[update]add ocnli

README.md

@@ -14,16 +14,18 @@ size_categories:
 | 数据 | 语言 | 原始数据/项目地址 | 样本个数 | 原始数据描述 | 替代数据下载地址 |
 | :--- | :---: |     :---:      |   :---: |    :---:    |     :---:     |
 | ChineseSTS | 汉语 | [ChineseSTS](https://github.com/IAdmireu/ChineseSTS) | 24.7K | STS 中文文本语义相似度（使用时注意打乱数据集） | [ChineseSTS](https://huggingface.co/datasets/tiansz/ChineseSTS) |
-| ccks2018_task3 | 汉语 | [CCKS2018_3](https://www.biendata.xyz/competition/CCKS2018_3/data/) | 100K | CCKS 2018 微众银行智能客服问句匹配大赛 |  |
+| ccks2018_task3 | 汉语 | [BQ_corpus](http://icrc.hitsz.edu.cn/info/1037/1162.htm); [CCKS2018_3](https://www.biendata.xyz/competition/CCKS2018_3/data/) | TRAIN: 100K, VALID: 10K, TEST: 10K | CCKS 2018 微众银行智能客服问句匹配大赛 | [BQ_corpus](https://github.com/IceFlameWorm/NLP_Datasets/tree/master/BQ_corpus) |
 | DIAC2019 | 汉语 | [DIAC2019](https://www.biendata.xyz/competition/2019diac/data/) | 6K | 以问题组的形式提供，每组问句又分为等价部分和不等价部分，等价问句之间互相组合可以生成正样本，等价问句和不等价问句之间互相组合可以生成负样本。我们提供6000组问句的训练集。 |  |
-| LCQMC | 汉语 | [LCQMC](https://www.luge.ai/#/luge/dataDetail?id=14); [C18-1166.pdf](https://aclanthology.org/C18-1166.pdf) | TRAIN: 238766, VALID: 8802, TEST: 12500 | 百度知道领域的中文问题匹配数据集，目的是为了解决在中文领域大规模问题匹配数据集的缺失。该数据集从百度知道不同领域的用户问题中抽取构建数据。| [lcqmc_data](https://github.com/xiaohai-AI/lcqmc_data) |
-| AFQMC | 汉语 | [AFQMC](https://tianchi.aliyun.com/dataset/106411) | TRAIN: 34334, VALID: 4316, TEST: 3861 | 蚂蚁金融语义相似度数据集，用于问题相似度计算。即：给定客服里用户描述的两句话，用算法来判断是否表示了相同的语义。 |  |
+| LCQMC | 汉语 | [LCQMC](http://icrc.hitsz.edu.cn/Article/show/171.html); [LCQMC](https://www.luge.ai/#/luge/dataDetail?id=14); [C18-1166.pdf](https://aclanthology.org/C18-1166.pdf) | TRAIN: 238766, VALID: 8802, TEST: 12500 | 百度知道领域的中文问题匹配数据集，目的是为了解决在中文领域大规模问题匹配数据集的缺失。该数据集从百度知道不同领域的用户问题中抽取构建数据。| [lcqmc_data](https://github.com/xiaohai-AI/lcqmc_data) |
