Evaluating Lexical Similarity to build Sentiment Similarity

Evaluating Lexical Similarity to Build Sentiment Similarity Grégoire Jadi∗, Vincent Claveau†, Béatrice Daille∗, Laura Monceaux-Cachard∗ ∗ LINA - Univ. Nantes {gregoire.jadi beatrice.daille laura.monceaux}@univ-nantes.fr † IRISA–CNRS, France vincent.claveau@irisa.fr Abstract In this article, we propose to evaluate the lexical similarity information provided by word representations against several opinion resources using traditional Information Retrieval tools. Word representation have been used to build and to extend opinion resources such as lexicon, and ontology and their performance have been evaluated on sentiment analysis tasks. We question this method by measuring the correlation between the sentiment proximity provided by opinion resources and the semantic similarity provided by word representations using different correlation coefficients. We also compare the neighbors found in word representations and list of similar opinion words. Our results show that the proximity of words in state-of-the-art word representations is not very effective to build sentiment similarity. Keywords: opinion lexicon evaluation, sentiment analysis, spectral representation, word embedding 1. Introduction The goal of opinion mining or sentiment analysis is to iden- tify and classify texts according to the opinion, sentiment or emotion that they convey. It requires a good knowledge of the sentiment properties of each word. This properties can be either discovered automatically by machine learning al- gorithms, or embedded into language resources. On the one hand, there exist lexical resources (opinion lex- icons) in which words are associated with their valence (e.g. positive or negative opinion) or regrouped in senti- ment classes (e.g. { positive, negative, objective/neutral }, { fear, domination, sympathy, ... }). Many sentiment anal- ysis systems exploit such resources, but one major issue is to complete or adapt them to the domain/context. On the other hand, rich word representations such as word embeddings are also widely used in sentiment analysis. These word representations encode a semantic similarity meaning that two semantically close words are close in the representation space. They can be used by classification systems for sentiment analysis, or to build and enrich opin- ion lexicons. In any case, the underlying hypothesis is that the semantic similarity as encoded by the word representa- tion is correlated to the sentiment or emotional similarity. In this paper, our main goal is to question this hypothesis by evaluating the convergence between lexical similarity and opinion or sentiment-based similarity, that is: if two words are considered as similar from a semantic point-of-view, are they similar in terms of valence or sentiment values. 2. Related work Lexical similarity has been used in opinion mining systems in different ways. Some studies have proposed to use exist- ing resources to build or extend opinion lexicons. Among them, WordNet is certainly one of the most known seman- tic lexicon for English. Words are regrouped into synsets (sets of synonyms) which are linked by different seman- tic relations such as hypernymy, or meronymy. SentiWord- Net is an extension of WordNet that assigns to each synsets three sentiment scores : positivity, negativity and objectiv- ity. Those resources have been used extensively to build opinion analysis system. They can be used …

Content truncated to 500 words