Multi-layer network community mining algorithm combined with non-negative matrix factorization

Abstract

The invention provides a multilayer network community mining algorithm combined with non-negative matrix factorization. The invention comprises the following steps: 1, acquiring an experimental data set; 2, evaluating indexes; 3, performing an algorithm process; 4, updating and optimizing the algorithm; 5, performing algorithm evaluation. According to the method, the sparsity constraint is added in each step of matrix decomposition process, so that the non-negative matrix decomposition of the algorithm core reduces the sensitivity to an initial value in the optimization process, and the influence of noise and an abnormal value is weakened in each iteration, so the good stability and average performance can be maintained, and the algorithm is not liable to fall into local optimum; and the multi-layer network community mining performance and stability of the algorithm under the NF-CCE framework can be improved.

Description

technical field [0001] The invention belongs to the field of multi-layer network community mining and multi-view data clustering, and in particular relates to a multi-layer network community mining algorithm of joint non-negative matrix decomposition. Background technique [0002] As the basis of network science, complex network is an important tool for analyzing complex systems and relationships. It analyzes the network topology by abstracting entities into nodes and connecting entities with abstract nodes and edges. Community structure (Community) is an important topological characteristic of complex networks, which has the characteristics of close connection between nodes of the same type and sparse connection between different nodes. Community mining is to find out the community structure based on the hidden topological information in the complex network. It can be applied to information labeling, virus prevention, behavior prediction, etc. It is also useful for research...

AI Technical Summary

Problems solved by technology

[0009] (1) The core of CSNMF is symmetric non-negative matrix decomposition, which is sensitive to the initial value in the optimization process, so the decomposition result is easy to fall into local optimum, and it is easy to produce floating;

[0010] (2) Each row of the strict community indicator matrix represents a node. The value of the community number corresponding to the node is 1, and the value of the rest of the positions is 0. In CSNMF, the community indicator matrix is ​​approximated by the consensus matrix H, which cannot be guaranteed except that the community corresponds to The values ​​outside the position are as many as 0 as possible, so it will affect the final result

[0011] This makes the results of the CSNMF algorithm fluctuate greatly when performing community mining, and it is not easy to achieve the global optimum and obtain the best community division effect

Images