Hit rate is a binary top-k metric: it equals 1 when at least one relevant item appears in the recommendation list and 0 otherwise. section 4.6.1
Use hit rate to answer whether a recommendation list surfaced anything relevant.
- Prefer it when users typically engage with only one item per session; knowing that any relevant item appeared can matter more than counting every match. section 4.6.1
- Interpret it alongside precision and recall, since increasing k generally raises the chance of a hit while also adding irrelevant recommendations. section 4.6.1
- Compare it against random, popularity, and simple personalization baselines rather than treating the score as meaningful in isolation. section 4.10
Single-item session
For a session whose test set contains one relevant item, assign hit rate@k = 1 if that item appears anywhere in the top-k recommendations; otherwise assign 0. section 4.6.1
Aggregate hit rate can hide unequal user experiences, so inspect its distribution across users and evaluate it with ranking, coverage, and diversity metrics in the full pipeline. section 4.10 section 4.8
