u51fa": 1229,
+    "\u949b": 1230,
+    "\u6297": 1231,
+    "\u5e7f\u6cdb\u5e94\u7528": 1232,
+    "\u5851\u6027": 1233,
+    "\u9700": 1234,
+    "\u6d41\u53d8": 1235,
+    "\u7f3a\u70b9": 1236,
+    "\u4f5c": 1237,
+    "\u8bc4\u8ff0": 1238,
+    "Ti": 1239,
+    "6Al": 1240,
+    "4V": 1241,
+    "TiC": 1242,
+    "TiB": 1243,
+    "\u589e\u5f3a": 1244,
+    "\u5747\u5300": 1245,
+    "\u6cbf": 1246,
+    "\u521d\u59cb": 1247,
+    "\u03b2": 1248,
+    "\u7ec6\u5316": 1249,
+    "\u8d8b\u52bf": 1250,
+    "\u7ec4\u6210": 1251,
+    "\u5f3a\u5ea6": 1252,
+    "\u4e0b\u964d": 1253,
+    "\u2103": 1254,
+    "s": 1255,
+    "\u5cf0\u503c": 1256,
+    "\u7531": 1257,
+    "MPa": 1258,
+    "5": 1259,
+    "\u672c\u6784": 1260,
+    "\u65b9\u7a0b": 1261,
+    "6": 1262,
+    "7": 1263,
+    "\u56fe": 1264,
+    "\u5bf9\u5e94": 1265,
+    "\u6700\u4f73": 1266,
+    "\u6570\u636e\u5904\u7406": 1267,
+    "\u5145\u5b9e": 1268,
+    "\u5b66\u79d1\u4e13\u4e1a": 1269,
+    "\u4e13\u4e1a\u6280\u80fd": 1270,
+    "\u7ee7\u7535\u5668": 1271,
+    "\u4e3a\u9898": 1272,
+    "\u56fa\u6709\u9891\u7387": 1273,
+    "\u7c27\u7247": 1274,
+    "\u5b89\u88c5": 1275,
+    "\u9650\u4f4d": 1276,
+    "\u7247": 1277,
+    "\u9650\u5236": 1278,
+    "\u4e13\u95e8\u77e5\u8bc6": 1279,
+    "\u949b\u5408\u91d1": 1280,
+    "\u58f0": 1281,
+    "\u53d1\u5c04": 1282,
+    "\u83b7\u53d6": 1283,
+    "\u635f\u4f24": 1284,
+    "\u91d1\u5c5e": 1285,
+    "\u76d1\u6d4b": 1286,
+    "\u8bc6\u522b\u65b9\u6cd5": 1287,
+    "\u63a8\u8350": 1288,
+    "\u00b0": 1289,
+    "\u4e09\u76f8": 1290,
+    "\u89e3": 1291,
+    "\u96f6": 1292,
+    "\u4f1a": 1293,
+    "\u5bfc\u81f4": 1294,
+    "\u4ee3\u66ff": 1295,
+    "SVPWM": 1296,
+    "\u590d\u6742\u5ea6": 1297,
+    "\u5f53\u4eca": 1298,
+    "\u4e16\u754c": 1299,
+    "\u4fa6\u67e5": 1300,
+    "\u7740": 1301,
+    "\u4fb5\u6743": 1302,
+    "\u53ef\u80fd": 1303,
+    "\u635f\u5bb3": 1304,
+    "\u7a0b\u5e8f": 1305,
+    "\u89c4\u5236": 1306,
+    "\u73b0\u884c": 1307,
+    "\u6551\u6d4e": 1308,
+    "\u53f8\u6cd5": 1309,
+    "\u4ee5\u6b64": 1310,
+    "\u610f\u8bc6": 1311,
+    "\u5f3a\u70c8": 1312,
+    "\u4e4b\u4e0a": 1313,
+    "\u53ef\u64cd\u4f5c\u6027": 1314,
+    "\u4f7f\u5f97": 1315,
+    "\u786c\u4ef6": 1316,
+    "B": 1317,
+    "\u51cf\u5c0f": 1318,
+    "\u89c2": 1319,
+    "\u89c6\u57df": 1320,
+    "\u719f\u6089": 1321,
+    "\u57fa\u672c\u539f\u7406": 1322,
+    "\u8bed\u8a00\u8868\u8fbe": 1323,
+    "\u987a\u7545": 1324,
+    "PEO": 1325,
+    "PLA": 1326,
+    "\u56fa\u6001": 1327,
+    "\u805a\u5408\u7269": 1328,
+    "\u7535\u89e3\u8d28": 1329,
+    "\u9502\u79bb\u5b50": 1330,
+    "\u7535\u6c60": 1331,
+    "\u7269": 1332,
+    "LCI": 1333,
+    "\u63ba\u6742": 1334,
+    "\u81ea\u52a8": 1335,
+    "\u7535\u5bfc\u7387": 1336,
+    "\u7ea4\u7ef4": 1337,
+    "\u5374": 1338,
+    "\u4ee5\u4e0a": 1339,
+    "\u8be6\u5b9e": 1340,
+    "\u591a\u5b54": 1341,
+    "\u5f15\u8d77": 1342,
+    "\u9676\u74f7": 1343,
+    "\u6d82\u5c42": 1344,
+    "\u5b8c\u5168": 1345,
+    "\u67e5\u8be2": 1346,
+    "\u7b26\u5408\u5b9e\u9645": 1347,
+    "\u8fc7\u5ea6": 1348,
+    "\u533b\u7597": 1349,
+    "\u8d23\u4efb": 1350,
+    "\u7ef4\u62a4": 1351,
+    "\u60a3\u8005": 1352,
+    "\u7b2c": 1353,
+    "\u4e13\u5bb6": 1354,
+    "\u5b9e\u65bd": 1355,
+    "\u7edf\u4e00": 1356,
+    "\u4e0a\u8ff0": 1357,
+    "\u65b0\u610f": 1358,
+    "\u503c\u5f97": 1359,
+    "\u6838\u5fc3\u90e8\u4ef6": 1360,
+    "\u4e4b\u4e00": 1361,
+    "\u6da1\u8f6e": 1362,
+    "\u589e\u538b\u5668": 1363,
+    "\u6210\u4e3a": 1364,
+    "\u70ed\u70b9": 1365,
+    "\u8bd5\u9a8c\u53f0": 1366,
+    "\u4f9b\u6c14": 1367,
+    "\u51fa\u53e3": 1368,
+    "\u538b\u529b": 1369,
+    "\u6210": 1370,
+    "\u7535\u6c14": 1371,
+    "\u6574\u4e2a": 1372,
+    "\u6709\u5f85": 1373,
+    "\u7535\u529b": 1374,
+    "\u5b66": 1375,
+    "\u4fe1\u53f7\u5904\u7406": 1376,
+    "\u4e00\u4f53\u5316": 1377,
+    "\u81ea\u7531\u5ea6": 1378,
+    "\u5dee\u52a8": 1379,
+    "\u8ba1\u7b97\u516c\u5f0f": 1380,
+    "\u7535\u78c1": 1381,
+    "\u4f4d\u79fb": 1382,
+    "\u8f6d": 1383,
+    "\u8ba1\u7b97\u7ed3\u679c": 1384,
+    "\u53d7": 1385,
+    "\u504f\u79fb": 1386,
+    "\u6267\u884c\u5668": 1387,
+    "\u4f9d\u636e": 1388,
+    "\u7535\u5e73": 1389,
+    "\u5f00\u5173": 1390,
+    "\u56f4\u7ed5": 1391,
+    "\u56fa\u4f53": 1392,
+    "\u5fae\u6ce2": 1393,
+    "\u653e\u5927": 1394,
+    "\u6d88\u7eb3": 1395,
+    "\u5206\u5e03\u5f0f": 1396,
+    "\u7535\u6e90": 1397,
+    "\u7535\u4ef7": 1398,
+    "\u7ecf\u6d4e\u6027": 1399,
+    "\u5229\u7528\u7387": 1400,
+    "\u9002\u4e2d": 1401,
+    "\u5e72\u6270": 1402,
+    "\u79fb\u52a8\u673a\u5668\u4eba": 1403,
+    "\u4fe1\u606f": 1404,
+    "\u673a\u5668\u4eba": 1405,
+    "\u76fe\u6784": 1406,
+    "\u5730\u94c1": 1407,
+    "\u96a7\u9053": 1408,
+    "\u5efa\u8bbe": 1409,
+    "\u4ec0\u4e48": 1410,
+    "\u6216\u8005": 1411,
+    "\u91cd\u5927": 1412,
+    "\u7535\u5de5\u94a2": 1413,
+    "\u65cb\u8f6c": 1414,
+    "\u78c1\u5316": 1415,
+    "\u78c1\u6ede": 1416,
+    "\u5e38\u7528": 1417,
+    "Jiles": 1418,
+    "Atherton": 1419,
+    "\u521d\u59cb\u503c": 1420,
+    "k": 1421,
+    "\u5bc6\u5ea6": 1422,
+    "\u628a": 1423,
+    "\u4e00\u53f0": 1424,
+    "\u94c1\u5fc3": 1425,
+    "\u78c1\u573a\u5f3a\u5ea6": 1426,
+    "\u8bb8\u591a": 1427,
+    "\u5e0c\u671b": 1428,
+    "\u878d\u5408": 1429,
+    "\u5de5\u5b66": 1430,
+    "\u7acb\u8bba": 1431,
+    "\u6709\u636e": 1432,
+    "\u7ae0\u8282": 1433,
+    "\u73af\u6c27\u6811\u8102": 1434,
+    "\u6cbf\u9762": 1435,
+    "\u95ea\u7edc": 1436,
+    "\u6570\u636e\u5206\u6790": 1437,
+    "\u7535\u6781": 1438,
+    "\u538b\u5f3a": 1439,
+    "\u501f\u52a9": 1440,
+    "\u626b\u63cf": 1441,
+    "\u7834\u574f": 1442,
+    "\u504f\u78c1\u4e0b": 1443,
+    "\u5341\u5206": 1444,
+    "\u504f\u78c1": 1445,
+    "\u6709\u65e0": 1446,
+    "\u53d6\u5411": 1447,
+    "\u7845\u94a2\u7247": 1448,
+    "\u5bf9\u79f0": 1449,
+    "\u516c\u5f0f": 1450,
+    "\u53d8": 1451,
+    "\u54cd\u5e94": 1452,
+    "\u62df\u5408": 1453,
+    "\u5f15\u7528": 1454,
+    "\u865a\u62df": 1455,
+    "\u53cc\u9988": 1456,
+    "\u53d8\u6362\u5668": 1457,
+    "\u6d88\u9664": 1458,
+    "\u5206\u91cf": 1459,
+    "\u8c10\u6ce2": 1460,
+    "Matlab": 1461,
+    "\u7f3a\u5c11": 1462,
+    "\u8fbd\u5b81\u7701": 1463,
+    "\u79d1\u6280\u56ed": 1464,
+    "\u524d\u6cbf": 1465,
+    "\u56fd\u5185": 1466,
+    "\u7c92\u5b50": 1467,
+    "\u6307\u51fa": 1468,
+    "\u652f\u6491": 1469,
+    "\u6587\u7b14\u6d41\u7545": 1470,
+    "\u5f0f": 1471,
+    "\u8c03\u901f": 1472,
+    "DS": 1473,
+    "\u4f18\u70b9": 1474,
+    "\u78c1\u8def": 1475,
+    "\u6ce8\u5165": 1476,
+    "\u95ed\u73af": 1477,
+    "\u5236\u4f5c": 1478,
+    "\u6837\u673a": 1479,
+    "\u6709\u7740": 1480,
+    "\u9ad8\u9891": 1481,
+    "\u9002\u5408": 1482,
+    "\u907f\u7a0e": 1483,
+    "\u8d27\u5e01\u653f\u7b56": 1484,
+    "\u7ecf\u8425": 1485,
+    "\u4e1a\u7ee9": 1486,
+    "\u57fa\u672c\u6982\u5ff5": 1487,
+    "\u4ee3\u7406": 1488,
+    "\u66f4\u4e3a": 1489,
+    "\u878d\u8d44": 1490,
+    "\u7ea6\u675f": 1491,
+    "\u5b9e\u52a1": 1492,
+    "\u6765\u770b": 1493,
+    "GIL": 1494,
+    "\u5e03\u7f6e": 1495,
+    "\u5fae\u7c92": 1496,
+    "\u7279\u8272": 1497,
+    "\u5176\u4e2d": 1498,
+    "\u65bd\u52a0": 1499,
+    "\u8f68\u8ff9": 1500,
+    "\u8986\u819c": 1501,
+    "\u539a\u5ea6": 1502,
+    "\u7406\u8bba\u6307\u5bfc": 1503,
+    "\u884c\u6587": 1504,
+    "AGV": 1505,
+    "\u907f\u969c": 1506,
+    "\u884c\u9a76": 1507,
+    "\u969c\u788d\u7269": 1508,
+    "\u51cf\u5c11": 1509,
+    "\u5e8f\u5217": 1510,
+    "\u8f6c\u89d2": 1511,
+    "\u6a21\u7cca\u63a7\u5236": 1512,
+    "\u89c4\u5219": 1513,
+    "\u7efc\u4e0a": 1514,
+    "\u6b63\u5f26": 1515,
+    "\u6fc0\u52b1": 1516,
+    "\u8be6\u7ec6\u5206\u6790": 1517,
+    "\u6ce2": 1518,
+    "\u63a5\u5934": 1519,
+    "\u56fa\u5b9a": 1520,
+    "\u7ba1\u8def": 1521,
+    "\u8fde\u63a5": 1522,
+    "\u5411": 1523,
+    "\u4ecb\u8d28": 1524,
+    "\u9632\u6b62": 1525,
+    "\u4e32\u8054": 1526,
+    "\u6cc4\u6f0f": 1527,
+    "\u6d41\u573a": 1528,
+    "Fluent": 1529,
+    "\u7740\u91cd": 1530,
+    "\u5bc6\u5c01": 1531,
+    "\u95f4\u9699": 1532,
+    "\u63a5\u5730": 1533,
+    "\u5355\u76f8\u63a5\u5730": 1534,
+    "\u5f88\u5927": 1535,
+    "\u6bd4\u4f8b": 1536,
+    "\u94c1\u78c1": 1537,
+    "\u9009\u7ebf": 1538,
+    "\u96f6\u5e8f": 1539,
+    "\u76f8\u4f4d": 1540,
+    "\u5355\u4e00": 1541,
+    "\u51c6\u786e\u5ea6": 1542,
+    "\u6c14\u9699": 1543,
+    "\u5982\u679c": 1544,
+    "\u9519\u522b\u5b57": 1545,
+    "\u533b\u9662": 1546,
+    "\u7528\u6237": 1547,
+    "DC": 1548,
+    "\u5e26\u6765": 1549,
+    "\u6307\u5bfc\u610f\u4e49": 1550,
+    "\u80a1\u4efd": 1551,
+    "\u8d44\u91d1": 1552,
+    "\u4f1a\u8ba1": 1553,
+    "\u6df7\u4e71": 1554,
+    "\u5c24\u5176": 1555,
+    "\u503a\u5238": 1556,
+    "\u5c5e\u6027": 1557,
+    "\u6307\u5bfc": 1558,
+    "\u70bc\u94a2": 1559,
+    "\u8fde\u94f8": 1560,
+    "\u5b9a\u4e49": 1561,
+    "\u89c4\u5212": 1562,
+    "\u6700\u5c0f\u5316": 1563,
+    "\u5206\u6563": 1564,
+    "\u641c\u7d22\u7b97\u6cd5": 1565,
+    "\u6392\u5e8f": 1566,
+    "\u9057\u4f20\u7b97\u6cd5": 1567,
+    "\u968f\u673a": 1568,
+    "\u540c\u5b66": 1569,
+    "\u672c\u95e8": 1570,
+    "\u5bbd\u5e7f": 1571,
+    "\u6dd8\u6d17": 1572,
+    "\u601d\u8def\u6e05\u6670": 1573,
+    "\u6e29\u5347": 1574,
+    "\u9608\u503c": 1575,
+    "\u6a21\u6001": 1576,
+    "means": 1577,
+    "\u805a\u7c7b": 1578,
+    "\u79bb\u6563": 1579,
+    "\u5c11": 1580,
+    "\u63d0\u70bc": 1581,
+    "\u6821\u6b63": 1582,
+    "FV520B": 1583,
+    "\u4e0d\u9508\u94a2": 1584,
+    "\u94dd\u5408\u91d1": 1585,
+    "\u786b\u5316\u7269": 1586,
+    "\u542b": 1587,
+    "\u5dee\u5f02": 1588,
+    "\u6c27\u5316": 1589,
+    "\u819c": 1590,
+    "\u949d\u5316": 1591,
+    "\u6eb6\u89e3": 1592,
+    "\u88c2\u7eb9": 1593,
+    "\u9633\u6781": 1594,
+    "\u5f00\u88c2": 1595,
+    "\u6269\u5c55": 1596,
+    "\u786c\u5ea6": 1597,
+    "\u53f6\u8f6e": 1598,
+    "\u8870\u51cf": 1599,
+    "\u8d8a": 1600,
+    "\u9ad8\u6027\u80fd": 1601,
+    "\u660e\u663e": 1602,
+    "\u52a0\u5feb": 1603,
+    "\u5f3a\u5316": 1604,
+    "\u57fa\u4f53": 1605,
+    "\u9ad8\u4e8e": 1606,
+    "\u6982\u7387": 1607,
+    "\u7269\u6d41\u914d\u9001": 1608,
+    "\u53cc\u5c42": 1609,
+    "\u7a97": 1610,
+    "\u6ee1\u610f\u5ea6": 1611,
+    "\u6392\u7248": 1612,
+    "\u6253\u78e8": 1613,
+    "\u5de5\u4ef6": 1614,
+    "\u81ea\u52a8\u5316": 1615,
+    "\u67d4\u6027": 1616,
+    "\u5b9a\u4f4d": 1617,
+    "\u53f0": 1618,
+    "\u82af": 1619,
+    "\u5939\u5177": 1620,
+    "\u65b9\u6848\u8bbe\u8ba1": 1621,
+    "\u5339\u914d": 1622,
+    "\u8fd9\u79cd": 1623,
+    "\u8fd0\u52a8\u5b66": 1624,
+    "\u5173\u8282": 1625,
+    "\u8c03\u8bd5": 1626,
+    "\u78c1\u82af": 1627,
+    "C": 1628,
+    "\u7b49\u6548": 1629,
+    "\u8ba1\u7b97\u65b9\u6cd5": 1630,
+    "\u7ba1\u9053": 1631,
+    "\u79bb\u5fc3\u6cf5": 1632,
+    "\u51b2\u538b": 1633,
+    "\u710a\u63a5": 1634,
+    "\u53f6\u7247": 1635,
+    "\u7814\u53d1": 1636,
+    "\u79fb\u52a8": 1637,
+    "\u8679\u819c": 1638,
+    "\u4e3a\u6b64": 1639,
+    "\u5b66\u79d1\u524d\u6cbf": 1640,
+    "\u767d": 1641,
+    "\u9152\u74f6": 1642,
+    "\u74f6\u53e3": 1643,
+    "\u5728\u7ebf": 1644,
+    "\u96be\u9898": 1645,
+    "\u6700\u5c0f": 1646,
+    "\u7ea6": 1647,
+    "\u8f68\u9053\u4ea4\u901a": 1648,
+    "\u65ad\u8def\u5668": 1649,
+    "\u540c\u6b65\u63a7\u5236": 1650,
+    "\u64cd\u52a8": 1651,
+    "\u6536\u655b": 1652,
+    "\u69fd": 1653,
+    "\u7535\u67a2": 1654,
+    "\u5e73\u5747": 1655,
+    "\u4e60\u8fd1\u5e73": 1656,
+    "\u65f6\u4ee3": 1657,
+    "\u4e2d\u56fd": 1658,
+    "\u793e\u4f1a\u4e3b\u4e49": 1659,
+    "\u5f53\u4ee3": 1660,
+    "\u524d\u4eba": 1661,
+    "\u542b\u6c34\u7387": 1662,
+    "\u5783\u573e": 1663,
+    "\u5404\u79cd": 1664,
+    "\u73af\u4fdd": 1665,
+    "\u5b9a\u91cf": 1666,
+    "\u4f9b\u5e94\u94fe": 1667,
+    "H": 1668,
+    "\u628a\u63e1": 1669,
+    "\u5f00\u91c7": 1670,
+    "\u76d8": 1671,
+    "\u94e3\u524a": 1672,
+    "\u87ba\u6746\u6cf5": 1673,
+    "\u5373": 1674,
+    "\u5207\u524a\u529b": 1675,
+    "\u81ea\u7531": 1676,
+    "\u957f\u5ea6": 1677,
+    "\u52a0\u5de5\u8fc7\u7a0b": 1678,
+    "\u540e\u7eed": 1679,
+    "\u6280\u80fd": 1680,
+    "\u9ad8\u6821": 1681,
+    "\u793e\u56e2": 1682,
+    "\u79cd\u7c7b": 1683,
+    "\u96be\u4ee5": 1684,
+    "\u7ba1\u7406\u7cfb\u7edf": 1685,
+    "\u67b6\u6784": 1686,
+    "\u4fdd\u7406": 1687,
+    "\u5408\u540c": 1688,
+    "\u5f53\u4e8b\u4eba": 1689,
+    "\u6743\u5229\u4e49\u52a1": 1690,
+    "\u6001\u5ea6": 1691,
+    "\u7814\u7a76\u8bfe\u9898": 1692,
+    "\u7f13\u51b2": 1693,
+    "\u521a\u5ea6": 1694,
+    "\u53cd": 1695,
+    "\u9ad8\u6548": 1696,
+    "\u8272\u8c31": 1697,
+    "\u8003\u5bdf": 1698,
+    "\u6d41\u52a8": 1699,
+    "\u6d41\u901f": 1700,
+    "\u67f1": 1701,
+    "mm": 1702,
+    "\u00d7": 1703,
+    ":": 1704,
+    "min": 1705,
+    "\u65e0\u7ebf": 1706,
+    "\u5e7f\u9614": 1707,
+    "\u5206\u7c07": 1708,
+    "\u8282\u70b9": 1709,
+    "\u514d\u75ab": 1710,
+    "\u6240\u6709": 1711,
+    "\u5b83": 1712,
+    "\u751f\u5b58": 1713,
+    "\u63a8\u7406": 1714,
+    "\u884c\u6587\u6d41\u7545": 1715,
+    "\u5145\u7535": 1716,
+    "\u5408\u9002": 1717,
+    "\u4f9b\u7535\u7cfb\u7edf": 1718,
+    "\u539f\u8fb9": 1719,
+    "\u95f4\u8ddd": 1720,
+    "\u8fdb\u800c": 1721,
+    "\u5e76\u8054": 1722,
+    "\u6587\u7406": 1723,
+    "\u9502\u7535\u6c60": 1724,
+    "Fe": 1725,
+    "Ga": 1726,
+    "\u88c5\u7f6e": 1727,
+    "\u60ac\u81c2": 1728,
+    "\u8f93\u51fa\u529f\u7387": 1729,
+    "\u4f38\u7f29": 1730,
+    "\u60ac\u81c2\u6881": 1731,
+    "\u5f2f\u66f2": 1732,
+    "\u5200\u5177": 1733,
+    "\u4f4e\u901f": 1734,
+    "\u7406\u89e3": 1735,
+    "\u5c0f\u578b": 1736,
+    "\u5e94\u8be5": 1737,
+    "\u51e0\u4e4e": 1738,
+    "\u8fc7\u4e8e": 1739,
+    "\u5219": 1740,
+    "\u8026\u5408": 1741,
+    "\u5173\u952e\u6280\u672f": 1742,
+    "\u5c55\u793a": 1743,
+    "\u8f83\u5dee": 1744,
+    "\u7a7a\u6c14": 1745,
+    "\u5e73\u8861": 1746,
+    "\u2014": 1747,
+    "\u5546\u4e1a\u79d8\u5bc6": 1748,
+    "\u6297\u8fa9": 1749,
+    "\u5256\u6790": 1750,
+    "\u73b0\u72b6\u53ca": 1751,
+    "\u610f\u601d": 1752,
+    "\u7ee9\u6548\u8003\u6838": 1753,
+    "\u7ef4\u5ea6": 1754,
+    "\u5386\u7a0b": 1755,
+    "\u8230\u8f7d": 1756,
+    "\u6297\u51b2\u51fb": 1757,
+    "\u7f13\u51b2\u5668": 1758,
+    "0.3": 1759,
+    "\u56e0\u5b50": 1760,
+    "\u6587\u672c": 1761,
+    "\u5176\u4ed6": 1762,
+    "\u6324\u538b": 1763,
+    "\u7ecf\u5178": 1764,
+    "\u78c1\u963b": 1765,
+    "\u4e00\u6b3e": 1766,
+    "\u954d": 1767,
+    "\u77ed\u8def": 1768,
+    "\u52a8\u529b": 1769,
+    "\u6ce2\u5f62": 1770,
+    "\u4f53\u79ef": 1771,
+    "\u7535\u78c1\u573a": 1772,
+    "\u52a0\u8f7d": 1773,
+    "\u5f62": 1774,
+    "\u5e94\u53d8": 1775,
+    "\u73af\u7f51\u67dc": 1776,
+    "\u6b63": 1777,
+    "\u5f27\u5149": 1778,
+    "\u6700": 1779,
+    "\u5c31\u662f": 1780,
+    "\u4ecd\u7136": 1781,
+    "\u6a61\u80f6": 1782,
+    "\u9ad8\u538b": 1783,
+    "\u6db2\u4f53": 1784,
+    "\u529b\u5b66": 1785,
+    "\u78e8\u635f": 1786,
+    "\u4e01\u8148\u6a61\u80f6": 1787,
+    "\u6c1f\u6a61\u80f6": 1788,
+    "\u73af\u5df1\u70f7": 1789,
+    "\u8001\u5316": 1790,
+    "\u6eb6\u80c0": 1791,
+    "\u6469\u64e6": 1792,
+    "\u6d53\u5ea6": 1793,
+    "\u9884": 1794,
+    "\u8bd5\u6837": 1795,
+    "\u65ad\u88c2": 1796,
+    "\u76f8\u540c": 1797,
+    "\u6469\u64e6\u7cfb\u6570": 1798,
+    "\u56fe\u6807": 1799,
+    "\u5171\u4eab": 1800,
