Abstract :

Session-based subsequent item a user will engage with, based on a sequence of her prior interactions with previous things. This machine learning topic addresses a fundamental scenario in e-commerce systems, which seek to propose appealing goods for purchase to customers navigating the site. Session-based recommenders are challenging to scale because of their exponentially vast input space of possible sessions. This hinders offline precomputation of recommendations and necessitates the maintenance of state throughout the online calculation of subsequent item recommendations. We present VMIS-kNN, an adaption of a cutting-edge nearest neighbor methodology for session-based recommendation, which utilizes a preconstructed index to provide next item suggestions with minimal latency in contexts involving hundreds of millions of clicks to sift through. Utilizing this methodology, we develop and deploy the scalable session-based recommender system Serenade