Anomaly Research Based On Evolutional Mechanisms Of Dynamic Social Networks

With the rapid development of various online social application services,the scale of existing dynamic social networks based on social application services has been expanding,which makes the social connection between people more and more close.In this context,the sharing speed of information is accelerated,the communication scope is widening,a nd the number of participants is increasing.The effects of social emergencies become more uncontrollable,which may cause dramatic changes in the social network structure and even affect the harmony and stability of the society.The anomaly research of dynamic social networks built on real social data is of great significance to guide the development of social networks in a healthy and stable way.The existing anomaly research of dynamic social networks is mainly divided into two categories: text-based analysis methods and network-based feature statistics methods.The former processes text information from different dimensions and extracts valuable information of anomalies.However,the social relations in real dynamic social networks have diverse expression forms.Due to the social way or the protection of user privacy,dynamic social networks cannot offer valid text information in many cases,which greatly limits the application scope of this method.The latter selects specific network feature parameters,analyzes the changes in the values of network feature parameters,and tracks the evolution state of social networks.However,there are many existing network feature parameters,and it is difficult to make a fair selection of potential network feature parameters.At the same time,dynamic social networks usually follow specific evolutional mechanisms in their evolutional processes.As an important attribute of the dynamic social network,the evolutional mechanism is an intrinsic driving force for the structural change in the dynamic social network.It is acknowledged that many structural characteristics of real dynamic social networks are determined by their evolutional mechanisms.Many social network evolution models are proposed to simulate the evolutional process of real dynamic social networks.Different from the existing methods,this dissertation,for the first time,analyzes the anomalies of dynamic social networks from the perspective of evolutional mechanisms,and calls such anomalies as evolutional anomalies of dynamic social networks.The following three major issues are addressed.(1)The existing evolutional mechanisms of dynamic social networks are mostly abstract descriptions in theory.For large-scale dynamic social networks,how to effectively analyze their evolutional mechanisms,and detect and evaluate their evolutional anomalies?(2)The existing analysis of dynamic social networks usually assumes that all nodes in the social network follow the same evolutional mechanism.However,different nodes in a real social network often follow different evolutional mechanisms.How to evaluate the diversity of the nodes' evolutional mechanisms in dynamic social networks and conduct an effective demonstration?(3)In the evolutional process of dynamic social networks,how to fit the evolutional mechanisms of nodes at the micro-level,and detect and evaluate their evolutional anomalies? Therefore,the main work of this dissertation is divided into the following three parts.(1)The detection and evaluation of the evolutional anomalies of the macro dynamic social network;(2)The study on the diversity of the nodes' evolutional mechanisms in the dynamic social network;(3)The detection and evaluation of the evolutional anomalies of micro nodes in the dynamic social network.Furthermore,the second part of our work is the bridge between the first part and the third part.Only when the diversity of the nodes' evolutional mechanisms has been reasonably demonstrated,it is necessary to analyze the evolutional anomalies of dynamic social networks from macro social networks to micro nodes.The contributions of this dissertation mainly have the following four aspects.(1)It introduces and defines three problems of the anomaly research in the dynamic social network from the perspective of evolutional mechanisms.They are the problem of the detection and evaluation of the evolutional anomalies of macro dynamic social networks,the problem of the diversity of the nodes' evolutional mechanisms in the dynamic social network,and the problem of the detection and evaluation of the evolutional anomalies of micro nodes in the dynamic social network.(2)It proposes an uncertain evolutional superposition state construction algorithm for the macro dynamic social network.Different link prediction algorithms are used to analyze the evolutional mechanism of the macro dynamic social network,and the evolutional superposition state is constructed to represent the evolutional state of the macro dynamic social network.The optimal evolutional observation algorithm is proposed to maximally reflect the evolutional fluctuations in the evolutional processes of social networks,detect the anomalous fluctuations,and evaluate the anomalous degrees.(3)It improves the existing link prediction algorithm from the perspective of the edge's nodes,and proposes the edge generation coefficient to evaluate the existence rationality of the edges of a node.Different link prediction algorithms are applied to indirectly explain the rationality of the existence of different edges in the dynamic social network,and the optimal link prediction algorithm is determined for each edge.The assumption of the diversity of the nodes' evolutional mechanisms is formally proposed,and the evolutional diversity distance is defined to quantify the diversity of the nodes' evolutional mechanisms in the dynamic social network.In addition,this dissertation proposes a link prediction algorithm based on the diversity of the nodes' evolutional mechanisms,which indirectly demonstrates the rationality of the diversity of the nodes' evolutional mechanisms.(4)From the perspective of micro nodes,the existing theoretical framework and evaluation methods of link prediction are introduced to fit the evolutional mechanisms of nodes relfected in the edge generation and removal processes in dynamic social networks.The node's evolutional fitting vectors in different periods are constructed to track the changes in evolutional mechanisms of nodes,and detect the evolutional anomalies of nodes.The generation and removal processes of the node's edges with anomalous evolutional mechanisms are seen as a disturbance to the dynamic social network structure.The further disturbance analysis is carried out on the adjacency matrix of the current structure of the dynamic social network,and the structural disturbance index is proposed to quantitatively evaluate the effect of nodes with evolutional anomalies.