Recognition of Affect, Judgment, and Appreciation in Text

Proceedings of the 23rd International Conference on Computational Linguistics (Coling 2010), pages 806–814, Beijing, August 2010 Recognition of Affect, Judgment, and Appreciation in Text Alena Neviarouskaya University of Tokyo lena@mi.ci.i.u- tokyo.ac.jp Helmut Prendinger Nat. Institute of Informatics Tokyo helmut@nii.ac.jp Mitsuru Ishizuka University of Tokyo ishizuka@i.u- tokyo.ac.jp Abstract The main task we address in our research is classification of text using fine-grained attitude labels. The developed @AM sys- tem relies on the compositionality prin- ciple and a novel approach based on the rules elaborated for semantically distinct verb classes. The evaluation of our me- thod on 1000 sentences, that describe personal experiences, showed promising results: average accuracy on the fine- grained level (14 labels) was 62%, on the middle level (7 labels) – 71%, and on the top level (3 labels) – 88%. 1 Introduction and Related Work With rapidly growing online sources aimed at encouraging and stimulating people’s discussions concerning personal, public or social issues (news, blogs, discussion forums, etc.), there is a great need in development of a computational tool for the analysis of people’s attitudes. Ac- cording to the Appraisal Theory (Martin and White, 2005), attitude types define the specifics of appraisal being expressed: affect (personal emotional state), judgment (social or ethical ap- praisal of other’s behaviour), and appreciation (evaluation of phenomena). To analyse contextual sentiment of a phrase or a sentence, rule-based approaches (Nasukawa and Yi, 2003; Moilanen and Pulman, 2007; Sub- rahmanian and Reforgiato, 2008), a machine- learning method using not only lexical but also syntactic features (Wilson et al., 2005), and a model of integration of machine learning ap- proach with compositional semantics (Choi and Cardie, 2008) were proposed. With the aim to recognize fine-grained emotions from text on the level of distinct sentences, researchers have em- ployed a keyword spotting technique (Chuang and Wu, 2004; Strapparava et al., 2007), a tech- nique calculating emotion scores using Pointwise Mutual Information (PMI) (Kozareva et al., 2007), an approach inspired by common-sense knowledge (Liu et al., 2003), rule-based linguis- tic approaches (Boucouvalas, 2003; Chaumartin, 2007), machine-learning methods (Alm, 2008; Aman and Szpakowicz, 2008; Strapparava and Mihalcea, 2008), and an ensemble based multi- label classification technique (Bhowmick et al., 2009). Early attempts to focus on distinct attitude types in the task of attitude analysis were made by Taboada and Grieve (2004), who determined a potential value of adjectives for affect, judge- ment and appreciation by calculating the PMI with the pronoun-copular pairs ‘I was (affect)’, ‘He was (judgement)’, and ‘It was (apprecia- tion)’, and Whitelaw et al. (2005), who used a machine learning technique (SVM) with fine- grained semantic distinctions in features (attitude type, orientation) in combination with “bag of words” to classify movie reviews. However, the concentration only on adjectives expressing ap- praisal and their modifiers greatly narrows the potential of the Whitelaw et al. (2005) approach. In this paper we introduce our system @AM (ATtitude Analysis Model), which (1) classifies sentences according to the fine-grained attitude labels (nine affect categories (Izard, 1971): ‘an- ger’, ‘disgust’, ‘fear’, …

Content truncated to 500 words