| There are often a large number of community structures in real-world networks.The problem of searching for a particular node’s community in a large network is called as community search problem.In real networks such as social networks,protein interaction networks,and collaborative networks,nodes often have attributes,so the practical application of attribute community search is more extensive and important.The influence is the probability of propagation between nodes calculated according to the propagation model in the network,that is,the degree of influence between two nodes.The influence has important applications in some scenarios,such as community recommendations for cold start,trend monitoring issues,and more.However,influence is rarely considered in existing attribute community search studies.Most community search methods do not take into account factors such as the tightness of the community’s structure,attribute aggregation and the degree of community influence synthetically.Community search that only considers attributes cannot guarantee the scale-out of the community;community search that only considers the influence cannot guarantee the community’s attribute aggregation.In addition,the existing methods are not applicable in the case of community search with new node as query nodes.In view of the above problems,this paper studies an influence-based attribute community search method,which comprehensively considers the influence of community and attributes,the main work and innovations of the thesis are as follows:Firstly,this paper proposes an attribute pkd-truss community model that comprehensively considers the influence within the community and the cohesion of node attributes in the community.The model defines the maximum influence path’s threshold p between query node and nodes in the community and(k,d)-truss.And this paper adopts this model as a standard for subsequent community search.Secondly,this paper proposes an attribute community search method based on influence.On the basis of giving the formal definition of the influence-based attribute community search problem,the basic ideas and detailed steps of the method are elaborated,and the complexity analysis and comparative discussion of the algorithm are calculated.The method uses a graph refining method to remove the nodes in the original graph that are less influence-correlated with the query node,which greatly saves the time spent on the subsequent search process and is suitable for effective influential attribute community search in large networks.The method also adopts a score function that comprehensively evaluates the influence of outside the community and attribute aggregation.It can evaluate the candidate communities in the search process and iteratively search for the pkd-truss community with the highest score through greedy thought.In addition,a top N community sorting algorithm is presented,which can output the optimal top N communities and provides a solution to the attribute community search problem.Finally,experiments are conducted with Amazon and DBLP datasets with real communities.The results of the experiment were compared,analyzed and discussed,the correctness and feasibility of the proposed model and method are verified. |
