Analysis on Opinion Mining Using Combining Lexicon-Based Method and Multinomial Naive Bayes

摘要

Opinion mining is the analysis on opinions which is done by looking at the sentiments, behaviors, or emotions contained in a product. Some of the opinion mining methods are using the lexicon-based and supervised learning. Lexicon-based method has a low recall, while supervised learning has good accuracy but requires a long training period. Therefore this paper will discuss lexicon-based method with one of the supervised learning methods namely Multinomial Naive Bayes for the English language. These methods are used to classify opinions based on the sentiments, i.e., positive and negative. This research employed the feature extractions: unigram, POS-Tagging, and score-based feature on lexicon. The output of the system is the polarity of each document and the performance will be calculated using Precision, Recall, and F-measure. By implementing the opinion mining using the combining lexicon-based method and Multinomial Naive Bayes, the accuracy obtained was 0.637.