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。
{
"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": ""… See the full description on the dataset page:
https://huggingface.co/datasets/NEUDM/arts.
