The Research Of Link Prediction And Signed Network Community Detection

In recent years, with the rapid development of the Internet as the representative of the network of information technology, human society has entered the era of complex networks. In life, many complex systems can be abstracted into complex network, and then convert into a graph in order to study. In recent years,community structure as one of the most important properties of the network has attracted more and more research people. Understanding the community structure of the network can not only help to analyze the topological structure of the network but also have important practical value, for example, in the social network we can recommend to each other who belong to the same community but don’t know each other. Link prediction as a bridge which connect the complex network and the information science has attracted more and more attention. The related research of the link prediction can not noly promote the development of the network science and the information science theory, but also has great value of the practical application, for example, it also can be used to guide the protein-protein interaction experiment and the online social recommendation.However, now most link prediction algorithms consider only the local information or path information of the network and in general rarely take the structure information into account. This paper proposes a link prediction algorithm which considers the structure of the network based on dynamics and apply this algorithm on the complex networks. I also study the structure of the signed network and put forward a community detection algorithm based on the improved modularity. The main work of this paper is as follows:1.In order to study the influence of the structure property to the link prediction results, mainly the relationship between the clustering coefficient and the link prediction, the test was carried out in different network models respectively. The experimental results show that the link prediction results increase with the clustering coefficient of the network improved.2.Based on the Kuramoto model, the continuous phase differential equation is changed into discrete phase equation, we propose a new link prediction algorithm based on phase similarity. Because in the real social networks, the phase of two nodes which have no edge in the same community is relatively large and the phase of two nodes belong to different community which have edge is relatively small, Therefore, the effect of the link prediction only based on phase similarity is not ideal. To overcome this drawback, connecting phase similarity and CN, that is to say, the whole structure information and local information, we propose a hybrid link prediction algorithm based on dynamics and the structure of the network and do experiment in the real network and artificially generated network, compared with the existing link prediction algorithm, it proves that the method proposed in this paper is effective.3.Putting forward a community detection algorithm for signed network. Based on the modularity and properties of the signed network, we define a similarity function which special for signed network. According to the similarity function, we find the most suitable nodes which have the largest similarity to merge. In judging whether the nodes can be merged and the merge of the termination conditions, we define a new condition to judge whether the nodes can be merged by improving the existing modularity of the signed network.