Edge community discovery algorithm based on deep sparse auto-encoder

Abstract

The invention discloses an edge community discovery algorithm based on a deep sparse auto-encoder. The method mainly comprises the following steps: constructing an edge adjacency matrix; calculating an edge similarity matrix; carrying out dimension reduction on an edge similarity matrix based on a stack type depth sparse auto-encoder; performing edge clustering based on a fast density peak searchclustering algorithm; and carrying out node community division based on the node membership degree. According to the invention, a side community discovery algorithm based on a deep sparse auto-encoderis used, side community division and node overlapping community division are realized; a novel and efficient algorithm is provided for overlapped community discovery; compared with the existing algorithm, the method has the following advantages: (1) the algorithm considers the connection relationship between the edges and the topological structure relationship between the common neighbor edges between the edges to construct the edge similarity matrix, and uses the depth sparse auto-encoder to reduce the dimension of the edge similarity matrix, so that the complexity of the edge clustering process is reduced; and (2) the algorithm uses membership degrees to divide node communities, so that a better evaluation metric value can be achieved.

Description

technical field [0001] The invention belongs to the field of complex networks, in particular to an edge community discovery algorithm based on a deep sparse autoencoder. Background technique [0002] Edge community detection algorithm is an important class of overlapping community detection algorithms in the field of complex networks, and it is one of the important means to study complex networks. Its purpose is to discover edge communities through similarity relations to obtain meaningful node communities. The nodes inside the node community are closely connected, and the nodes outside the community are sparsely connected. A good node community structure can help people better understand various types of complex networks such as social networks, information networks, and biological networks. The research and application of edge community discovery has important theoretical and application value for completing the task of overlapping community discovery. [0003] At prese...

AI Technical Summary

Problems solved by technology

In addition, when converting edge communities to node communities, more overlapping nodes will be generated, resulting in poor accuracy of edge-based community discovery algorithms

In summary, there are two main problems in edge-based community discovery algorithms: 1) The size of the edge similarity matrix is ​​large; 2) In the process of converting edge communities into node communities, too many overlapping nodes are generated

Images