Research On Analysis Of Dynamics Oriented Link Prediction Methods In Social Networks

As one direction of the data m ining field, link prediction has been extensively studied by many researchers. It predicts existent yet unknown links and future links for mining hidden network relations and analyzing the evolution of network structure by the network structure information. The link pred iction methods that have been known just always use network structure inform ation, however, because of that social netw ork is dynamic, so the performance of the methods based on static network structure is bad.Based on dynam ics of social network, this article summa rizes the m ain characteristics of m obile social networ k and scientific collaboration network. Considering the history inform ation of the node’ connects, network tim e and network structure and other factors, this article proposed Human Behavioral Rhythms Analysis(HBRA) and Sub-graph Evolution(SE) li nk prediction methods by introducing local information-based similarity index and sub-graph theory of evolution, aims at seeking to the balance between accuracy and effi ciency which building the foundation for high-reliable link Prediction.The interaction between nodes th at is often closely associ ated with their previous pattern of behavior over tim e is producing the links. Based on the idea of com bining historical network structure inform ation with now, this article proposed a m odel for mobile social network and link prediction, an d then defined the shortest path betw een nodes to extract network topology. This paper defined the connection gravity based on the model and converted the link predic tion problem to connection gravity value calculation that was solved by an algorithm in continuous tim e in dynam ic social network. The experiments prove that the H BRA algorithm can predict links between nodes accurately in mobile social networks.The collaborations between scientists often come about in the for m of teamwork, in addition, the relationship between two scientists having been affected, so predicting links in scientific collaboration network have to consider the impact of the evolution of node. Since the triangular relationship of thr ee scientists is the sim plest form that reflecting the influence and the trivial sub-graphs composed by three nodes are sensitive to the changes of network stru cture, so this ar ticle presents a link p rediction method based on SE. First, in order to get a network with high clustering coefficient, filtering some nodes that is not useful by analyzing the rhythms of scientific collaboration. Then, we calculate the transition probability matrix of sub-graph to determ ine the possibility of links. Because of the division of sub- graph, SE link prediction m ethod which is effective and stable gets the prediction accuracy with efficiency.