Keeping Sensitive Data Safe Using Recommendation Systems
summary
Claim: repurposed recommendation models flag unauthorized database access by scoring each user‑table query and alerting low‑score outliers. Implementation uses implicit ALS via the implicit Python library, trains on six months of de‑duplicated daily queries with alpha=40, filters rare users/tables, and counts repeated same‑day queries once. Evaluation uses ~500 test events and 500 users with two unseen tables each, yields separation of distributions and AUC ~0.9. A production scoring+alert pipeline is built and a patent filing at PayPal documents the idea.
AI-written summary · about
chapters (11)
- 0:00Sensitive data protection intro
- 1:25Recommendation systems for security
- 4:13Recommendation systems basics
- 6:39Implicit feedback and matrix factorization
- 11:10ALS algorithm in Python
- 14:00Database access recommendations
- 16:26Anomaly detection with reversed recs
- 19:13Model accuracy validation
- 22:24Thresholds and anomaly alerts
- 23:46Challenges and future apps
- 24:48Conclusion
related talks
14mALERT: Machine Learning-Enhanced Risk Estimation for Databases Supporting...
01h 01mD1T2 - Applying Machine Learning to User Behavior Anomaly Analysis
17mWhat data should I protect?: recommender and planning support for data security analysts
33mAssurance Scoring Using Machine Learning and Analytics to Reduce Risk in the Public Sector
55mAI Security in Practice: Protecting Your AI-Powered Applications
23mFrom Data to Recommendations: Building an Intelligent System with ROR
18mTransforming Financial Decision Intelligence Through Advanced NNs
27mEnhancing Personalization Using Recommendation Systems & Multi-Armed Bandits