Sentiment Classification by Combining Triplet Belief Functions

Sentiment analysis is an emerging technique that caters for semantic orientation and opinion mining. It is increasingly used to anal- yse online product reviews for identifying customers’ opinions and atti- tudes to products or services in order to improve business performance of companies. This paper presents an innovative approach to combining outputs of sentiment classifiers under the framework of belief functions. The approach is composed of the formulation of outputs of sentiment classifiers in the triplet structure and adoption of its formulas to combin- ing simple support functions derived from triplet functions by evidential combination rules. The empirical studies have been conducted on the performance of sentiment classification individually and in combination, the experimental results show that the best combined classifiers made by these combination rules outperform the best individual classifiers over the MP3 and Movie-Review datasets.