Complex Networks Field Model And Application

Complex networks are abstract representation of real-world systems and are graphs, where nodes represent entities and edges denote the relationship between entities. In this way, the generated complex networks are always static so that relevant researches on complex networks mainly focus on the topological features and always neglect the dy-namic characteristics. But the formation of complex networks is the result of all kinds of interactions (information exchange) between entities. In general, these interactions have certain strengths, directions and always have sources. In physics, this kind of in-teractions can be described by the conception of field, while field equations specifically describe the generation and propagation of field. So by endowing complex networks with field structure, we can restore the interactions in networks and comprehend com-plex networks from dynamic view. Based on the above reasons, we generalize the field theory which is in the background of space in physics to complex networks. According to the basic principle and ideas of field theory, we build the field model on complex networks which may provide a new viewpoint and method to understand complex net-works and deal with relevant problems.Based on the built field model, this dissertation further applies it to solve two hot issues in complex networks science at present:community detection and link prediction problem.Considering that most of existing community detection algorithms make use of the static topology structure but neglect the dynamic characteristics of community structure, we start from the sense that community is the basic functional module and organized group, realize that community structure is suitable for interaction (information medium) spreading quickly, by simulating the propagation of interaction in "virtual field" to de-tect community structure. In this way, we endow community detection procedure with chronology and propose the field-based community detection algorithm IFA (Informa-tion Flow Algorithm). This algorithm not only provides an effective method to find out communities, but also simulates the dynamic spreading process of information in network at the same time. Compared with classical community detection algorithms, experiments on artificial benchmarks and real-world networks show that the proposed community detection algorithm gives better results which supports the reasonableness of field model and feasibility of its application on community detection problem.In link prediction problem, classical prediction algorithms always focus on static topology to estimate node similarity to measure the link probability and seldom take the network dynamic process into consideration. We consider the interaction procedure that may happen in network, build "virtual field" and make the interaction (information medium) spread in the network. We believe that the stronger two nodes interact, the more likely a link would appear between them and propose two field-based link pre-diction algorithm:GFlow and LFlow (Global Flow and Local Flow) algorithms. These algorithms at the same time simulate the information flow process on networks. Ex-periments on real-world networks show that the field-based link prediction algorithm has better performance. Especially on some certain networks, the proposed algorithms maintain consistent performance which also supports the reasonableness of field model at the same time.