Fake listing detection accuracy declines — рост fake

Ситуация: Авито detection алгоритм fake listings: precision 92% historically. Now 85%, fake reports rising. Что произошло?

Fake patterns: too-good prices, stolen photos, mass-uploaded. Detection: ML + rules + reports.

Доступные данные

• listings: id, posted_at, status (active/removed/fake)
• detections: listing_id, ts, model_version, score
• user_reports: listing_id, reason, reported_at
• manual_reviews: listing_id, ts, decision

Задачи

• Precision degradation: where missed.
• New fake patterns emerging?
• Model + rules audit.
• Recommendations: retrain / new signals / process.

Все кейсы для подготовки → ← Все кейсы