Sentiment Feature Selection Algorithm for Chinese Micro-blog

Abstract

Sentiment analysis is to extract the opinion of the user from of the text documents. Sentiment classification using machine learning methods face problem of handing huge number of unique terms in a feature vector for the classification. Therefore, a feature selection method is required to eliminate the irrelevant and noisy features from the feature vector for efficient working of ML algorithms. Rought set Theory based feature selection method is not good at Chinese text although it did well in English text. In this paper, improved feature selection methods are proposed which are based on rough set theory and adapt to the Chinese micro blog. We name them as IGAR (IG and Rough set) and CHIAR (CHI and Rough set). The performance of the improved feature selection methods are compared with Information Gain (IG) method which has been identified as one of the best feature selection method for sentiment classification. Experimentation of improved feature selection methods was performed on two datasets which are extracted from Sina microblog. Experimental results show that improved feature selection methods outperform other feature selection.