Finding Emotion Holder from Bengali Blog Texts—An Unsupervised Syntactic Approach

Finding Emotion Holder from Bengali Blog Texts –An Unsupervised Syntactic Approach* Dipankar Das and Sivaji Bandyopadhyay Department of Computer Science and Engineering, Jadavpur University, 188, Raja S.C. Mullick Road, Kolkata -700032, West Bengal dipankar.dipnil2005@gmail.com, sivaji_cse_ju@yahoo.com Abstract. This paper presents two different approaches for identifying emotion holders from Bengali blog sentences. Two types of strategies yield average agreement measures of 0.78 and 0.80 for annotating emotion holders with respect to all emotion classes. The baseline model is developed based on the combinations of various part-of-speech (POS) features extracted from the phrase-based similarities. The syntactic model is based on the argument structure of the sentences with respect to the verbs. If the acquired argument structure of a Bengali blog sentence with respect to its verb matches with any of the frame syntax retrieved for its equivalent English verb of identical sense from VerbNet, the holder role associated with the English VerbnNet frame is mapped to the appropriate slot in Bengali sentence. The syntactic model with an average F-score of 60.03% outperforms the baseline model with an average F-score of 50.85% on 500 test sentences. Keywords: Emotion Holder, Kappa, Argument Structure, Syntax, VerbNet. * The work reported in this paper was supported by a grant from the India-Japan Cooperative Programme (DST- JST) 2009 Research project entitled “Sentiment Analysis where AI meets Psychology” funded by Department of Science and Technology (DST), Government of India. Copyright 2010 by Dipankar Das and Sivaji Bandyopadhyay 1 Introduction Researches on emotion holder extraction are important for discriminating emotions that are viewed from different perspectives (Seki, 2007). Emotion holder inscribed in natural language texts plays an important role with respect to the reader or writer. A wide range of natural language processing (NLP) tasks such as tracking users’ emotion about reviews or events or politics as expressed in online forums or news, customer relationship management all are using the emotional information. Blogs on the other hand are the communicative and informative repository of text based emotional contents in the Web 2.0 (Lin et al., 2007). Sometimes, blog posts are annotated by other bloggers. The utilization of blog medium containing users’ emotional contents is therefore considered as an affective substrate to analyze the reaction of emotion catalyst like emotion holder. In the present task, identification of emotion holder is attempted for Bengali; a less privileged, less computerized and morphologically rich language. There is no existing emotion holder annotated corpus in Bengali. Manual annotation of emotion holder and the successive inter- annotator agreements have been carried out on a small set of 500 sentences of the Bengali blog corpus (Das and Bandyopadhyay, 2009). The corpus is tagged with Ekman’s (1993) six emotion types at sentence level. The phrase based similarity clues containing different part-of-speech (POS) combinations of the blog sentences are considered as the probable candidates of emotion holder for the baseline model. PACLIC 24 Proceedings 621 Developing syntactic model based on argument structure satisfies the demands of the baseline model as well. The pivotal hypothesis considered in the syntactic model …

Content truncated to 500 words