In this conversation, Hua from TikTok discusses the complexities of large transaction modelling and fraud detection. He emphasises the importance of combining traditional rules with advanced models, the challenges of managing numerous signals, and the evolving role of AI in detecting fraud. The discussion also included detection models.

Key Takeaways

• Fraud detection combines traditional rules and advanced models.
• Rules are quicker and easier to explain than models.
• Managing over 10,000 signals is a challenge in fraud detection.
• Deep learning enhances the ability to detect sophisticated fraud.
• Models are currently better than humans at detecting fraud.
• Building effective measurement systems is crucial for success.
• The economics of AI in fraud detection is still challenging.
• Data quality is essential for effective model training.
• Future models will leverage more features and signals.
• The industry is moving towards large-scale models for fraud detection.

Chapters

00:00 Introduction to Large Transaction Modeling

05:16 Managing Signals in Fraud Detection

10:06 AI vs. Human Decision Making in Fraud Detection

15:09 The Future of AI in Fraud Detection

20:13 The Need for Better Measurement Systems

#frauddetection #largetransactionmodeling #AI #machinelearning #signals #deeplearning #dataquality #measurementsystems #TikTok #paymentrisks