Multilingual aspect clustering for sentiment analysis

摘要

In the last few years, there has been growing interest in aspect-based sentiment analysis, which deals with extracting, clustering, and rating the overall opinion about the features of the entity being evaluated. Techniques for aspect extraction can produce an undesirably large number of aspects — with many of those relating to the same product feature. Hence, aspect clustering becomes necessary. Current solutions for aspect clustering are monolingual, but in many practical situations, reviews for a given entity are available in several languages, calling for multilingual integration. In this article, we address the novel task of multilingual aspect clustering, which aims at grouping semantically related aspects extracted from reviews written in several languages. Our method is unsupervised and relies on the contextual information of the aspects, which is represented by word embeddings. This representation allied with a suitable similarity measure allows clustering related aspects. Our experiments on two datasets with five languages each showed that our unsupervised clustering technique achieves results that outperform monolingual baselines adapted to work with multilingual data. We also show the benefits of the multilingual approach compared to using languages in isolation.