+    "\u5c1a\u53ef": 1801,
+    "\u5f97": 1802,
+    "\u7b2c\u56db\u7ae0": 1803,
+    "\u56db\u4e2a": 1804,
+    "\u867d\u7136": 1805,
+    "\u7b2c\u4e09\u7ae0": 1806,
+    "\u8fd8\u662f": 1807,
+    "\u7528\u8bed": 1808,
+    "\u5fae\u6e90": 1809,
+    "\u662f\u5426": 1810,
+    "\u8868": 1811,
+    "\u94a2\u7bb1": 1812,
+    "\u62f1\u6865": 1813,
+    "\u65bd\u5de5": 1814,
+    "\u62f1": 1815,
+    "\u6263\u5854": 1816,
+    "\u6c11\u4e3b\u515a\u6d3e": 1817,
+    "\u653f\u6cbb": 1818,
+    "\u65e0\u6cd5": 1819,
+    "\u56e0\u4e3a": 1820,
+    "\u5fc5\u987b": 1821,
+    "\u6709\u673a": 1822,
+    "\u519c\u4e1a": 1823,
+    "\u751f\u6001": 1824,
+    "\u79d1\u6280\u8fdb\u6b65": 1825,
+    "\u51fa\u8be5": 1826,
+    "\u5927\u90e8\u5206": 1827,
+    "\u8868\u8fbe\u6e05\u6670": 1828,
+    "\u8f66": 1829,
+    "\u8def\u9762": 1830,
+    "\u8f66\u8f86": 1831,
+    "\u566a\u58f0": 1832,
+    "\u54c1\u724c": 1833,
+    "\u5e38\u89c4": 1834,
+    "\u5355\u4f4d": 1835,
+    "\u4f8b\u5982": 1836,
+    "\u6458\u8981": 1837,
+    "\u4e4b": 1838,
+    "\u652f\u6301": 1839,
+    "\u91c7\u8d2d": 1840,
+    "\u7269\u6599": 1841,
+    "\u98ce\u9669\u7ba1\u7406": 1842,
+    "\u5546\u4e1a\u94f6\u884c": 1843,
+    "\u5ba1\u8ba1": 1844,
+    "GS": 1845,
+    "\u7b26\u53f7": 1846,
+    "\u6811": 1847,
+    "ANSYS": 1848,
+    "\u5ea6": 1849,
+    "\u76f4\u6d41\u7535\u673a": 1850,
+    "\u5f53": 1851,
+    "\u529f\u7387\u5bc6\u5ea6": 1852,
+    "\u65e0\u5237": 1853,
+    "\u6c11\u65cf": 1854,
+    "\u624d": 1855,
+    "\u7cbe\u795e": 1856,
+    "\u4ef7\u503c\u89c2": 1857,
+    "\u5c5e\u4e8e": 1858,
+    "\u76f8\u6bd4": 1859,
+    "\u4f18\u52bf": 1860,
+    "\u9694\u79bb": 1861,
+    "\u4e3b": 1862,
+    "\u53cc\u76ee": 1863,
+    "\u7acb\u4f53": 1864,
+    "\u6293\u53d6": 1865,
+    "\u5b66\u672f\u7814\u7a76": 1866,
+    "\u76f8\u673a": 1867,
+    "\u6807\u5b9a": 1868,
+    "\u6ee4\u6ce2": 1869,
+    "\u4f18\u4e8e": 1870,
+    "\u5706\u67f1\u5f62": 1871,
+    "\u5404\u7c7b": 1872,
+    "\u503a\u52a1": 1873,
+    "\u627e\u5230": 1874,
+    "\u5047\u8bbe": 1875,
+    "\u7a33\u5065\u6027": 1876,
+    "\u94a2": 1877,
+    "\u8f6f\u96c6": 1878,
+    "\u76f4\u89c9": 1879,
+    "\u6570": 1880,
+    "Choquet": 1881,
+    "\u903c\u8fd1": 1882,
+    "\u7406\u60f3": 1883,
+    "TOPSIS": 1884,
+    "\u51b3\u7b56": 1885,
+    "\u79ef\u5206": 1886,
+    "\u96b6\u5c5e": 1887,
+    "\u51b3\u7b56\u95ee\u9898": 1888,
+    "\u800c\u4e14": 1889,
+    "\u5f55\u7528": 1890,
+    "\u5b66\u672f\u8bba\u6587": 1891,
+    "\u9762\u5411": 1892,
+    "\u5bfc\u5e08": 1893,
+    "\u64cd\u4f5c\u6280\u80fd": 1894,
+    "\u524d\u63d0": 1895,
+    "\u81ea\u7136\u79d1\u5b66": 1896,
+    "\u5de5\u7a0b\u6280\u672f": 1897,
+    "\u5173\u952e\u95ee\u9898": 1898,
+    "\u7535\u70ed": 1899,
+    "\u55b7\u5634": 1900,
+    "\u6781": 1901,
+    "\u56de\u6d41": 1902,
+    "\u5782\u76f4": 1903,
+    "\u96fe\u5316": 1904,
+    "\u89d2": 1905,
+    "\u9897\u7c92": 1906,
+    "3D": 1907,
+    "\u771f\u5b9e": 1908,
+    "\u81c2": 1909,
+    "\u8fde\u6746": 1910,
+    "\u5f39\u6027": 1911,
+    "\u5feb\u901f": 1912,
+    "\u524d\u9988": 1913,
+    "\u8bef\u5dee": 1914,
+    "\u89c2\u6d4b\u5668": 1915,
+    "\u89d2\u901f\u5ea6": 1916,
+    "\u8de8\u5883": 1917,
+    "\u5e76\u8d2d": 1918,
+    "\u7ba1\u58f3": 1919,
+    "\u6362\u70ed\u5668": 1920,
+    "\u5931\u6548": 1921,
+    "\u5f15\u53d1": 1922,
+    "\u673a\u68b0\u5de5\u7a0b": 1923,
+    "\u6761\u7406\u5206\u660e": 1924,
+    "\u5b66\u672f\u89c2\u70b9": 1925,
+    "\u5206\u4e3a": 1926,
+    "\u592a": 1927,
+    "\u51b7\u85cf\u8f66": 1928,
+    "\u836f\u54c1": 1929,
+    "\u8fdc\u7a0b": 1930,
+    "\u8f66\u8f7d": 1931,
+    "\u4e3b\u63a7": 1932,
+    "CPU": 1933,
+    "\u5b9a\u4f4d\u7cfb\u7edf": 1934,
+    "\u670d\u52a1\u5668": 1935,
+    "\u7ec8\u7aef": 1936,
+    "\u8bf4": 1937,
+    "\u76f4\u9a71": 1938,
+    "\u50a8\u6cb9\u7f50": 1939,
+    "\u8584\u819c": 1940,
+    "\u87ba\u7eb9": 1941,
+    "\u5fae\u52a8": 1942,
+    "\u6a2a\u5411": 1943,
+    "\u87ba\u6813": 1944,
+    "\u63a5\u89e6\u9762": 1945,
+    "\u6ed1\u79fb": 1946,
+    "\u9884\u7d27": 1947,
+    "\u529b\u77e9": 1948,
+    "\u673a\u55b7\u6746": 1949,
+    "\u55b7\u6746": 1950,
+    "\u4e8c": 1951,
+    "\u5c1a\u672a": 1952,
+    "\u4e0d\u80fd": 1953,
+    "\u5149\u4f0f": 1954,
+    "\u8bbe\u8ba1\u65b9\u6848": 1955,
+    "MMC": 1956,
+    "\u4e2d\u5c0f\u4f01\u4e1a": 1957,
+    "\u4e00\u7c7b": 1958,
+    "\u8fd9\u662f": 1959,
+    "\u8fd9\u4e2a": 1960,
+    "\u7275\u5f15": 1961,
+    "\u4f20\u52a8\u7cfb\u7edf": 1962,
+    "\u53d1\u6ce1": 1963,
+    "\u8f66\u95f4": 1964,
+    "\u82f1\u6587": 1965,
+    "\u5e7f\u4e49": 1966,
+    "\u7ebf\u6027\u5316": 1967,
+    "\u901f\u7387": 1968,
+    "\u9c81\u68d2": 1969,
+    "\u221e": 1970,
+    "\u8fd9\u6837": 1971,
+    "\u6881\u6f31\u6e9f": 1972,
+    "\u5112\u5b66": 1973,
+    "\u8f6c\u53d8": 1974,
+    "\u54f2\u5b66": 1975,
+    "\u8bb2": 1976,
+    "\u529b\u5ea6": 1977,
+    "\u76d1\u7763": 1978,
+    "\u53d7\u9650": 1979,
+    "\u4eba\u673a\u4ea4\u4e92": 1980,
+    "\u4ea4\u4e92": 1981,
+    "\u9ad8\u901f": 1982,
+    "\u60ac\u67b6": 1983,
+    "\u672a\u80fd": 1984,
+    "\u70ae\u957f": 1985,
+    "\u7784\u51c6\u955c": 1986,
+    "\u4e13\u5bb6\u7cfb\u7edf": 1987,
+    "\u6027\u80fd\u6307\u6807": 1988,
+    "\u90e8\u4ef6": 1989,
+    "\u7ef4\u4fee": 1990,
+    "\u7cbe\u5bc6": 1991,
+    "\u5355\u8f74": 1992,
+    "\u5168\u5c40": 1993,
+    "\u9012\u5f52": 1994,
+    "\u55b7": 1995,
+    "\u6bd4\u5982": 1996,
+    "\u65e0\u4eba\u673a": 1997,
+    "\u8054\u7cfb\u5b9e\u9645": 1998,
+    "\u7cbe\u76ca": 1999,
+    "\u8ba2\u5355": 2000,
+    "\u9762\u4e34": 2001,
+    "\u751f\u6d3b": 2002,
+    "PLC": 2003,
+    "\u5185\u6a21": 2004,
+    "\u8fdb": 2005,
+    "\u9ad8\u7cbe\u5ea6": 2006,
+    "\u6570\u63a7\u673a\u5e8a": 2007,
+    "\u63a8\u529b": 2008,
+    "\u8868\u8fbe\u5f0f": 2009,
+    "\u57fa\u91d1": 2010,
+    "\u5de8\u5927": 2011,
+    "\u4e89\u8bae": 2012,
+    "\u5b9a\u4ef7": 2013,
+    "\u56de\u987e": 2014,
+    "\u671f\u6743": 2015,
+    "\u914d\u9001": 2016,
+    "\u6539\u6027": 2017,
+    "\u7269\u8d28": 2018,
+    "ACM": 2019,
+    "HPVC": 2020,
+    "TPE": 2021,
+    "\u8010\u70ed": 2022,
+    "\u6cb9": 2023,
+    "\u56de\u5f39": 2024,
+    "\u7efc\u5408\u6027": 2025,
+    "\u6709\u6240": 2026,
+    "\u6750\u6599\u79d1\u5b66": 2027,
+    "\u79d1\u5b66\u6027": 2028,
+    "\u6309\u7167": 2029,
+    "CAD": 2030,
+    "CAM": 2031,
+    "\u6563\u70ed\u5668": 2032,
+    "\u6e29\u5ea6\u573a": 2033,
+    "\u6c14\u6d41": 2034,
+    "\u6708": 2035,
+    "\u65e5": 2036,
+    "\u8f6f\u4ef6\u7cfb\u7edf": 2037,
+    "\u8102\u80aa\u9187": 2038,
+    "\u805a\u6c27\u4e59\u70ef\u919a": 2039,
+    "=": 2040,
+    "OFDM": 2041,
+    "\u4e00\u5e26": 2042,
+    "\u4e00\u8def": 2043,
+    "\u6cbf\u7ebf": 2044,
+    "\u6d6e\u52a8": 2045,
+    "\u6cb9\u5c01": 2046,
+    "\u96f6\u4ef6": 2047,
+    "O": 2048,
+    "\u5bc6\u5c01\u5708": 2049,
+    "\u6ed1\u52a8": 2050,
+    "EBE": 2051,
+    "\u9884\u5904\u7406": 2052,
+    "\u8fed\u4ee3": 2053,
+    "\u504f\u5fc3": 2054,
+    "\u52a9\u884c": 2055,
+    "\u6f14\u53d8": 2056,
+    "\uff1f": 2057,
+    "\u5f62\u6001": 2058,
+    "\u5e2e\u52a9": 2059,
+    "\uff0e": 2060,
+    "\u58f0\u5b66": 2061,
+    "\u91cd\u5efa": 2062,
+    "MKR": 2063,
+    "MTR": 2064,
+    "\u5947\u5f02": 2065,
+    "\u9762": 2066,
+    "\u4f8b\u5b50": 2067,
+    "\u73b0\u4ee3": 2068,
+    "\u7269\u6d41": 2069,
+    "\u8fd0\u8425": 2070,
+    "\u7269\u7406": 2071,
+    "\u4f9b\u7ed9": 2072,
+    "\t": 2073,
+    "\u5206\u6790\u6a21\u578b": 2074,
+    "\u7565\u663e": 2075,
+    "\u77f3\u58a8": 2076,
+    "\u70ef": 2077,
+    "\u8d1f\u6781": 2078,
+    "\u591a\u5c42": 2079,
+    "\u73b0\u91d1": 2080,
+    "\u6df1": 2081,
+    "\u8f67\u5236": 2082,
+    "\u79bb\u5fc3": 2083,
+    "\u538b\u7f29\u673a": 2084,
+    "\u6709\u8bef": 2085,
+    "\u4e3a\u4ec0\u4e48": 2086,
+    "\u8bbe\u5b9a": 2087,
+    "A": 2088,
+    "\u68d2\u6750": 2089,
+    "\u77eb\u76f4": 2090,
+    "\u5f2f\u77e9": 2091,
+    "\u66f2\u7387": 2092,
+    "\u8f8a": 2093,
+    "\u6700\u4f18\u63a7\u5236": 2094,
+    "\u5ea7\u6905": 2095,
+    "\u5f2f\u7ba1": 2096,
+    "\u4ee5\u5185": 2097,
+    "\u663e\u7136": 2098,
+    "\u57f9\u80b2": 2099,
+    "\u5706\u5468": 2100,
+    "\u5206\u79bb\u673a": 2101,
+    "\u789f\u7247": 2102,
+    "\u76d2": 2103,
+    "\u8868\u793a": 2104,
+    "BM": 2105,
+    "\u5168": 2106,
+    "\u5e38\u6001": 2107,
+    "\u8f6c\u5411": 2108,
+    "BIM": 2109,
+    "\u4e13\u4e1a\u5b66\u4f4d": 2110,
+    "Zr50Cu40Al10": 2111,
+    "\u526a\u5207": 2112,
+    "\u6790\u51fa": 2113,
+    "\u6837\u54c1": 2114,
+    "\u4fdd\u6301": 2115,
+    "\u7eb9": 2116,
+    "\u6676\u4f53": 2117,
+    "\u03b1": 2118,
+    "GO": 2119,
+    "PVDF": 2120,
+    "\u5fae": 2121,
+    "\u6ee4\u819c": 2122,
+    "\u6297\u6c61\u67d3": 2123,
+    "\u5ba2\u89c2": 2124,
+    "\u6e05\u6d01": 2125,
+    "\u540c\u6b65\u673a": 2126,
+    "\u5bf9\u63a5": 2127,
+    "\u5e76\u672a": 2128,
+    "\u78c1\u969c": 2129,
+    "\u5e94\u529b\u573a": 2130,
+    "\u666e\u901a": 2131,
+    "\u7247\u5f0f": 2132,
+    "\u81ea\u8eab": 2133,
+    "Cu": 2134,
+    "\u6210\u5f62": 2135,
+    "\u78e8\u524a": 2136,
+    "\u667a\u80fd\u5316": 2137,
+    "\u4f5c\u52a8\u5668": 2138,
+    "\u5149\u6805": 2139,
+    "\u77ac\u6001": 2140,
+    "\u504f\u597d": 2141,
+    "\u5e8f": 2142,
+    "\u556e\u5408": 2143,
+    "\u8f6e\u9f7f": 2144,
+    "\u9644\u8fd1": 2145,
+    "\u94dd\u57fa": 2146,
+    "\u78b3\u5316\u7845": 2147,
+    "\u6700\u65b0": 2148,
+    "\u8230\u8239": 2149,
+    "\u5148\u8fdb\u6027": 2150,
+    "\u6c34\u6ce5": 2151,
+    "\u5e26\u6709": 2152,
+    "\u5eb7\u590d": 2153,
+    "\u8ba1\u7b97\u673a": 2154,
+    "\u4e94\u8f74": 2155,
+    "\u4e49\u9f7f": 2156,
+    "\u53ef\u89c6\u5316": 2157,
+    "\u4f9d\u6cd5\u6cbb\u56fd": 2158,
+    "\u7eb5\u5411": 2159,
+    "\u5728\u4e8e": 2160,
+    "\u7a00\u758f": 2161,
+    "\u4e0d\u9519": 2162,
+    "\u7fa4\u4f17\u89c2": 2163,
+    "\u8df5\u884c": 2164,
+    "\u9014\u5f84": 2165,
+    "\u77ff\u7269": 2166,
+    "\u65cb\u6d41": 2167,
+    "\u4f18\u826f": 2168,
+    "\u4eba\u4eec": 2169,
+    "\u6743\u5229": 2170,
+    "\u5185\u6db5": 2171,
+    "\u7a7a\u5316": 2172,
+    "\u649e\u51fb": 2173,
+    "\u78b0\u649e": 2174,
+    "15.0": 2175,
+    "\u5316\u5b66\u53cd\u5e94": 2176,
+    "\u5bfb\u627e": 2177,
+    "\u51e0\u4f55": 2178,
+    "\u673a\u5e8a": 2179,
+    "\u5e95": 2180,
+    "\u4e0d\u5f53": 2181,
+    "\u8fd1": 2182,
+    "\u65ad\u5c42": 2183,
+    "\u6709\u9650": 2184,
+    "\u6a21\u62df\u8ba1\u7b97": 2185,
+    "\u79bb\u5b50": 2186,
+    "\u5206\u79bb": 2187,
+    "\u805a\u919a": 2188,
+    "\u8fde\u7eed": 2189,
+    "\u78b3\u7ea4\u7ef4": 2190,
+    "\u51c6\u5219": 2191,
+    "\u4f20\u52a8\u8f74": 2192,
+    "\u52a0\u70ed": 2193,
+    "\u4e2a\u4eba": 2194,
+    "\u4fe1\u606f\u7f51\u7edc": 2195,
+    "\u51b3\u7b56\u8005": 2196,
+    "\u4f9b\u5e94\u5546": 2197,
+    "\u4e3a\u4f8b": 2198,
+    "\u5730\u65b9": 2199,
+    "\u8bae\u8bba": 2200,
+    "\u6f5c\u6cb9": 2201,
+    "\u7ec4": 2202,
+    "\u53d8\u7535\u7ad9": 2203,
+    "\u4e94\u9632": 2204,
+    "\u8870\u9000": 2205,
+    "\u5bf9\u6bd4\u5ea6": 2206,
+    "B1": 2207,
+    "\u53cd\u6f14": 2208,
+    "\u5e9f\u65e7": 2209,
+    "\u5c1a": 2210,
+    "\u4f7f\u7528\u8005": 2211,
+    "\u91d1\u878d": 2212,
+    "\u58f0\u97f3": 2213,
+    "\u4efb\u52a1": 2214,
+    "\u5355\u76ee": 2215,
+    "\u6444\u50cf\u673a": 2216,
+    "\u4ee3\u4ef7": 2217,
+    "\u805a\u5408": 2218,
+    "\u4e0d\u52a8\u4ea7": 2219,
+    "\u9884\u544a": 2220,
+    "\u767b\u8bb0": 2221,
+    "Cf": 2222,
+    "\u5c42\u72b6": 2223,
+    "\u8d85\u58f0": 2224,
+    "\u9884\u5236": 2225,
+    "TC4": 2226,
+    "\u710a\u7f1d": 2227,
+    "\u7194\u6c60": 2228,
+    "\u6307\u8282": 2229,
+    "\u9ad8\u65af": 2230,
+    "\u6dec\u706b": 2231,
+    "\u5438\u6ce2\u5242": 2232,
+    "\u5438\u6ce2": 2233,
+    "\u672c\u8eab": 2234,
+    "\u5e94\u5f53": 2235,
+    "Ni": 2236,
+    "\u9884\u5e94\u529b": 2237,
+    "\u5bc6\u5b9e": 2238,
+    "\u518d\u751f": 2239,
+    "\u6df7\u51dd\u571f": 2240,
+    "\u9aa8\u6599": 2241,
+    "\u6881": 2242,
+    "\u793e\u533a": 2243,
+    "\u5408\u4f5c": 2244,
+    "\u5ca9\u77f3": 2245,
+    "\u6309\u952e": 2246,
+    "\u4e8c\u7ea7": 2247,
+    "\u6eda\u5200": 2248,
+    "\u6da1\u6d41": 2249,
+    "\u6cb9\u7eb8": 2250,
+    "\u8f6e\u6bc2": 2251,
+    "\u8230\u8f7d\u673a": 2252,
+    "\u7275\u5f15\u8f66": 2253,
+    "\u5236\u52a8": 2254,
+    "\u4eba\u5de5\u667a\u80fd": 2255,
+    "\u4eba\u8138\u8bc6\u522b": 2256,
+    "\u4e73\u80f6\u624b\u5957": 2257,
+    "\u7ecf\u8425\u6743": 2258,
+    "\u571f\u5730": 2259,
+    "\u627f\u5305\u4eba": 2260,
+    "\u9000\u706b": 2261,
+    "\u65f6\u95f4\u5e38\u6570": 2262,
+    "\u70ed\u7535\u5076": 2263,
+    "\u5206\u62e3": 2264,
+    "\u6c14\u6ce1": 2265,
+    "\u7384\u6b66\u5ca9": 2266,
+    "\u884c\u661f": 2267,
+    "\u672c\u7ae0": 2268,
+    "\u534f\u540c\u6548\u5e94": 2269,
+    "\u6676\u5706": 2270,
+    "\u4e8c\u6781\u7ba1": 2271,
+    "\u80a1\u6743": 2272,
+    "\u6587\u5316\u9057\u4ea7": 2273,
+    "\u4e3b\u9898": 2274,
+    "\u76f8\u4f3c": 2275,
+    "\u77e5\u8bc6\u4ea7\u6743": 2276,
+    "\u9875": 2277,
+    "\u4f5c\u54c1": 2278,
+    "\u671f\u9650": 2279,
+    "\u8f6e": 2280,
+    "\u78c1\u529b": 2281,
+    "\u6bd4\u7387": 2282,
+    "\u6e20\u9053": 2283,
+    "\u7535\u4e3b\u8f74": 2284,
+    "\u673a\u68b0\u624b": 2285,
+    "\u94a2\u7ba1": 2286,
+    "\u725b\u817f": 2287,
+    "\u6ce2\u6d6a": 2288,
+    "\u6ed1\u7fd4": 2289,
+    "\u6c34\u9762": 2290,
+    "\u6d6e\u4f53": 2291,
+    "\u4eba\u58eb": 2292,
+    "\u534f\u4f5c": 2293,
+    "\u4eba\u673a": 2294,
+    "\u6c89\u79ef": 2295,
+    "\u62a5\u8b66": 2296,
+    "\u5b6a\u6676": 2297,
+    "\u6cb9\u70df": 2298,
+    "\u901a\u8baf": 2299,
+    "LED": 2300,
+    "LLC": 2301,
+    "\u73b0\u91d1\u6d41": 2302,
+    "\u73b0\u91d1\u6d41\u91cf": 2303,
+    "\u9884\u7b97": 2304,
+    "\u670d\u52a1": 2305,
+    "\u6c34\u4e0b": 2306,
+    "\u91cd\u5fc3": 2307,
+    "\u87ba\u65cb\u6868": 2308,
+    "\u60ef\u6027": 2309,
+    "\u9006\u5224": 2310,
+    "\u5c04\u9891": 2311,
+    "\u91cd\u4f20": 2312,
+    "\u6bcd\u6750": 2313,
+    "\u9886\u5934": 2314,
+    "\u9e1f": 2315,
+    "\u9a6c\u6c0f\u4f53": 2316,
+    "HAZ": 2317,
+    "\u970d\u59c6\u65af": 2318,
+    "\u7ecf\u9a8c\u8bba": 2319,
+    "\u4f01\u4e1a\u5bb6": 2320,
+    "\u5361\u5c14\u66fc\u6ee4\u6ce2": 2321,
+    "\u4e09\u4e09\u5236": 2322,
+    "\u653f\u6743": 2323,
+    "\u653f\u6743\u5efa\u8bbe": 2324,
+    "\u78b3\u5316\u787c": 2325,
+    "\u884c\u6ce2": 2326,
+    "\u819d\u5173\u8282": 2327,
+    "\u4e91": 2328,
+    "SCADA": 2329,
+    "CPs": 2330,
+    "\u6728\u6750": 2331,
+    "\u5c0f\u65f6": 2332,
+    "Ca": 2333,
+    "\u71c3\u673a": 2334,
+    "\u9762\u95f4": 2335,
+    "\u7acb\u4f53\u4ed3\u5e93": 2336,
+    "\u51fa\u5e93": 2337,
+    "\u7eb5\u6881": 2338,
+    "\u7fc5\u7247\u7ba1": 2339,
+    "\u6a2a\u6446": 2340,
+    "\u626b\u67e5": 2341,
+    "\u7194\u8986": 2342,
+    "\u533b\u836f\u516c\u53f8": 2343,
+    "\u62d3\u5c55": 2344,
+    "\u56fa\u4f53\u5e9f\u7269": 2345,
+    "\u7a7a\u8c03": 2346,
+    "\u6cb9\u6e29": 2347,
+    "\u5de5\u65f6": 2348,
+    "\u70ad\u819c": 2349,
+    "\u5806\u77f3": 2350,
+    "\u52a0\u5f3a\u7b4b": 2351,
+    "\u6cb9\u7bb1": 2352,
+    "\u8003\u6838": 2353,
+    "\u524a\u5cf0": 2354,
+    "\u586b\u8c37": 2355,
+    "\u5438\u5f15": 2356,
+    "\u5e72\u70ed\u5ca9": 2357,
+    "\u82b1\u5c97\u5ca9": 2358,
+    "\u4e09\u8f74": 2359,
+    "\u03b5": 2360,
+    "\u5750\u57ab": 2361,
+    "\u6c14\u6d6e": 2362,
+    "\u4e92\u8865": 2363,
+    "\u4f9d\u5b58": 2364,
+    "<pad>": 0,
+    "<start>": 1,
+    "<end>": 2,
+    "<unk>": 3
+}

