Association Rules

Association Rules

If you bought x, you prob also bought y (Not necessarily the other way around)

Total set - all products in store Subset - a transaction Co-occurences (or frequent itemset) - looking across these subsets, what tend to be purchased together?   Frequent sequential pattern - first buy a laptop, then a cover, then a mouse

Applications Product placement (pop to pop) Product bundling (pop to indy) User profiling (indy to indy)

Notes Association rules are not causal Anticipate ripple effects

---

Rule: left -> right Evaluate   Suport: how often do they appear together? (this does not change if we swap left and right)   Confidence: given left, how often do we see right? (this does change if we swap left and right)   Lift: how does the confidence compare to random chance?    Lift = 1 means random

Apriori method (Parameters we set when finding rules) Phase 1 - Min support (need to have enough in common) Phase 2 - Min confidence (one leading to the other)

---

In this example, the combo of c + d + e was frequent. By definition, all subsets (less detail) above must all be frequent with each other. Image

---

The combo a + b was infrequent. All resulting supersets (greater detail) must also be infrequent Image