Improving Affective Event Classification with Multi-Perspective Knowledge Injection

Improving Affective Event Classification with Multi-Perspective Knowledge Injection Wenjia Yi, Yanyan Zhao∗, Jianhua Yuan, Weixiang Zhao, Bing Qin Research Center for Social Computing and Information Retrieval Harbin Institute of Technology, China {wjyi, yyzhao, jhyuan, wxzhao, qinb}@ir.hit.edu.cn Abstract In recent years, many researchers have recognized the importance of associating events with sentiments. Previous approaches focus on generalizing events and extracting sentimental in- formation from a large-scale corpus. However, since context is absent and sentiment is often implicit in the event, these methods are limited in comprehending the semantics of the event and capturing effective sentimental clues. In this work, we propose a novel Multi-perspective Knowledge-injected Interaction Network (MKIN) to fully understand the event and accurately predict its sentiment by injecting multi-perspective knowledge. Specifically, we leverage con- texts to provide sufficient semantic information and perform context modeling to capture the semantic relationships between events and contexts. Moreover, we also introduce human emo- tional feedback and sentiment-related concepts to provide explicit sentimental clues from the perspective of human emotional state and word meaning, filling the reasoning gap in the senti- ment prediction process. Experimental results on the gold standard dataset show that our model achieves better performance over the baseline models. 1 Introduction Affective Event Classification (AEC) aims at predicting the sentiment polarity of the given event. We consider events that have positive effects on people who experience them as positive events. For instance, typically positive events include having a new crush, going to the bonfire, seeing a rainbow. On the con- trary, events that have negative effects on people who experience them are treated as negative events, such as breaking a marriage, going to the funeral, hearing a loud noise. Since events often trigger senti- ments and sentiments are often implicit, recognizing affective events is of great values to various natural language processing applications, covering dialogue systems (Shi and Yu, 2018), question-answering systems (Oh et al., 2012), implicit sentiment analysis (Zhou et al., 2021) and opinion mining (Xu et al., 2022b). The challenges of AEC lie in the limited context and the implicit sentiment of the event. To be specific, we often rely on the context to analyze sentiments, but there is no rich context to understand the event. Besides, traditional sentiment analysis methods rely on the occurrence of explicit sentiment words, but there are few explicit sentimental clues in the event. Many previous approaches have been devoted to cope with the challenges by extracting sentimental information from a large-scale corpus (Ding and Riloff, 2018; Saito et al., 2019; Zhuang et al., 2020). However, such attempts are not effective enough to understand the event and capture sentimental clues due to weak context modeling and insufficient sentimental information. We believe that this task would benefit from multi-perspective knowledge injection. Specifically, con- text can provide additional semantic information to understand the event, and human emotional feedback as well as sentiment-related concepts can provide explicit sentimental information from two perspectives to fill the reasoning gap. Figure 1 shows an example of an affective event, demonstrating …

Content truncated to 500 words