data/ai_challenger_fsauor2018_testa_20180816.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a768be12cb62b4f70d99816f2c023237716129fc93785cde67ea1c777537c299
+size 6923945

data/ai_challenger_fsauor2018_trainingset_20180816.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:715f342e7be701d133abb21219fa1c2ef248de730aed14c5e1fdf5318dfef5f1
+size 49569368

data/ai_challenger_fsauor2018_validationset_20180816.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b65b4cee8749c1608e782712aac13896a5a81d023451832220ac64a01bd2abd8
+size 7115966

data/ai_challenger_sentiment_analysis_testa_20180816/README.txt

@@ -0,0 +1,2 @@
+sentiment_analysis_testa.csv 为测试集A数据文件，共15000条评论数据
+protocol.txt 为数据集下载协议

data/ai_challenger_sentiment_analysis_testa_20180816/protocol.txt

@@ -0,0 +1,12 @@
+			   数据集下载协议
+
+您（以下称“研究者”）正在请求举办方授予您访问、下载并使用数据集（以下简称“数据集”）的权利（以下简称“授权”），作为获得该等授权的条件，您同意遵守以下条款：
+
+1、研究者同意仅为非商业性的科学研究或课堂教学目的使用数据集，并不得将数据集用于任何商业用途；
+2、我们不享有数据集中使用的图片、音频、文字等内容的知识产权，对前述内容不作任何保证，包括但不限于不侵犯他人知识产权或可将前述内容用于任何特定目的；
+3、我们不承担因数据集使用造成的任何形式的损失或伤害，不会对任何因使用比赛数据产生的法律后果承担任何责任；
+4、 与数据集使用有关的任何法律责任均由研究者承担，如研究者或其员工、代理人、分支机构使用数据集的行为给我们造成声誉或经济损失，研究者应当承担赔偿责任；
+5、研究者可以授权其助手、同事或其他合作者访问和使用数据集，但应确保前述人员已经认真阅读并同意接受本协议约束；
+6、如果研究者受雇于以盈利为目的的商业主体，应确保使用数据集仅用于非商业目的，且其雇主同样受本协议约束，研究者确认其签订本协议前已经取得雇主的充分授权。
+7、我们有权随时取消或撤回对研究者使用数据集的授权，并有权要求研究者删除已下载数据集；
+8、凡因本合同引起的或与本合同有关的任何争议，均应提交中国国际经济贸易仲裁委员会，按照申请仲裁时该会现行有效的仲裁规则，并适用中华人民共和国法律解决进行仲裁。仲裁语言应为中文。

data/ai_challenger_sentiment_analysis_testa_20180816/sentiment_analysis_testa.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:807246836a9dc6974860bdb18a33be07cb850d2ed5e28c87170514fe16df33d2
+size 15608586

data/ai_challenger_sentiment_analysis_testa_20180816/sentiment_analysis_testa2.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f853d7296a74269ec3da9a2bf3ed1493cd422d40bd799dbb41ec9fd09543b1f1
+size 200185

data/ai_challenger_sentiment_analysis_trainingset_20180816/README.txt

@@ -0,0 +1,3 @@
+sentiment_analysis_trainingset.csv 为训练集数据文件，共105000条评论数据
+sentiment_analysis_trainingset_annotations.docx 为数据标注说明文件
+protocol.txt 为数据集下载协议

