Evaluating Context Selection Strategies to Build Emotive Vector Space Models

Evaluating Context Selection Strategies to Build Emotive Vector Space Models Lucia C. Passaro, Alessandro Lenci CoLing Lab, Dipartimento di Filologia, Letteratura e Linguistica, University of Pisa (Italy) lucia.passaro@for.unipi.it, alessandro.lenci@unipi.it Abstract In this paper we compare different context selection approaches to improve the creation of Emotive Vector Space Models (VSMs). The system is based on the results of an existing approach that showed the possibility to create and update VSMs by exploiting crowdsourcing and human annotation. Here, we introduce a method to manipulate the contexts of the VSMs under the assumption that the emotive connotation of a target word is a function of both its syntagmatic and paradigmatic association with the various emotions. To study the differences among the proposed spaces and to confirm the reliability of the system, we report on two experiments: in the first one we validated the best candidates extracted from each model, and in the second one we compared the models’ performance on a random sample of target words. Both experiments have been implemented as crowdsourcing tasks. Keywords: Emotive lexicon, Context selection, Vector Space Models 1. Introduction With the proliferating use of social media, textual emotion analysis is becoming increasingly important. Emotion de- tection can be useful in several applications: for instance, in Customer Relationship Management (CRM) it can be used to track sentiments towards companies and their services, products or others target entities. In Government Intelli- gence, it can be used to collect people’s emotions and points of views about government decisions. Emotion lexica, in which lemmas are associated to the emo- tions they evoke, are knowledge sources that can help the development of detection algorithms and prediction sys- tems. In recent years great attention has been given to sen- timent polarity recognition, but a new trend is leading to the development of novel methods to automatically clas- sify the emotions expressed in an opinionated piece of text (Turney and Littman, 2003; Pang and Lee, 2008; Liu, 2012; Liu, 2015) as well as to the building of annotated lexical resources like SentiWordNet (Esuli and Sebastiani, 2006; Das and Bandyopadhyay, 2010), WordNet Affect (Strappa- rava and Valitutti, 2004) or EmoLex (Mohammad and Tur- ney, 2013). One major bottleneck of research on emotion detection is the lack of emotive resources. This problem is even more pressing for Italian. In Passaro et al. (2015) we proposed a language-independent distributional semantic method to semi-automatically build an emotive lexicon starting from a small number of seed terms and by exploiting crowd- sourcing methods. The output of this methodology is ItEM (Italian EMotive lexicon), a high-coverage emotion lexi- con for Italian. As a follow-up study, in this paper we compare different context selection approaches to improve emotive VSMs by exploiting the syntagmatic and paradig- matic properties of the target words. Distributional semantics is grounded on Harris’s distribu- tional hypothesis (Harris, 1954), which states that semanti- cally similar words tend to appear in similar contexts. From a computational point of view, each word is represented by a weighted feature vector, where features …

Content truncated to 500 words