Predicting burst in threads of bulletin board systems based on “Connoisseurs”

There have been a large amount of research on topic tracking and detecting bursts in topics. However, there has been little research on predicting bursts in the future. We propose a method for predicting bursts of threads in a Bulletin Board System (BBS) using a kind of user. We call the kind of the user a ldquoconnoisseurrdquo. A connoisseur is a user who comments early in a thread which will burst in the future. We believe a thread in which connoisseurs appear has a high possibility of bursting. We calculated the degree of a user becoming a connoisseur from past data. Then we predicted bursts of threads based on the comments of connoisseurs. We evaluated the performance of the proposed method using data from ldquoYahoo! bulletin boardrdquo and showed its usefulness.