Back to the problems

1. How to measure repeat usage for anonymous user sessions?

   The model already implies repeat-buying in aggregated data.

2. Which co-occurrences qualify as non-random?

   The model is based on strict independence assumptions. It estimates the probability that a product combination is used together $r$-times by chance.

   We expect non-random choices (complementarity between two products) to occur more often than the model would predict.

Index: Contents