Unsupervised subjectivity-lexicon generation based on vector space model for multi-dimensional opinion analysis in blogosphere

Abstract

This paper presents an unsupervised framework to generate a vector-space-modeled subjectivity-lexicon for multi-dimensional opinion mining and sentiment analysis, such as criticism analysis, for which the traditional polarity analysis alone is not adequate. The framework consists of four major steps: first, creating a dataset by crawling blog posts of fiction reviews; secondly, creating a "subjectivity-term to object" matrix, with each subjectivity-term being modeled as a dimension of a vector space; thirdly, feature-transforming each subjectivity-term into the new feature-space to create the final multi-dimensional subjectivity-lexicon (MDSL); and fourthly, using the generated MDSL for opinion analysis. In the experiments, it shows that the improvement by the feature transform can be up to 31% in terms of the entropy of features. In addition, the subjectivity-terms and objects are also successfully and reasonably clustered in the demonstration of fiction review (literary criticism) analysis. © 2010 Springer-Verlag Berlin Heidelberg.