자연어 처리를 이용한 유튜브 댓글 여론 변화 분석

Abstract

As information in online media has increased, various studies have been conducted on the effect of comments on public opinion. However, in the existing studies, experiments and surveys were conducted on individuals. This cannot be considered appropriate in that the situation of writing comments in an actual online environment is anonymous. In this paper, in order to overcome the limitations of existing research papers, we chose a method of selecting a medium in which public opinion has hardened to some extent and observing it. Among many online media, we selected YouTube, in which opinion sharing is most actively performed recently and information on comments is easy to obtain and specified the videos on topics in which public opinion could be changed over time. After that, we crawled comments on the videos and measured the degree of the emotion of the comments by conducting a natural language processing-based sentiment analysis using the KcELECTRA model. In addition, we visualized this and analyzed how much public opinion of the latter period is affected by early public opinion. The analysis result of this paper show that the probability of public opinion reversal is very low, which means that the probability that the initial public opinion will become the final public opinion is very high. Based on the conclusion of this study, we expect the public to realize that the public can be one of the users who are swept by public opinion, and to accept information from a more critical and objective perspective with caution.