Recommender Systems for Online Classi ed Advertisements Vasily A. Leksin vleksin@avito.ru 0 Recommender Systems Unit, AVITO.ru , Moscow , Russia

This talk is about our journey from simple linear models to neural networks in the development of Recommender Systems in Avito and the current recommendations architecture. I share successes and failures in building highly loaded Recommendation Systems. We discuss the gap between the latest scienti c approaches and methods applicable in production systems and why we claim that the \keep it simple" principle works well in production systems.

 Reccomnder Systems Classi ed Advertisements high load 1. Rubtsov , V. , Kamenshchikov , M. , Valyaev , I. , Leksin , V. , Ignatov , D.I.: A hybrid two-stage recommender system for automatic playlist continuation . In: Proceedings of the ACM Recommender Systems Challenge 2018 . RecSys Challenge ' 18 , New York, NY, USA, ACM ( 2018 ) 16 : 1 { 16 : 4 2. Leksin , V. , Ostapets , A. , Kamenshikov , M. , Khodakov , D. , Rubtsov , V. : Combination of content-based user pro ling and local collective embeddings for job recommendation . In: CEUR Workshop Proceeding , Vol. 1968 , International Workshop on Experimental Economics and Machine Learning (EEML 2017 ). CEUR-ws ( 2017 ) 9 { 17 3. Leksin , V. , Ostapets , A. : Job recommendation based on factorization machine and topic modelling . In: Proceedings of the Recommender Systems Challenge. RecSys Challenge '16 , New York, NY, USA, ACM ( 2016 ) 6:1{6:4 4. Leksin , V.A. , Nikolenko , S.I. : Semi-supervised tag extraction in a web recommender system . In Brisaboa, N., Pedreira , O. , Zezula , P., eds.: Similarity Search and Applications , Berlin, Heidelberg, Springer Berlin Heidelberg ( 2013 ) 206 { 212