Applications of Graph in Fraud Detection
summary
Graphs convert PlayStation tables into networks at 62M consoles and 100M+ users and extract concrete signals — PageRank and degree centrality, weakly connected components and label propagation — fed alongside tabular features into existing models and rules. A real‑time graph feature service computes scores at click time, flags anomalies, temporarily bans nodes for 24 hours and escalates cases. The deployed approach reduces unseen fraud despite delayed chargeback labels and adversarial pattern drift.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 2:58Agenda
- 5:15What is a graph
- 8:29Graph structure
- 12:27Graph algorithms
- 14:52Importance algorithms
- 16:18PageRank explained
- 21:19Subgraph algorithms
- 23:32Fraud detection basics
- 27:27PlayStation fraud flow
- 30:15Fraud detection challenges
- 32:24Graph for fraud detection
- 35:45Graph features for ML
- 36:49Anomaly detection with graph
- 39:37Real-time fraud prevention
- 41:21Label propagation
- 42:25Q&A
related talks
28mA Quantum Leap In Fraud Detection Accuracy Using Graph-based Features
26mAdvancing Financial Fraud Detection with Graph Dbs
28mFraud Graph with Amazon Neptune and Machine Learning
56mGraph Analytic Systems
46mGraph Algorithms: Predict Real-World Behavior
12mCloud-Native Friendly Fraud Detection
23mGraph ML – The Next Level Of Machine Learning; Learn How To Use It And When Not To
31mLet's do graph machine learning on Cloud