data/ai_challenger_sentiment_analysis_trainingset_20180816/protocol.txt

@@ -0,0 +1,12 @@
+			   数据集下载协议
+
+您（以下称“研究者”）正在请求举办方授予您访问、下载并使用数据集（以下简称“数据集”）的权利（以下简称“授权”），作为获得该等授权的条件，您同意遵守以下条款：
+
+1、研究者同意仅为非商业性的科学研究或课堂教学目的使用数据集，并不得将数据集用于任何商业用途；
+2、我们不享有数据集中使用的图片、音频、文字等内容的知识产权，对前述内容不作任何保证，包括但不限于不侵犯他人知识产权或可将前述内容用于任何特定目的；
+3、我们不承担因数据集使用造成的任何形式的损失或伤害，不会对任何因使用比赛数据产生的法律后果承担任何责任；
+4、 与数据集使用有关的任何法律责任均由研究者承担，如研究者或其员工、代理人、分支机构使用数据集的行为给我们造成声誉或经济损失，研究者应当承担赔偿责任；
+5、研究者可以授权其助手、同事或其他合作者访问和使用数据集，但应确保前述人员已经认真阅读并同意接受本协议约束；
+6、如果研究者受雇于以盈利为目的的商业主体，应确保使用数据集仅用于非商业目的，且其雇主同样受本协议约束，研究者确认其签订本协议前已经取得雇主的充分授权。
+7、我们有权随时取消或撤回对研究者使用数据集的授权，并有权要求研究者删除已下载数据集；
+8、凡因本合同引起的或与本合同有关的任何争议，均应提交中国国际经济贸易仲裁委员会，按照申请仲裁时该会现行有效的仲裁规则，并适用中华人民共和国法律解决进行仲裁。仲裁语言应为中文。

data/ai_challenger_sentiment_analysis_trainingset_20180816/sentiment_analysis_trainingset.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b9fe39cb75d3cbe8cb72871b963678fd56885d55fe6a80182838d3ef71955fea
+size 112621685

data/ai_challenger_sentiment_analysis_validationset_20180816/README.txt

@@ -0,0 +1,3 @@
+sentiment_analysis_validationset.csv 为验证集数据文件，共15000条评论数据
+sentiment_analysis_validationset_annotations.docx 为数据标注说明文件
+protocol.txt 为数据集下载协议

data/ai_challenger_sentiment_analysis_validationset_20180816/protocol.txt

@@ -0,0 +1,12 @@
+			   数据集下载协议
+
+您（以下称“研究者”）正在请求举办方授予您访问、下载并使用数据集（以下简称“数据集”）的权利（以下简称“授权”），作为获得该等授权的条件，您同意遵守以下条款：
+
+1、研究者同意仅为非商业性的科学研究或课堂教学目的使用数据集，并不得将数据集用于任何商业用途；
+2、我们不享有数据集中使用的图片、音频、文字等内容的知识产权，对前述内容不作任何保证，包括但不限于不侵犯他人知识产权或可将前述内容用于任何特定目的；
+3、我们不承担因数据集使用造成的任何形式的损失或伤害，不会对任何因使用比赛数据产生的法律后果承担任何责任；
+4、 与数据集使用有关的任何法律责任均由研究者承担，如研究者或其员工、代理人、分支机构使用数据集的行为给我们造成声誉或经济损失，研究者应当承担赔偿责任；
+5、研究者可以授权其助手、同事或其他合作者访问和使用数据集，但应确保前述人员已经认真阅读并同意接受本协议约束；
+6、如果研究者受雇于以盈利为目的的商业主体，应确保使用数据集仅用于非商业目的，且其雇主同样受本协议约束，研究者确认其签订本协议前已经取得雇主的充分授权。
+7、我们有权随时取消或撤回对研究者使用数据集的授权，并有权要求研究者删除已下载数据集；
+8、凡因本合同引起的或与本合同有关的任何争议，均应提交中国国际经济贸易仲裁委员会，按照申请仲裁时该会现行有效的仲裁规则，并适用中华人民共和国法律解决进行仲裁。仲裁语言应为中文。

data/ai_challenger_sentiment_analysis_validationset_20180816/sentiment_analysis_validationset.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a72efc85000b735c4a1e795e3f854c880887bf0a25216c8e81e5d99ddb6979f8
+size 16016113

data/protocol.txt

@@ -0,0 +1,12 @@
+			   数据集下载协议
+
+您（以下称“研究者”）正在请求举办方授予您访问、下载并使用数据集（以下简称“数据集”）的权利（以下简称“授权”），作为获得该等授权的条件，您同意遵守以下条款：
+
+1、研究者同意仅为非商业性的科学研究或课堂教学目的使用数据集，并不得将数据集用于任何商业用途；
+2、我们不享有数据集中使用的图片、音频、文字等内容的知识产权，对前述内容不作任何保证，包括但不限于不侵犯他人知识产权或可将前述内容用于任何特定目的；
+3、我们不承担因数据集使用造成的任何形式的损失或伤害，不会对任何因使用比赛数据产生的法律后果承担任何责任；
+4、 与数据集使用有关的任何法律责任均由研究者承担，如研究者或其员工、代理人、分支机构使用数据集的行为给我们造成声誉或经济损失，研究者应当承担赔偿责任；
+5、研究者可以授权其助手、同事或其他合作者访问和使用数据集，但应确保前述人员已经认真阅读并同意接受本协议约束；
+6、如果研究者受雇于以盈利为目的的商业主体，应确保使用数据集仅用于非商业目的，且其雇主同样受本协议约束，研究者确认其签订本协议前已经取得雇主的充分授权。
+7、我们有权随时取消或撤回对研究者使用数据集的授权，并有权要求研究者删除已下载数据集；
+8、凡因本合同引起的或与本合同有关的任何争议，均应提交中国国际经济贸易仲裁委员会，按照申请仲裁时该会现行有效的仲裁规则，并适用中华人民共和国法律解决进行仲裁。仲裁语言应为中文。

data/test/test_data.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e98ad61b049ebf83bc1bbcb6fdb1cfc5c044f65eed77bdcdf3983a1ed5669ba3
+size 609792

data/test/test_data_old.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b44d2ea97a6beca9822e6cd2f977ba90d7ee30cd821f1248d6939312fca9ce4
+size 158194

data/total.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a702dda1b1abec204cb0a389c39bcd3968fcbfc4e802ae263b72b8fe772f9c98
+size 3102984

data/train/train_data.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d470333dade24f7f5bb484c64aa66372206dddf89c0a152a9d23489aabf248e
+size 1873975

data/train/train_data2.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4fa7c2fc2712066a55f1367fc4421e3e7206400e315b68df7defd594e1a34030
+size 468835

data/train/train_data_old.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:97cd8de21112574716c34b2dcaf7bfede6b32b2f3dfe19a06367c477c09aa159
+size 469082

data/valid/val_data.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1834db140b02f3f530cd85272a2fbfc219b30bc9fb19e7a5893ddc81c79cfd0
+size 616683

data/valid/val_data_old.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c826978817a4677a8fbab1195e2ac7ff9741aa653da16f1dd7a7f3f56f844520
+size 154559

data/valid/专家评语.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:55621fde0640b7d171df93ff215ec75f74bd3c3b0c7605f2e6ace9e9a7992a57
+size 535168

data_gen.py

@@ -0,0 +1,249 @@
+import itertools
+
+import jieba
+import numpy as np
+from torch.utils.data import Dataset
+import  config
+from tqdm import tqdm
+from utils import *
+import jieba_cut
+
+
+def to_categorical(y, num_classes):
+    """ 1-hot encodes a tensor """
+    return np.eye(num_classes, dtype='uint8')[y]
+
+
+# Meaning	    Positive	Neutral	    Negative	Not mentioned
+# Old labels    1	        0 	        -1	        -2
+# New labels    3           2           1           0                      2.1.0
+def map_sentimental_type(value):
+    #return value + 2
+    return value + 1
+
+
+def parse_user_reviews(user_reviews):
+    samples = []
+    for i in range(len(user_reviews)):
+        content = user_reviews['content'][i]
+        label_tensor = np.empty((num_labels,), dtype=np.int32)
+        for idx, name in enumerate(label_names):
+            sentimental_type = user_reviews[name][i]
+            y = map_sentimental_type(sentimental_type)
+            # label_tensor[:, idx] = to_categorical(y, num_classes)
+            # CrossEntropyLoss does not expect a one-hot encoded vector as the target, but class indices.
+            label_tensor[idx] = y
+        samples.append({'content': content, 'label_tensor': label_tensor})
+    #print("samples",samples)
+    return samples
+
+PAD, CLS, SEP = '[PAD]', '[CLS]','[SEP]'  # padding符号, bert中综合信息符号
+def parse_user_reviews_bert(user_reviews):
+    samples = []
+    pad_size=config.pad_size
+    sen_pad_size=config.sentence_pad_size
+    for i in range(len(user_reviews)):
+        content = user_reviews['content'][i]
+        sentences=jieba_cut.cut_sentence(content)
+        label_tensor = np.empty((num_labels,), dtype=np.int32)
+        sentences_ids=[]
+        sentences_mask=[]
+        for sentence in sentences:
+            token = config.tokenizer.tokenize(sentence)
+            ids = config.tokenizer.convert_tokens_to_ids(token)
+            if sen_pad_size:
+                if len(token) < sen_pad_size:
+                    sentence_mask = [1] * len(ids) + [0] * (sen_pad_size - len(token))
+                    ids += ([0] * (sen_pad_size - len(token)))
+                else:
+                    sentence_mask = [1] * sen_pad_size
+                    ids = ids[:sen_pad_size]
+                    seq_len = sen_pad_size
+            #print("每个句子的长度",len(ids))
+            sentences_ids.append(ids)
+            sentences_mask.append(sentence_mask)
+        #print(sentences_ids,len(sentences_ids))
+
+        for idx, name in enumerate(label_names):
+            sentimental_type = user_reviews[name][i]
+            y = map_sentimental_type(sentimental_type)
+            # label_tensor[:, idx] = to_categorical(y, num_classes)
+            # CrossEntropyLoss does not expect a one-hot encoded vector as the target, but class indices.
+            label_tensor[idx] = y
+        token = config.tokenizer.tokenize(content)
+        #token = [CLS] + ["标题", "选题角度与价值", "文献综述归纳总结情况", "论文工作量", "是否掌握基础（专业）知识", "是否具备科研能力", "格式规范", "行文表达", "逻辑性", "研究方法", "研究结论", "创新性及论文价值", "理论深度", "学术端正性"]+token
+        aspect_token=config.tokenizer.tokenize("标题, 选题角度与价值, 文献综述归纳总结情况, 论文工作量, 是否掌握基础（专业）知识, 是否具备科研能力, 格式规范, 行文表达, 逻辑性, 研究方法, 研究结论, 创新性及论文价值, 理论深度, 学术端正性")
+        #print(aspect_token)
+        token = [CLS] + token
+        #token = [CLS] + token+ [SEP]+aspect_token+[SEP]
+        #print("token",token)
+        seq_len = len(token)
+        #print(seq_len)
+        mask = []
+        #这一步只是根据字典将字符转换为对应ID，还没有自注意力的运算
+        token_ids = config.tokenizer.convert_tokens_to_ids(token)         #config.tokenizer=BertTokenizer.from_pretrained(bert_path).convert_tokens_to_ids
+
+        #print("token_ids",token_ids)
+        if pad_size:
+            if len(token) < pad_size:
+                mask = [1] * len(token_ids) + [0] * (pad_size - len(token))
+                token_ids += ([0] * (pad_size - len(token)))
+            else:
+                mask = [1] * pad_size
+                token_ids = token_ids[:pad_size]
+                seq_len = pad_size
+        #contents.append((token_ids, int(label), seq_len, mask))
+        samples.append({'content': token_ids, 'label_tensor': label_tensor,'mask':mask,'sentences':sentences_ids,'sentences_mask':sentences_mask})
+    #print("samples", samples)
+    return samples
+
+
+def zeroPadding(l, fillvalue=PAD_token):
+    return list(itertools.zip_longest(*l, fillvalue=fillvalue))
+
+
+# Returns padded input sequence tensor and lengths
+def inputVar(indexes_batch):
+    lengths = torch.tensor([len(indexes) for indexes in indexes_batch])
+    padList = zeroPadding(indexes_batch)
+    padVar = torch.LongTensor(padList)
+    return padVar, lengths
+
+
+# Returns all items for a given batch of pairs
+def batch2TrainData(pair_batch):
+    pair_batch.sort(key=lambda x: len(x[0]), reverse=True)
+    input_batch, output_batch = [], []
+    for pair in pair_batch:
+        input_batch.append(pair[0])
+        output_batch.append(pair[1])
+    inp, lengths = inputVar(input_batch)
+    output = torch.LongTensor(output_batch)
+    return inp, lengths, output
+
+def batch2TrainData_bert(pair_batch):
+    pair_batch.sort(key=lambda x: len(x[0]), reverse=True)
+    input_batch, output_batch,mask,sentences,sentences_mask = [], [],[],[],[]
+    for pair in pair_batch:
+        input_batch.append(pair[0])
+        output_batch.append(pair[1])
+        mask.append(pair[2])
+        sentences.append(pair[3])
+        sentences_mask.append(pair[4])
+    inp, lengths = inputVar(input_batch)
+    output = torch.LongTensor(output_batch)
+    return inp, lengths, output,mask,sentences,sentences_mask
+
+class SaDataset(Dataset):
+    def __init__(self, split, voc,flag):
+        self.split = split
+        self.voc = voc
+        self.flag=flag
+        #assert self.split in {'train', 'valid'}   #断言语句，用于检查self.split的取值是否在{'train', 'valid'}中。
+
+        if split == 'train':
+            filename = os.path.join(train_folder, train_filename)
+        elif split == 'valid':
+            filename = os.path.join(valid_folder, valid_filename)
+        else:
+            filename = os.path.join(test_a_folder, test_a_filename)
+
+        if flag=='bert':
+            user_reviews = pd.read_csv(filename)
+            self.samples = parse_user_reviews_bert(user_reviews)   #到这一步都没问题，samples中含有mask
+            self.num_chunks = len(self.samples) // chunk_size
+            #print(self.samples,"sampless!!!!")
+        else:
+            # user_reviews = pd.read_csv(filename,encoding='gbk')
+            user_reviews = pd.read_csv(filename)
+            self.samples = parse_user_reviews(user_reviews)
+            self.num_chunks = len(self.samples) // chunk_size
+
+
+    def __getitem__(self, i):
+
+        pair_batch = []
+        if self.flag=='bert':
+            for i_chunk in range(chunk_size):
+                idx = i * chunk_size + i_chunk
+                content = self.samples[idx]['content']
+                input_indexes = content
+                # print(list(seg_list))
+                label_tensor = self.samples[idx]['label_tensor']
+                mask = self.samples[idx]['mask']
+                sentences=self.samples[idx]['sentences']   #分句的列表
+                sentences_mask=self.samples[idx]['sentences_mask']  #分句的mask
+                pair_batch.append((input_indexes, label_tensor,mask,sentences,sentences_mask))
+
+            return batch2TrainData_bert(pair_batch)
+
+        else:
+            for i_chunk in range(chunk_size):
+                idx = i * chunk_size + i_chunk
+                content = self.samples[idx]['content']
+                content = content.strip()
+                seg_list = jieba.cut(content)
+                input_indexes = encode_text(self.voc.word2index, list(seg_list))
+                # print(list(seg_list))
+                label_tensor = self.samples[idx]['label_tensor']
+                pair_batch.append((input_indexes, label_tensor))
+
+            return batch2TrainData(pair_batch)
+
+
+        # pair_batch = []
+        #
+        # for i_chunk in range(chunk_size):
+        #     idx = i * chunk_size + i_chunk
+        #     content = self.samples[idx]['content']
+        #     # print(list(seg_list))
+        #     label_tensor = self.samples[idx]['label_tensor']
+        #     mask=self.samples[idx]['mask']
+        #     pair_batch.append((content, label_tensor,mask))
+        #
+        # return pair_batch
+
+
+    def __len__(self):
+        return self.num_chunks
+
+
+
+
+
+def build_dataset(config):
+
+    def load_dataset(path, pad_size=32):
+        contents = []
+
+        with open(path, 'r', encoding='UTF-8') as f:
+            for line in tqdm(f):
+                #print(line)
+                lin = line.strip()
+                #print(lin)
+                if not lin:
+                    continue
+                content, label = lin.split('\t')
+                #print(label)
+                #content, label = lin.split(',')
+                token = config.tokenizer.tokenize(content)
+                #print(token)
+                token = [CLS] + token
+                seq_len = len(token)
+                mask = []
+                token_ids = config.tokenizer.convert_tokens_to_ids(token)
+
+                if pad_size:
+                    if len(token) < pad_size:
+                        mask = [1] * len(token_ids) + [0] * (pad_size - len(token))
+                        token_ids += ([0] * (pad_size - len(token)))
+                    else:
+                        mask = [1] * pad_size
+                        token_ids = token_ids[:pad_size]
+                        seq_len = pad_size
+                contents.append((token_ids, int(label), seq_len, mask))
+        return contents
+    train = load_dataset(config.train_path, config.pad_size)
+    dev = load_dataset(config.dev_path, config.pad_size)
+    test = load_dataset(config.test_path, config.pad_size)
+    return train, dev, test

data_make.py

@@ -0,0 +1,22 @@
+import pandas as pd
+import numpy as np
+from sklearn.model_selection import train_test_split
+import chardet
+
+# 检测文件的编码方式
+with open('data/total.csv', 'rb') as f:
+    result = chardet.detect(f.read())
+
+# 使用检测到的编码方式读取文件
+#data = pd.read_csv('专家评语.csv', encoding=result['encoding'])
+
+data = pd.read_csv('data/total.csv',encoding=result['encoding'])
+
+# 划分数据集为训练集、验证集和测试集
+train_data, test_data = train_test_split(data, test_size=0.2, random_state=42)
+train_data, val_data = train_test_split(train_data, test_size=0.25, random_state=42)
+
+# 将数据保存到新的CSV文件
+train_data.to_csv('train_data.csv', index=False)
+val_data.to_csv('val_data.csv', index=False)
+test_data.to_csv('test_data.csv', index=False)

ddparser_test.py

@@ -0,0 +1,8 @@
+from ddparser import DDParser
+# -*- coding: utf-8 -*-
+
+ddp = DDParser(use_pos=True)
+str = '本文基于单目视觉方法，研究了一种动态环境下的视觉SLAM系统，利用光流检测算法剔除动态特征点，从而利用静态特征点估计相机位姿，实现单目视觉的SLAM定位功能；课题具有较为重要的理论意义与实用价值。论文结构较为严谨，工作量满足学位论文要求，研究成果反映了作者专业理论知识基础较为扎实，具有一定的从事科学研究能力。'
+result = ddp.parse(str)
+print(result)
+

