Stratified evaluation measures recommendation performance separately across user groups, revealing who benefits and who remains underserved. section 4.7
Evaluate across user-lifecycle segments:
- Cold-start users and users with a few interactions. section 4.7
- Active users and long-term users. section 4.7
Compare segment-level results
A model may improve active-user metrics while worsening cold-start performance. Compare each segment with an appropriate baseline: popularity is a strong benchmark for cold-start users, while power users expect more personalization. section 4.7
Figure 4.19 The different stages of a user's lifecycle, and which types of recommendations can be used at each stage.

Use the lifecycle diagram to relate user stages to recommendation types. section 4.7
