Smart Sentiment Analyzer for Bengali-English based on Hybrid Model

Abstract

This paper describes an intelligent sentiment analyzer incorporating linguistic knowledge. Our proposed method comprises three stages i) corpus construction, ii) classification using machine learning tools, and iii) linguistic knowledge integration in post processing stage if any misclassification occurs due to minor problem. Here, we applied nine supervised and ensemble learning approaches. From the experiments it is observed that the naïve Bayesian classifier with linguistic knowledge provides better accuracy (93%) on an average over the other classifiers using the 4-fold cross validation. Positive or negative or confused emotions with corresponding emoticons for both English and Bengali languages are determined and the results are demonstrated via an easy to use web based interface.