Semantic Spaces for Sentiment Analysis

This article presents a new semi-supervised method for documentlevel sentiment analysis. We employ a supervised state-of-the-art classification approach and enrich the feature set by adding word cluster features. These features exploit clusters of words represented in semantic spaces computed on unlabeled data. We test our method on three large sentiment datasets (Czech movie and product reviews, and English movie reviews) and outperform the current state of the art. To the best of our knowledge, this article reports the first successful incorporation of semantic spaces based on local word co-occurrence in the sentiment analysis task.