Sentiment Analysis using Naive Bayes

Sentiment analysis is a challenging and interesting natural language processing task, if only because it naturally lends itself to domain adaptation. We study sentiment analysis using Naive Bayes and essentially reproducing the results from [1]. We start by describing the Naive Bayes model we use, then we describe the experimental setup and finally we discuss our observations and results. The Naive Byes assumption asserts that the features of a review are independent conditioned on the classification. The merit of making this assumption is the robustness of the trained model and the ease and efficiency of the parameters estimation. Formally, the probability that a review ~x is classified c is then given by the expression