Sentiment Analysis in Twitter Messages Using Constrained and Unconstrained Data Categories

This paper describes a system to answer a specific sentiment analysis problem described in 2015 iteration of SemEval (Semantic Evaluation series), Sentiment Analysis in Twitter as the base challenge to be improved. When it comes to sentiment analysis competitions and shared tasks, this is the most popular to date with more than 40 teams participating each year from its inception in 2013. Only subtask B (Message Polarity Classification) was considered under main task (task 10) in this model, as it was a return from previous years and remained highly challenging and competitive among teams from around the world. The proposed model performed exceedingly well, notably getting best results (1st) against 2015 test set, 2nd best results for evaluations against 2013 and 2014 test sets. We performed evaluation using both constrained and unconstrained data under two major classification techniques, single classifier based approach and ensemble approach. For single classifier based approach, classifiers such as Support Vector Machines, Logistic regression, BayesNet and Artificial Neural Networks and for ensemble approach algorithms such as Bagging, Ada Boosting, Random forest and Voting techniques were used. Several key contributions of this research including enhanced feature extraction algorithms, newly created sentiment lexicon, exhaustive analysis by means of various classification techniques using both constrained and unconstrained data clearly prove to be effective in addressing the given task..