Graph Technology for Realtime Fraud Detection and Prevention
Speaker: Chenxu Ma
Time: 15:45 - 16:45
Location: Lecture Hall Pi
Zoom: link
Abstract: Conventional fraud detection methods typically rely on simplistic systems that block users based on predetermined rules, but these systems often fail to take into account the vast amounts of historical data and relationships between data points. By representing our data in the form of a graph, we are able to retain this information and make more informed insights.
At Booking.com, we addressed this issue by designing a system that generates graphical representations of our transaction data, and uses these graph features to train machine learning models. Additionally, we have created a visualization tool named GNet, which aids in identifying new fraud patterns and allows us to explore new connecting identifiers and graph algorithms, thus enabling us to formulate new graph features. We also built Dolos, a system that makes experimentation of new graph features very fast by allowing us to reconstruct them for historical data and train a model offline for quick impact analysis .
In this presentation, we will discuss the process of creating this in-house technological ecosystem, including how the system ingests historical and real time data, how it computes features offline and online, and our experimentation with various types of features. We will also share details on some of the particularly interesting features we discovered using our system and the things we wish to explore next using these set of technologies
Applications of (Graph) Machine Learning Techniques in Finance
Speakers: Afrasiab Kadhum and Aratrika Das
Time: 16:45 - 17:45
Location: Lecture Hall Pi
Zoom: link
Abstract: Ortec Finance is a fintech company specialized in creating mathematical modeling tools for financial decision-making. Within its R&D department, we conduct research into both classical (econometric) and machine-learning-based techniques to improve our models. During this talk, we will dive into various domains within machine learning that are relevant to our financial applications and discuss how these can intersect with graph-based approaches. Moreover, one of our former graduate students, Aratrika Das, will give an overview of her thesis on applying graph neural networks in the context of real estate valuation.