Text sentiment classification based on a genetic algorithm and word and document co-clustering

Resumo

A new text sentiment analysis method based on the computation of the weights of the sentiment words is proposed. This method allows us to automatically recognize a positive or negative sentiment expressed in the text with respect to some object. The problem of determining the weight of the sentiment words is considered as an optimization problem by the criterion of the maximization of the chosen quality metric of the sentiment analysis. In order to reduce the search space of the optimal weights of the sentiment words, co-clustering is used in the proposed method; as a result of co-clustering, groups of highly related sentiment words and text documents are obtained. The weights are optimized based on the genetic algorithm independently for each cluster. The experiments on the text collections of the Russian Information Retrieval Evaluation seminar (ROMIP) confirm the effectiveness of the proposed method. The computer support for different research studies, including the analysis of opinions—sociology, political science, and marketing—is a practical application of the method.