Mean Average Precision (MAP)
What you'll learn: How MAP provides a single, comprehensive score that measures recommendation quality across multiple users and ranking positions.
The Core Idea
Imagine you're evaluating a recommender system for 100 different users. For each user, you calculated Precision@K at various cutoff points (like P@1, P@2, P@3, etc.). But how do you summarize all these numbers into one meaningful score?
Mean Average Precision (MAP) solves this by:
- Computing Average Precision (AP) for each user/query
- Averaging those AP scores across all users
Understanding Average Precision
For a single user's recommendation list, Average Precision is the mean of precision values calculated at each position where a relevant item appears.
If a user has 5 relevant items and your system recommends 10 items:
- Calculate Precision@K only at positions where you correctly recommended a relevant item
- Average those precision values
- This rewards systems that rank relevant items higher
Why MAP Matters
MAP aggregates performance across:
- Multiple ranking positions (early positions count more because you only compute precision when hitting relevant items)
- Multiple queries/users (the final "mean" averages AP across everyone)
Unlike simple Precision@K which only looks at one cutoff, MAP considers the order of relevant items throughout the entire ranked list. A system that puts all relevant items at the top gets a higher MAP than one that scatters them throughout.
Key Takeaway: MAP is a single metric that rewards recommender systems for ranking relevant items highly and consistently across all users, making it ideal for comparing overall system performance.