Temporal-spatial three-way granular computing for dynamic text sentiment classification

Abstract

In dynamic and open environments, the traditional static sentiment analysis or opinion mining model is unsuitable for continuous computation and classification of human opinions, sentiments and emotions when training and testing data increase over time. Based on multi-granularity computing with pertinent data and parameters, this study conducted a temporal-spatial three-way multi-granularity learning framework for dynamic text sentiment classification to continually address dynamic data uncertainty. It dynamically updated the proposed model with the evolving text using a sequential three-way sentiment classification. Under a temporal-spatial multi-granularity structure, this model gradually tackled uncertain samples in the boundary region according to the monotonous variation of coarser-to-finer granularity. Subsequently, this study combined a novel dynamic sentiment classification model with balancing the performances and costs by considering four benchmark models: fastText, TextCNN, TextRNN, and TextRCNN. Finally, the comparative results of experiments on three public datasets are reported to verify the efficiency of the proposed models.