1. Definition

  • Catalog Coverage measures the proportion of the entire item catalog that appears in recommendations across all users.
  • It reflects how much of the available catalog is being exposed to users by the recommender system.

Note: In many papers, Catalog Coverage is used interchangeably with Item Coverage, but sometimes it emphasizes the whole catalog exposure, while Item Coverage may be restricted (e.g., Top-N list).

2. Formula

Let:

  • $I$ = full item catalog.
  • $R(U)$ = set of all unique items recommended to all users.

$\text{Catalog Coverage} = \frac{|R(U)|}{|I|}$

  • Same mathematical form as Item Coverage.
  • The distinction is mostly interpretation: Catalog Coverage focuses on system-wide use of catalog, not just per-user lists.

3. Example

Suppose:

  • Catalog has 10,000 movies.
  • Across all users, recommendations include 2,000 distinct movies.

$\text{Catalog Coverage} = \frac{2000}{10000} = 0.20 \; (20\%)$

So the system is leveraging only 20% of its catalog.

4. Intuition

  • High catalog coverage → system surfaces a wide range of items, good for discovery and long-tail exploration.
  • Low catalog coverage → system concentrates on a small subset (e.g., popular blockbusters), ignoring niche items.

5. Variants

  • Top-N Catalog Coverage: restrict measurement to top-N recommendations per user.
  • Temporal Catalog Coverage: how much of the catalog is covered over time.
  • Weighted Coverage: weights items by importance or popularity.

6. Trade-offs

  • Systems with high accuracy often have low catalog coverage (they keep recommending the same popular items).
  • Improving catalog coverage increases novelty and diversity, but may slightly reduce accuracy.
  • Balance is key: users like accurate results but also appreciate freshness and variety.

7. Relation to Item Coverage

  • Item Coverage = distinct items recommended / total items.
  • Catalog Coverage = same formula, but explicitly emphasizes coverage of the full catalog across the entire system.
  • Many authors use them interchangeably, but Catalog Coverage is often reported in system-wide evaluation, while Item Coverage is sometimes reported at per-user granularity (Top-N lists).

Summary:
Catalog Coverage = fraction of the catalog that appears in recommendations across all users. It measures how much of the available catalog is utilized by the system, complementing accuracy metrics by focusing on variety and fairness of exposure.