+| AFQMC | 汉语 | [AFQMC](https://tianchi.aliyun.com/dataset/106411) | TRAIN: 34334, VALID: 4316, TEST: 3861 | 蚂蚁金融语义相似度数据集，用于问题相似度计算。即：给定客服里用户描述的两句话，用算法来判断是否表示了相同的语义。 | [ATEC](https://huggingface.co/datasets/shibing624/nli_zh); [ATEC](https://github.com/IceFlameWorm/NLP_Datasets/tree/master/ATEC) |
 | BUSTM | 汉语 | [BUSTM](https://tianchi.aliyun.com/competition/entrance/531851/information); [BUSTM](https://github.com/xiaobu-coai/BUSTM) | 总样本数为：177173，其中，匹配样本个数为：54805，不匹配样本个数为：122368 | 小布助手对话短文本语义匹配比赛数据集 | [BUSTM](https://github.com/CLUEbenchmark/FewCLUE/tree/main/datasets/bustm) |
 | CHIP2019 | 汉语 | [CHIP2019](https://www.biendata.xyz/competition/chip2019/) | 2万 | 平安医疗科技疾病问答迁移学习比赛数据集 |  |
 | COVID-19 | 汉语 | [COVID-19](https://tianchi.aliyun.com/competition/entrance/231776/information) |  | 天池新冠疫情相似句对判定大赛 | [COVID-19](https://gitee.com/liangzongchang/COVID-19-sentence-pair/) |
 | Chinese-MNLI | 汉语 | [Chinese-MNLI](https://github.com/pluto-junzeng/CNSD) | TRAIN: 390K, VALID: 12K, TEST: 13K | 通过翻译加部分人工修正的方法，从英文原数据集生成（原数据是：蕴含，中性，冲突，的句子推理数据集，已转换为句子对）。 |  |
 | Chinese-SNLI | 汉语 | [Chinese-SNLI](https://github.com/pluto-junzeng/CNSD) | TRAIN: 550K, VALID: 10K, TEST: 10K | 通过翻译加部分人工修正的方法，从英文原数据集生成（原数据是：蕴含，中性，冲突，的句子推理数据集，已转换为句子对）。 |  |
 | OCNLI | 汉语 | [OCNLI](https://github.com/CLUEbenchmark/OCNLI) | TRAIN: 50K, VALID: 3K, TEST: 3K | 原生中文自然语言推理数据集，是第一个非翻译的、使用原生汉语的大型中文自然语言推理数据集。 |  |
+| STS-B | 汉语 | [STS-B](https://adapterhub.ml/explore/sts/sts-b/); [STS Benchmark](https://ixa2.si.ehu.eus/stswiki/index.php/STSbenchmark) | TRAIN: 5749, VALID: 1500, TEST: 1379 | 语义文本相似性基准测试 | [STS-B](https://pan.baidu.com/s/10yfKfTtcmLQ70-jzHIln1A?pwd=gf8y#list/path=%2F); [STS-B](https://huggingface.co/datasets/shibing624/nli_zh/viewer/STS-B) |
+| PAWSX-ZH | 汉语 | [PAWSX](https://paperswithcode.com/paper/paws-x-a-cross-lingual-adversarial-dataset/review/) | TRAIN: 49.4K, VALID: 2K, TEST: 2K | 从 PAWSX翻译成中文的数据集 | [PAWSX](https://pan.baidu.com/share/init?surl=ox0tJY3ZNbevHDeAqDBOPQ&pwd=mgjn); [PAWSX](https://huggingface.co/datasets/shibing624/nli_zh/viewer/PAWSX) |
 
