Japanese Election Prediction Based on Sentiment Analysis of Twitter Replies to Candidates

Election prediction based on Web information has attracted much research interest. One previous research for Japanese election prediction proposed two kinds of indicators including familiarity degree and support degree and performed machine learning to achieve the goal. In this paper, we explore new features including the number of Twitter replies to election candidates and the sentimental values of reply texts, and develop models that have higher performance. The results of experiments conducted for the Japanese House of Councillors election in 2019 show that our method can obtain a higher F-measure than the previous method, which indicates that the usage of reply texts and sentiment analysis can contribute to election prediction.