Feature selection approach for twitter sentiment analysis and text classification based on chi-square and naïve bayes

Abstract

With the rapid growth of web and mobile technology, Social networking services like Twitter are widely used, resulting in large amounts of data being generated daily in social networking sites. Efficient Sentiment analysis of such data is very important for a range of applications and improvement of accuracy in detecting sentiment is the main aim of this research. This report examines the combination of a Chi-Squared feature selection algorithm, k-mean clustering and TF-IDF for attribute weighting based on Naïve Bayes, for classification of text and sentiment in communications generated on Twitter. This approach is compared with other approaches based on Naïve Bayes to give an account of their relative strengths and weaknesses. When running experiments on multi-domain twitter datasets, results indicate that the proposed method shows superior performance across a range of. The main aim of this research is to enhance the performance of the Naïve Bayes classifier using a feature selection technique.