Movie Review Analysis Using Text Sentiment Analysis and Designing a Strategy of Selecting a Representative Review

Abstract

In this paper, the goal is to find a compromise between consumers seeking objective information and providers seeking to provide as much good information as possible about movie. This requires identifying movie information preference patterns, reliable review intervals, and consumption trends according to good reviews. Based on these purpose, I designed three research questions and conducted experiments on each question. In the process of constructing experiments, I crawled data from movie review sites to collect and extract emotional features of each review sentences using sentiment analysis techniques. The testbed was produced in unity and deployed online, with 72 participants. As a result, I observed the distribution of what emotions were dominant for each score range, and identified what characteristics exist in reviews that deviate from representative emotions. Results from user study identified which factors (star rating and review) influenced more in each of two cases in binary selection and single environment. Furthermore, I interpreted user selections in a single environment through reliable review intervals, presenting a strategy to the movie provider on which reviews should be selected as representative reviews.