PayPal is at the forefront of applying large-scale graph processing and machine learning algorithms to keep fraudsters at bay. While deep learning algorithms have found enormous success in the application to unstructured data, their application to structured data, such as graphs, is in the nascent stage of research. One of the emerging research area is graph-based representation learning where features are learned automatically from network structure.
In this talk, I'll present how this emerging area can help to automatically learn features from a large scale payment network. Learned features are used to build machine learning models that are used to prevent complex fraudster activities where they collude with several other actors to commit fraud. I'll elaborate on specific challenges in applying this state-of-the-art machine learning algorithm to large scale payment network. Results from experiments conducted on a very large graph data set containing billions of edges and vertices will be presented.