@article {10.3844/jcssp.2026.1552.1568,
article_type = {journal},
title = {Optimizing Bias Detection in Tweets Using Bayesian Probabilistic Model},
author = {Rao, Prasanth G and Chigurupati, Harsha and Hashia, Krish and J, Thriveni and Shenoy, P Deepa and R, Venugopal K},
volume = {22},
number = {5},
year = {2026},
month = {May},
pages = {1552-1568},
doi = {10.3844/jcssp.2026.1552.1568},
url = {https://thescipub.com/abstract/jcssp.2026.1552.1568},
abstract = {Content moderation on social media faces persistent challenges from inconsistent evaluation shaped by subjective judgment and subtle semantic variations. This work proposes a Bayesian probabilistic framework for detecting bias in tweets using WordNet-based vocabulary filtering, statistical normalization via z-scores, and threshold optimization. The system is stateless, scalable, and dataset-agnostic, requiring no session-specific information. Unlike complex models such as Support Vector Machines (SVM), Multi-Layer Perceptrons (MLP), and AdaBoost, which tend to exhibit skewed classification patterns, the proposed approach achieves balanced confusion matrices and competitive F1 scores. Experimental evaluation across three benchmark datasets covering hate speech, political partisanship, and racial and gender-based discrimination demonstrates accuracy ranging from 71 to 82.4%, with the highest F1 score of 0.859 on Dataset 1, confirming the framework’s effectiveness for interpretable and balanced bias detection.},
journal = {Journal of Computer Science},
publisher = {Science Publications}
}