Leaderboard Overview

The ORBIT Leaderboard provides a comprehensive evaluation of recommendation models across multiple benchmark datasets and evaluation metrics. Our goal is to offer a transparent, reproducible, and fair comparison of models across domains and settings.

ORBIT enforces fixed train/validation/test splits, standardized evaluation metrics, and uniform candidate pools for all models. This ensures reproducibility and fair comparison, addressing the inconsistency challenges found in flexible toolkits like RecBole and Elliot. Every result shown on the leaderboard is computed under the same evaluation protocol and candidate set, making model rankings trustworthy and comparable.