Dataset recommender system helps choose evaluation data for recommender algorithms
FINALLY: A Dataset Recommender System for Recommender-Systems Research
Information Retrieval
Summary
Choosing the right datasets to test recommendation algorithms is important but challenging. The authors created FINALLY, a web tool that helps pick groups of datasets meeting specific rules and goals. They tested it thoroughly and found the tool consistently selects datasets as intended, including diverse and non-diverse groups. However, they note this does not prove which selections are best for all uses. The tool helps system builders set up fair and varied testing scenarios for recommendation algorithms.
What this means in practice
- •For recommendation engineers: Construct offline evaluation dataset sets that meet multiple constraints for testing recommender algorithms.
- •For machine learning platform teams: Provide configurable dataset selection tools to streamline benchmarking and algorithm comparison workflows.
Authors
Louis Owie
Abstract
Dataset selection shapes the empirical conditions under which recommender-system algorithms are evaluated, yet existing tools provide limited support for constructing complete dataset sets that jointly satisfy experimental constraints and set-level selection objectives. To address this problem, I developed FINALLY, a web-based dataset recommender for constructing configurable dataset sets for offline recommender-systems evaluations. FINALLY combines required datasets, candidate-pool restrictions, metadata filters, configurable target-set sizes, Random selection, and diverse and non-diverse strategies based on adapted Effective Covariance and Convex Hull objectives. I evaluated FINALLY through 420 recommendation runs across ten systematically varied configurations. All evaluated dataset sets satisfied the applicable target-size, duplicate-avoidance, snapshot-membership, required-dataset, and metadata-filter requirements. All 40 deterministic strategy--configuration combinations were reproducible. Both the Effective-Covariance-based and Convex-Hull-based strategies produced the expected diverse-versus-non-diverse score ordering in all ten configurations. Under their corresponding objectives, the diverse strategies produced scores above all 30 configuration-specific Random results, whereas the non-diverse strategies produced scores below all 30 Random results. These results establish technical consistency for the evaluated FINALLY workflow and show that the implemented strategies follow their intended optimization directions within the investigated configuration space. They do not establish the scientific suitability, global optimality, or practical superiority of the generated selections.