Knowledge graph-based data abnormal transmission detection method

Abstract

The application discloses a data abnormal transmission detection method based on a knowledge graph, relates to the technical field of data transmission detection, and comprises the following steps: inputting a context-associated heterogeneous graph subgraph into a pre-trained heterogeneous graph neural network, performing aggregated calculation on the features of multiple types of nodes and relationships in the subgraph through a heterogeneous attention mechanism, generating a semantic embedding vector of the transmission event, triggering an abnormality alarm if an abnormality score exceeds a preset determination threshold, and performing reverse backtracking based on the attention weight output by the heterogeneous graph neural network, generating and outputting an explainable semantic association path corresponding to the abnormal transmission event; the method automatically performs reverse backtracking with the target transmission node as the starting point based on the node-level attention weight coefficient output by the heterogeneous graph neural network, retains the entity nodes and corresponding relationship edges with the highest contribution to abnormality determination in each hop, and splices them in accordance with the business causal logic sequence of identity authentication, behavior triggering and data flow.