Credit fraud detection method and its model training method, device and server

Abstract

The present invention provides a method, device and server for training a credit fraud detection method and a model thereof, wherein, in the training method for a credit fraud detection model, a customer statistical feature vector and a customer relationship map are determined by using acquired credit customer data; According to the state label data corresponding to the credit customer data, the node sequence in the customer relationship graph is converted into a node label sequence; then the skip-gram algorithm is used to convert the node label sequence into the feature vector corresponding to the customer relationship graph; according to the statistical feature vector and The feature vector of the customer relationship graph is used to train the initial model until the number of iterations of the training meets the preset number of iterations threshold, and the credit fraud detection model is obtained. The credit fraud detection model obtained by the invention utilizes the relational graph network structure, the label feature and the statistical feature vector, which can better identify the customer's fraud risk, thereby improving the reliability of the risk detection result.

Description

technical field [0001] The present invention relates to the technical field of Internet credit risk control, in particular to a credit fraud detection method and a training method, device and server for its model. Background technique [0002] With the development of knowledge graph and social network technology and its wide application in the Internet field, many Internet financial institutions have also begun to apply relationship graphs to the Internet credit field; for example, visual display of character relationships to assist manual pre-loan approval; Artificially summarize several derived variables from the map based on expert experience, such as first-degree correlation features, second-degree correlation features, etc., or extract such things as out-degree, in-degree, centrality, and PageRank (page level) from social networks based on graph theory features for application scoring or anti-fraud; these methods are based on the network structure of the relationship gr...

AI Technical Summary

Problems solved by technology

[0002] With the development of knowledge graph and social network technology and its wide application in the Internet field, many Internet financial institutions have also begun to apply relationship graphs to the Internet credit field; for example, visual display of character relationships to assist manual pre-loan approval; Artificially summarize several derived variables from the map based on expert experience, such as first-degree correlation features, second-degree correlation features, etc., or extract such things as out-degree, in-degree, centrality, and PageRank (page level) from social networks based on graph theory features for application scoring or anti-fraud; these methods are based on the network structure of the relationship graph to extract certain information features. When faced with a large number of complex networks, there must be certain information missing in this process, and it cannot Fully express the structural information of the network, resulting in Internet credit detection only by the network structure of the relationship graph, and the reliability of the obtained results is poor

Images