A forum is a social network that consists of posters and the following comments made by netizens. Generally speaking, forum topics are evolving over time dynamically. In this paper, based on time series analysis and matrix modularity analysis, a novel prediction method is proposed through investigating the correlating influence of three key measurements: relationship strength, pillars, and change frequency of a forum topic. The method demonstrates that there exist some macroscopic and potential laws for forum situation prediction. Extensive experiments over large many datasets show the efficiency and effectiveness of the algorithms.