demo.py

@@ -0,0 +1,80 @@
+# import the necessary packages
+import json
+import os
+import random
+
+import jieba
+import pandas as pd
+import torch
+
+from config import device, save_folder, valid_folder, valid_filename
+from data_gen import parse_user_reviews, batch2TrainData
+from utils import Lang, encode_text
+
+if __name__ == '__main__':
+    voc = Lang('data/WORDMAP.json')
+    print("voc.n_words: " + str(voc.n_words))
+
+    checkpoint = '{}/BEST_checkpoint.tar'.format(save_folder)  # model checkpoint
+    print('checkpoint: ' + str(checkpoint))
+
+    # Load model
+    checkpoint = torch.load(checkpoint)
+    encoder = checkpoint['encoder']
+    print(encoder)
+
+    # Use appropriate device
+    encoder = encoder.to(device)
+
+    # Set dropout layers to eval mode
+    encoder.eval()
+
+    filename = os.path.join(valid_folder, valid_filename)
+    user_reviews = pd.read_csv(filename)
+    samples = parse_user_reviews(user_reviews)
+
+    #samples = random.sample(samples, 10)  #随机选取10个来验证
+    pair_batch = []
+    result = []
+    label=[]
+    for i, sample in enumerate(samples):
+        #print(i,sample)
+        content = sample['content']
+
+        result.append({'content': content})
+        #result.append({'ori_label': ori_label})  #把真实标签写入结果，方便对比查看
+
+        label_tensor = sample['label_tensor']
+        #label.append({'label':label_tensor})
+        content = content.strip()
+        seg_list = jieba.cut(content)
+        input_indexes = encode_text(voc.word2index, list(seg_list))
+        label_tensor = sample['label_tensor']
+        pair_batch.append((input_indexes, label_tensor))
+
+    #print(pair_batch[0])
+    test_data = batch2TrainData(pair_batch)
+    #print(test_data)
+    input_variable, lengths, _ = test_data
+    input_variable = input_variable.to(device)
+    lengths = lengths.to(device)
+    outputs = encoder(input_variable, lengths)
+
+    _, outputs = torch.max(outputs, 1)    #_下划线表示不需要，max函数返回最大值的索引，也就是模型预测的结果
+    print(outputs.shape)
+    print('outputs.size(): ' + str(outputs.size()))
+    outputs = outputs.cpu().numpy()
+
+
+
+    for i in range(len(samples)):
+        result[i]['labels'] = (outputs[i] - 1).tolist()
+        result[i]['ori_label'] =(samples[i]['label_tensor']-1).tolist()  #转换成list方便保存成json，-1是因为做标签映射的时候+1了
+
+        #print(result[i])
+        #print("origin",label[i]['label']-1)
+       # print("result",result[i]['labels'])
+
+    with open('result.json', 'w') as file:
+        json.dump(result, file, indent=4, ensure_ascii=False)
+

extract.py

@@ -0,0 +1,31 @@
+import os
+import zipfile
+
+train_folder = 'data/ai_challenger_fsauor2018_trainingset_20180816'
+valid_folder = 'data/ai_challenger_fsauor2018_validationset_20180816'
+test_a_folder = 'data/ai_challenger_fsauor2018_testa_20180816'
+
+
+def ensure_folder(folder):
+    if not os.path.exists(folder):
+        os.makedirs(folder)
+
+
+def extract(folder):
+    filename = '{}.zip'.format(folder)
+    print('Extracting {}...'.format(filename))
+    with zipfile.ZipFile(filename, 'r') as zip_ref:
+        zip_ref.extractall('data')
+
+
+if __name__ == '__main__':
+    ensure_folder('data')
+
+    if not os.path.isdir(train_folder):
+        extract(train_folder)
+
+    if not os.path.isdir(valid_folder):
+        extract(valid_folder)
+
+    if not os.path.isdir(test_a_folder):
+        extract(test_a_folder)

gbk2utf.py

@@ -0,0 +1,13 @@
+import codecs
+
+with codecs.open('./data/valid/val_data.csv', 'r', 'gbk') as f:
+    content = f.read()
+
+
+
+content_utf8 = content.encode('utf-8')
+
+#print(content)
+
+with codecs.open('./data/valid/val_data.csv', 'w', 'utf-8') as f:
+    f.write(content_utf8.decode('utf-8'))

jieba_cut.py

@@ -0,0 +1,71 @@
+# -*- coding: utf-8 -*-
+import jieba
+import pandas as pd
+import config
+
+def cut_sentence(text):
+    sentence_list = []
+    seg_list = jieba.cut(text, cut_all=False)
+    seg_list = [word for word in seg_list if word.strip()]  # 去除空字符
+    start = 0
+    for i, word in enumerate(seg_list):
+        if word in ['。', '！', '？','；']:
+            sentence = ''.join(seg_list[start:i+1])
+            sentence_list.append(sentence)
+            start = i + 1
+    if start < len(seg_list):
+        sentence = ''.join(seg_list[start:])
+        sentence_list.append(sentence)
+    return sentence_list
+
+filename="data/train/train_data.csv"
+filename2="data/total.csv"
+user_reviews = pd.read_csv(filename)
+#print(user_reviews)
+max_len=0
+# '''统计下最长句子的长度'''
+# for i in range(len(user_reviews)):
+#     content = user_reviews['content'][i]
+#     sentences = cut_sentence(content)
+#     for sentence in sentences:
+#         length=len(sentence)
+#         #print(sentence)
+#         if length>max_len:
+#             max_len=length
+# print("最大句子长度",max_len)
+'''统计下前5长的句子的长度'''
+lengths = []  # 用于存储句子长度的列表
+conten_lengths=[]
+for i in range(len(user_reviews)):
+    content = user_reviews['content'][i]
+    con_len=len(content)
+    conten_lengths.append(con_len)
+    sentences = cut_sentence(content)
+    for sentence in sentences:
+        length = len(sentence)
+        lengths.append(length)
+        #print(sentence)
+        # if length==518:
+        #     print(sentence)
+# 对句子长度进行排序
+lengths.sort(reverse=True)
+# 获取前5个最长句子的长度
+top_5_lengths = lengths[:300]
+# 对整段长度进行排序
+conten_lengths.sort(reverse=True)
+# 获取前5个最长段落的长度
+count = 0  # 用于计数大于512的个数
+
+for length in conten_lengths:
+    if length > 512:
+        count += 1
+
+print("大于512的评语数:", count)
+top_5_content_lengths = conten_lengths[:300]
+print("前5个最长句子的长度:", top_5_lengths)
+print("前5个最长段落的长度:", top_5_content_lengths)
+
+
+
+# text="再制造是一项融入节约资源、环保等理念的绿色制造模式！随着工业技术不断更新换代，大量机床面临淘汰，机床再制造已经成为我国重点扶持的一项产业。本课题以某废旧机床再制造企业为背景，深入研究了废旧机床再制造的逆向物流过程。论文选题针对性明确，实用性明显。主要研究成果：1、提出了基于GM（1,1）和傅立叶级数的废旧机床回收量预测模型，用傅立叶级数对GM（1,1）模型预测值与原数值的残差序列进行拟合，以提高废旧机床回收量的预测精度。2、构建了基于碳税约束条件下废旧机床再制造逆向物流网络模型，从碳排放总量、网络总成本两个方面对逆向物流网络进行规划设计，并进行了求解。3、运用序关系法与逼近理想解排序法的混合评价方法，建立了第三方物流供应商评价指标和体系。论文对国内外该研究领域的研究现状调研分析较全面，参考文献尚充实，研究工作结合实际，有作者的独立见解。但是论文逻辑性较差，写作不规范，数学模型的推导过程欠严谨，符号变量表示较混乱，写作质量较差，表述不清和错别字很多，标点符号问题较多，实验研究工作量小，英文写作质量较差。论文工作表明该生基本能够掌握本学科领域的基础理论和专业知识，有一定独从事科学研究工作的能力。论文尚能够达到硕士学位论文的水平和要求，同意答辩。"
+# sentences = cut_sentence(text)

log.txt

@@ -0,0 +1,287 @@
+voc.n_words: 2365
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_CNN_BiGru_parms 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_CNN_BiGru_parms 0.02 ========================================
+[2024-04-23 09:48:24] Epoch: [0][0/230]	Batch Time 7.092 (7.092)	Loss 1.0954 (0.1095)	Accuracy 35.714 (3.571)
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_CNN_BiGru_parms 0.02 ========================================
+[2024-04-23 09:55:35] Epoch: [0][0/230]	Batch Time 6.703 (6.703)	Loss 1.0695 (0.1070)	Accuracy 42.857 (4.286)
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_CNN_BiGru_parms 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_CNN_BiGru_parms 0.02 ========================================
+[2024-04-23 10:24:34] Epoch: [0][0/230]	Batch Time 6.640 (6.640)	Loss 1.1402 (0.1140)	Accuracy 33.036 (3.304)
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_CNN_BiGru_parms 0.02 ========================================
+[2024-04-23 10:25:11] Epoch: [0][0/230]	Batch Time 7.017 (7.017)	Loss 1.1596 (0.1160)	Accuracy 31.250 (3.125)
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_CNN_BiGru_parms 0.02 ========================================
+[2024-04-23 10:26:06] Epoch: [0][0/230]	Batch Time 6.974 (6.974)	Loss 1.1505 (0.1151)	Accuracy 32.143 (3.214)
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_CNN_BiGru_parms 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_CNN_BiGru_parms 0.02 ========================================
+[2024-04-23 10:32:54] Epoch: [0][0/230]	Batch Time 6.740 (6.740)	Loss 1.1093 (0.1109)	Accuracy 42.857 (4.286)
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_CNN_BiGru_parms 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_CNN_BiGru_parms 0.02 ========================================
+[2024-04-23 10:49:56] Epoch: [0][0/230]	Batch Time 6.495 (6.495)	Loss 1.1974 (0.1197)	Accuracy 35.714 (3.571)
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_CNN_BiGru_parms 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_CNN_BiGru_parms 0.02 ========================================
+[2024-04-23 10:53:17] Epoch: [0][0/230]	Batch Time 6.078 (6.078)	Loss 1.2826 (0.1283)	Accuracy 25.000 (2.500)
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_CNN_BiGru_parms 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_CNN_BiGru_parms 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_CNN_BiGru_parms 0.02 ========================================
+[2024-04-23 13:56:12] Epoch: [0][0/230]	Batch Time 6.209 (6.209)	Loss 1.1109 (0.1111)	Accuracy 33.929 (3.393)
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_gru_totalSentence 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_gru_totalSentence 0.02 ========================================
+[2024-04-23 16:00:01] Epoch: [0][0/229]	Batch Time 26.299 (26.299)	Loss 1.1046 (0.1105)	Accuracy 34.821 (3.482)
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_gru_totalSentence 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert 0.02 ========================================
+[2024-04-23 16:03:18] Epoch: [0][0/613]	Batch Time 8.473 (8.473)	Loss 1.1432 (0.1143)	Accuracy 28.571 (2.857)
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert 0.02 ========================================
+[2024-04-23 16:45:49] Epoch: [0][0/613]	Batch Time 6.602 (6.602)	Loss 1.0679 (0.1068)	Accuracy 30.952 (3.095)
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert 0.02 ========================================
+[2024-04-23 17:01:25] Epoch: [0][0/613]	Batch Time 5.616 (5.616)	Loss 1.1528 (0.1153)	Accuracy 28.571 (2.857)
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert 0.02 ========================================
+[2024-04-24 13:59:24] Epoch: [0][0/613]	Batch Time 6.501 (6.501)	Loss 1.2672 (0.1267)	Accuracy 26.190 (2.619)
+voc.n_words: 2365
+voc.n_words: 2365
+voc.n_words: 2365
+voc.n_words: 2365
+voc.n_words: 2365
+voc.n_words: 2365
+voc.n_words: 2365
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert 0.02 ========================================
+[2024-04-24 14:45:52] Epoch: [0][0/613]	Batch Time 12.326 (12.326)	Loss 1.3152 (0.1315)	Accuracy 23.810 (2.381)
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert 0.02 ========================================
+[2024-04-24 14:48:30] Epoch: [0][0/613]	Batch Time 9.587 (9.587)	Loss 1.0519 (0.1052)	Accuracy 42.857 (4.286)
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert 0.02 ========================================
+[2024-04-24 14:50:03] Epoch: [0][0/613]	Batch Time 9.403 (9.403)	Loss 1.2586 (0.1259)	Accuracy 26.190 (2.619)
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_CNN_BiGru 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_CNN_BiGru 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_CNN_BiGru 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_sentence 0.02 ========================================
+[2024-04-24 15:41:59] Epoch: [0][0/613]	Batch Time 11.057 (11.057)	Loss 1.2508 (0.1251)	Accuracy 19.048 (1.905)
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_sentence 0.02 ========================================
+
+ * ACCURACY - 0.000, LOSS - 0.000
+
+
+Epochs since last improvement: 1
+
+
+ * ACCURACY - 0.000, LOSS - 0.000
+
+
+Epochs since last improvement: 2
+
+
+ * ACCURACY - 0.000, LOSS - 0.000
+
+
+Epochs since last improvement: 3
+
+
+ * ACCURACY - 0.000, LOSS - 0.000
+
+
+Epochs since last improvement: 4
+
+
+ * ACCURACY - 0.000, LOSS - 0.000
+
+
+Epochs since last improvement: 5
+
+
+ * ACCURACY - 0.000, LOSS - 0.000
+
+
+Epochs since last improvement: 6
+
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_sentence 0.02 ========================================
+
+ * ACCURACY - 0.000, LOSS - 0.000
+
+
+Epochs since last improvement: 1
+
+
+ * ACCURACY - 0.000, LOSS - 0.000
+
+
+Epochs since last improvement: 2
+
+
+ * ACCURACY - 0.000, LOSS - 0.000
+
+
+Epochs since last improvement: 3
+
+
+ * ACCURACY - 0.000, LOSS - 0.000
+
+
+Epochs since last improvement: 4
+
+
+ * ACCURACY - 0.000, LOSS - 0.000
+
+
+Epochs since last improvement: 5
+
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_sentence 0.02 ========================================
+
+ * ACCURACY - 0.000, LOSS - 0.000
+
+
+Epochs since last improvement: 1
+
+
+ * ACCURACY - 0.000, LOSS - 0.000
+
+
+Epochs since last improvement: 2
+
+
+ * ACCURACY - 0.000, LOSS - 0.000
+
+
+Epochs since last improvement: 3
+
+
+ * ACCURACY - 0.000, LOSS - 0.000
+
+
+Epochs since last improvement: 4
+
+
+ * ACCURACY - 0.000, LOSS - 0.000
+
+
+Epochs since last improvement: 5
+
+
+ * ACCURACY - 0.000, LOSS - 0.000
+
+
+Epochs since last improvement: 6
+
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_sentence 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_sentence 0.02 ========================================
+[2024-04-24 16:05:07] Epoch: [0][0/613]	Batch Time 6.894 (6.894)	Loss 1.2131 (0.1213)	Accuracy 33.333 (3.333)
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_sentence 0.02 ========================================
+[2024-04-24 16:05:40] Epoch: [0][0/613]	Batch Time 5.485 (5.485)	Loss 1.1047 (0.1105)	Accuracy 42.857 (4.286)
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_sentence 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_sentence 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_sentence3 0.02 ========================================
+voc.n_words: 2365
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_sentence3 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_sentence3 0.02 ========================================
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_sentence3 0.02 ========================================
+[2024-05-17 09:18:27] Epoch: [0][0/613]	Batch Time 7.952 (7.952)	Loss 1.1274 (0.1127)	Accuracy 30.952 (3.095)
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_sentence3 0.02 ========================================
+[2024-05-17 09:33:19] Epoch: [0][0/613]	Batch Time 7.035 (7.035)	Loss 1.1134 (0.1113)	Accuracy 33.333 (3.333)
+voc.n_words: 2365
+Building optimizers ...
+================================= Bert_sentence3 0.02 ========================================
+[2024-05-17 09:35:55] Epoch: [0][0/613]	Batch Time 7.327 (7.327)	Loss 1.0643 (0.1064)	Accuracy 47.619 (4.762)

