Sentiment Analysis Model for Online Forums to Classify Code-Mixed Text
Abstract
The rapid increase of online forums has led to a surge in multilingual and C-Mx text, where users blend two or more languages within a single post or comment. Traditional sentiment analysis models, predominantly trained on monolingual data, struggle to accurately interpret such hybrid content, resulting in decreased classification performance. The work proposes a novel sentiment analysis framework specifically designed to handle Code-Mixed (C-Mx) text from online forums. The proposed model integrates language-aware embeddings, contextual transformers, and data augmentation techniques to capture the semantic and syntactic nuances of mixed-language posts. By leveraging a combination of supervised learning and transfer learning strategies, the system effectively identifies positive, negative, and neutral sentiments across diverse language pairs. Experimental results on multilingual forum datasets demonstrate significant improvements in accuracy, F1-score, and robustness compared to existing baselines. The study highlights the importance of specialized approaches for C-Mx content and provides a scalable methodology for sentiment analysis in multilingual digital communities.