Research Of Dynamic Link And Cluster Prediction For Offline Ephemeral Social Networks

Offline ephemeral social networks is kind of offline social networks which refers to a specific time, designed for a specific purpose, short duration and formed by face to face contact. Unlike online social networks, the offline ephemeral social networks can provide more reliable information, but the social relationships in the networks are transient and volatile. For the prediction problems of the networks, the traditional methods of the online social networks cannot work effectively. Therefore, how to effectively use the information in the offline ephemeral social networks to solve the prediction problem is a major challenge that the researchers are facing with.Firstly, the related works of the offline ephemeral social networks are analyzed, which has not adequately taken the characteristics of the networks into account. Then, the related works of link prediction, community and clique discovery are introduced and analyzed in detail.Aiming at the mutual link prediction problem between the offline ephemeral and the online social networks, a unified prediction framework based on the factor graph model is proposed. First of all, we analyze the correlation between the two kinds of social networks based on a real data set, to verify some basic assumptions: users’ behaviors have similarities in the offline ephemeral and the online social networks, they are related heterogeneous networks; if the users has a friend relationship in the online social networks, they maybe encounter more frequently in the offline ephemeral social networks; if users encounter more frequently in the offline ephemeral social networks, they may be easier to become friends in the online social networks; the encounter behaviors of users are not only affected by friendship but also by their own past behaviors. According to the correlative characteristics in the two networks, the influencing factors of users behavior are extracted, and different factors are modeled as factor functions, then the a mutual link prediction model based on factor graph is establisheded. Finally, experiments on a real data set are completed, to verify the feasibility and effectiveness of the model.Aiming to the characteristics such as dynamic change and short duration time of location proximity, this paper intends to study the problem of multi-user location proximity in the offline ephemeral social networks. First of all, the paper puts forward relevant concepts in the offline ephemeral social networks and gives the problem definition. Then, it designs the overall framework for multi-user location proximity prediction, including collecting network segments, constructing overlay networks, network filtering and dynamic cluster prediction algorithm. The paper puts forward a maximal close subgraph mining algorithm based on the idea of network splitting, to help predict multi-user location proximity. The mining algorithm uses the weighted edge betweenness as the basis of splitting, and uses the aggregate density as the termination condition of spitting, which can effectively solve the uncertain problems of the number of location proximity relations and the number of users in each location proximity relation. Finally, the experiments on two real datasets verify the feasibility and efficiency of the suggested prediction strategy.