A Comparison of Domain-based Word Polarity Estimation using different Word Embeddings

A Comparison of Domain-based Word Polarity Estimation using different Word Embeddings Aitor Garc´ıa-Pablos, Montse Cuadros, German Rigau Vicomtech-IK4, Vicomtech-IK4, IXA research group agarciap@vicomtech.org, mcuadros@vicomtech.org, german.rigau@ehu.eus Abstract A key point in Sentiment Analysis is to determine the polarity of the sentiment implied by a certain word or expression. In basic Sentiment Analysis systems this sentiment polarity of the words is accounted and weighted in different ways to provide a degree of positivity/negativity. Currently words are also modelled as continuous dense vectors, known as word embeddings, which seem to encode interesting semantic knowledge. With regard to Sentiment Analysis, word embeddings are used as features to more complex supervised classification systems to obtain sentiment classifiers. In this paper we compare a set of existing sentiment lexicons and sentiment lexicon generation techniques. We also show a simple but effective technique to calculate a word polarity value for each word in a domain using existing continuous word embeddings generation methods. Further, we also show that word embeddings calculated on in-domain corpus capture the polarity better than the ones calculated on general-domain corpus. Keywords: sentiment lexicon, sentiment analysis, word embedding 1. Introduction A key point in Sentiment Analysis is to determine the po- larity of the sentiment implied by a certain word or ex- pression (Taboada et al., 2011). In basic Sentiment Anal- ysis systems this sentiment polarity of the words is ac- counted and weighted in different ways to provide a degree of positivity/negativity of, for example, a customer review. In more sophisticated systems, word polarity is employed as an additional feature for machine learning algorithms. This polarity value can be a categorical value (e.g. pos- itive/neutral/negative) or a real value within a range (e.g. from -1.0 to +1.0), and can be plugged in supervised classi- fication algorithms together with other lexical and seman- tic features to help discriminating the overall polarity of an expression or a sentence. Currently words are also mod- elled as continuous dense vectors, known as word embed- dings, which seem to encode interesting semantic knowl- edge. The word vectors are usually computed using very big corpora of texts, like the English Wikipedia. One of the best known systems to obtain a dense continuous rep- resentation of words is Word2Vec (Mikolov et al., 2013c). But Word2Vec is not the only one, and in fact there are already a lot of variants and many researchers working on different kinds of word embeddings (Le and Mikolov, 2014; Iacobacci et al., 2015; Ji et al., 2015; Hill et al., 2014; Schwartz et al., 2014). With regard to Sentiment Analy- sis, word embeddings are used as features to more complex supervised classification systems to obtain very precise sen- timent classifiers (Tang et al., 2014a; Socher et al., 2013). In this paper we compare a set of existing static sentiment lexicons and dynamic sentiment lexicon generation tech- niques. We also show a simple but competitive technique to calculate a word polarity value for each word in a do- main using continuous word embeddings. Our objective is …

Content truncated to 500 words