An Exploration of Features for Recognizing Word Emotion

Proceedings of the 23rd International Conference on Computational Linguistics (Coling 2010), pages 922–930, Beijing, August 2010 An Exploration of Features for Recognizing Word Emotion Changqin Quan Faculty of Engineering University of Tokushima quan-c@is.tokushima-u.ac.jp Fuji Ren Faculty of Engineering University of Tokushima ren@is.tokushima-u.ac.jp Abstract Emotion words have been well used as the most obvious choice as feature in the task of textual emotion recognition and auto- matic emotion lexicon construction. In this work, we explore features for rec- ognizing word emotion. Based on Ren- CECps (an annotated emotion corpus) and MaxEnt (Maximum entropy) model, sev- eral contextual features and their com- bination have been experimented. Then PLSA (probabilistic latent semantic anal- ysis) is used to get semantic feature by clustering words and sentences. The ex- perimental results demonstrate the effec- tiveness of using semantic feature for word emotion recognition. After that, “word emotion components” is proposed to describe the combined basic emotions in a word.�A significant performance improvement over contextual and seman- tic features was observed after adding word emotion components as feature. 1 Introduction Textual emotion analysis is becoming increas- ingly important due to augmented communication via computer mediated communication (CMC). A possible application of textual emotion recogni- tion is online chat system. An emotion feedback system can recognize users’ emotion and give ap- propriate responses. Another application exam- ple is weblog emotion recognition and prediction. Blogspace consists of millions of users who main- tain their online diaries, containing frequently- updated views and personal remarks about a range of issues. An emotion recognition and predic- tion system can understand the public’s reaction to some social issues and predict emotion changes. It would be helpful for solving some psychological problems or giving early warnings, such as suicide or terrorism. Textual emotion analysis also can improve the accuracy of other nonverbal modalities like speech or facial emotion recognition, and to im- prove human computer interaction systems. How- ever, automatic recognition of emotion meaning from texts presents a great challenge. One of the reasons is the manifoldness of expressed emotions in words. Emotion words have been well used as the most obvious choice as feature in the task of tex- tual emotion recognition and automatic emotion lexicon construction (Virginia and Pablo, 2006; Tokuhisa et al., 2008, etc.). And there are many lexical resources developed for these tasks, such as GI (Stone et al., 1966), WordNet-Affect (Strap- parava and Valitutti, 2004), NTU Sentiment Dic- tionary (Ku et al., 2006), Hownet (Dong and Dong, 2003), SentiWordnet (Esuli and Sebastiani, 2006). In these sentimental or affective lexicons, the words usually bear direct emotions or opin- ions, such as happy or sad, good or bad. Al- though they play a role in some applications, sev- eral problems of emotion expression in words have been ignored. Firstly, there are a lot of sentences can evoke emotions without direct emotion words. For ex- ample, (1) ���������������� ��(Spring is in children’s eyes, and in their hearts.) In sentence (1), we may feel joy, love or expect delivered by the writer. …

Content truncated to 500 words