Research On Methods Of Context-Aware Event Recommendation For Groups In Event-Based Social Networks

With the development of Internet and mobile communication technology,Event-Based Social Network(EBSN)platforms,such as Meetup,Plancast and Douban Events,are increasingly favored by users.As EBSNs are growing popular,there arises the problem of information overload,which makes it difficult for users to find their favorite events from the massive published events.Recommender systems have been an effective method to address the problem by retrieving the most relevant information from the data and providing personalized information service for users.In recent years,EBSN event recommendation has gained the attention of researchers and many achievements have been obtained.However,most of the existing works focus on recommending events for individual users.In practice,people often attend offline events together as a group.This motivates the research of recommending events for groups in EBSNs.In addition,there are abundant contextual information in EBSNs,Exploiting these contextual information is an effective mean to alleviate the cold start problem,and it also can help to improve the accuracy of group preference acquisition.Therefore,the research on context-aware event recommendation for groups in EBSN has important theoretical significance and practical value for enriching EBSN group recommendation theory and improving the quality of group recommendation service.Different from recommendation for individual users,group recommendation faces challenges such as how to aggregate different user preferences,how to acquire group preferences accurately,and how to model the influences of various contexts effectively.To tackle above challenges,this dissertation focuses on group preference aggregation,group preference acquisition,and context influence modeling.The main work and contributions are summarized as follows:1.A group recommendation method considering the impact of preference consistency is proposed.To address the problem of existing group aggregation strategies ignoring the influence of member preference consistency,the characteristics of time,location and semantic context in EBSNs are analyzed firstly.Then,event feature extraction methods under the influence of these contexts are given,and the acquisition methods of contextual preferences are also proposed.Secondly,a group preference aggregation strategy based on random walk with restarts is defined.In order to consider the influence of consistency between member preferences and group preferences when setting restart vectors,a concept of Consistency Decision Weight of User is defined,and the weight computing problem is converted into a convex quadratic programming problem.Experiments on real EBSN datasets show that the proposed group recommendation method has better recommendation performance than baseline methods.2.An acquisition method for the preferences of group unexperienced events is proposed.To address the problem of existing work focusing only on the recommendation of familiar group events,how to make use of the social relation beyond the group to acquire the preferences of group unexperienced events is investigated.Firstly,the implicit trust relations are mined based on users' online and offline social behaviors and the user trust network is constructed.Then,to obtain members' preferences for unexperienced events,a probabilistic method based on random walk is used to simulate the dynamic process of members' consultation with their friends.Experiments on real EBSN datasets show that compared with the baseline group recommendation methods,our recommendation method based on the proposed preference acquisition method has improved the accuracy of event recommendation,especially the accuracy of the unexperienced event recommendation.3.A deep context-aware group recommendation method based on attention mechanism is proposed.Aiming at the limitation that heuristic methods could only model the shallow influences of contexts,a multi-layer neural network with nonlinear functions is constructed to effectively capture the complex and nonlinear influences of contexts on users,groups and events.And a novel neural attention mechanism is developed to capture the dynamic change of context weights with events.To address the problem that the pre-defined method leads to the lack of adaptability of group aggregation strategies,an adaptable group preference aggregation strategy based on neural attention mechanism is proposed,which can automatically adjust the decision rule according to the current events concerned by the group.Experiments on real EBSN datasets show that the proposed recommendation method has better recommendation performance than baseline methods.