Empirical comparison of sentiment analysis techniques for social media

Nowadays the excessive use of internet produces a huge amount of data due to the social networks such as Twitter, Facebook, Orkut and Tumbler. These are microblogging sites and are used to share the people opinions and suggestions on daily basis relevant to the certain topic. These are beneficial for decision making or extracting conclusions. Analysis of these feeds aims to assess the thinking and comments of people about some personality or topic. Sentiment analysis is a type of text classification and is performed by various techniques such as Machine Learning Techniques and shows that the text is negative, positive or neutral. In this work, we provide a comparison of most recent sentiment analysis techniques such as Naive Bayes, Bagging, Random Forest, Decision Tree, Support Vector Machine and Maximum entropy. The purpose of the study is to provide an empirical analysis of existing classification techniques for social media for analyzing the good performance and better information retrieval. A comprehensive comparative framework is designed to compare these techniques. Various benchmark datasets (UCI, KEEL) available in different repositories are used for comparison purpose. We presented an empirical analysis of six classifiers. The analysis results that Support Vector Machine performs much better as compared to other. Efforts are made to provide a conclusion about different algorithms on the basis of numerical and graphical metrics to conclude that which algorithm is optimal. (C) 2018 The Authors. Published by IASE.