Fuzzy Sentiment Membership Determining for Sentiment Classification

Traditional support vector machine treats all samples using the same weight. Therefore it is very sensitive to noisy data. While the fuzzy support vector machine assigns lower weights to the samples which make small contributions to classification, thus it is beneficial to reduce the effects of noisy and unimportant data on the classification accuracy rate. In this paper, we propose a novel fuzzy sentiment membership determining method for solving sentiment classification task. We assume that strong intensity texts make more contributions to sentiment classification, while weak intensity texts are unimportant for the classification. In order to get the fuzzy sentiment membership of review texts, this paper proposes a three-layer sentiment propagation model. Firstly, we calculate the sentiment score of texts by the interrelations of the texts, topics and words, and ensure that the absolute value of sentiment score as the fuzzy sentiment membership degree of texts. Then, we train a fuzzy support vector machine to classify the samples from the test data sets. Finally, we conduct some experiments on four English reviews data sets from Amazon shopping websites. The experimental results show that the proposed method can improve the accuracy of sentiment classification effectively.