Analyzing Influence of Emotional Tweets on User Relationships by Naive Bayes Classification and Statistical Tests

As an SNS, Twitter is popular because users can post their emotions as a short message easily. Emotional tweets may influence user relationships. In our previous study, we found that positive users construct mutual relationships in Twitter. Keyword matching with emotional word dictionaries was used to detect positive users. The problem of keyword matching is the limitation of word number. To solve this problem, we use machine learning, specifically Naive Bayes Classification, to classify emotions of tweets. We analyze whether there is a difference in user relationships between the positive and negative groups by the Brunner-Munzel test. The result shows that the relationships of positive users increase more than that of negative users in the followee fluctuation, follower fluctuation and mutual follow fluctuation, which means that a positive user is more active to construct user relationships than a negative user.