Extracting Features for Sentiment Classification: in the Perspective of Statistical Natural Language Processing

Abstract

In order to improve the accuracy of sentiment classification especially for Chinese online reviews, this paper proposed an approach to extract features for sentiment classification effectively and efficiently. The proposed method is based on supervised machine learning and follows the basic procedure of statistical natural language processing (SNLP). Through feature selection, feature extraction and feature weighting, the significant features for sentiment classification are obtained while the insignificant ones are removed. Comparative experiments have been made on mobile phone online reviews in Chinese, and improvement of three extraction algorithms (DF, IG and CHI) have also been made to obtain more accurate result. This research enriches the studies on sentiment classification theoretically, and is helpful for future work.