towe / README.md
xinxngxin's picture
Add researched SOTA result
698a9b1
metadata
language:
  - en

上述数据集为ABSA(Aspect-Based Sentiment Analysis)领域数据集,基本形式为从句子中抽取:方面术语、方面类别(术语类别)、术语在上下文中情感极性以及针对该术语的观点词,不同数据集抽取不同的信息,这点在jsonl文件的“instruction”键中有分别提到,在此我将其改造为了生成任务,需要模型按照一定格式生成抽取结果。

以acos数据集中抽取的jsonl文件一条数据举例:

{
    "task_type": "generation",
    "dataset": "acos", 
    "input": ["the computer has difficulty switching between tablet and computer ."], 
    "output": "[['computer', 'laptop usability', 'negative', 'difficulty']]",
    "situation": "none", 
    "label": "", 
    "extra": "", 
    "instruction": "    
        Task: Extracting aspect terms and their corresponding aspect categories, sentiment polarities, and opinion words. 
        Input: A sentence
        Output: A list of 4-tuples, where each tuple contains the extracted aspect term, its aspect category, sentiment polarity, and opinion words (if any). Supplement: \"Null\" means that there is no occurrence in the sentence.
        Example:  
            Sentence: \"Also it's not a true SSD drive in there but eMMC, which makes a difference.\"  
            Output: [['SSD drive', 'hard_disc operation_performance', 'negative', 'NULL']]' 
    "
}

此处未设置label和extra,在instruction中以如上所示的字符串模板,并给出一个例子进行one-shot,ABSA领域数据集(absa-quad,acos,arts,aste-data-v2,mams,semeval-2014,semeval-2015,semeval-2016,towe)每个数据集对应instruction模板相同,内容有细微不同,且部分数据集存在同一数据集不同数据instruction内容不同的情况。

原始数据集

当前SOTA

数据来自论文

  • 评价指标:F1-Score
  • 模型:IOG
    • laptop14:71.35
    • restuarant14:80.02
    • restuarant15:73.25
    • restuarant16:81.69
  • Paper:Paper:Target-oriented Opinion Words Extraction with Target-fused Neural Sequence Labeling
  • 说明:TOWE论文贡献为提出ABSA新的子任务(TOWE),并创建了新的数据集,但是据我在google scholar的初步调研发现虽然后较多工作引用该论文,但是均未使用该TOWE数据集,因此暂且将提出TOWE数据集的论文模型作为SOTA模型