Generating a Word-Emotion Lexicon from #Emotional Tweets

Proceedings of the Third Joint Conference on Lexical and Computational Semantics (*SEM 2014), pages 12–21, Dublin, Ireland, August 23-24 2014. Generating a Word-Emotion Lexicon from #Emotional Tweets Anil Bandhakavi1 Nirmalie Wiratunga1 Deepak P2 Stewart Massie1 1 IDEAS Research Institute, Robert Gordon University, Scotland, UK 2 IBM Research - India, Bangalore, India {a.s.bandhakavi,n.wiratunga}@rgu.ac.uk deepaksp@acm.org, s.massie@rgu.ac.uk Abstract Research in emotion analysis of text sug- gest that emotion lexicon based features are superior to corpus based n-gram fea- tures. However the static nature of the general purpose emotion lexicons make them less suited to social media analysis, where the need to adopt to changes in vo- cabulary usage and context is crucial. In this paper we propose a set of methods to extract a word-emotion lexicon automati- cally from an emotion labelled corpus of tweets. Our results confirm that the fea- tures derived from these lexicons outper- form the standard Bag-of-words features when applied to an emotion classification task. Furthermore, a comparative analysis with both manually crafted lexicons and a state-of-the-art lexicon generated using Point-Wise Mutual Information, show that the lexicons generated from the proposed methods lead to significantly better classi- fication performance. 1 Introduction Emotion mining or affect sensing is the compu- tational study of natural language expressions in order to quantify their associations with different emotions (e.g. anger, fear, joy, sadness and sur- prise). It has a number of applications for the in- dustry, commerce and government organisations, but uptake has arguably been slow. This in part is due to the challenges involved with modelling sub- jectivity and complexity of the emotive content. However, use of qualitative metrics to capture emotive strength and extraction of features from these metrics has in recent years shown promise (Shaikh, 2009). A general-purpose emotion lexi- con (GPEL) is a commonly used resource that al- lows qualitative assessment of a piece of emotive text. Given a word and an emotion, the lexicon provides a score to quantify the strength of emo- tion expressed by that word. Such lexicons are carefully crafted and are utilised by both super- vised and unsupervised algorithms to directly ag- gregate an overall emotion score or indirectly de- rive features for emotion classification tasks (Mo- hammad, 2012a), (Mohammad, 2012b). Socio-linguistics suggest that social media is a popular means for people to converse with individ- uals, groups and the world in general (Boyd et al., 2010). These conversations often involve usage of non-standard natural language expressions which consistently evolve. Twitter and Facebook were credited for providing momentum for the 2011 Arab Spring and Occupy Wall street movements (Ray, 2011),(Skinner, 2011). Therefore efforts to model social conversations would provide valu- able insights into how people influence each other through emotional expressions. Emotion analysis in such domains calls for automated discovery of lexicons. This is so since learnt lexicons can in- tuitively capture the evolving nature of vocabulary in such domains better than GPELs. In this work we show how an emotion la- belled corpus can be leveraged to generate a word- emotion lexicon automatically. …

Content truncated to 500 words