Auto Content Moderation in C2C e-Commerce
summary
Mercari, inc. deploys an ML pipeline for C2C item moderation since late 2017 that cuts operation cost and catches policy-violating listings rule-based checks miss, claiming 554.8% coverage increase versus rules. The system runs 50+ production models, uses image/text/category fusion and gated matching to reduce false positives, and introduces an online “offer evaluations” flow that shortens model rollout to about one business week while improving precision@K (example: +23.2%).
AI-written summary · about
chapters (8)
- 0:00Introduction to moderation
- 1:56Necessity of moderation
- 3:40Rule-based vs machine learning
- 5:32Data pipeline design
- 7:18Model training strategy
- 9:52Evaluation methodology
- 14:16System architecture
- 17:37System performance results
related talks
44mRobust AI Search Ranking for Radical C2C Marketplace Growth
33mMachine Learning to moderate ads
41mBuilding MLOps Infrastructure at Japan's Largest C2C E-Commerce Site
13mHarnessing Machine Learning to Combat E-commerce Fraud
58mReal Data Science Examples at OLX's Online Marketplace
38mE-Commerce Product Categorization at Scale
22mMachine Learning: The Engine Behind Modern E-commerce Search
22mFighting against Marketplace Scams using Community-driven AI