Sentiment Analysis of Twitter Data with Hybrid Learning for Recommender Applications

This paper proposes a sentiment analysis approach to extract sentiments of tweets based on their polarity and subjectivity, classify them and visualize results graphically. This helps to understand opinions of existing users that can be helpful in future recommendations. Our proposed approach entails a hybrid learning method for classification of tweets based on a Bayesian probabilistic method for sentence level models given partially labeled training data. For implementation, we use AWS to extract data from Twitter, store extracted data in MySQL databases and code Python scripts in order to implement the analyzer. The graphical models are viewed using IPython Notebook. The results of this work would be helpful in providing recommendations to users for product reviews, political campaigns, stock predictions, urban policy decisions etc. The novelty of this research lies mainly in the hybrid learning method for sentiment analysis. We present our approach along with its implementation, evaluation and applications.