|
--- |
|
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 |
|
- 数据[链接](https://alt.qcri.org/semeval2016/task5/) |
|
- Paper：[SemEval-2016 Task 5: Aspect Based Sentiment Analysis](https://aclanthology.org/S16-1002/) |
|
- 说明：数据分为Laptop和restaurant两个主题的数据，分别在两个文件夹中放置。两个主题的数据抽取的元素不同。 |
|
|
|
#### 当前SOTA |
|
*数据来自[PaperWithCode](https://paperswithcode.com/sota)* |
|
|
|
- SemEval2016-Laptop |
|
未调研到相关评测工作 |
|
|
|
- [SemEval2016-Restaurant](https://paperswithcode.com/sota/aspect-based-sentiment-analysis-on-semeval-2) |
|
- 评价指标：Accuracy(抽取的分类准确率) |
|
- 模型：BERT-IL Finetuned (**88.70**) |
|
- Paper：[Does BERT Understand Sentiment? Leveraging Comparisons Between Contextual and Non-Contextual Embeddings to Improve Aspect-Based Sentiment Models](https://paperswithcode.com/paper/does-bert-understand-sentiment-leveraging) |
|
- 信息来源：[SemEval-2016](https://paperswithcode.com/sota/aspect-based-sentiment-analysis-on-semeval-2) |
