Performance Improvement of Naïve Bayes Classifier for Sentiment Estimation in Ambiguous Tweets of US Airlines

Abstract

Airline deals with wide scale of customers, including family members, business class, business deals, men, women, and youngsters. Every kind of customer is involved in this domain. So their feedbacks are very important; direct feedback of customers is always positive but analyzing their tweets is important, how the Tweets are? Whenever we move toward personal tweets, analysis becomes so much hectic because of customers in very high volume. Most of the times tweets are ambiguous; it depends on thinker. If the thinker is positive, then he always gives positive tweets and negative thinker gives negative tweets. So ultimately, our work is to find sentiments of descriptive tweets and words and address their expression in quantities format whether they are happy or not. The novel factor inside this work is to examine ambiguous tweets and neutralize them according to proposed algorithm. The complete work is based on Twitter dataset for US Airlines and performs different level of mining and processing for most accurate results. Here, improved sentiment analysis model has been proposed based on Naïve Bayes classifier to classify tweets based on sentiments and neutralized tweets from ambiguous to positive or negative. A Java tool has been developed to implement the proposed solution that is tested through manual testing and performance parameters. This work considers recall/accuracy, f-score, and computation time as performance parameter for proposed solution. The complete solution is compared with existing work and concluded that it performs better than conventional algorithms.