Research On Social Event Organization And Recommendation In Social Networks

With the rapid development of the Internet technology,online social networks have become important media for people to obtain and share information.In recent years,event-based social networks(EBSNs)that combine online and offline user interactions have developed rapidly.Meanwhile,the amount of social event data in EBSNs has increased dramatically.Based on rich social event data,the research on social event aiming at satisfying the requirements of social event organizers and participants has attracted wide attention.For event organizers,how to attract event participants and how to select right participants are the key issues to organize an event successfully.For event participants,how to find the events that they are interested in from a large number of events is the main issue they face.Therefore,the research on social event organization and recommendation for these problems is critical.Although domestic and foreign researchers have made a series of research results on social event organization and recommendation,the existing researches still have some shortcomings in terms of lacking full consideration of realistic factors,insufficient information utilization and poor model effect.Therefore,aiming at providing efficient intelligent services for users,this thesis considers the shortcomings of existing researches,to study three key issues in social event organization and recommendation:social event influence maximization for social event organizers;willingness maximization for social group for social event organizers;personalized social event recommendation for social event participants.The main contributions of this thesis are as follows.(1)In the aspect of social event influence maximization,we propose a fair-aware approach for competitive social event influence maximization.The existing studies ignore the demand of maximizing the influence for multiple social events under a competitive environment in real-world scenarios.To bridge this gap,we study a novel problem of competitive event influence maximization,for maximizing influence spread and ensuring fairness for multiple social events.To solve this problem,we propose a fair-aware approach.Considering the characteristics of offline social events,we propose a propagation model to describe the information propagation process of social events in competitive situation in EBSNs.Based on the propagation model,we propose a randomized algorithm based on the cross entropy method to find influential users(seed nodes)for each social event,so as to maximize the influence spread of social event on the premise of fairness.It can optimally allocate the node selection probability for each node,so as to effectively approach to the maxima.We conduct extensive experiments using two real-world datasets and experimental results demonstrate that the approach proposed improves the influence spread and reduces the running time under the premise of fairness.(2)In the aspect of willingness maximization for social group,we propose an approach for multi-role willingness maximization for social group.Focusing on that existing studies have not taken the role requirements of participants in social event into consideration,we study a novel problem of multi-role willingness maximization for social group.To solve this problem,we propose a continuous relaxation technique based algorithm.Firstly,it converts this problem into an equivalent unconstrained discrete problem by using exact penalty approach.Secondly,based on the concept of Lovasz extensions,it converts the unconstrained discrete problem to an equivalent unconstrained continuous problem.Finally,it employs a continuous optimization algorithm to find a local optimal solution.In order to achieve a better tradeoff between performance and running time,we propose a parallelization heuristic algorithm based on improved PageRank.This algorithm greatly reduces the running time by selecting social event participants with different roles in parallel.Two real-word datasets are used to verify the effectiveness of the proposed algorithms.The experimental results demonstrate the effectiveness and efficiency of the above two algorithms.(3)In the aspect of personalized social event recommendation,we propose an approach fusing the positive and negative feedback information of users for social event recommendation.The existing studies utilize the context information of social events that target user participated in,i.e.,positive feedback information,for learning user's preference.They ignore the importance of the context information of social events that target user refused,i.e.,negative feedback information,for characterizing user's preference,which leads to the problem of low accuracy of event recommendation.Therefore,considering the positive and negative feedback information of users and social influence,we propose a social event recommendation approach.In this approach,firstly,we modify the Long Short-Term Memory(LSTM)to learn the preference of users by using the positive and negative feedback information of users.Secondly,we propose a time-aware attention mechanism.Then we introduce it into Graph Neural Network(GNN)to learn the influence of friends'preference to target user's decision for participating in an event.Finally,we aggregate target user's preference and friends'preference influence to predict the social event that the target user will participate in.We conduct extensive experiments using real-world datasets to verify the approach.The experimental results show that our approach can effectively improve the accuracy of personalized social event recommendation.