Predicting the popularity of tags in StackExchange QA communities

StackExchange is one of the most popular Question and Answering (QA) websites, where each community address the questions on specific domain, e.g., programming, math, game, and so on. In these communities, users can use tags to label questions, which facilitates the search of questions and recommendation of experiments. Some tags are frequently used and thus get more and more popular with time, while some others are seldom used and finally diminish. The goal of this study is to find out the features that affect the future usage of tags and then design the popularity prediction algorithms. We investigate structural and non-structural features of tags, and using machine learning methods to classify popular and unpopular tags. The results show that, in general, the prediction models based on both structural and non-structural features indeed behaves better than those just based on one type of features, and the random forest (RF) method behaves the best among all the four considered machine learning methods.