A Memic algorithm-based network representation learning method

Abstract

The invention discloses a Memic algorithm-based network representation learning method. The method comprises the following steps of: randomly generating real number coding individuals to form an initial population P0 according to the number n of network nodes and a vector dimension d; evaluating a fitness function value of an initial population P0 real number coding individual; performing iterative optimization on each real number coding individual of the initial population P0 by adopting a Memic algorithm on the basis of the fitness function value of the real number coding individual; outputting the real number coding individual with the highest fitness function value in the population after iterative optimization as a representation vector set of the network node; wherein the iterative optimization comprises performing cross probability iterative optimization, population random number variation probability iterative optimization and local search iterative optimization on real numbercoding individuals in an initial population P0 in sequence; experimental results show that the method can effectively encode the network structure information, especially the community structure information, into the representation vectors, and can be used for tasks such as node classification, community detection and visualization.

Description

technical field [0001] The invention belongs to the technical field of social network calculation and representation learning, and in particular relates to a network representation learning method based on a Memetic algorithm, which can be used for tasks such as node classification, community detection and visualization. Background technique [0002] Network representation learning is a technology that embeds each node in the network into a low-dimensional dense vector space to obtain a vector representation of the network. Due to the shortcomings of traditional network representation methods, such as adjacency matrix, which are sparse and difficult to reflect the potential relationship between nodes, network representation learning technology has increasingly attracted the attention of relevant experts and scholars. Compared with the traditional network representation, by preserving the topology information of the network, the network representation obtained by embedding th...

AI Technical Summary

Problems solved by technology

However, the network representation obtained by this method has no obvious distinction, that is, the distance between classes is small, which will increase the difficulty of subsequent model analysis networks, and the linear sequence acquisition method of random walk makes it difficult for these methods to maintain the nonlinearity of network nodes characteristics, such as the community of the network

[0004] Since the above-mentioned network representation learning method only considers the low-order similarity, the learned representation vectors are not clearly differentiated, resulting in limitations in the performance of network analysis tasks.

Images