Research On Context-aware Group Recommendation Approaches In Event-based Social Networks

With the development of mobile Internet,the interactions between users have become increasing convenient and there is a new trend to combine online and offline interactions between users.Event-based social networks(EBSNs)develop rapidly in recent years by combining online and offline interactions.However,the rapid development of EBSNs brings serious information over-load problem,where finding the attractive events for users has become increas-ing difficult.Event recommender systems are effective tools to alleviate the problem of information overload.Compared with the recommendation in tra-ditional domains,event recommendation inherently faces cold-start problem.Various contextual information in EBSNs is required to be exploited to allevi-ate the cold-start problem in event recommendation.Moreover,the previous study on event recommendation mainly focuses on recommend events for in-dividual users and ignores suggesting events for a group of users,i.e.,group recommendation.However,the diversity of user preferences makes it difficult to find events satisfying all users in a group,which brings more challenges for group recommendation in EBSNs.To this end,this thesis aims to study on context-aware group recommendation in EBSNs,and provides more effective event recommendation services to users and groups.The main contributions and innovations of this thesis are summarized as follows:(1)We formalize group event recommendation problem as a learning-to-rank problem and propose a group event recommendation framework based on learning-to-rank technique.Specifically,we first analyze different contextual influences on user' s event attendance,and extract preference of user to event considering each contextual influence.Then,the preference scores of the users in a group are taken as the features for learning-to-rank to model the preference of the group.Moreover,Bayesian group ranking algorithm is proposed to learn ranking model for each group.The results of extensive experiments on two real-world datasets show the appealing performance of our method on both accuracy and time efficiency.(2)We first discover the correlation between organizer and textual content,i.e.,the events held by the same organizer tend to have more similar content than those held by different organizers.Based on this observation,we present a probabilistic generative model to exploit correlation between organizer and content to alleviate the sparsity of textual content,and extract group preference by discovering content topic and venue topic.Finally,a group event recom-mendation method using our model is proposed.The experimental results on two real-world datasets demonstrate that the proposed model outperforms the state-of-the-art methods that suggest upcoming events for groups.Besides,our model can learn semantically coherent latent topics which are useful to explain recommendations.(3)We investigate the periodicity of the events held by same group,and then propose a group-aware periodic topic model for cold-start event recom-mendation.Both periodic and non-periodic influences can be captured by our model.For periodic events,we assume that their timestamps are generated from a Gaussian mixture model.A multinomial distribution is used to model the timestamps of non-periodic events.Moreover,our proposed model jointly discovers user preferences on event and venue from users' event attendance be-haviors.The experimental results on the real-world dataset show the significant improvements of our model over other comparison methods.(4)We propose a fairness and diversity aware group recommendation method.We define a weighted coverage as a submodular function on event similar-ity graph,where the similarity between events is computed based on multiple context.Then,the fairness and diversity aware group event recommendation problem is transformed to maximization of weighted coverage on event similar-ity graph to improve the accuracy,fairness and diversity simultaneously.We exploit greedy algorithm to maximize weighted coverage.The experimental results on the real-world dataset show that our proposed method find a better trade-off between accuracy,fairness and diversity compared with the state-of-the-art methods.