Research Of Role Discovery Algorithms Based On Complex Network Node Structural Features

In complex network research,compared to community detection and link prediction,role discovery is an emerging area that allows analysis of complex network in an intuitive way.Role discovery refers to the process of dividing nodes into and equivalent node class.Roles represent node-level connectivity patterns such as star-center nodes,star-edge nodes,near-cliques or bride nodes.Roles have been mainly of interest of sociologists,but more recently,roles have become increasingly useful in other domains.The methods for role discovery is mainly divided into two classes,graph-based and feature-based role discovery.In graph-based role discovery method,roles are computed directly from graph representation,which is typically in the form of adjacency matrix.This is contrast to feature-based roles which are computed from the graph by transforming the graph representation into feature-vector representation,and nodes are assigned to the same role if they share similar feature vector.This paper studies methods based on feature-based role discovery.Firstly,we introduce an role discovery algorithm which studies the feature-based role discovery problem.Then,this paper aims at the shortcomings that most feature-based role discovery methods only consider the local and neighbor structural information of network,ignoring the global structural information.This paper introduce both local and global structural features to build the node feature representation,which allows our algorithm better performs than other role discovery algorithms in the global properties.At the same time,this paper proposed a role discovery algorithm evaluation method combining the NodeSense method based on the evaluation framework proposed by RID?Rs algorithm.Our method uses different node importance properties to evaluate the performance of role discovery algorithm in an intuitive way.Finally,this paper compares proposed role discovery algorithm to other role discovery algorithms using proposed evaluation method on different social network datasets.Experiments shows that our algorithm performs better than other algorithms in both local and global properties.