A combination of lexicon and machine learning approaches for sentiment analysis on facebook

Abstract

The increase of user-generated content (UGC) on the Internet has led previous studies to propose various sentiment analysis approaches to understand public opinion. The primary goal is to enhance engagement through social media by analyzing various feedback. Sentiment analysis is performed based on two approaches i.e. machine learning and lexicon-based. Since approaches based on machine learning require costly preparation of training dataset and the approaches based on lexicon produce unsatisfactory performance, in this paper, both approaches are combined to perform sentiment analysis on Facebook comments. The importance of a lexicon-based approach to automatically construct the labeled data for machine learning sentiment classification is discussed in this paper. Experiments performed using the Universiti Utara Malaysia (UUM) Facebook posts show that using the combined lexicon-based and machine learning approach on two classifiers i.e. Naïve Bayes and Support Vector Machine outperform the single approaches to produce more accurate sentiment classifications.