DOMAIN-ADAPTIVE SENTIMENT ANALYSIS THROUGH BAYESIAN NETWORK–LOGISTIC REGRESSION-BASED PROBABILISTIC DEPENDENCY MODELING

Cross-domain sentiment classification presents persistent challenges in opinion mining due to vocabulary drift, contextual ambiguity, and polarity inconsistency across product domains. Traditional classifiers trained on a single domain often fail to generalize, reducing performance when exposed to new, structurally distinct datasets. This research introduces a probabilistically grounded sentiment classification framework—Bayesian Network–Logistic Regression (BN-LR)—designed to address these challenges within multi-domain online review environments. BN-LR integrates a Bayesian Network to model conditional dependencies among sentiment-bearing features, capturing latent inter-feature relationships across syntactic structures. These probabilistic insights are dynamically incorporated into a logistic regression classifier, enabling adaptive feature weighting and uncertainty-aware sentiment inference. As an IT contribution, BN-LR offers a scalable, interpretable, and statistically principled solution suitable for intelligent recommendation engines, feedback analytics, and sentiment-based decision systems across digital platforms. Evaluated on Amazon reviews across four domains, BN-LR consistently delivers high accuracy without requiring domain-specific retraining or external lexicons. The proposed framework enhances real-world information systems by enabling robust cross-domain sentiment generalization, fulfilling a critical need in adaptive text analytics for IT-driven e-commerce intelligence