- Baba Hallah 1*, Aliyu A. A. 1, Muhammad A. A 1, Mohammed I. 1, Abdulkadir S. 2, Abubakar M. A. 2
- 1 Department of Secure Computing, Kaduna State University., 2 Department of Informatics, Kaduna State University.
- DOI:10.5281/zenodo.16436985
Sentiment analysis is a key research area for opinion classification and prediction, especially on social media platforms like Twitter. The Naïve Bayes classifier is widely used due to its simplicity and efficiency, but its performance is often hindered by inadequate negation handling during preprocessing. This research proposes an enhanced negation handling technique that integrates part-of-speech (POS) tagging to improve the accuracy of Naïve Bayes classifiers in Twitter opinion mining. The methodology refines preprocessing through normalization, stop-word removal, spelling correction, targeted negation handling, lemmatization, and Laplace smoothing. Experimental results on Twitter datasets show that the enhanced model outperforms both standard Naïve Bayes and existing negation handling approaches, achieving higher precision, recall, F-score, and overall accuracy. The findings highlight the importance of sophisticated preprocessing for effective sentiment analysis and offer a robust approach for real-time opinion mining applications.
