SEMGraph: Incorporating Sentiment Knowledge and Eye Movement into Graph Model for Sentiment Analysis

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 7521 - 7531 December 7-11, 2022 ©2022 Association for Computational Linguistics SEMGraph: Incorporating Sentiment Knowledge and Eye Movement into Graph Model for Sentiment Analysis Bingbing Wang1,2, Bin Liang1,2, Jiachen Du1,2, Min Yang3, and Ruifeng Xu1,2,4∗ 1 Harbin Insitute of Technology, Shenzhen, China 2 Guangdong Provincial Key Laboratory of Novel Security Intelligence Technologies 3 SIAT, Chinese Academy of Sciences, Shenzhen, China 4 Peng Cheng Laboratory, Shenzhen, China {bingbing.wang,bin.liang}@stu.hit.edu.cn, xuruifeng@hit.edu.cn, jacobvan199165@gmail.com, min.yang@siat.ac.cn Abstract This paper investigates the sentiment analysis task from a novel perspective by incorporat- ing sentiment knowledge and eye movement into a graph architecture, aiming to draw the eye movement-based sentiment relationships for learning the sentiment expression of the context. To be specific, we first explore a lin- guistic probing eye movement paradigm to ex- tract eye movement features based on the close relationship between linguistic features and the early and late processes of human reading be- havior. Furthermore, to derive eye movement features with sentiment concepts, we devise a novel weighting strategy to integrate sentiment scores extracted from affective commonsense knowledge into eye movement features, called sentiment-eye movement weights. Then, the sentiment-eye movement weights are exploited to build the sentiment-eye movement guided graph (SEMGraph) model, so as to model the intricate sentiment relationships in the context. Experimental results on two sentiment anal- ysis datasets with eye movement signals and three sentiment analysis datasets without eye movement signals show that the proposed SEM- Graph achieves state-of-the-art performance, and can also be directly generalized to those sentiment analysis datasets without eye move- ment signals. 1 Introduction As one of the hot research directions in Natural Language Processing (NLP), sentiment analysis (SA), aiming to extract knowledge and analyze the sentiment of texts, is crucial in a wide range of applications across multiple domains such as customer opinion identification (Kumar et al., 2019; Bose et al., 2019), product recommendation (Dang et al., 2021), and mental health monitoring based on patients’ social media posts (Rajput, 2019), etc. Some pioneering research efforts in cognitive science affirm the association between cognitive ∗ Corresponding Authors processing and emotional manifestations (Isen and Means, 1983; Storbeck and Clore, 2007). Intu- itively, humans are quite apt in extracting subjec- tive content and emphases from texts when reading a context. Based on the intuition that the cogni- tive data which covers eye movement signals or electroencephalography (EEG) signals might bring substantial improvements to sentiment analysis, re- cent research on sentiment analysis relying on eye movement signals has been attached importance (Mishra et al., 2017; Long et al., 2021; Chen et al., 2022; Joshi et al., 2014; Hollenstein et al., 2018). Despite promising progress made by existing eye movement-based sentiment analysis studies, they largely rely on the data containing eye move- ment signals. In the real-world scenario, however, the acquisition of eye movement signals is expen- sive, and many commonly used sentiment analysis datasets lack eye movement signals. In addition, existing eye movement-based sentiment analysis work ignores the role …

Content truncated to 500 words