Machine Learning to moderate ads
summary
Claims a deployed system that moderates 100+ million classifieds per month using concrete choices: feature hashing (about 200,000 buckets for description and 50,000 for title), SVM-light style files, XGBoost models and automated pipelines with Luigi pipeline + Docker deployment. Batching predictions via Kafka, Flask model servers, sampling live predictions for human review, controlling error via probability thresholds, and daily monitoring of hundreds of country-specific models are the things that prove it works.
AI-written summary · about
chapters (7)
- 0:00ML ad moderation intro
- 4:20Problem statement and OLX
- 10:11Offline model creation
- 15:40Model performance metrics
- 18:56Productionizing ML model
- 23:58Monitoring in production
- 26:28Q&A
related talks
48mMachine Learning in the MAIL.RU Advertising System
27mLarge Scale Response Prediction to Serve Ads on LinkedIn
19mMachine Learning for Privacy-Preserving Ads
54mOn How Machine Learning and Auction Theory Power Facebook Advertising
58mReal Data Science Examples at OLX's Online Marketplace
29mAI and ML for Computational Advertising
39mDeveloping Machine Learning for Impact
25mML model from an idea to production with the help...