Sentiment classification in Persian: Introducing a mutual information-based method for feature selection

With the enormous growth of online reviews in Internet, sentiment analysis has received more and more attention in information retrieval and natural language processing community. Up to now there are very few researches conducted on sentiment analysis for Persian documents. This paper considers the problem of sentiment classification for online customer reviews in Persian language. One of the challenges of Persian language is using of a wide variety of declensional suffixes. Another common problem of Persian text is word spacing. In Persian in addition to white space as interwords space, an intra-word space called pseudo-space separates word's part. One more noticeable challenge in customer reviews in Persian language is that of utilizing many informal or colloquial words in text. In this paper we study these challenges by proposing a model for sentiment classification of Persian review documents. The proposed model is based on a lemmatization approach for Persian language and is employed Naive Bayes learning algorithm for classification. Additionally we present a new feature selection method based on the mutual information method to extract the best feature collection from the initial extracted features. Finally we evaluate the performance of the model on a manually gathered collection of cellphone reviews, where the results show the effectiveness of the proposed model.