Multipolarity-Based Sentiment Classification Using Hybrid HDL-Fuzzy-RMDL Network

Many companies and organizations extract valuable data from online reviews through the opinions and thoughts of people about a particular product. So, aspect-based sentiment analysis has become a popular method for identifying significant characteristics in review sentiment categorization. Occasionally, while evaluating document phrases or text, we encounter multipolarity concerns that impact the overall sentiment categorization. Hence, there is a need to handle this challenge of multipolarity that misleads due to conflicting or contradicting sentiments in the text. This research article implements the HDL-Fuzzy-RMDL model, which includes hierarchical deep learning for text classification, fuzzy sets, and random multimodel deep learning for multipolarity-based sentimental classification. The bidirectional encoder representations from transformers initially convert the input sentence tokens. The bidirectional long short-term memory performs the multiword aspect term extraction. The hybrid model utilizes significant keywords or aspects for multipolarity sentiment classification. The efficacy of the proposed approach is evaluated based on the accuracy concerning ATE, precision, recall, and F1-score. Furthermore, the HDL-Fuzzy-RMDL-based multipolarity sentiment classification model attained significant improvements in performance as compared with the existing baseline models.