A hybrid approach for sentiment analysis using fuzzy logic and DistilBERT to address ambiguity and long range dependencies

Abstract

The opinion categorization process known as sentiment analysis is aimed at identifying the polarity of sentiment in user-generated information on social media, which might include visual, aural, or textual components such as emojis. The rapid expansion of social networks like Facebook, Instagram, and Twitter has increased demand for automated systems that can effectively gather public opinion. Organizations and individuals can use opinion mining as a vital tool to assess general sentiment, whether favorable, unfavorable, or neutral, about particular events, goods, or subjects. Although methods have progressed from lexicon-based approaches to machine learning and deep learning models, two enduring issues still plague current approaches: natural language naturally contains ambiguities that make interpretation more difficult, and traditional deep learning architectures frequently fall short in capturing long-range contextual dependencies in data. In order to overcome these drawbacks, this paper suggests a hybrid framework that combines the DistilBert transformer model for efficient long-range dependency detection with fuzzy logic for managing linguistic ambiguity and this framework is evaluated on both the IMDB and Yelp datasets. Compared to state-of-the-art techniques, the experimental findings show notable gains in accuracy (97.68%), F1 score (96.74%), recall (97.99%) and precision (97.05%), confirming the framework’s capacity to increase sentiment classification robustness.