Calculating by IP: How to Fight Spam on a Social Network
summary
Badoo’s anti-spam system for its dating social network: how to detect and block spammers, scammers and adult-content spreaders by classifying billions of user actions in real time. The talk explains the use of machine learning models for detection and the techniques used to reduce false positives.
AI-written summary · about
chapters (20)
- 0:00Introduction
- 1:25Spam and scammer types
- 7:25Simple spam defenses
- 10:11Manual moderation limits
- 15:25IP and attribute stats
- 21:00User clustering at signup
- 24:29Message text analysis
- 28:02Proxy and anonymizer detection
- 31:50Why signup is critical
- 32:32Avoiding false positives
- 34:37Anomaly and false negative search
- 37:09Why not one big model
- 43:13System evolution and new challenges
- 47:07Detecting new spam types
- 48:39Testing new models and features
- 50:09Spammers adapting to defenses
- 51:35System response time
- 53:21Silent Hill sandbox for bots
- 55:30Particularity coefficient in action
- 58:43Q&A
related talks
32mHow Badoo Moderates 1 Billion Photos a Year
12mWelcome to Spam Fighting @Scale
34mSpam-fighting Strategy
48mWhat fighting spammers in Beeline Anti-Fraud looks like through a Data Scientist's eyes
35mTwo clicks to cluster and block thousands of spammers
34mArchitecture of an Anti-Bot Protection
47mAlgorithmic bias is everywhere (especially at Breeze) - what can we do about it?
31mStrategies for Identifying and Mitigating Bot Threats