The Event-based Social Networks(EBSN)connect online social relationships and offline local events together,encouraging the integration of both online and offline interactions among users.But the growing amount of events published on EBSNs brings severe information overloading problem to users,making them hard to find interesting events to participate.Recommender systems are the effective solutions to alleviate such problem.The thesis studies on local event recommendation by personalized recommendation algorithms,aiming to further improve the recommendation performance.The main contributions from this paper are summarized as follows:(1)A new personalized local event recommendation algorithm based on latent correlation learning,called CCRL,is proposed.The CCRL algorithm explores and exploits the latent correlations between event participation preferences of users and contextual information,modeling these latent correlations and implementing personalized local event recommendation based on collective contextual relation learning.Comparative experiments on Meetup dataset demonstrate that CCRL outperforms existing recommendation algorithms on many metrics.(2)A novel personalized local event recommendation algorithm based on latent revenue relation learning,called PCFM,is proposed.Influential factors for users' event participation behaviors can be categorized into preference factors and constraint factors,which make up the latent revenue relation of event participation.The PCFM algorithm estimates the influence on event participation behaviors from constraint factors like venue and start time of events,and the preference and constraint factor model is proposed to incorporate these factors and model event participation behaviors of users for personalized local event recommendation.Experimental results on Meetup and Douban Events datasets show the improvement of recommendation performance by the algorithm.(3)A new personalized local event recommendation algorithm based on latent competitive relation learning,called LCRL,is proposed.The event participation decision process of users can be regarded as the competition among revenues of participating different events.The LCRL algorithm models such competitive process by the proposed latent competitive relation learning loss function,aiming to simulate decision process of event participants and train the users' preference model.Many metrics in the evaluating experiments on real-world datasets demonstrate the effectiveness of LCRL algorithm. |