KC-ISA: An Implicit Sentiment Analysis Model Combining Knowledge Enhancement and Context Features

Proceedings of the 29th International Conference on Computational Linguistics, pages 6906–6915 October 12–17, 2022. 6906 KC-ISA: An Implicit Sentiment Analysis Model Combining Knowledge Enhancement and Context Features Minghao Xu, Daling Wang∗, Shi Feng, Zhenfei Yang, Yifei Zhang Northeastern University, Shenyang, China 2071825@stu.neu.edu.cn, {wangdaling, fengshi}@cse.neu.edu.cn 2001851@stu.neu.edu.cn, zhangyifei@cse.neu.edu.cn Abstract Sentiment analysis has always been an impor- tant research direction in natural language pro- cessing. The research can be divided into ex- plicit sentiment analysis and implicit sentiment analysis according to whether there are senti- ment words in language expression. There have been many research results in explicit sentiment analysis. However, implicit sentiment analysis is rarely studied. Compared with explicit senti- ment expression, implicit sentiment expression usually omits a lot of knowledge and common sense, and context also has an important impact on implicit sentiment expression. In this paper, we use a knowledge graph to supplement im- plicit sentiment expression and propose a novel Implicit Sentiment Analysis model combining Knowledge enhancement and Context features (dubbed KC-ISA). The KC-ISA model can effectively integrate external knowledge and contextual features by the coattention mecha- nism. Finally, we conduct experiments on the SMP2019 implicit sentiment analysis dataset. Moreover, to verify the generality of the model, we also conduct experiments on two common sentiment analysis datasets. The results on three datasets show that our proposed KC-ISA model can achieve better results on text senti- ment analysis. 1 Introduction With the rapid development of the Internet, the In- ternet has become the media form with the widest audience and the fastest reflection of social dynam- ics in today’s society. Online social media has be- come an excellent platform in which all users can easily participate, share and communicate. There- fore, a large number of texts that reflect the user’s subjective consciousness have begun to appear, and the texts with the user’s sentiment tendencies are also increasing. Text sentiment analysis is the process of analyz- ing, summarizing, and reasoning subjective senti- ∗ Corresponding author ment texts. From the expression level of text, it can be divided into explicit sentiment text and im- plicit sentiment text according to whether the text contains explicit sentimental words. Liu (2012) firstly divided sentiment into subjective opinion and factual implication opinion. The former is a subjective statement that gives the polarity of sen- timent, while the latter is an implicit expression of sentiment through an objective statement. Liao et al. (2019) defined implicit sentiment sentences as "language fragments that express subjective sen- timent but do not contain explicit sentiment words". As the basic research of sentiment analysis, the analysis of explicit sentiment text has abundant relevant research results. However, the emotions reflected by people’s behavior and feelings towards objective things are often rich and abstract and are usually expressed in subtle ways. For example, using objective statements to express one’s senti- ment, or using rhetorical expressions to describe them vividly, is called implicit sentiment expres- sion. That means the implicit sentiment expression depends on something other than the sentiment words. So implicit sentiment analysis first …

Content truncated to 500 words