model2.py

@@ -0,0 +1,298 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from torch.autograd import Variable
+
+import config
+from config import num_labels, num_classes, batch_first, pad_size, chunk_size, hidden_size,filter_sizes
+from pytorch_pretrained import BertModel, BertTokenizer
+
+import torch
+import torch.nn as nn
+
+import torch
+import torch.nn as nn
+
+import torch
+import torch.nn as nn
+
+class Bert_add_aspect(nn.Module):
+    def __init__(self, input_size, bert_path,hidden_size, n_layers=1, dropout=0):
+        super(Bert_add_aspect, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.bert = BertModel.from_pretrained(bert_path)
+        for param in self.bert.parameters():
+            param.requires_grad = True
+        self.fc = nn.Linear(hidden_size, num_labels * num_classes)
+
+
+    def forward(self, input_variable,mask, hidden=None):
+        # samples=samples.T
+        # print(samples)
+        context = input_variable.T  # 输入的句子
+        #print(context)
+        mask = torch.tensor(mask).to("cuda")  # 对padding部分进行mask，和句子一个size，padding部分用0表示，如：[1, 1, 1, 1, 0, 0]
+        #print('mask',mask)
+        _, pooled = self.bert(context, attention_mask=mask,output_all_encoded_layers=False )  #output_all_encoded_layers=False
+        # print(pooled.size())
+        out = self.fc(pooled)
+        out = out.view((-1, num_classes,num_labels))
+        # outputs = [batch size, num_classes, num_labels]
+        out = F.log_softmax(out, dim=1)
+        return out
+class NonLocalBlock(nn.Module):
+    def __init__(self, in_channels, reduction=2):
+        super(NonLocalBlock, self).__init__()
+
+        self.in_channels = in_channels
+        self.reduction = reduction
+
+        self.theta = nn.Conv1d(in_channels, in_channels // reduction, kernel_size=1)
+        self.phi = nn.Conv1d(in_channels, in_channels // reduction, kernel_size=1)
+        self.g = nn.Conv1d(in_channels, in_channels // reduction, kernel_size=1)
+        self.out_conv = nn.Conv1d(in_channels // reduction, in_channels, kernel_size=1)
+        self.softmax = nn.Softmax(dim=-1)
+
+    def forward(self, x):
+        # Reshape input tensor for convolution
+        batch_size, seq_length, hidden_dim = x.size()
+        x = x.view(batch_size, hidden_dim, seq_length)
+
+        # Compute query, key, and value
+        theta = self.theta(x)
+        phi = self.phi(x)
+        g = self.g(x)
+
+        # Compute similarity scores
+        theta = theta.permute(0, 2, 1)
+        similarity_scores = torch.matmul(theta, phi)
+        similarity_scores = similarity_scores / (self.in_channels // self.reduction) ** 0.5
+        attention_weights = self.softmax(similarity_scores)
+
+        # Compute the weighted sum of values
+        g = g.permute(0, 2, 1)
+        weighted_sum = torch.matmul(attention_weights, g)
+        weighted_sum = weighted_sum.permute(0, 2, 1)
+
+        # Apply 1x1 convolution and add skip connection
+        out = self.out_conv(weighted_sum)
+        out = out + x
+
+        # Reshape output tensor to original shape
+        out = out.view(batch_size, seq_length, hidden_dim)
+
+        return out
+
+'''加入了NonLocalBlock'''
+class Bert_CNN_CLS_BiGru2(nn.Module):
+    def __init__(self, input_size, bert_path,hidden_size, n_layers=1, dropout=0):
+        super(Bert_CNN_CLS_BiGru2, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.bert = BertModel.from_pretrained(bert_path)
+        for param in self.bert.parameters():
+            param.requires_grad = True
+        self.convs = nn.ModuleList(
+            [nn.Conv2d(in_channels=1, out_channels=256, kernel_size=(k, hidden_size)) for k in filter_sizes])
+        self.dropout = nn.Dropout(0.5)
+        self.non = NonLocalBlock(hidden_size)
+        self.gru=nn.GRU(input_size=hidden_size,hidden_size=hidden_size,num_layers=1,batch_first=True,bidirectional=True)
+        self.fc = nn.Linear(768+1536, num_labels * num_classes)
+
+    def conv_and_pool(self, x, conv):
+        x=conv(x)
+        #print("conv",x.size())
+        x = F.relu(x).squeeze(3)
+        #print("x2",x.size())
+        x = F.max_pool1d(x, x.size(2)).squeeze(2)
+        #x = F.adaptive_max_pool1d
+        #print("max_pool1d",x.size())
+        return x
+
+    def forward(self, input_variable,mask, hidden=None):
+        # samples=samples.T
+        # print(samples)
+        context = input_variable.T  # 输入的句子
+        #print(context)
+        mask = torch.tensor(mask).to("cuda")  # 对padding部分进行mask，和句子一个size，padding部分用0表示，如：[1, 1, 1, 1, 0, 0]
+        #print('mask',mask)
+        bert_out, pooled = self.bert(context, attention_mask=mask,output_all_encoded_layers=False )  #output_all_encoded_layers=False
+        #print(_out.size(),"bert_out")
+        #print("bert_out",bert_out.size())
+        #print("pooled_out",pooled.size())
+        bert_pool_out=pooled.unsqueeze(0)
+        non_out=self.non(bert_out)
+        #print("non_out",non_out.size())
+        out = non_out.unsqueeze(1)    #增加一个维度
+        cnn_out = torch.cat([self.conv_and_pool(out, conv) for conv in self.convs], 1)
+
+        cnn_out = self.dropout(cnn_out)    #不知道在哪里dropout比较好，有待研究   [8,768]
+        cnn_out=cnn_out.unsqueeze(0)      #给CNN的输出增加一个维度，方便拼接，增加后的形状为[1, 8, 768]
+        #print("cnn_out",cnn_out.size())
+        gru_out,gru_h=self.gru(bert_out)
+        #print("gru_h",gru_h.size())
+        split_tensors = torch.split(gru_h, split_size_or_sections=1, dim=0)
+        #print("split_tensors",split_tensors[0].size())
+        cat_output = torch.cat(split_tensors, dim=2)  #拼接正向输出和反向输出
+        #print('cat_output',cat_output.size())
+        #cat_cnn_gru=torch.cat([cat_output,cnn_out],dim=2)
+        cat_cnn_gru=torch.cat([cat_output,cnn_out],dim=2)
+        #print("cat_cnn_gru",cat_cnn_gru.size())
+        #out = torch.cat([bert_out,out],dim=1)   #在第二个维度做拼接    [8,501,768]
+        #print("cat",out.size())
+        #out,h_n = self.gru(out)    #[8,501,768]
+        #h_n=h_n.unsqueeze(0)
+        #print(h_n.size(),"gru_out")
+        out = self.fc(cat_cnn_gru)    #[768,12*3]
+        #print(out.size(),"fc_out")
+        out = out.view((-1, num_classes,num_labels))
+        #print(out)
+
+        # outputs = [batch size, num_classes, num_labels]
+        out = F.log_softmax(out, dim=1) #[0,0,1]  [1,0,0]   [1,0,01]
+        return out
+
+class Bert_gru_totalSentence(nn.Module):
+    def __init__(self, input_size, bert_path, hidden_size, n_layers=1, dropout=0):
+        super(Bert_gru_totalSentence, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.bert = BertModel.from_pretrained(bert_path)
+        for param in self.bert.parameters():
+            param.requires_grad = True
+        self.gru = nn.GRU(input_size=hidden_size, hidden_size=100, num_layers=1, batch_first=True,
+                          bidirectional=False)
+        self.gru2=nn.GRU(input_size=500*100, hidden_size=hidden_size, num_layers=1, batch_first=True,
+                          bidirectional=False)
+        self.fc = nn.Linear(hidden_size, num_labels * num_classes)
+
+    def forward(self, input_variable, mask, hidden=None):
+        # samples=samples.T
+        # print(samples)
+        context = input_variable.T  # 输入的句子
+        # print(context)
+        mask = torch.tensor(mask)  # 对padding部分进行mask，和句子一个size，padding部分用0表示，如：[1, 1, 1, 1, 0, 0]
+        # print('mask',mask)
+        _out, pooled = self.bert(context, attention_mask=mask,
+                                 output_all_encoded_layers=False)  # output_all_encoded_layers=False
+        #print(_out.size(), "bert_out")
+        out, h_n = self.gru(_out)
+        #print(h_n.size(), "hn_out", out.size(), "gru_out")
+        out = out.reshape(8, 50000)
+        out, h_n = self.gru2(out)
+        gru_out2=out.unsqueeze(0)
+        #print(h_n.size(), "hn_out2", out.size(), "gru_out2")
+        out = self.fc(gru_out2)
+        # print(out.size(),"fc_out")
+        out = out.view((-1, num_classes, num_labels))
+        # outputs = [batch size, num_classes, num_labels]
+        out = F.log_softmax(out, dim=1)
+        return out
+
+class Bert_my_vocab(nn.Module):
+    def __init__(self, input_size, bert_path,hidden_size, n_layers=1, dropout=0):
+        super(Bert_my_vocab, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.bert = BertModel.from_pretrained(bert_path)
+        for param in self.bert.parameters():
+            param.requires_grad = True
+        self.fc = nn.Linear(hidden_size, num_labels * num_classes)
+
+
+    def forward(self, input_variable,mask, hidden=None):
+        # samples=samples.T
+        # print(samples)
+        context = input_variable.T  # 输入的句子
+        #print(context)
+        mask = torch.tensor(mask).to("cuda")  # 对padding部分进行mask，和句子一个size，padding部分用0表示，如：[1, 1, 1, 1, 0, 0]
+        #print('mask',mask)
+        _, pooled = self.bert(context, attention_mask=mask,output_all_encoded_layers=False )  #output_all_encoded_layers=False
+        # print(pooled.size())
+        out = self.fc(pooled)
+        out = out.view((-1, num_classes,num_labels))
+        # outputs = [batch size, num_classes, num_labels]
+        out = F.log_softmax(out, dim=1)
+        return out
+
+'''Bert-CNN-BiGRU冻结bert参数'''
+class Bert_CNN_BiGru_parms(nn.Module):
+    def __init__(self, input_size, bert_path,hidden_size, n_layers=1, dropout=0):
+        super(Bert_CNN_BiGru_parms, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.bert = BertModel.from_pretrained(bert_path)
+        #self.att=nn.MultiheadAttention()
+        # 加载保存的参数字典
+        #saved_params = torch.load('models/checkpoint_8_80.956.tar',map_location=torch.device('cpu'))  # 假设保存的参数在'saved_model.pth'文件中
+        # 创建一个新的空字典，用于存储要加载到模型中的参数
+        #new_state_dict = {}
+        # 从保存的参数字典中筛选出与BERT模型的参数名称匹配的键值对
+        # for key in saved_params.keys():
+        #     if key.startswith('bert.'):
+        #         new_state_dict[key] = saved_params[key]
+        # self.bert.load_state_dict(new_state_dict)
+        for param in self.bert.parameters():
+            param.requires_grad = False
+        self.convs = nn.ModuleList(
+            [nn.Conv2d(in_channels=1, out_channels=256, kernel_size=(k, hidden_size)) for k in filter_sizes])
+        for param in self.convs.parameters():
+            param.requires_grad = True
+        self.dropout = nn.Dropout(0.5)
+        self.gru=nn.GRU(input_size=hidden_size,hidden_size=hidden_size,num_layers=1,batch_first=True,bidirectional=True)
+        self.fc = nn.Linear(2304, num_labels * num_classes)
+        for param in self.gru.parameters():
+            param.requires_grad = True
+        for param in self.fc.parameters():
+            param.requires_grad = True
+
+
+    def conv_and_pool(self, x, conv):
+        x=conv(x)
+        #print("conv",x.size())
+        x = F.relu(x).squeeze(3)
+        #print("x",x)
+        #print("x2",x.size())
+        x = F.max_pool1d(x, x.size(2)).squeeze(2)
+        #print("max_pool1d",x.size())
+        return x
+
+    def forward(self, input_variable,mask, hidden=None):
+        # samples=samples.T
+        # print(samples)
+        context = input_variable.T  # 输入的句子
+        #print(context)
+        mask = torch.tensor(mask)  # 对padding部分进行mask，和句子一个size，padding部分用0表示，如：[1, 1, 1, 1, 0, 0]
+        #print('mask',mask)
+        bert_out, pooled = self.bert(context, attention_mask=mask,output_all_encoded_layers=False )  #output_all_encoded_layers=False
+        #print(_out.size(),"bert_out")
+        #print("bert_out",bert_out.size())
+        out = bert_out.unsqueeze(1)    #增加一个维度
+        #print("bert",out.size())
+        cnn_out = torch.cat([self.conv_and_pool(out, conv) for conv in self.convs], 1)
+        #print("cnn_out",cnn_out.size())
+        cnn_out = self.dropout(cnn_out)    #不知道在哪里dropout比较好，有待研究   [8,768]
+        cnn_out=cnn_out.unsqueeze(0)      #给CNN的输出增加一个维度，方便拼接，增加后的形状为[1, 8, 768]
+        gru_out,gru_h=self.gru(bert_out)
+        #print("gru_h",gru_h.size(),"gru_out",gru_out.size())
+        split_tensors = torch.split(gru_h, split_size_or_sections=1, dim=0)
+        #print("split_tensors",split_tensors[0].size())
+        cat_output = torch.cat(split_tensors, dim=2)  #拼接正向输出和反向输出
+        #print('cat_gru_output',cat_output.size())
+        cat_cnn_gru=torch.cat([cat_output,cnn_out],dim=2)
+        #print("cat_cnn_gru",cat_cnn_gru.size())
+        #out = torch.cat([bert_out,out],dim=1)   #在第二个维度做拼接    [8,501,768]
+        #print("cat",out.size())
+        #out,h_n = self.gru(out)    #[8,501,768]
+        #h_n=h_n.unsqueeze(0)
+        #print(h_n.size(),"gru_out")
+        out = self.fc(cat_cnn_gru)    #[768,12*3]
+        #print("fc_out",out.size())
+        out = out.view((-1, num_classes,num_labels))
+        #print("view_out",out.size())
+
+        # outputs = [batch size, num_classes, num_labels]
+        out = F.log_softmax(out, dim=1) #[0,0,1]  [1,0,0]   [1,0,01]
+        #print("log_softmax_out",out.size())
+        return out

