Detection of Flaming Participants Using Account Information and Stylistic Features of Posts

With the rapid development of SNS in recent years, the number of SNS users has increased rapidly, and people can easily communicate interactively with an unspecified large number of people. With these changes in the information society, a phenomenon known as "flaming", in which critical comments flood SNS, has become a frequent occurrence. In recent years, various studies on flaming have been conducted, but most of them are concerned with those who receive a large number of critical comments, not on those who write critical comments, called "flaming participants". In this study, we examine the characteristics of flaming participants on Twitter by using machine learning to classify them into two groups: flaming participants and normal users. For the classification features, we use account information, i.e., statistical data for each account, and stylistic features of the postings, i.e., (1, n)-grams of the part-of-speech tags of the postings. The experimental results show that these features are effective in detecting Twitter flaming participants. Furthermore, we found that flaming participants use more quote tweets than the normal user, and that there are word patterns that are characteristic of flaming participants.