RecSys 2021 - Papers and Talks to Chew on
Eugene Yan
RecSys 2021 conference (September 27 - October 1) featured several papers on recommendation systems, including work on higher-order interactions for collaborative filtering, comparisons between matrix factorization and neural approaches, serverless deployment patterns, and transformer-based sequential recommendation models. Key findings included that simple linear models with higher-order interactions matched deep learning baselines on datasets like MovieLens-20M, matrix factorization outperformed neural collaborative filtering on accuracy metrics, and that focusing on data quality and managed cloud services improved deployment efficiency. The research advances practical methods for building, operating, and improving recommendation systems at scale.
Why it matters
Simple baselines, ideas, tech stacks, and packages to try.
