No abstract

Abstract.A standard approach for supervised sentiment analysis with n-grams features cannot correctly identify complex sentiment expressions due to the loss of information when representing a text using the bagof-words model. In our research, we propose to use subgraphs from the dependency tree of a parsed sentence as features for sentiment classification. We represent a text with a feature vector based on extracted subgraphs and use state of the art SVM classifier to identify the polarity of the given text. Our experimental evaluations on the movie-review dataset show that using our proposed features outperforms the standard bag-of-words and n-gram models. In this paper, we work with English, however most of our techniques can be easily adapted for other languages.