Training method and device of unsupervised risk prevention and control model, updating method and device of unsupervised risk prevention and control model

Abstract

The embodiment of the invention discloses a training method and device of an unsupervised risk prevention and control model, an updating method and device of an unsupervised risk prevention and control model. The method comprises the following steps: sampling service data of an online system in a target service based on a preset time interval; performing a resampling operation based on the sampledservice data to obtain target service data, wherein the resampling operation at least comprises a data trusted filtering operation; when the target service data which do not participate in training reaches a preset number, based on the preset number of target service data which do not participate in training, carrying out incremental training on a current risk prevention and control model of theonline system in the target service to obtain a target risk prevention and control model, the current risk prevention and control model being an isolated forest model; and replacing the current risk prevention and control model of the online system in the target service with the target risk prevention and control model.

Description

technical field [0001] The present application relates to the technical field of computer software, in particular to a training method, updating method and device for an unsupervised risk prevention and control model. Background technique [0002] In order to attract new users and promote active users to cultivate user stickiness, trading platforms often spend a lot of money on marketing. The activities involved are varied, including red envelopes, rewards, coupons and more. These funds will attract some illegal users to maliciously obtain marketing funds from the platform. For this type of risk, the source of the risk is usually the operation of the account owner himself, rather than malicious theft, which is an active risk. [0003] This type of risk has three characteristics: 1. It involves a large amount of capital, and a large amount of marketing funds are invested in a short period of time. Once the prevention and control is not good, it will cause huge losses to the...

AI Technical Summary

Problems solved by technology

[0003] This type of risk has three characteristics: 1. It involves a large amount of funds, and a large amount of marketing funds are invested in a short period of time. Once the prevention and control is not good, it will cause huge losses to the platform.

2. For black samples that have not been reported, it is difficult for the risk control team to observe the current risk level and carry out targeted prevention and control. Existing risk control methods can only characterize some transactions through rules afterwards to determine whether they are fraudulent marketing funds

3. Techniques mutate quickly, and the entire risk confrontation is constantly mutating for a long time

[0004] In the current risk control system, the risk prevention and control models are trained offline, and can be updated and iteratively deployed online at the fastest T+1. For risk confrontation, the response speed is far from enough

Images