Keyword Discovery by Measuring Influence Rates on Bulletin Board Services

Abstract

In this paper, we focus on relations between comments on Tree-style Bulletin Board Services (BBSs), and propose a method for discovering keywords by measuring influence rates thereon. Our method is based on an extended model of Influence Diffusion Model (IDM) proposed by N. Matsumura et al. in 2002, where they discussed the influence diffusion of a term in a comment to all succeeding comments that include that term and reply to that comment. Here we additionally consider the influence diffusion of a term over comments that include that term and all reply to a same comment, as well as the influence diffusion of a term over nearby comments that include that term, regardless of their reply relation. Evaluation results using Tree-style BBS data related to Massively Multiplayer Online Games (MMOGs) show that the proposed method has higher precision and recall rates than IDM and a classical method based on term frequencies. As a result, keywords discovered by the proposed method can be effectively used by MMOG publishers for incorporating users’ needs into game contents.