This repository contains our solution for the
Recommender Systems Challenge 2025,
organized as part of the Recommender Systems course at Politecnico di Milano.
The project was developed by a team of two and achieved:
- 🥇 1st place at the first deadline
- 🥈 2nd place at the final deadline
out of 71 participating teams.
The application domain is movie recommendation. The provided dataset contains interactions of users with movies. The main goal of the competition is to discover which new movies a user will interact with.
Our solution follows a two-stage recommendation pipeline, combining multiple collaborative filtering models for candidate generation and a learning-to-rank model for final re-ranking.
The implementation is largely based on the official course repository, which provides the core recommender models and evaluation utilities.
The dataset contains implicit user–item interactions:
- ~3.0M interactions
- ~27k users
- ~7k items
A set of target users is provided for which a list of 20 recommendations must be generated.
Evaluation is performed using Recall@20.