Construction of Emotional Lexicon Using Potts Model

International Joint Conference on Natural Language Processing, pages 674–679, Nagoya, Japan, 14-18 October 2013. Construction of Emotional Lexicon Using Potts Model Braja Gopal Patra*, Hiroya Takamura+, Dipankar Das#, Manabu Okumura+ and Sivaji Bandyopadhyay* *Dept. of Computer Science & Engineering, Jadavpur University, Kolkata, India #Dept. of Computer Science & Engineering, NIT Meghalaya, India +Precision and Intelligence Laboratory, Tokyo Institute of Technology, Japan brajagopal.cse@gmail.com, takamura@pi.titech.ac.jp, dipankar.dipnil2005@gmail.com, oku@pi.titech.ac.jp, sivaji_cse_ju@yahoo.com Abstract Emotion is an instinctive state of mind aroused by some specific objects or situation. Ex- change of textual information is an important medium for communication and contains a rich set of emotional expressions. The compu- tational approaches to emotion analysis in tex- tual data require annotated lexicons with po- larity tags. In this paper we propose a novel method for constructing emotion lexicon an- notated with Ekman‟s six basic emotion clas- ses (anger, disgust, fear, happy, sad and sur- prise). We adopt the Potts model for the prob- ability modeling of the lexical network. The lexical network has been constructed by con- necting each pair of words in which one of the two words appears in the gloss of the other. Starting with a small number of emotional seed words, the emotional categories of other words have been determined. With manual checking of top 200 words from each class an average precision of 85.41% has been achieved. 1 Introduction Sentiment analysis and classification from elec- tronic text is a hard semantic disambiguation problem (Das and Bandyopadhyay, 2010). Many recent researches have been conducted in the fields of sentiment extraction (Kim et al., 2012; Taboada et al., 2011), opinion mining, summari- zation (Aman and Szpakowicz, 2007; Das et al., 2012; Yang et al., 2007; Quan and Ren, 2010) and each of which has a variety of potentially valuable applications. For example, we can effi- ciently collect people‟s opinion on a new rule enforced by the Government from Blog sites and at the same time be able to grasp their emotion without having to read their comments. An im- perative resource for such kind of emotional analysis is an emotion lexicon annotated with several emotional classes like happy, sad, fear, anger, surprise and disgust. In the previous ex- ample, frequent appearance of words from the happy class in a blog document would imply that the writer of the comment is quite happy with the new rule proposed by the Government. Several works have been conducted on build- ing emotional corpora in different languages such as in English (Aman and Szpakowicz, 2007), Chinese (Yang et al., 2007; Quan and Ren, 2010), and Bengali (Das and Bandyopadh- yay, 2010) etc. All these works have focused on developing sentiment lexicon with three senti- ment classes. For instance, Takamura et al. (2005) have developed a lexicon of emotion words tagged with the classes desirable and un- desirable using Spin model. A number of other polarity sentiment lexicons are available in Eng- lish such as SentiWordNet 3.0 (Esuli et al., 2010), Subjectivity Word List (Wilson et al., 2005), WordNet-Affect list …

Content truncated to 500 words