A comprehensive sentiment analysis method based on Sentiment Multi-label and Probabilistic Hesitant Fuzzy Decision-making
Qiu Gongda, Zhang Guixin, Shi Hui, Deng Liqiong · Proceedings of the 2020 4th International Conference on Electronic Information Technology and Computer Engineering · 2020
Aiming at the difficulty of analysis on social text with complex sentiment and lacking of the comprehensive sentiment analysis based on historical social text set, a sentiment portrait analysis method based on multi-sentiment analysis and probabilistic hesitant fuzzy decision-making was proposed. Firstly, the BERT model was appropriately adapted to the hesitant fuzzy problem. Topic class and sentiment label of user text are acquired by topic classification and text sentiment multi-label model based on BERT model, Meanwhile the hesitant and fuzzy sentiment of different topics text are preserved by multi-level and multi-label sentiment. Secondly, the probabilistic hesitation fuzzy set is constructed based on the hesitation fuzzy and probability of text, in which the probabilistic hesitation fuzzy elements of each attribute are integrated by PHFWA operator. Finally, the modified sentiment entropy was proposed based on the statistical distribution of sentiment and information entropy, combined with its score function, deviation degree function and modified sentiment entropy, three types of sentiment distribution are defined and judged. In the experiment, four public Weibo accounts were selected and their status data were collected for analysis, which verified the effectiveness of the method in this paper.