 
 <details>
@@ -33,5 +35,7 @@ https://github.com/liucongg/NLPDataSet
 
 https://huggingface.co/datasets/tiansz/ChineseSTS
 https://zhuanlan.zhihu.com/p/454173790
+
+https://huggingface.co/datasets/shibing624/nli_zh
 </code></pre>
 </details>

data/pawsx_zh.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b9cf16ca90456852dec844396a64dd120db5acf1fa7d4dc600ea00cebaac8379
+size 16737660

data/sts_b.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:38f45fcff68071a92e45a4893d829a2f85d6b2fbb8f898729ba6c379d25aad22
+size 1789990

examples/preprocess/process_pawsx_zh.py

@@ -0,0 +1,76 @@
+#!/usr/bin/python3
+# -*- coding: utf-8 -*-
+import argparse
+import json
+import os
+import sys
+
+pwd = os.path.abspath(os.path.dirname(__file__))
+sys.path.append(os.path.join(pwd, '../../'))
+
+from datasets import load_dataset
+from tqdm import tqdm
+
+from project_settings import project_path
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument("--dataset_path", default="shibing624/nli_zh", type=str)
+    parser.add_argument("--dataset_name", default="PAWSX", type=str)
+    parser.add_argument(
+        "--dataset_cache_dir",
+        default=(project_path / "hub_datasets").as_posix(),
+        type=str
+    )
+    parser.add_argument(
+        "--output_file",
+        default=(project_path / "data/pawsx_zh.jsonl"),
+        type=str
+    )
+
+    args = parser.parse_args()
+    return args
+
+
+def main():
+    args = get_args()
+
+    dataset_dict = load_dataset(
+        path=args.dataset_path,
+        name=args.dataset_name,
+        cache_dir=args.dataset_cache_dir,
+    )
+    print(dataset_dict)
+
+    with open(args.output_file, "w", encoding="utf-8") as f:
+        for split, dataset in dataset_dict.items():
+            for sample in dataset:
+                sentence1 = sample["sentence1"]
+                sentence2 = sample["sentence2"]
+                label = sample["label"]
+
+                label = str(int(label))
+
+                if label not in ("0", "1", None):
+                    print(label)
+                    raise AssertionError
+
+                row = {
+                    "sentence1": sentence1,
+                    "sentence2": sentence2,
+                    "label": label,
+                    "category": None,
+                    "data_source": "PAWSX-ZH",
+                    "split": split
+                }
+
+                row = json.dumps(row, ensure_ascii=False)
+                f.write("{}\n".format(row))
+
+    return
+
+
+if __name__ == '__main__':
+    main()

examples/preprocess/process_sts_b.py

@@ -0,0 +1,76 @@
+#!/usr/bin/python3
+# -*- coding: utf-8 -*-
+import argparse
+import json
+import os
+import sys
+
+pwd = os.path.abspath(os.path.dirname(__file__))
+sys.path.append(os.path.join(pwd, '../../'))
+
+from datasets import load_dataset
+from tqdm import tqdm
+
+from project_settings import project_path
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument("--dataset_path", default="shibing624/nli_zh", type=str)
+    parser.add_argument("--dataset_name", default="STS-B", type=str)
+    parser.add_argument(
+        "--dataset_cache_dir",
+        default=(project_path / "hub_datasets").as_posix(),
+        type=str
+    )
+    parser.add_argument(
+        "--output_file",
+        default=(project_path / "data/sts_b.jsonl"),
+        type=str
+    )
+
+    args = parser.parse_args()
+    return args
+
+
+def main():
+    args = get_args()
+
+    dataset_dict = load_dataset(
+        path=args.dataset_path,
+        name=args.dataset_name,
+        cache_dir=args.dataset_cache_dir,
+    )
+    print(dataset_dict)
+
+    with open(args.output_file, "w", encoding="utf-8") as f:
+        for split, dataset in dataset_dict.items():
+            for sample in dataset:
+                sentence1 = sample["sentence1"]
+                sentence2 = sample["sentence2"]
+                label = sample["label"]
+
+                label = "1" if label >= 3 else "0"
+
+                if label not in ("0", "1", None):
+                    print(label)
+                    raise AssertionError
+
+                row = {
+                    "sentence1": sentence1,
+                    "sentence2": sentence2,
+                    "label": label,
+                    "category": None,
+                    "data_source": "STS-B",
+                    "split": split
+                }
+
+                row = json.dumps(row, ensure_ascii=False)
+                f.write("{}\n".format(row))
+
+    return
+
+
+if __name__ == '__main__':
+    main()

sentence_pair.py

@@ -20,6 +20,8 @@ _URLS = {
     "diac2019": "data/diac2019.jsonl",
     "lcqmc": "data/lcqmc.jsonl",
     "ocnli": "data/ocnli.jsonl",
+    "pawsx_zh": "data/pawsx_zh.jsonl",
+    "sts_b": "data/sts_b.jsonl",
 
 }