Selective Co-occurrences for Word-Emotion Association

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 1579–1590, Osaka, Japan, December 11-17 2016. Selective Co-occurrences for Word-Emotion Association Ameeta Agrawal and Aijun An Department of Electrical Engineering and Computer Science York University, Toronto, Canada {ameeta,aan}@cse.yorku.ca Abstract Emotion classification from text typically requires some degree of word-emotion association, either gathered from pre-existing emotion lexicons or calculated using some measure of semantic relatedness. Most emotion lexicons contain a fixed number of emotion categories and provide a rather limited coverage. Current measures of computing semantic relatedness, on the other hand, do not adapt well to the specific task of word-emotion association and therefore, yield average results. In this work, we propose an unsupervised method of learning word-emotion association from large text corpora, called Selective Co-occurrences (SECO), by leveraging the property of mutual exclusivity generally exhibited by emotions. Extensive evaluation, using just one seed word per emotion category, indicates the effectiveness of the proposed approach over three emotion lexicons and two state-of-the-art models of word embeddings on three datasets from different domains. 1 Introduction Emotion detection from text is the task of identifying emotions from natural language data such as user reviews, blogs, news articles, etc. (Alm et al., 2005; Aman and Szpakowicz, 2007). Although there is no strict definition for emotion, most researchers agree that it is a particular feeling that characterizes the state of mind such as happiness, anger, sadness and so on. Emotion analysis is extremely popular in the field of market research, where brands tap into their customer base by analyzing user-generated data, readily available nowadays due to the rapid growth of social media (De Bondt et al., 2013; Shrum et al., 2013). Mohammad and Turney (2013) present a more extensive list of applications. Emotion detection from text typically requires some degree of word-emotion association, i.e., knowing which words are more appropriately associated with which emotions. For example, the word “accident” can be considered associated with the emotion SADNESS. This association can be obtained from a pre- compiled emotion lexicon or calculated using some measure of semantic relatedness. The currently available manually annotated emotion lexicons tend to be restricted in size due to the expensive process of human annotation. This limited coverage leads to the undesirable effect of leaving too many words unassociated with any emotion category. Moreover, most lexicons contain a small fixed set of emotions which is unsuitable for a larger (Du et al., 2014) or a newly defined set of emotions (Facebook, 2016). On the other hand, while automatically computing word-emotion association scores from text corpora possibly provides a better coverage and more flexibility, the current techniques are ill-suited to the task of emotion detection and therefore, tend to yield average results. To address these issues, in this work, we propose an unsupervised method of learning word-emotion association scores from text corpora, which we call Selective Co-occcurences (SECO). By modifying conventional co-occurrence-based methods, we compute a uni-directional asymmetric association be- tween a given word and an emotion seed word. The …

Content truncated to 500 words