Advancing sentiment analysis by addressing negation handling challenge via unsupervised mathematical approach

Sentiment analysis, a critical component of Natural Language Processing (NLP), focuses on the automatic detection and classification of emotions in text. With the vast increase in user-generated content on digital platforms, manual sentiment analysis has become impractical, necessitating advanced automated methods for efficiently processing and interpreting large-scale sentiment data. Traditional sentiment analysis techniques, which often rely on machine learning models and large pre-trained datasets, face challenges such as handling linguistic complexities like negation and adapting across different domains and languages. To address these limitations, we propose the “Evaluation based on Distance from Average Solution Optimization Technique-based Negation Handling Tagger” (EOT-NH), an innovative unsupervised sentiment classification model. The EOT-NH model incorporates multi-criteria decision-making (MCDM) principles and concepts from game theory to handle negation effectively and operate without extensive pre-training. This framework extracts key textual features—word count, context, and emotion score—while maintaining language and domain independence. Experimental evaluations demonstrate the model’s competitive performance, achieving a classification accuracy of 91% on the IMDB English reviews dataset The EOT-NH model presents a robust solution for sentiment analysis, providing significant advancements in negation handling and language flexibility, making it highly applicable to dynamic, multilingual environments.