A financial anti-fraud method and device based on heterogeneous graph deconstruction

Abstract

The present invention provides a financial anti-fraud method and device based on heterogeneous graph deconstruction, and relates to the technical field of financial anti-fraud. The method includes the following steps: extracting entities that need to be associated in sample data, constructing relationships between entities, and Convert to a heterogeneous graph; construct at least one meta-path based on the entity, perform node walk sampling on the heterogeneous graph according to the meta-path, and deconstruct the heterogeneous graph to obtain sampling samples of nodes; wherein, the meta-path is composed of symmetrical entities, and the meta-path The entities in the path correspond to the nodes of the meta path; construct at least one isomorphic graph based on sampling samples; extract the graph features of the isomorphic graph, and build an anti-fraud model based on the graph features; input the data to be processed into the anti-fraud model, and get the anti-fraud model. The anti-fraud analysis results output by the fraud model, the invention fully utilizes the entity information contained in the heterogeneous graph, effectively replaces the expert experience to formulate rules, and constructs a general adaptive financial anti-fraud solution.

Description

technical field [0001] The invention relates to the technical field of financial anti-fraud, in particular to a financial anti-fraud method and device based on heterogeneous graph construction. Background technique [0002] In the field of Internet financial anti-fraud, different from traditional anti-fraud, various transaction behaviors on the Internet are virtual, so for Internet financial anti-fraud, the processing of data is particularly critical. Especially in anti-financial fraud fields such as financial loans and transaction fraud, it is usually necessary to verify user information for fraud risk management and control. For example, anti-fraud financial fields such as financial loans and transaction fraud usually require verification, analysis and mining of user information for fraud risk management and control. The existing loan business verification mode usually adopts manual combined with expert system or artificial intelligence model. The above two modes have som...

AI Technical Summary

Problems solved by technology

In recent years, there have been many studies on heterogeneous graphs. The utilization of heterogeneous graphs is mainly divided into two ways. One is to use graph representation algorithms to directly calculate the relevant characteristics of nodes in heterogeneous graphs. The problems involved are: It is generally difficult for the representation algorithm to achieve real-time update and real-time calculation of the features on the entire graph, and the graph representation algorithm generally consumes more resources, and there are many difficulties in the application perspective; the second is to use expert experience to formulate some rules for heterogeneous The relationship or nodes in the graph are reduced to convert it into an isomorphic graph for the next calculation, but expert experience is not universal in different products and scenarios, and the rules that may be known for different products are not applicable. Involving the development of new rules requires a lot of manpower and material resources, and the expert experience generally only focuses on existing entities. For the addition of unknown entities, the rules need to be redeveloped. The process is cumbersome and there may be errors. Most of the reductions use only one The entity-related isomorphic graph ignores the information that may be implied by other entity isomorphic graphs. Most of the reduction rules are also regarded as commercial secrets, which are difficult to implement in practice.

Images