Transaction fraud behavior monitoring method and device

Abstract

The invention provides a transaction fraud behavior monitoring method and a device, and the method comprises the steps: obtaining a target transaction information group of a target user; applying a preset feature mapping model to obtain a transaction feature vector group corresponding to the target transaction information group; inputting the transaction feature vector group into a current transaction anti-fraud model, taking the output of the current transaction anti-fraud model as a fraud behavior monitoring result of the target user, wherein the current transaction anti-fraud model is obtained by performing online updating on a previous transaction anti-fraud model by applying a previous historical transaction feature vector group, and the previous transaction anti-fraud model is an unsupervised learning model obtained by pre-training. On the basis of improving the application accuracy of the transaction anti-fraud model, the application real-time performance of the transaction anti-fraud model can be effectively improved, and then the real-time performance and accuracy of transaction fraud behavior monitoring can be effectively improved.

Description

technical field [0001] The present application relates to the technical field of artificial intelligence, in particular to a method and device for monitoring transaction fraud. Background technique [0002] With the rapid development of the financial industry, fraudulent behaviors targeting the transaction process have also come along, leading to increasingly serious crimes and affecting the safety and interests of financial sectors such as banks. Therefore, monitoring transaction fraud is a challenge faced by financial sectors such as banks. Long and arduous task. [0003] The current method of monitoring transaction fraud is mainly to use the traditional anti-fraud model for monitoring. The anti-fraud model is a machine learning model based on the classification algorithm. The classification algorithm can be logistic regression and random forest algorithm, etc., and requires manual periodic Anti-fraud model training for a large number of accounts. [0004] The applicatio...

AI Technical Summary

Problems solved by technology

[0004] The application of the classification algorithm needs to label the samples in advance, so the training process of the anti-fraud model is an offline training process. With the continuous updating of fraud means, the anti-fraud model trained offline may not be able to obtain the latest information in time and cannot adapt to real-time In addition, as online payments become more frequent, the transaction behavior of a single account may also change over time. In the technology, artificially conduct anti-fraud model training on a large number of accounts periodically, ignoring the change of transaction behavior of a single account over time, and the time-consuming and labor-intensive efficiency is low, which leads to low accuracy of transaction fraud detection

Images