Group Detection And Group Recommendation In Event Based Social Networks

Event-based social networks(EBSN)provides an online platform for people to create,publish and organize social events to help users with similar interests interact online and offline.Unlike traditional social networks,the network is driven by events,with groups as the main form,and contains two types of interaction between the online and the offline.Traditional methods have no effect on the group discovery and group recommendation problem.So,we focus on group detection and group recommendation strategy in EBSN.In order to solve the problem of group discovery and recommendation based on EBSN,it is necessary to first extract the user's implicit preference and predict the user's ratings on unknown events.User's preference detection is the basis of group detection and group recommendation.Therefore,considering user's action information in online and offline,this paper proposes deep neural network based user's preference detection model.To achieve this ideas,we use event's description text,time and location information to fuse online group's labels and semantic information by deep neural network.To get user ratings for events,we extract users' features and events' features respectively,and map them into the same space by deep semantic network,so as to get users' ratings of events.Aiming at the problem of group detection in EBSN,this paper proposes the concept of k-maximal heterogeneous group,which satisfies the characteristics of small scale,similar preferences among members,frequent communication in online and offline.Then,we propose a recursive based algorithm for mining the largest heterogeneous group.The algorithm extracts all the maximal k-heterogenous groups in the time complexity of exponential.Finally,based on maximal k-heterogenous groups detection algorithm,this paper proposes a classification based approximate algorithm of k-maximal heterogeneous group.The algorithm selects the initial user according to the classification probability of the classifier and determines whether to prune every step in the search process,which leads to higher search efficiency.In this paper,the validity of the algorithm is tested on a real dataset.Aiming at the group recommendation problem in EBSN,this paper transforms it into a deep neural network and uses the BP algorithm to solve it according to the user's repeated communication model.Considering the complex network structureand long training time,this paper use AP clustering and auto encoder to get the appropriate number of neurons in hidden layers,which reduces the network complexity and achieves better performance.Finally,the experiment is carried out on the real data set of Meetup to verify the effectiveness of the algorithm.