Lightweight Model for Session-Based Recommender Systems with Seasonality Information in the Fashion Domain
Nicola Della Volpe, Lorenzo Mainetti, Alessio Martignetti, Andrea Menta, Riccardo Pala, Giacomo Polvanesi, Francesco Sammarco, Fernando Benjamín Pérez Maurera, Cesare Bernardis, Maurizio Ferrari Dacrema · 2022
This paper presents the solution designed by the team “Boston Team Party” for the ACM RecSys Challenge 2022. The competition was organized by Dressipi and was framed under the session-based fashion recommendations domain. Particularly, the task was to predict the purchased item at the end of each anonymous session. Our proposed two-stage solution is effective, lightweight, and scalable. First, it leverages the expertise of several strong recommendation models to produce a pool of candidate items. Then, a Gradient-Boosting Decision Tree model aggregates these candidates alongside several hand-crafted features to produce the final ranking. Our model achieved a score of 0.18800 in the public leaderboard. To aid in the reproducibility of our findings, we open-source our materials.