情绪分析
计算机科学
笔记本电脑
任务(项目管理)
光学(聚焦)
自然语言处理
人工智能
情报检索
经济
操作系统
物理
管理
光学
作者
YuXuan Liu,Zhong Ping Jiang
标识
DOI:10.1007/978-3-031-46674-8_7
摘要
Aspect-Category-Opinion-Sentiment quadruple extraction (ACOS) is the novel and challenging sentiment analysis task, which aims to analyze the full range of emotional causes. Existing approaches focus on solving explicit sentiment, but struggle with analyzing implicit sentiment reviews. In this paper, to address the issue, we propose SI-TS, a framework that takes implicit sentiment extraction into account. Specifically, we design target structure (TS) to capture implicit sentiment by converting sentiment elements into a structured format. Furthermore, to adaptively generate appropriate TS according to different sentiment scenarios, we design an prompt template based sentiment instructor(SI). It assists the framework in effectively extracting implicit sentiment elements from the reviews. Extensive experiments were conducted on two widely used ACOS benchmarks, and improvements in F1 values were observed. Specifically, we achieved a 1.05% and 1.28% improvement in F1 values for Laptop-ACOS and Restaurant-ACOS, respectively. Notably, significant results were achieved in extracting implicit sentiment.
科研通智能强力驱动
Strongly Powered by AbleSci AI