Han Zhichao: real time risk control practice of eBay based on graph neural network

This paper mainly introduces the application of graph neural network in some real-time scenarios of payment risk control , Focusing on the practice of point and edge construction at the data level , And skills at the model architecture level . Graph operators are completely used in the model to construct the concept of space-time , It can be reused in most risk control fields , Or a decision-making scenario with a certain time interval , And all relatively new developments in the field of graphics can be embedded in the framework , Pre training including graph structure , New network architecture , And the scheme of large-scale graph cutting at the deployment level .

Sharing includes the following parts ：

• Anti fraud scenario in e-commerce payment

• The challenge of real-time image experiment composition

• Feasible end-to-end solutions at this stage

• Work summary and outlook for the future