Coverage

Coverage metrics assess the extent to which a recommender system is able to recommend items from the entire catalog. They measure the diversity of the items recommended and the proportion of the item space that the system can effectively explore. High coverage suggests that the system can offer a wide range of recommendations beyond just the most popular items.

API Reference

For class signatures and source code, see the Coverage Metrics API Reference.

ItemCoverage

Item Coverage (ItemCoverage@K). Measures the number of unique items recommended in the top-K positions across all users, indicating catalog coverage.

\[ \text{ItemCoverage@}K = \left|\bigcup_{u \in \mathcal{U}} \mathcal{L}_u^K\right| \]

evaluation:
    top_k: [10, 20, 50]
    metrics: [ItemCoverage]

UserCoverage

User Coverage (UserCoverage@K). Calculates the number of users with at least one recommended item in their top-K recommendations, indicating reach and usefulness.

\[ \text{UserCoverage@}K = |\{u \in \mathcal{U} : |\mathcal{L}_u^K| > 0\}| \]

evaluation:
    top_k: [10, 20, 50]
    metrics: [UserCoverage]

NumRetrieved

Number of Retrieved Items (NumRetrieved@K). Counts the total number of distinct items retrieved in the top-K recommendations across all users.

For further details, please refer to this link.

evaluation:
    top_k: [10, 20, 50]
    metrics: [NumRetrieved]

UserCoverageAtN

User Coverage At N (UserCoverageAtN). Measures the number of users for whom the recommender retrieves at least N items, reflecting system responsiveness or minimum output capability.

evaluation:
    top_k: [10, 20, 50]
    metrics: [UserCoverageAtN]