UT-LDA Based Similarity Computing in Microblog

With the widespread of microblog, the microblog platform like Sina has become information provider to release mass information. Mining the valuable information from microblog and making personalized recommendations have become a hotspot of recent research. But most researches about microblog primarily focus on user´s social circle. The recommendation is mainly based on the user´s friends or the users they follow, and the study on how to find users with similar hobbies from unfamiliar social circles is rare. Therefore, we propose a microblog generation model UT-LDA (User Topic Latent Dirichlet Allocation) to compute users´ similarities. This model takes one user´s microblog as a document, and gets the different topic distributions on different categories in each microblog. According to the users´ topic distributions and the profile information of each user, we can calculate the similarities between users. The experimental results show that our UT-LDA model works effectively in recommending similar users.