A Diverse Models Ensemble for Fashion Session-Based Recommendation
Benedikt Schifferer, Jiwei Liu, Sara Rabhi, Gilberto Titericz, Chris Deotte, Gabriel de Souza Pereira Moreira, Ronay Ak, Kazuki Onodera · 2022
Session-based recommendation is an important task for domains like e-commerce, that suffer from the user cold-start problem due to anonymous browsing and for which users preferences might change considerably over time. The RecSys Challenge 2022, organized by Dressipi, is focused on the session-based recommendation problem for the fashion e-commerce domain. In this paper, the NVIDIA RAPIDS and NVIDIA Merlin teams present their solution that placed 3rd in the challenge. Among the most effective techniques we found sessions augmentation and ensembling a very diverse set of statistical, machine learning and deep learning models. Our recommendation pipeline is composed of three stages, where the first level is focused on candidate generation and the others refine the recommendation ranking for more robust and accurate recommendations.