METHOD FOR INTELLECTUAL ANALYSIS OF TEXTUAL INFORMATION EMOTIONAL TONALITY FOR DETERMINE THE BEHAVIORAL INTENTIONS BY NEURAL NETWORKS MEANS

Abstract

In this paper, based on analysis results of current state of problem of intellectual analysis of the information texts emotional tonality, was determined that the use of neural networks for the analysis of texts emotional tonality is relevant, as it provides higher classification accuracy than alternative approaches. It was proposed to use the neural network of BERT architecture as one of the most accurate for the intellectual analysis of emotional tonality of texts, while it was proposed to use its modification RoBERTa for analysis of short documents.

The paper proposes the method for intellectual analysis of textual information emotional tonality for determine the behavioral intentions of users of socially oriented services and electronic commerce tools by neural networks means. The method uses neural network of the RoBERTa architecture. Conducted studies of the effectiveness of the method established that for combined multilingual texts it was possible to obtain an accuracy of 0.92, while the loss function had value of 0.29. This method should be used to determine the emotional tonality of short texts up to 500 words long, presented in Ukrainian language. At the same time, texts may contain surzhik and foreign words. The obtained results confirm the possibility and effectiveness of using method for intellectual analysis of textual information emotional tonality by neural networks means for determine the behavioral intentions of users, in particular, in socially oriented services and electronic commerce tools.