Maxim Kashirin, Data Science Unit Lead, Avito.
The speaker has worked in Avito moderation for more than five years: he joined as a DS engineer and later became head of the unit. The talk is about moderation platformization, primarily from an ML engineer's point of view. Plan: the context of moderation, how things were historically, what problems this caused, and four clusters of solutions: data collection (in anti-fraud and moderation the sample is biased), universal text and image embeddings with classifiers on top, automatic online threshold selection for models (distributions in the flow change), and real-time monitoring of model behavior.
Google or a one-time email code. No password.