models.py

@@ -0,0 +1,873 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from torch.autograd import Variable
+
+
+from config import num_labels, num_classes, batch_first, pad_size, chunk_size, hidden_size,filter_sizes
+from pytorch_pretrained import BertModel, BertTokenizer
+
+
+class EncoderRNN(nn.Module):
+    def __init__(self, input_size, hidden_size, n_layers=1, dropout=0):
+        super(EncoderRNN, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.embedding = nn.Embedding(input_size, hidden_size)
+
+        # Initialize GRU; the input_size and hidden_size params are both set to 'hidden_size'
+        #   because our input size is a word embedding with number of features == hidden_size
+        self.gru = nn.GRU(hidden_size, hidden_size, n_layers,
+                          dropout=(0 if n_layers == 1 else dropout), bidirectional=True)
+        self.fc = nn.Linear(hidden_size, num_labels * num_classes)
+
+    def forward(self, input_seq, input_lengths, hidden=None):
+        # input_seq = [sent len, batch size]
+        # Convert word indexes to embeddings
+        embedded = self.embedding(input_seq)
+        # embedded = [sent len, batch size, hidden size]
+        # Pack padded batch of sequences for RNN module
+        packed = torch.nn.utils.rnn.pack_padded_sequence(embedded, input_lengths)
+        # Forward pass through GRU
+        outputs, hidden = self.gru(packed, hidden)
+        # Unpack padding
+        outputs, _ = torch.nn.utils.rnn.pad_packed_sequence(outputs)
+        # Sum bidirectional GRU outputs
+        outputs = outputs[:, :, :self.hidden_size] + outputs[:, :, self.hidden_size:]
+        # outputs = [sent len, batch size, hidden size]
+        # outputs = outputs[-1]
+
+        # Extract the outputs for the last timestep of each example
+        idx = (input_lengths - 1).view(-1, 1).expand(
+            len(input_lengths), outputs.size(2))
+        time_dimension = 1 if batch_first else 0
+        idx = idx.unsqueeze(time_dimension)
+        # Shape: (batch_size, rnn_hidden_dim)
+        outputs = outputs.gather(
+            time_dimension, Variable(idx)).squeeze(time_dimension)
+
+        # outputs = [batch size, hidden size]
+        outputs = self.fc(outputs)
+        # outputs = [batch size, num_labels * num_classes]
+        outputs = outputs.view((-1, num_classes, num_labels))
+        # outputs = [batch size, num_classes, num_labels]
+        outputs = F.log_softmax(outputs, dim=1)
+        # outputs = [batch size, num_classes, num_labels]
+
+        # Return output
+        return outputs
+
+
+class Bert(nn.Module):
+    def __init__(self, input_size, bert_path,hidden_size, n_layers=1, dropout=0):
+        super(Bert, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.bert = BertModel.from_pretrained(bert_path)
+        for param in self.bert.parameters():
+            param.requires_grad = True
+        self.fc = nn.Linear(hidden_size, num_labels * num_classes)
+
+
+    def forward(self, input_variable,mask, hidden=None):
+        # samples=samples.T
+        # print(samples)
+        context = input_variable.T  # 输入的句子
+        #print(context)
+        mask = torch.tensor(mask).to("cuda")  # 对padding部分进行mask，和句子一个size，padding部分用0表示，如：[1, 1, 1, 1, 0, 0]
+        #print('mask',mask)
+        _, pooled = self.bert(context, attention_mask=mask,output_all_encoded_layers=False )  #output_all_encoded_layers=False
+        # print(pooled.size())
+        out = self.fc(pooled)
+        out = out.view((-1, num_classes,num_labels))
+        # outputs = [batch size, num_classes, num_labels]
+        out = F.log_softmax(out, dim=1)
+        return out
+
+class Bert_sentence(nn.Module):
+    def __init__(self, input_size, bert_path,hidden_size, n_layers=1, dropout=0):
+        super(Bert_sentence, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.bert = BertModel.from_pretrained(bert_path)
+        for param in self.bert.parameters():
+            param.requires_grad = True
+        self.bigru=nn.GRU(input_size=hidden_size,hidden_size=hidden_size,num_layers=1,batch_first=True,bidirectional=False)
+        self.fc = nn.Linear(hidden_size*2, num_labels * num_classes)
+        self.att=nn.MultiheadAttention(embed_dim=768,num_heads=8,batch_first=True)
+
+
+    def forward(self, input_variable,mask,sentences,sentences_mask, hidden=None):
+        # samples=samples.T
+        # print(samples)
+        #print("sentences_mask",sentences_mask,len(sentences_mask))
+        context = input_variable.T  # 输入的句子
+        mask = torch.tensor(mask).to("cuda")  # 对padding部分进行mask，和句子一个size，padding部分用0表示，如：[1, 1, 1, 1, 0, 0]
+        #sentences_mask = torch.tensor(sentences_mask).to("cuda")  # 对padding部分进行mask，和句子一个size，padding部分用0表示，如：[1, 1, 1, 1, 0, 0]
+        _, pooled = self.bert(context, attention_mask=mask,output_all_encoded_layers=False)  #output_all_encoded_layers=False
+        sentence_feature=[]
+        for paragraph,sentence_mask in zip(sentences,sentences_mask):
+            s_pooled_list = []  # 创建一个空列表用于保存 s_pooled
+            for sentence,sen_mask in zip(paragraph,sentence_mask):
+                sentence=torch.tensor(sentence).to("cuda")
+                sen_mask=torch.tensor(sen_mask).to("cuda")
+                sentence=torch.unsqueeze(sentence,dim=0)
+                sen_mask=torch.unsqueeze(sen_mask,dim=0)
+                s_, s_pooled = self.bert(sentence,sen_mask,output_all_encoded_layers=False)
+                # 将 s_pooled 添加到列表中
+                s_pooled_list.append(s_pooled)
+                #print("s_pooled",s_pooled.size(),s_pooled,"s_",s_.size(),s_)
+            # 在循环结束后，将 s_pooled_list 中的张量连接为一个张量
+            s_pooled_tensor = torch.cat(s_pooled_list, dim=0)
+            att_out, attention_weights=self.att(s_pooled_tensor,s_pooled_tensor,s_pooled_tensor)
+            print("att_out",att_out.size(),"attention_weights",attention_weights.size(),attention_weights)
+            gru_out,h_n=self.bigru(att_out)
+            sentence_feature.append(h_n)
+            print("gru_out",gru_out.size(),"h_n", h_n.size())
+        sentence_feature_tensor = torch.cat(sentence_feature, dim=0)
+        #s_, s_pooled = self.bert(sentences,output_all_encoded_layers=False )  #output_all_encoded_layers=False
+        #print("sentence_out",s_pooled)
+        cat_out=torch.cat((sentence_feature_tensor,pooled),dim=1)
+        out = self.fc(cat_out)
+        out = out.view((-1, num_classes,num_labels))
+        # outputs = [batch size, num_classes, num_labels]
+        out = F.log_softmax(out, dim=1)
+        return out
+
+'''得到句子的CLS后不使用GRU，而是用最大池化'''
+class Bert_sentence3(nn.Module):
+    def __init__(self, input_size, bert_path,hidden_size, n_layers=1, dropout=0):
+        super(Bert_sentence3, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.bert = BertModel.from_pretrained(bert_path)
+        for param in self.bert.parameters():
+            param.requires_grad = True
+        self.bigru=nn.GRU(input_size=hidden_size,hidden_size=hidden_size,num_layers=1,batch_first=True,bidirectional=False)
+        self.fc = nn.Linear(hidden_size, num_labels * num_classes)
+        self.att=nn.MultiheadAttention(embed_dim=768,num_heads=12,batch_first=True)
+
+
+    def forward(self, input_variable,mask,sentences,sentences_mask, hidden=None):
+        # samples=samples.T
+        # print(samples)
+        #print("sentences_mask",sentences_mask,len(sentences_mask))
+        context = input_variable.T  # 输入的句子
+        mask = torch.tensor(mask)  # 对padding部分进行mask，和句子一个size，padding部分用0表示，如：[1, 1, 1, 1, 0, 0]
+        #sentences_mask = torch.tensor(sentences_mask).to("cuda")  # 对padding部分进行mask，和句子一个size，padding部分用0表示，如：[1, 1, 1, 1, 0, 0]
+        _, pooled = self.bert(context, attention_mask=mask,output_all_encoded_layers=False)  #output_all_encoded_layers=False
+        sentence_feature=[]
+        for paragraph,sentence_mask in zip(sentences,sentences_mask):
+            s_pooled_list = []  # 创建一个空列表用于保存 s_pooled
+            for sentence,sen_mask in zip(paragraph,sentence_mask):
+                sentence=torch.tensor(sentence)
+                sen_mask=torch.tensor(sen_mask)
+                sentence=torch.unsqueeze(sentence,dim=0)
+                sen_mask=torch.unsqueeze(sen_mask,dim=0)
+                s_, s_pooled = self.bert(sentence,sen_mask,output_all_encoded_layers=False)
+                # 将 s_pooled 添加到列表中
+                s_pooled_list.append(s_pooled)
+                #print("s_pooled",s_pooled.size(),s_pooled,"s_",s_.size(),s_)
+            # 在循环结束后，将 s_pooled_list 中的张量连接为一个张量
+            s_pooled_tensor = torch.cat(s_pooled_list, dim=0)
+            att_out, attention_weights=self.att(s_pooled_tensor,s_pooled_tensor,s_pooled_tensor)
+            # 最大池化操作
+            att_out_max, _ = torch.max(att_out, dim=0, keepdim=True)  # 在第0维度上获取最大值，并保持维度为[1, 768]
+            # print("att_out",att_out.size(),"attention_weights",attention_weights.size(),attention_weights)
+            # gru_out,h_n=self.bigru(att_out)
+            sentence_feature.append(att_out_max)
+        sentence_feature_tensor = torch.cat(sentence_feature, dim=0)
+        #print("sentence_feature_tensor",sentence_feature_tensor.size())
+        #s_, s_pooled = self.bert(sentences,output_all_encoded_layers=False )  #output_all_encoded_layers=False
+        #print("sentence_out",s_pooled)
+        #cat_out=torch.cat((sentence_feature_tensor,pooled),dim=1)
+        out = self.fc(sentence_feature_tensor)
+        #print(self.fc)
+        out = out.view((-1, num_classes,num_labels))
+        #print(out.size())
+        # outputs = [batch size, num_classes, num_labels]
+        out = F.log_softmax(out, dim=1)
+        return out
+
+class Bert_GRU(nn.Module):
+    def __init__(self, input_size, bert_path,hidden_size, n_layers=1, dropout=0):
+        super(Bert_GRU, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.bert = BertModel.from_pretrained(bert_path)
+        for param in self.bert.parameters():
+            param.requires_grad = True
+        self.gru=nn.GRU(hidden_size,hidden_size,1,batch_first=True,bidirectional=False)
+        self.fc = nn.Linear(hidden_size, num_labels * num_classes)
+
+
+    def forward(self, input_variable,mask, hidden=None):
+        # samples=samples.T
+        # print(samples)
+        context = input_variable.T  # 输入的句子
+        #print(context)
+        mask = torch.tensor(mask).to("cuda")  # 对padding部分进行mask，和句子一个size，padding部分用0表示，如：[1, 1, 1, 1, 0, 0]
+        #print('mask',mask)
+        _, pooled = self.bert(context, attention_mask=mask,output_all_encoded_layers=False )  #output_all_encoded_layers=False
+        print(pooled.size())
+        _,out = self.gru(pooled)
+        print(out.size())
+        out = self.fc(out)
+        print(out.size())
+        out = out.view((-1, num_classes,num_labels))
+        # outputs = [batch size, num_classes, num_labels]
+        out = F.log_softmax(out, dim=1)
+        return out
+
+'''Bert还未池化的输出后面接gru，相当于用bert编码，用gru提起特征'''
+class Bert_GRU2(nn.Module):
+    def __init__(self, input_size, bert_path,hidden_size, n_layers=1, dropout=0):
+        super(Bert_GRU2, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.bert = BertModel.from_pretrained(bert_path)
+        for param in self.bert.parameters():
+            param.requires_grad = True
+        # for param in self.parameters():
+        #     param.requires_grad = True
+        self.gru=nn.GRU(input_size=hidden_size,hidden_size=hidden_size,num_layers=1,batch_first=True,bidirectional=False)
+        self.fc = nn.Linear(hidden_size, num_labels * num_classes)
+
+
+    def forward(self, input_variable,mask, hidden=None):
+        # samples=samples.T
+        # print(samples)
+        context = input_variable.T  # 输入的句子
+        #print(context)
+        mask = torch.tensor(mask).to("cuda")  # 对padding部分进行mask，和句子一个size，padding部分用0表示，如：[1, 1, 1, 1, 0, 0]
+        #print('mask',mask)
+        _out, pooled = self.bert(context, attention_mask=mask,output_all_encoded_layers=False )  #output_all_encoded_layers=False
+        print(_out.size(),"bert_out")
+        out,h_n = self.gru(_out)
+        print(h_n.size(),"hn_out",out.size(),"gru_out")
+        out = self.fc(h_n)
+        #print(out.size(),"fc_out")
+        out = out.view((-1, num_classes,num_labels))
+        # outputs = [batch size, num_classes, num_labels]
+        out = F.log_softmax(out, dim=1)
+        return out
+
+
+'''Bert还未池化的输出后面接Bigru,自定义一个可学习的权重参数，用于计算BiGRU前向输出和后向输出在加权求和中的权重'''
+class Bert_BiGRU(nn.Module):
+    def __init__(self, input_size, bert_path,hidden_size, n_layers=1, dropout=0):
+        super(Bert_BiGRU, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.bert = BertModel.from_pretrained(bert_path)
+        for param in self.bert.parameters():
+            param.requires_grad = True
+        #创建一个大小为2的参数，且可以被学习
+        self.weights = nn.Parameter(torch.randn(2))
+        self.gru=nn.GRU(input_size=hidden_size,hidden_size=hidden_size,num_layers=1,batch_first=True,bidirectional=False)
+        self.Bigru=nn.GRU(input_size=hidden_size,hidden_size=hidden_size,num_layers=1,batch_first=True,bidirectional=True)
+        self.fc = nn.Linear(hidden_size, num_labels * num_classes)
+        self.global_avg_pooling = nn.AdaptiveAvgPool1d(768)
+
+    def forward(self, input_variable,mask, hidden=None):
+        # samples=samples.T
+        # print(samples)
+        context = input_variable.T  # 输入的句子
+        #print(context)
+        mask = torch.tensor(mask).to("cuda")  # 对padding部分进行mask，和句子一个size，padding部分用0表示，如：[1, 1, 1, 1, 0, 0]
+        #print('mask',mask)
+        _out, pooled = self.bert(context, attention_mask=mask,output_all_encoded_layers=False )  #output_all_encoded_layers=False
+        #print(_out.size(),"bert_out")
+        Biout,Bih_n = self.Bigru(_out)
+        # 加权求和前向输出和后向输出
+        weighted_sum = self.weights[0] * Bih_n[0] + self.weights[1] * Bih_n[1]
+        #weighted_output = torch.einsum('sbh,h->sbh', Bih_n, self.weights)
+        #print(Bih_n.size(),"hn_out",Biout.size(),"gru_out")
+        #out,h_n = self.gru(_out)
+        #print(h_n.size(),"hn_out",out.size(),"gru_out")
+        #out = self.fc(h_n)
+        Biout = self.fc(weighted_sum)
+        Biout=Biout.unsqueeze(0)
+        #print(Biout.size(),"Biout-fc")
+        #print(out.size(),"fc_out")
+        Biout = Biout.view((-1, num_classes,num_labels))
+        # outputs = [batch size, num_classes, num_labels]
+        out = F.log_softmax(Biout, dim=1)
+        return out
+
+
+'''Bert编码后用CNN提取特征，CNN层的输出与Bert层输出拼接，再输入到GRU'''
+class Bert_CNN_Gru(nn.Module):
+    def __init__(self, input_size, bert_path,hidden_size, n_layers=1, dropout=0):
+        super(Bert_CNN_Gru, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.bert = BertModel.from_pretrained(bert_path)
+        for param in self.bert.parameters():
+            param.requires_grad = True
+        self.convs = nn.ModuleList(
+            [nn.Conv2d(in_channels=1, out_channels=256, kernel_size=(k, hidden_size)) for k in filter_sizes])
+        self.dropout = nn.Dropout(0.5)
+        self.gru=nn.GRU(input_size=hidden_size,hidden_size=hidden_size,num_layers=1,batch_first=True,bidirectional=False)
+        self.fc = nn.Linear(hidden_size, num_labels * num_classes)
+
+    def conv_and_pool(self, x, conv):
+        x=conv(x)
+        #print("conv",x.size())
+        x = F.relu(x).squeeze(3)
+        #print("x2",x.size())
+        x = F.max_pool1d(x, x.size(2)).squeeze(2)
+        #print("max_pool1d",x.size())
+        return x
+
+    def forward(self, input_variable,mask, hidden=None):
+        # samples=samples.T
+        # print(samples)
+        context = input_variable.T  # 输入的句子
+        #print(context)
+        mask = torch.tensor(mask).to("cuda")  # 对padding部分进行mask，和句子一个size，padding部分用0表示，如：[1, 1, 1, 1, 0, 0]
+        #print('mask',mask)
+        bert_out, pooled = self.bert(context, attention_mask=mask,output_all_encoded_layers=False )  #output_all_encoded_layers=False
+        #print(_out.size(),"bert_out")
+        #print("bert_out",bert_out.size())
+        out = bert_out.unsqueeze(1)    #增加一个维度
+        out = torch.cat([self.conv_and_pool(out, conv) for conv in self.convs], 1)
+        out = self.dropout(out)    #不知道在哪里dropout比较好，有待研究   [8,768]
+        out=out.unsqueeze(1)     #[8,1,768]在第二维度上添加一个维度，以跟bert的输出做拼接cat
+        #print("dropout",out.size())
+        out = torch.cat([bert_out,out],dim=1)   #在第二个维度做拼接    [8,501,768]
+        #print("cat",out.size())
+        out,h_n = self.gru(out)    #[8,501,768]
+        #h_n=h_n.unsqueeze(0)
+        #print(h_n.size(),"gru_out")
+        out = self.fc(h_n)    #[768,12*3]
+        print(out.size(),"fc_out")
+        out = out.view((-1, num_classes,num_labels))
+        print(out)
+
+        # outputs = [batch size, num_classes, num_labels]
+        out = F.log_softmax(out, dim=1) #[0,0,1]  [1,0,0]   [1,0,01]
+        return out
+
+
+'''Bert-CNN-BiGRU'''
+class Bert_CNN_BiGru(nn.Module):
+    def __init__(self, input_size, bert_path,hidden_size, n_layers=1, dropout=0):
+        super(Bert_CNN_BiGru, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.bert = BertModel.from_pretrained(bert_path)
+        for param in self.bert.parameters():
+            param.requires_grad = True
+        self.convs = nn.ModuleList(
+            [nn.Conv2d(in_channels=1, out_channels=256, kernel_size=(k, hidden_size)) for k in filter_sizes])
+        self.dropout = nn.Dropout(0.5)
+        self.gru=nn.GRU(input_size=hidden_size,hidden_size=hidden_size,num_layers=1,batch_first=True,bidirectional=True)
+        self.fc = nn.Linear(2304, num_labels * num_classes)
+
+    def conv_and_pool(self, x, conv):
+        x=conv(x)
+        #print("conv",x.size())
+        x = F.relu(x).squeeze(3)
+        #print("x2",x.size())
+        x = F.max_pool1d(x, x.size(2)).squeeze(2)
+        #print("max_pool1d",x.size())
+        return x
+
+    def forward(self, input_variable,mask, hidden=None):
+        # samples=samples.T
+        # print(samples)
+        context = input_variable.T  # 输入的句子
+        #print(context)
+        mask = torch.tensor(mask).to("cuda")  # 对padding部分进行mask，和句子一个size，padding部分用0表示，如：[1, 1, 1, 1, 0, 0]
+        #print('mask',mask)
+        bert_out, pooled = self.bert(context, attention_mask=mask,output_all_encoded_layers=False )  #output_all_encoded_layers=False
+        #print(_out.size(),"bert_out")
+        print("bert_out",bert_out.size())
+        out = bert_out.unsqueeze(1)    #增加一个维度
+        cnn_out = torch.cat([self.conv_and_pool(out, conv) for conv in self.convs], 1)
+        print("cnn_out",cnn_out.size())
+        cnn_out = self.dropout(cnn_out)    #不知道在哪里dropout比较好，有待研究   [8,768]
+        cnn_out=cnn_out.unsqueeze(0)      #给CNN的输出增加一个维度，方便拼接，增加后的形状为[1, 8, 768]
+        gru_out,gru_h=self.gru(bert_out)
+        print("gru_h",gru_h.size(),"gru_out",gru_out.size())
+        split_tensors = torch.split(gru_h, split_size_or_sections=1, dim=0)
+        #print("split_tensors",split_tensors[0].size())
+        cat_output = torch.cat(split_tensors, dim=2)  #拼接正向输出和反向输出
+        print('cat_gru_output',cat_output.size())
+        cat_cnn_gru=torch.cat([cat_output,cnn_out],dim=2)
+        print("cat_cnn_gru",cat_cnn_gru.size())
+        #out = torch.cat([bert_out,out],dim=1)   #在第二个维度做拼接    [8,501,768]
+        #print("cat",out.size())
+        #out,h_n = self.gru(out)    #[8,501,768]
+        #h_n=h_n.unsqueeze(0)
+        #print(h_n.size(),"gru_out")
+        out = self.fc(cat_cnn_gru)    #[768,12*3]
+        print("fc_out",out.size())
+        out = out.view((-1, num_classes,num_labels))
+        print("view_out",out.size())
+
+        # outputs = [batch size, num_classes, num_labels]
+        out = F.log_softmax(out, dim=1) #[0,0,1]  [1,0,0]   [1,0,01]
+        print("log_softmax_out",out.size())
+        return out
+
+
+'''Bert-CNN-CLS-BiGRU'''
+class Bert_CNN_CLS_BiGru(nn.Module):
+    def __init__(self, input_size, bert_path,hidden_size, n_layers=1, dropout=0):
+        super(Bert_CNN_CLS_BiGru, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.bert = BertModel.from_pretrained(bert_path)
+        for param in self.bert.parameters():
+            param.requires_grad = True
+        self.convs = nn.ModuleList(
+            [nn.Conv2d(in_channels=1, out_channels=256, kernel_size=(k, hidden_size)) for k in filter_sizes])
+        self.dropout = nn.Dropout(0.5)
+        self.gru=nn.GRU(input_size=hidden_size,hidden_size=hidden_size,num_layers=1,batch_first=True,bidirectional=True)
+        self.fc = nn.Linear(3072, num_labels * num_classes)
+
