Aff2Vec: Affect–Enriched Distributional Word Representations

Proceedings of the 27th International Conference on Computational Linguistics, pages 2204–2218 Santa Fe, New Mexico, USA, August 20-26, 2018. 2204 Aff2Vec: Affect–Enriched Distributional Word Representations Sopan Khosla, Niyati Chhaya, and Kushal Chawla Big Data Experience Lab Adobe Research Bangalore, India {skhosla,nchhaya,kchawla}@adobe.com Abstract Human communication includes information, opinions and reactions. Reactions are often cap- tured by the affective-messages in written as well as verbal communications. While there has been work in affect modeling and to some extent affective content generation, the area of af- fective word distributions is not well studied. Synsets and lexica capture semantic relationships across words. These models, however, lack in encoding affective or emotional word interpreta- tions. Our proposed model, Aff2Vec, provides a method for enriched word embeddings that are representative of affective interpretations of words. Aff2Vec outperforms the state-of-the-art in intrinsic word-similarity tasks. Further, the use of Aff2Vec representations outperforms baseline embeddings in downstream natural language understanding tasks including sentiment analysis, personality detection, and frustration prediction. 1 Introduction Affect refers to the experience of a feeling or an emotion (Scherer et al., 2010; Picard, 1997). This defini- tion includes emotions, sentiments, personality, and moods. The importance of affect analysis in human communication and interactions has been discussed by Picard (1997). Historically, affective computing has focused on studying human communication and reactions through multi-modal data gathered via various sensors. The study of human affect from text and other published content is an important topic in language understanding. Word correlation with social and psychological processes is discussed by Pen- nebaker (2011). Preotiuc-Pietro et al. (2017) studied personality and psycho-demographic preferences through Facebook and Twitter content. Sentiment analysis in Twitter, with a detailed discussion on hu- man affect (Rosenthal et al., 2017) and affect analysis in poetry (Kao and Jurafsky, 2012) have also been explored. Human communication not only contains semantic and syntactic information but also reflects the psychological and emotional states. Examples include the use of opinion and emotion words (Ghosh et al., 2017). The analysis of affect in interpersonal communication such as emails, chats, and longer articles is necessary for various applications including the study of consumer behavior and psychology, understanding audiences and opinions in computational social science, and more recently for dialogue systems and conversational agents. This is an open research space today. Traditional natural language understanding systems rely on statistical language modeling and seman- tic word distributions such as WORDNET (Miller, 1995) to understand relationships across different words. There has been a resurgence of research efforts towards creating word distributions that capture multi-dimensional word semantics (Mikolov et al., 2013a; Pennington et al., 2014). Sedoc et al. (2017b) introduce the notion of affect features in word distributions but their approach is limited to creating en- riched representations and no comments on the utility of the new word distribution is presented. Beyond word-semantics, deep learning research in natural language understanding is focused towards sentence representations using encoder-decoder models (Ahn et al., 2016), integration of symbolic knowledge to language models (Vinyals et al., 2015), and some recent works …

Content truncated to 500 words