A network characterization algorithm stability measurement method

Abstract

The invention relates to a network characterization algorithm stability measurement method, which comprises the following steps of: utilizing a network characterization algorithm to keep the characteristic of a relation between network nodes, and judging the stability by using whether a nearest neighbor set of nodes in a characterization space is consistent or not; Mapping a network into a multi-dimensional continuous dense vector space by adopting a certain representation method, and calculating the first K nodes closest to each node as the basis of stability measurement by utilizing cosine similarity; a plurality of vector spaces are generated for multiple times by adopting the same representation method for one network, and the similarity of nearest neighbor sets in the vector spaces iscalculated for one node, including the coincidence rate of the nodes and the ranking information of the coincidence nodes. According to the method, the stability performance of different algorithms on different network sets can be effectively measured, so that the influence of any factor on the stability of network characterization and the influence magnitude can be further disclosed.

Description

technical field [0001] The invention belongs to the field of network and graph representation algorithms, and relates to a method for measuring the stability of network representation algorithms. Background technique [0002] Graph-type data structures are basic discrete representations of social, biological, and information networks, but are difficult to generalize to machine learning tasks that require continuous features. So recently researchers have proposed a series of methods for learning continuous representation of nodes (retaining relationship information between nodes). Network representations have proven to be effective in downstream tasks such as node classification and clustering, link prediction, and network alignment. However, there are also certain problems in the network representation method. The Zügner method (refer to Zügner's method: Daniel Zügner, Amir Akbarnejad, and Stephan Günnemann.2018.Adversarial attacks on neural networks for graph data.InProc.o...

AI Technical Summary

Problems solved by technology

Through experiments, we found that this instability also exists in the network representation space. Compared with the word vector, the instability of the network representation algorithm has specific influencing factors, such as the DeepWalk method (refer to the DeepWalk method: Bryan Perozzi, Rami Al-Rfou , and StevenSkiena.2014.Deepwalk:Online learning of social representations.InProc.ofSIGKDD.ACM,701–710.) uses a random walk model to construct the interrelationships between nodes, which has an inherent uncertainty, which leads to instability in the representation space

Images