Skip to content

Glossary

Term Meaning
Donor A trained recommender whose scores over warm items are transferred to cold items.
Warm item An item available during donor training.
Cold item An item hidden from donor training and evaluated as a new item.
Pseudo-cold A warm historical item deliberately held out to emulate cold-start evaluation.
TransferInputs The bundle passed to ColdStartMethod.fit: scores, content, similarity, embeddings and optional val data.
requires Method-declared input kinds that are validated before fitting.
ModelAdapter Benchmark wrapper around a donor engine. It trains on warm interactions and emits scores.
DatasetLoader Benchmark component that loads interactions and item content into a Dataset.
Splitter Benchmark component that creates warm, val-cold and test-cold folds.
Grouped MP Grouped Most Popular baseline; popularity is computed within the cold item's content group.
Personalized Grouped MP Strong baseline using the user's affinity to the cold item's groups.
Score transfer Learning how content maps to donor scores, then applying that mapping to cold items.