Probabilistic Sense Sentiment Similarity through Hidden Emotions

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 983–992, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics Probabilistic Sense Sentiment Similarity through Hidden Emotions Mitra Mohtarami1, Man Lan2, and Chew Lim Tan1 1Department of Computer Science, National University of Singapore; 2Department of Computer Science, East China Normal University {mitra,tancl}@comp.nus.edu.sg;mlan@cs.ecnu.edu.cn Abstract Sentiment Similarity of word pairs reflects the distance between the words regarding their underlying sentiments. This paper aims to in- fer the sentiment similarity between word pairs with respect to their senses. To achieve this aim, we propose a probabilistic emotion- based approach that is built on a hidden emo- tional model. The model aims to predict a vec- tor of basic human emotions for each sense of the words. The resultant emotional vectors are then employed to infer the sentiment similarity of word pairs. We apply the proposed ap- proach to address two main NLP tasks, name- ly, Indirect yes/no Question Answer Pairs in- ference and Sentiment Orientation prediction. Extensive experiments demonstrate the effec- tiveness of the proposed approach. 1 Introduction Sentiment similarity reflects the distance be- tween words based on their underlying senti- ments. Semantic similarity measures such as La- tent Semantic Analysis (LSA) (Landauer et al., 1998) can effectively capture the similarity be- tween semantically related words like "car" and "automobile", but they are less effective in relat- ing words with similar sentiment orientation like "excellent" and "superior". For example, the fol- lowing relations show the semantic similarity between some sentiment words computed by LSA: ��: ��� ����������, ��������� = 0.40 < ��� ����������, ����� = 0.46 < ��� �����, ���� = 0.65 Clearly, the sentiment similarity between the above words should be in the reversed order. In fact, the sentiment intensity in "excellent" is closer to "superior" than "good". Furthermore, sentiment similarity between "good" and "bad" should be 0. In this paper, we propose a probabilistic ap- proach to detect the sentiment similarity of words regarding their senses and underlying sen- timents. For this purpose, we propose to model the hidden emotions of word senses. We show that our approach effectively outperforms the semantic similarity measures in two NLP tasks: Indirect yes/no Question Answer Pairs (IQAPs) Inference and Sentiment Orientation (SO) pre- diction that are described as follows: In IQAPs, answers do not explicitly contain the yes or no keywords, but rather provide con- text information to infer the yes or no answer (e.g. Q: Was she the best one on that old show? A: She was simply funny). Clearly, the sentiment words in IQAPs are the pivots to infer the yes or no answers. We show that sentiment similarity between such words (e.g., here the adjectives best and Funny) can be used effectively to infer the answers. The second application (SO prediction) aims to determine the sentiment orientation of individual words. Previous research utilized the semantic relations between words obtained from WordNet (Hassan and Radev, 2010) and semantic similari- ty measures (e.g. Turney and Littman, 2003) for this …

Content truncated to 500 words