Predicting popularity of articles on bulletin board system

In this work, we study how to predict the popularity of an article in the largest terminal-based bulletin board system (BBS) and also one of the most major social media, PTT (ptt.cc), in Taiwan. Given a specified article and a duration time after it is published, we want to predict the number of unique users that have ever commented the article and the degree of popularity based on the total number of comments of this article compared to other historical articles in the system. We first introduce the ecology of PTT and show our observations of the user posting behaviors. Since PTT has quite a different style compared to other commonly-known social media such as Facebook and Twitter, we show how to extract and integrate four sets of important and useful features, including the textual, author-wise, social and temporal ones, from the large-scale BBS data for predicting the article popularity with different classification models. Experiment results show the effectiveness of all these four types of features we extracted. Specifically, the temporal and social features help improve the prediction qualities most.