Probability adjustment Naïve Bayes algorithm based on nondomain-specific sentiment and evaluation word for domain-transfer sentiment analysis

In the research of sentiment analysis, some supervised learning algorithms play an important role. Among them, Naïve Bayes is often used in engineering application due to its low computational and space complexity. While traditional Naïve Bayes algorithm has been shown to perform very well in domain-specific sentiment classification, it often performs badly in domain-transfer problem. So we propose a probability adjust Naïve Bayes algorithm (PANB) to solve this problem. We use polarity D-value PointWise Mutual Information (PDPMI) method to obtain nondomain-specific words and their weight score, and then use the weight score to adjust probability of feature in training step. The result of experiment shows that our approach usually achieves better performance than traditional Naïve Bayes classifier.