liveBook · Manning

Mean average precision

As explained in AI Recommender Systems

Mean average precision is the mean across users or queries of average precision values computed at positions containing relevant items.

Mean average precision (MAP) measures how well a ranked recommendation list places relevant items near the top. For each user, calculate precision at every position containing a relevant item and average those values to get average precision (AP). MAP is the mean AP across users. section 4.6.3

Same recall, different ranking

Suppose two lists each contain three relevant items. One places them at positions 1, 2, and 3; the other at 1, 5, and 9. They have the same recall, but the first earns higher AP because its relevant items are clustered near the top. section 4.6.3

Use MAP alongside metrics that answer different questions: precision and recall measure whether relevant items appear, while mean reciprocal rank focuses on the first relevant item. Evaluate against a relevance set drawn from historical behavior, not an assumed objective ground truth. Compare MAP with baselines and inspect results across users; a mean alone can hide uneven experiences. section 4.6.3 section 4.10

In a model evaluation, calculate MAP with accuracy, other ranking, and beyond-accuracy metrics for both the candidate and its baselines. A ranking gain does not settle whether lower catalog coverage or diversity is acceptable; that decision depends on the product goal. section 4.8

How AI Recommender Systems covers it

How deep it goes
Defines per-user AP and cross-user MAP and illustrates their ranking effect; does not specify how to handle users with no relevant items.
Examples
Contrasts relevant items at positions 1, 2, and 3 with positions 1, 5, and 9 despite equal recall.
What it assumes
Builds AP from precision at relevant positions; assumes ranked recommendations and a historical relevance set for evaluation.
What you can do after
Evaluate/understand ranking gains with MAP while checking baselines, user-level variation, coverage, and diversity before advancing a model.
How it connects
Connects MAP to precision, recall, mean reciprocal rank, NDCG, relevance sets, and recommendation baselines.

Also explained in

Compare how 2 books explain mean average precision →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of AI Recommender Systems. The section links point to the parts of the book it draws on.