+    def conv_and_pool(self, x, conv):
+        x=conv(x)
+        #print("conv",x.size())
+        x = F.relu(x).squeeze(3)
+        #print("x2",x.size())
+        x = F.max_pool1d(x, x.size(2)).squeeze(2)
+        #print("max_pool1d",x.size())
+        return x
+
+    def forward(self, input_variable,mask, hidden=None):
+        # samples=samples.T
+        # print(samples)
+        context = input_variable.T  # 输入的句子
+        #print(context)
+        mask = torch.tensor(mask).to("cuda")  # 对padding部分进行mask，和句子一个size，padding部分用0表示，如：[1, 1, 1, 1, 0, 0]
+        #print('mask',mask)
+        bert_out, pooled = self.bert(context, attention_mask=mask,output_all_encoded_layers=False )  #output_all_encoded_layers=False
+        #print(_out.size(),"bert_out")
+        #print("bert_out",bert_out.size())
+        #print("pooled_out",pooled.size())
+        bert_pool_out=pooled.unsqueeze(0)
+        out = bert_out.unsqueeze(1)    #增加一个维度
+        cnn_out = torch.cat([self.conv_and_pool(out, conv) for conv in self.convs], 1)
+        #print("cnn_out",cnn_out.size())
+        cnn_out = self.dropout(cnn_out)    #不知道在哪里dropout比较好，有待研究   [8,768]
+        cnn_out=cnn_out.unsqueeze(0)      #给CNN的输出增加一个维度，方便拼接，增加后的形状为[1, 8, 768]
+        gru_out,gru_h=self.gru(bert_out)
+        #print("gru_h",gru_h.size())
+        split_tensors = torch.split(gru_h, split_size_or_sections=1, dim=0)
+        #print("split_tensors",split_tensors[0].size())
+        cat_output = torch.cat(split_tensors, dim=2)  #拼接正向输出和反向输出
+        #print('cat_output',cat_output.size())
+        #cat_cnn_gru=torch.cat([cat_output,cnn_out],dim=2)
+        cat_cnn_gru=torch.cat([cat_output,cnn_out,bert_pool_out],dim=2)
+        #print("cat_cnn_gru",cat_cnn_gru.size())
+        #out = torch.cat([bert_out,out],dim=1)   #在第二个维度做拼接    [8,501,768]
+        #print("cat",out.size())
+        #out,h_n = self.gru(out)    #[8,501,768]
+        #h_n=h_n.unsqueeze(0)
+        #print(h_n.size(),"gru_out")
+        out = self.fc(cat_cnn_gru)    #[768,12*3]
+        #print(out.size(),"fc_out")
+        out = out.view((-1, num_classes,num_labels))
+        #print(out)
+
+        # outputs = [batch size, num_classes, num_labels]
+        out = F.log_softmax(out, dim=1) #[0,0,1]  [1,0,0]   [1,0,01]
+        return out
+
+
+'''Bert-CNN-CLS-BiGRU，Self-attention'''
+class Bert_CNN_CLS_BiGru1(nn.Module):
+    def __init__(self, input_size, bert_path,hidden_size, n_layers=1, dropout=0):
+        super(Bert_CNN_CLS_BiGru1, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.bert = BertModel.from_pretrained(bert_path)
+        for param in self.bert.parameters():
+            param.requires_grad = True
+        self.convs = nn.ModuleList(
+            [nn.Conv2d(in_channels=1, out_channels=256, kernel_size=(k, hidden_size)) for k in filter_sizes])
+        self.dropout = nn.Dropout(0.5)
+        self.gru=nn.GRU(input_size=hidden_size,hidden_size=hidden_size,num_layers=1,batch_first=True,bidirectional=True)
+        self.attention=nn.MultiheadAttention(3072,num_heads=6)   #接在双头GRU后面，让模型自动注意哪部分权重比较重要
+        self.fc = nn.Linear(3072, num_labels * num_classes)
+
+    def conv_and_pool(self, x, conv):
+        x=conv(x)
+        #print("conv",x.size())
+        x = F.relu(x).squeeze(3)
+        #print("x2",x.size())
+        x = F.max_pool1d(x, x.size(2)).squeeze(2)
+        #print("max_pool1d",x.size())
+        return x
+
+    def forward(self, input_variable,mask, hidden=None):
+        # samples=samples.T
+        # print(samples)
+        context = input_variable.T  # 输入的句子
+        #print(context)
+        mask = torch.tensor(mask).to("cuda")  # 对padding部分进行mask，和句子一个size，padding部分用0表示，如：[1, 1, 1, 1, 0, 0]
+        #print('mask',mask)
+        bert_out, pooled = self.bert(context, attention_mask=mask,output_all_encoded_layers=False )  #output_all_encoded_layers=False   bert_out=[8,500,768]
+        #print(_out.size(),"bert_out")
+        #print("bert_out",bert_out.size())
+        #print("pooled_out",pooled.size())
+        bert_pool_out=pooled.unsqueeze(0)
+        out = bert_out.unsqueeze(1)    #增加一个维度
+        cnn_out = torch.cat([self.conv_and_pool(out, conv) for conv in self.convs], 1)
+        #print("cnn_out",cnn_out.size())
+        cnn_out = self.dropout(cnn_out)    #不知道在哪里dropout比较好，有待研究   [8,768]
+        cnn_out=cnn_out.unsqueeze(0)      #给CNN的输出增加一个维度，方便拼接，增加后的形状为[1, 8, 768]
+        gru_out,gru_h=self.gru(bert_out)
+        #print("gru_h",gru_h.size())     #[2, 8, 768]
+        split_tensors = torch.split(gru_h, split_size_or_sections=1, dim=0)
+        #print("split_tensors",split_tensors[0].size())
+        cat_output = torch.cat(split_tensors, dim=2)  #拼接正向输出和反向输出
+        #print('cat_output',cat_output.size())
+        #cat_cnn_gru=torch.cat([cat_output,cnn_out],dim=2)
+        cat_cnn_gru=torch.cat([cat_output,cnn_out,bert_pool_out],dim=2)   #拼接BiGRU的输出和CNN的输出以及bert_cls的输出
+        #print("cat_cnn_gru",cat_cnn_gru.size())
+        att_out,_=self.attention(cat_cnn_gru,cat_cnn_gru,cat_cnn_gru)   #[1, 8, 3072]
+        #print("att_out",att_out.size())
+        out = self.fc(att_out)
+        #print(out.size(),"fc_out")
+        out = out.view((-1, num_classes,num_labels))
+        #print(out)
+
+        # outputs = [batch size, num_classes, num_labels]
+        out = F.log_softmax(out, dim=1) #[0,0,1]  [1,0,0]   [1,0,01]
+        return out
+
+'''attention,准确率比较低(bert-attention-gru)'''
+class Bert_GRU_attention(nn.Module):
+    def __init__(self, input_size, bert_path,hidden_size, n_layers=1, dropout=0):
+        super(Bert_GRU_attention, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.bert = BertModel.from_pretrained(bert_path)
+        for param in self.bert.parameters():
+            param.requires_grad = True
+        self.gru=nn.GRU(input_size=hidden_size,hidden_size=hidden_size,num_layers=1,batch_first=True,bidirectional=False)
+        self.attention=nn.MultiheadAttention(embed_dim=768,num_heads=8,batch_first=True)
+        self.fc = nn.Linear(hidden_size, num_labels * num_classes)
+
+
+    def forward(self, input_variable,mask, hidden=None):
+        # samples=samples.T
+        # print(samples)
+        context = input_variable.T  # 输入的句子
+        #print(context)
+        mask = torch.tensor(mask).to("cuda")  # 对padding部分进行mask，和句子一个size，padding部分用0表示，如：[1, 1, 1, 1, 0, 0]
+        #print('mask',mask)
+        _out, pooled = self.bert(context, attention_mask=mask,output_all_encoded_layers=False )  #output_all_encoded_layers=False
+        _out,attention_weights =self.attention(_out,_out,_out)
+        print(_out.size(),"bert_out")
+        out,h_n = self.gru(_out)
+        print(h_n.size(),"hn_out",out.size(),"gru_out")
+        out = self.fc(h_n)
+        #print(out.size(),"fc_out")
+        out = out.view((-1, num_classes,num_labels))
+        # outputs = [batch size, num_classes, num_labels]
+        out = F.log_softmax(out, dim=1)
+        return out
+
+
+'''attention,(bert-gru-attention)'''
+class Bert_GRU_attention(nn.Module):
+    def __init__(self, input_size, bert_path,hidden_size, n_layers=1, dropout=0):
+        super(Bert_GRU_attention, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.bert = BertModel.from_pretrained(bert_path)
+        for param in self.bert.parameters():
+            param.requires_grad = True
+        self.gru=nn.GRU(input_size=hidden_size,hidden_size=hidden_size,num_layers=1,batch_first=True,bidirectional=False)
+        self.attention=nn.MultiheadAttention(embed_dim=768,num_heads=8,batch_first=True)
+        self.fc = nn.Linear(hidden_size, num_labels * num_classes)
+
+
+    def forward(self, input_variable,mask, hidden=None):
+        # samples=samples.T
+        # print(samples)
+        context = input_variable.T  # 输入的句子
+        #print(context)
+        mask = torch.tensor(mask).to("cuda")  # 对padding部分进行mask，和句子一个size，padding部分用0表示，如：[1, 1, 1, 1, 0, 0]
+        #print('mask',mask)
+        _out, pooled = self.bert(context, attention_mask=mask,output_all_encoded_layers=False )  #output_all_encoded_layers=False
+        #_out,attention_weights =self.attention(_out,_out,_out)
+        print(_out.size(),"bert_out")
+        out,h_n = self.gru(_out)
+        print(h_n.size(),"hn_out",out.size(),"gru_out")
+        out = self.fc(h_n)
+        #print(out.size(),"fc_out")
+        out = out.view((-1, num_classes,num_labels))
+        # outputs = [batch size, num_classes, num_labels]
+        out = F.log_softmax(out, dim=1)
+        return out
+
+'''bert未池化用BiGru提取特征'''
+class Bert_BiGRU1(nn.Module):
+    def __init__(self, input_size, bert_path,hidden_size, n_layers=1, dropout=0):
+        super(Bert_BiGRU1, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.bert = BertModel.from_pretrained(bert_path)
+        for param in self.bert.parameters():
+            param.requires_grad = True
+        #self.gru=nn.GRU(input_size=pad_size * hidden_size,hidden_size=hidden_size,num_layers=1,batch_first=True,bidirectional=False)
+        self.gru=nn.GRU(input_size=hidden_size,hidden_size=hidden_size,num_layers=1,batch_first=True,bidirectional=False)
+        self.fc = nn.Linear(hidden_size, num_labels * num_classes)
+        #self.attention=nn.MultiheadAttention()
+
+    def forward(self, input_variable,mask, hidden=None):
+        # samples=samples.T
+        # print(samples)
+        context = input_variable.T  # 输入的句子
+        #print(context)
+        mask = torch.tensor(mask).to("cuda")  # 对padding部分进行mask，和句子一个size，padding部分用0表示，如：[1, 1, 1, 1, 0, 0]
+        #print('mask',mask)
+        _out, pooled = self.bert(context, attention_mask=mask,output_all_encoded_layers=False )  #output_all_encoded_layers=False
+        print(_out,"bert_out")    #(10,350,768)
+        #out = _out.reshape(chunk_size, pad_size * hidden_size)
+        out,h_n = self.gru(_out)
+        #print(out.size(),"gru_out",h_n.size(),"h_n")    #(10,350,768*2)
+        print(out,"gru_out",h_n,"h_n")    #(10,350,768*2)
+        out = self.fc(h_n)
+        #print(out.size(),"fc_out")
+        out = out.view((-1, num_classes,num_labels))
+        # outputs = [batch size, num_classes, num_labels]
+        out = F.log_softmax(out, dim=1)
+        return out
+
+
+'''Bert还未池化的输出后面接gru，相当于用bert编码，用gru提起特征,再与cls做残差连接'''
+class Bert_GRU3(nn.Module):
+    def __init__(self, input_size, bert_path,hidden_size, n_layers=1, dropout=0):
+        super(Bert_GRU3, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.bert = BertModel.from_pretrained(bert_path)
+        for param in self.bert.parameters():
+            param.requires_grad = True
+        self.gru=nn.GRU(input_size=hidden_size,hidden_size=hidden_size,num_layers=1,batch_first=True,bidirectional=False)
+        self.fc = nn.Linear(hidden_size, num_labels * num_classes)
+
+
+    def forward(self, input_variable,mask, hidden=None):
+        # samples=samples.T
+        # print(samples)
+        context = input_variable.T  # 输入的句子
+        #print(context)
+        mask = torch.tensor(mask).to("cuda")  # 对padding部分进行mask，和句子一个size，padding部分用0表示，如：[1, 1, 1, 1, 0, 0]
+        #print('mask',mask)
+        _out, pooled = self.bert(context, attention_mask=mask,output_all_encoded_layers=False )  #output_all_encoded_layers=False
+        #print(_out.size(),"bert_out")
+        out,h_n = self.gru(_out)
+        out=h_n+pooled
+        #print(out.size(),"gru_out")
+        out = self.fc(out)
+        #print(out.size(),"fc_out")
+        out = out.view((-1, num_classes,num_labels))
+        # outputs = [batch size, num_classes, num_labels]
+        out = F.log_softmax(out, dim=1)
+        #out = F.softmax(out, dim=1)
+        return out
+
+'''尝试将方面词编码进bert'''
+class Bert_GRU4(nn.Module):
+    def __init__(self, input_size, bert_path,hidden_size, n_layers=1, dropout=0):
+        super(Bert_GRU4, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.bert = BertModel.from_pretrained(bert_path)
+        for param in self.bert.parameters():
+            param.requires_grad = True
+        self.gru=nn.GRU(input_size=hidden_size,hidden_size=hidden_size,num_layers=1,batch_first=True,bidirectional=False)
+        self.fc = nn.Linear(hidden_size, num_labels * num_classes)
+
+
+    def forward(self, input_variable,mask, hidden=None):
+        # samples=samples.T
+        # print(samples)
+        context = input_variable.T  # 输入的句子
+        #print(context)
+        mask = torch.tensor(mask).to("cuda")  # 对padding部分进行mask，和句子一个size，padding部分��0表示，如：[1, 1, 1, 1, 0, 0]
+        #print('mask',mask)
+        _out, pooled = self.bert(context, attention_mask=mask,output_all_encoded_layers=False )  #output_all_encoded_layers=False
+        #print(_out.size(),"bert_out")
+        out,h_n = self.gru(_out)
+        out=h_n+pooled
+        #print(out.size(),"gru_out")
+        out = self.fc(out)
+        #print(out.size(),"fc_out")
+        out = out.view((-1, num_classes,num_labels))
+        # outputs = [batch size, num_classes, num_labels]
+        out = F.log_softmax(out, dim=1)
+        #out = F.softmax(out, dim=1)
+        return out
+
+'''先用bert编码，再用GRU对编码进行特征提起，最后将bert编码和gru输出做残差连接'''
+class Bert_GRU_Add(nn.Module):
+    def __init__(self, input_size, bert_path,hidden_size, n_layers=1, dropout=0):
+        super(Bert_GRU_Add, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.bert = BertModel.from_pretrained(bert_path)
+        for param in self.bert.parameters():
+            param.requires_grad = True
+        self.gru=nn.GRU(hidden_size,hidden_size,1,bidirectional=False)
+        self.fc = nn.Linear(hidden_size, num_labels * num_classes)
+
+
+    def forward(self, input_variable,mask, hidden=None):
+        # samples=samples.T
+        # print(samples)
+        context = input_variable.T  # 输入的句子
+        #print(context)
+        mask = torch.tensor(mask).to("cuda")  # 对padding部分进行mask，和句子一个size，padding部分用0表示，如：[1, 1, 1, 1, 0, 0]
+        #print('mask',mask)
+        _, pooled = self.bert(context, attention_mask=mask,output_all_encoded_layers=False )  #output_all_encoded_layers=False
+        #print(pooled.size())
+        out,_ = self.gru(pooled)
+        out=out+pooled
+        out = self.fc(out)
+        out = out.view((-1, num_classes,num_labels))
+        # outputs = [batch size, num_classes, num_labels]
+        out = F.log_softmax(out, dim=1)
+        return out
+
+class BiGRU(nn.Module):
+    def __init__(self, input_size, hidden_size, num_layers,dropout, bidirectional=True):
+        super(BiGRU, self).__init__()
+        self.hidden_size = hidden_size
+        self.num_layers = num_layers
+        self.bidirectional = bidirectional
+
+        self.forward_gru = nn.GRU(input_size, hidden_size, num_layers,dropout=dropout, bidirectional=bidirectional)
+        self.backward_gru = nn.GRU(input_size, hidden_size, num_layers,dropout=dropout, bidirectional=bidirectional)
+
+    def forward(self, input):
+        forward_output, _ = self.forward_gru(input)
+
+        # Reverse the input sequence
+        reversed_input = torch.flip(input, [0])
+        backward_output, _ = self.backward_gru(reversed_input)
+
+        if self.bidirectional:
+            # Concatenate forward and backward outputs along the last dimension
+            output = torch.cat([forward_output, torch.flip(backward_output, [0])], dim=-1)
+        else:
+            # Use only the forward output
+            output = forward_output
+
+        return output
+class MyEncoderRNN(nn.Module):
+    def __init__(self, input_size, hidden_size, n_layers=1, dropout=0):
+        super(MyEncoderRNN, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.embedding = nn.Embedding(input_size, hidden_size)
+        # Initialize GRU; the input_size and hidden_size params are both set to 'hidden_size'
+        #   because our input size is a word embedding with number of features == hidden_size
+        self.gru = nn.GRU(hidden_size, hidden_size, n_layers,
+                          dropout=(0 if n_layers == 1 else dropout), bidirectional=True)
+        self.lstm=nn.LSTM(hidden_size,hidden_size)
+        self.rnn=nn.RNN(hidden_size,hidden_size)
+        self.fc = nn.Linear(hidden_size, num_labels * num_classes)
+    def forward(self, input_seq, input_lengths, hidden=None):
+        # input_seq = [sent len, batch size]
+        # Convert word indexes to embeddings
+        embedded = self.embedding(input_seq)
+        # embedded = [sent len, batch size, hidden size]
+        # Pack padded batch of sequences for RNN module
+        packed = torch.nn.utils.rnn.pack_padded_sequence(embedded, input_lengths)
+        # Forward pass through GRU
+        outputs, hidden = self.gru(packed, hidden)
+
+        # Unpack padding
+        outputs, _ = torch.nn.utils.rnn.pad_packed_sequence(outputs)
+        # Sum bidirectional GRU outputs
+        outputs = outputs[:, :, :self.hidden_size] + outputs[:, :, self.hidden_size:]
+
+        outputs,hidden_lstm = self.lstm(outputs)
+        outputs,_=self.rnn(outputs)
+        #print("type", type(outputs))
+        # outputs = [sent len, batch size, hidden size]
+        # outputs = outputs[-1]
+
+        # Extract the outputs for the last timestep of each example
+        idx = (input_lengths - 1).view(-1, 1).expand(
+            len(input_lengths), outputs.size(2))
+        time_dimension = 1 if batch_first else 0
+        idx = idx.unsqueeze(time_dimension)
+        # Shape: (batch_size, rnn_hidden_dim)
+        outputs = outputs.gather(
+            time_dimension, Variable(idx)).squeeze(time_dimension)
+
+        # outputs = [batch size, hidden size]
+        outputs = self.fc(outputs)
+        # outputs = [batch size, num_labels * num_classes]
+        outputs = outputs.view((-1, num_classes, num_labels))
+        # outputs = [batch size, num_classes, num_labels]
+        outputs = F.log_softmax(outputs, dim=1)
+        # outputs = [batch size, num_classes, num_labels]
+
+        # Return output
+        return outputs
+
+
+class AttentionEncoderRNN(nn.Module):
+    def __init__(self, input_size, hidden_size, n_layers=1, dropout=0):
+        super(AttentionEncoderRNN, self).__init__()
+        self.n_layers = n_layers
+        self.hidden_size = hidden_size
+        self.embedding = nn.Embedding(input_size, hidden_size)
+        # Initialize GRU; the input_size and hidden_size params are both set to 'hidden_size'
+        #   because our input size is a word embedding with number of features == hidden_size
+        self.gru = nn.GRU(hidden_size, hidden_size, n_layers,
+                          dropout=(0 if n_layers == 1 else dropout), bidirectional=True)
+        self.att=nn.MultiheadAttention(hidden_size, 8, batch_first=True)
+        self.fc = nn.Linear(hidden_size, num_labels * num_classes)
+    def forward(self, input_seq, input_lengths, hidden=None):
+        # input_seq = [sent len, batch size]
+        # Convert word indexes to embeddings
+        embedded = self.embedding(input_seq)
+        #print("embedded",embedded.size(),embedded)
+
+        # embedded = [sent len, batch size, hidden size]
+        # Pack padded batch of sequences for RNN module
+        packed = torch.nn.utils.rnn.pack_padded_sequence(embedded, input_lengths)
+        outputs, _ = torch.nn.utils.rnn.pad_packed_sequence(packed)
+        att,_ = self.att(outputs, outputs, outputs)
+        # Forward pass through GRU
+        outputs, hidden = self.gru(att, hidden)
+        # Unpack padding
+        #outputs, _ = torch.nn.utils.rnn.pad_packed_sequence(outputs)
+        # Sum bidirectional GRU outputs
+        outputs = outputs[:, :, :self.hidden_size] + outputs[:, :, self.hidden_size:]
+        #print('outputs',outputs.size(),outputs)
+        #print("type", type(outputs))
+        # outputs = [sent len, batch size, hidden size]
+        # outputs = outputs[-1]
+
+        # Extract the outputs for the last timestep of each example
+        idx = (input_lengths - 1).view(-1, 1).expand(
+            len(input_lengths), outputs.size(2))
+        time_dimension = 1 if batch_first else 0
+        idx = idx.unsqueeze(time_dimension)
+        # Shape: (batch_size, rnn_hidden_dim)
+        outputs = outputs.gather(
+            time_dimension, Variable(idx)).squeeze(time_dimension)
+
+        # outputs = [batch size, hidden size]
+        outputs = self.fc(outputs)
+        # outputs = [batch size, num_labels * num_classes]
+        outputs = outputs.view((-1, num_classes, num_labels))
+        # outputs = [batch size, num_classes, num_labels]
+        outputs = F.log_softmax(outputs, dim=1)
+        # outputs = [batch size, num_classes, num_labels]
+
+        # Return output
+        return outputs