Association Rules
Recommend items frequently co-purchased or co-interacted with
Association Rules-based recommendation uses frequent itemset mining (Apriori or FP-Growth) to discover items that are often purchased or viewed together. At inference time, it recommends items with high lift and confidence given the user's current session or basket.
When to use:
- Market basket analysis and cross-sell recommendations ("frequently bought together")
- Session-based recommendation based on the current browsing basket
- Retail, e-commerce, and catalog recommendation
Input: User's current basket or session items Output: Ranked list of items with high association rule confidence/lift
Model Settings (set during training, used at inference)
Min Support (default: 0.02) Minimum frequency threshold for itemsets.
Min Confidence (default: 0.5) Minimum conditional probability for a rule to be applied.
Min Lift (default: 1.2) Minimum lift for rule quality. Lift > 1 means co-occurrence is above chance.
Rule Metric (default: lift) Primary metric for ranking recommendation rules.
Inference Settings
No dedicated inference-time settings. Rules matching the input basket are looked up from the pre-mined rule store.