Heterogeneous network embedding method for reserving label information based on node signature

Abstract

The invention discloses a heterogeneous network embedding method for reserving label information based on node signature. During network embedding, high-dimensional graph data is mapped to low-dimensional vectors so that that a machine learning algorithm is adopted to effectively the high-dimensional graph data. The method comprises the following steps: carrying out prime number dictionary mappingon all label types in a network; extracting a neighborhood label set of network nodes; constructing a node signature vector; and building network node representations. The network structure, semantics and label information of heterogeneous graphs are comprehensively utilized. The idea of digital signature and the characteristics of prime numbers are used. A network representation learning framework of the heterogeneous graphs is constructed, reservation of network node and edge label information on the heterogeneous graph is realized, subsequent node clustering, classification, link prediction and other machine learning tasks are performed according to learned heterogeneous network node representations, and the existing homogeneous and heterogeneous network embedding method can be universally expanded and improved.

Description

technical field [0001] The invention belongs to the technical field of graph data processing. Background technique [0002] In today's information age, data is often represented by a network graph model in various applications. Effectively analyzing the structural information and label information in graph data helps to discover the internal relationship of complex network data, which in turn helps to make effective use of the hidden information in graph data, including but not limited to scientific search, personalized recommendation, etc. With the development of artificial intelligence technology, machine learning algorithms provide a general and effective means for data analysis and prediction. [0003] Due to the nature of graph data, it is difficult to perform data analysis directly on raw graphs. On the one hand, the traditional way of storing graph data in an adjacency matrix is ​​difficult to directly use as the input of machine learning algorithms for data analysi...

AI Technical Summary

Problems solved by technology

[0007] The purpose of the present invention is to solve the existing heterogeneous network embedding method's dependence on professional domain knowledge of network data, dependence on random walk strategy selection, and insufficient consideration of label information independence, and meet the requirements of improving the accuracy of downstream machine learning tasks Requirements, propose a heterogeneous network embedding method based on node signatures to preserve label information

Images