Universal Graph Embeddings for Transaction data

Here at Mastercard, we have a transaction data where an account (card holder) transacts at a merchant, this data can be viewed as a Bipartite Graph of account and merchant nodes, where each edge represent a transaction between them. We have multiple models/tasks that provide predictive intelligence at an account level, e.g., predicting an account likely to experience a fraud, predicting an account likely to raise a chargeback, and predicting an account likely to default in near future. Goal of this work is to learn universal embeddings of account and merchants that can be leveraged to improve multiple downstream models.

Faculty Supervisor:

Dionne Aleman

Student:

Partner:

Mastercard

Discipline:

Computer science

Sector:

Professional, scientific and technical services

University:

University of Toronto

Program: