BERT-RF Fusion Model Based on Bayesian optimization for Sentiment Classification

Abstract

Social media platforms have accumulated massive amounts of social text data. Mining people’s sentiment tendencies from these data is great significance. However, compared with ordinary text, social text data is shorter and more colloquial, and it is difficult to extract feature information, which affects the accuracy of sentiment classification. To improve the accuracy of sentiment classification, the BERT-RF fusion model based on Bayesian optimization for sentiment classification is proposed. Firstly, the key features in the social text are extracted through the deep structure of the BERT model, and the random forest model is used to replace the final output layer of the BERT, and the relevant social text is classified according to the key features. Hyperparameters of random forest are optimized using Bayesian method. The experimental results show that our model has better performance for sentiment classification.