On The Feature Extraction For Sentiment Analysis of Movie Reviews Based on SVM

Abstract

Watching a movie is one of the activities that reduce bored, so it is necessary to look for information about the movie, which is packaged in the form of a movie review to determine whether the movie considered for viewing or no. However, in searching for information through movie reviews, there are obstacles because there are many reviews conducted by reviewers. Therefore, sentiment analysis is needed aims to classify the movie review into positive and negative sentiments. Machine learning methods can use as a sentiment analysis classification because that can produce the best performance, the method called Support Vector Machine (SVM). That was a reason SVM classification used in sentiment analysis on movie review data. Use feature extraction of Term Frequency- Inverse Document Frequency (TF-IDF) was also carried out in the research this as a method of weighting words which then combined with the extraction of Latent features Dirichlet Allocation (LDA) as a method of modeling topics that can overcome the shortcomings of SVM. This research produced the best performance on a combination of TF-IDF and LDA, with 240 topics has 29792 features, which is 82.16%.