Hybrid CF + CB
Combines collaborative filtering with content-based signals for better coverage
Hybrid CF + CB blends collaborative filtering (user-item interaction patterns) with content-based filtering (item metadata similarity). This improves cold-start handling while maintaining personalization quality.
When to use:
- Mixed catalog with both established items (CF works well) and new items (content fills the gap)
- When neither pure CF nor pure content-based filtering is sufficient alone
- Balancing personalization depth with recommendation coverage
Input: User-item interactions + item metadata/content columns Output: Ranked list of recommended items blending both signals
Model Settings (set during training, used at inference)
CF Weight (default: 0.7)
Weight given to the collaborative filtering score. 1 - CF Weight is given to the content-based score.
CF Model (set during training) Which collaborative filtering method powers the CF component (e.g., SVD, KNN).
Content Similarity Metric (default: cosine) Metric for item content similarity.
Inference Settings
No dedicated inference-time settings. Final scores are a weighted combination of CF and content-based rankings.