Microblogger tag auto-annotation based on collective knowledge

In China, it is free for users to introduce themselves with tags on Sina weibo, the biggest Chinese microblogging platform. User tags provide much useful information for user retrieval and precision advertising. However, 64.2% of users in our study have not tagged themselves. This paper aims to providemicroblogger tag auto-annotation. First, we analyze the tagging behaviors of users to explore what information that is contained in the tagging. Second, based on the analyses and features of social networking, we propose two methods of user tag auto-annotation: one is Collective Filtering (CF) based method and the other is Probabilistic Latent Semantic Analysis (PLSA) based method. Specifically, the CF-based method annotates tags for one user according to the voting scheme by considering his/her neighboring users' tags. By contrast, the PLSA-based method annotates tags for one user according to the topic distribution of his/her neighboring users' tags. Experimental results demonstrate that: (1) the CF-based method outperforms the PLSA-based method at word-level; and (2) the PLSA-based method can obtain more accurate tag annotation results than the CF-based method at topic-level.