EmoDialect: Leveraging Fuzzy Matching and Dialect-Emotion Mapping for Sentiment Analysis

Abstract

Sentiment Analysis is a well-explored field in natural language processing, that relies on intricate textual features. However, recent models tend to overlook the influence of dialects, emotions, and their associations, leading to inaccurate classifications. This work presents EmoDialect, a novel fuzzy framework designed to enhance sentiment analysis by mapping dialect with emotions and hence, their coalition coined as EmoDialect. The introduced EmoDialect incorporates dialect-emotion associations in feature extraction and utilizes fuzzy matching for dialect identification. Further, it leverages tweaked term frequency-inverse document frequency and parts-of-speech tagged $\mathcal {N}-$grams to capture dialect-specific sentiment cues. This enhanced EmoDialect feature set enhances sentiment analysis by attuning to the unique linguistic and emotional characteristics of diverse English dialects. Tests conducted on diverse corpora spanning various domains demonstrate the remarkable superiority and consistency of EmoDialect in terms of weighted average F1-scores of 92%, 86.7%, and 93% in dialect, sentiment, and text classification respectively, overtaking its predecessors by a wide margin. Also, EmoDialect was extended to dialect translation, and the related examinations revealed the F1-score of 86.15% warranting its ability to aid cross-cultural communication.