Personalized Recommendation Algorithm on Microblogs
Ming Zhang · Jisuanji kexue yu tansuo · 2012
Microblogging community is different from conventional social networks and e-commerce systems for its low user activity,data sparsity and dynamic of user-interests.Because of these challenges,conventional recommendation algorithms cannot get desirable performance in microblogging community.This paper proposes a novel recommendation algorithm based on Bayesian personalized ranking(BPR) by modeling user's implicit feedbacks in microblogging community.The proposed algorithm collects implicit feedbacks in the form of microblogs pairs and uses them as training pairs to learn users'interest.This paper defines a confidential score for each microblogs pair based on the time user received it.Microblogs pairs with shorter interval time have higher confidential score and thus have much more impact on user's interest.Experiments on two real-world microblogs datasets show that the recommendation algorithm outperforms all the baselines.