Sentiment Analysis of Text to Speech with Emotion

Abstract

Abstract In sentiment analysis, the most accepted technique in statistical learning approaches is Bag-of-words (BOW). Though BOW is widely accepted technique, the elementary deficiencies in treating the polarity shift problem restrict the performance of BOW. In order to address this problem a model called Multiple sentiment analysis (MSA) for sentiment classification is proposed in this paper. For each training and test review a novel data expansion method by creating a sentiment reversed review is first introduced. To study a sentiment classifier, original as well as reversed training reviews are used in pairs. For this purpose, a multi-mood training algorithm is proposed. A multiple prediction algorithm is also proposed in order to classify test reviews with the consideration of two sides of one review.