Learning sentiment dependent Bayesian network classifier for online product reviews
Document Type
Article
Publication Title
Informatica (Slovenia)
Abstract
Analyzing sentiments for polarity classification has recently gained attention in the literature with different machine learning techniques performing moderately. The challenge is that sentiment-dependent information from multiple sources are not considered often in existing sentiment classification techniques. In this study, we propose a logical approach that maximizes the true sentiment class probabilities of the popular Bayesian Network for a more effective sentiment classification task using the individual word sentiment scores from SentiWordNet. We emphasize on creating dependency networks with quality variables by using a sentiment-dependent scoring technique that penalizes the existing Bayesian Network scoring functions such as K2, BDeu, Entropy, AIC and MDL. The outcome of this technique is called Sentiment Dependent Bayesian Network. Empirical results on eight product review datasets from different domains suggest that a sentiment-dependent scoring mechanism for Bayesian Network classifier could improve the accuracy of sentiment classification by 2% and achieve up to 86.7% accuracy on specific domains.
First Page
225
Last Page
235
Publication Date
6-1-2016
Recommended Citation
Orimaye, Sylvester Olubolu; Pang, Zi Yang; and Setiawan, Alvino Mandala Putra, "Learning sentiment dependent Bayesian network classifier for online product reviews" (2016). Global Population Health Faculty Publications. 18.
https://collections.uhsp.edu/global-population-health_pubs/18