Twitter Sentiment Analysis Based on Writing Style

摘要

This paper proposes a new method of sentiment analysis for Twitter. Tweets contain various expressions; e.g., use of emoticons. The usage of these expressions links to the user's identity and individual characters. Handling these characteristics is useful for the sentiment analysis. We focus on writing styles of each user. In this paper, we define three types of writing style; formal and two informal expressions. First, our method classifies each tweet into the three types. Then, it generates classifiers for each writing style. We apply our method to a positive / negative classification task of tweets. In the experiment, the accuracy of our method increased by approximately 3 points as compared with some baseline methods.