Abnormity detection method based on heterogeneous information network element learning architecture

Abstract

The invention discloses an anomaly detection method based on a heterogeneous information network element learning architecture. The anomaly detection method comprises the following steps: sampling sub-graphs from an original graph; sampling a plurality of support graph instances in the support set; iterating the sampling graph, generating structure node embedding in each sub-graph, and inputting the structure node embedding into a heterogeneous graph neural network module for training and updating; task embedding is obtained according to node embedding; calculating a support loss function and updating parameters; sampling a plurality of query graph instances, and learning node embedding and querying a loss function according to the same training process as the support set; task weights are calculated, anomaly detection is carried out, and anomaly labeling is carried out on corresponding products or nodes of the original graph. According to the invention, migration can be carried out among different heterogeneous information networks; a structure module, a heterogeneous module and a comparison module are adopted to capture structure information, heterogeneous features and unmarked information of sub-graphs respectively, and the technology is obviously superior to the latest technology on multiple heterogeneous information networks.

Description

technical field [0001] The invention belongs to the technical field of artificial intelligence, and in particular relates to an anomaly detection method based on a heterogeneous information network meta-learning framework. Background technique [0002] Heterogeneous Information Networks (HINs) are ubiquitous. Well-known examples are social networks, knowledge graphs, and recommender systems, all of which are composed of multiple types of nodes and edges. Unlike homogeneous networks, which assume that each node is a single type, heterogeneous information networks have more ways to describe the network. This provides a more efficient solution to data mining and knowledge discovery tasks, such as node classification, link prediction, and anomaly detection. [0003] Representation learning is a necessary prerequisite for mining heterogeneous information networks. Recent studies have achieved relatively good results using graph neural networks (GNNs). In heterogeneous informa...

AI Technical Summary

Problems solved by technology

In the present invention, the above-mentioned meta-learning framework for the few-sample learning problem on the heterogeneous information network is called META-HIN

Images