Implementation Of Heterogeneous Network Community Detection Methods

Community detection is an important unsupervised learning problem of complex network analysis,it can find all the closely connected node groups in complex networks,and it is very good to separate node groups in complex networks.Community detection plays an important role in recommendation systems and social network analysis.At present,many community detection algorithms are concentrated in homogenous networks with only one node type and one edge type.However,most networks in real life are heterogeneous networks with multiple node types and connection types.Therefore,the community detection algorithm of a heterogeneous network has greater application prospect and value.This paper studies and explores the existing heterogeneous network community detection algorithm,which mainly includes the following four parts:(1)The existing methods of heterogeneous network community detection are reviewed and divided into two categories,namely,the traditional heterogeneous network community detection method and the heterogeneous graph neural network clustering method based on deep learning.Among them,the traditional heterogeneous network community detection method is subdivided into two categories,namely,the heterogeneous network community detection method of bipartite network and the heterogeneous network community detection method of multidimensional network.Then we give a comparison of the applicability and characteristics of the existing heterogeneous network community detection algorithms and finally list the commonly used open-source data sets for researchers' reference.(2)In this paper,we propose a community detection method based on residual feature fusion and backbone degree(CDRFB),which first obtains the presentation vector with network structure information through the random-walk model,then combines the representation vector with the node's original property information to make the input characteristics of the model contains more network structure information;At the same time,we refer to the residual network,using the initial network structure embedding as the characteristics of jumping connection and the characteristics generated by the middle layer to stitch and fuse.In the case of,the distinction between nodes can still be guaranteed;We calculate the strength between nodes and communities in the network by backbone degree algorithm,so that the graph auto-encoder considers more community information when encoding,and finally uses three classic clustering algorithms for community detection.(3)In this paper,we extend the CDRFB algorithm to heterogeneous networks through the idea of meta-path and multi-path network,and propose a corresponding heterogeneous community detection method based on residual feature fusion and backbone degree(HCDRFB).HCDRFB converts the heterogeneous network into a multi-path network while preserving the semantic information of the heterogeneous information network as much as possible,and then the CDRFB algorithm is applied to each layer of the multi-path network to obtain semantic information from different meta-paths,and finally,the information from different layers is merged for community detection.(4)We designed and developed a visual analysis platform for community detection.The platform integrates the two algorithms proposed in this article and multiple comparison algorithms.At the same time,the platform implements normalized mutual information(NMI),macro-F1 scores,etc.Evaluation index.And apply D3.js visualization technology to large complex networks,visualize the results discovered by the community,and reveal the structure of complex networks in a more intuitive way.