Performance Evaluation of some Select Machine Learning Algorithm for Sentiment Analysis

Abstract

Background: Sentiment analysis determines the class that opinions about an instance or object belongs to. This class can be positive, negative or neutral which reflects satisfaction, dissatisfaction and indecision respectively. Computer algorithms are being used to automatically perform sentiment analysis. Aim: This study presents a comparative analysis of the performance of Naïve Bayes, K-Nearest Neighbor and Support Vector Machine in sentiment analysis. Method: Two large text dataset describing the opinions of customers on airline services and the 2015 super bowl show collected from Tweeter. The Scikit-Learn machine learning package on the Python programming language is used for training the models and measuring their performances in terms of precision, recall and accuracy. Results: the study shows that K-Nearest Neighbor has the best performance in terms of accuracy with measures of74.5% and 60.3% in the airline and deflategate datasets respectively. Naïve Bayes has the best performance in terms of precision and recall with a measure of 92% and 100% respectively in the airline dataset, and 82% and 96% respectively in the deflategate dataset.