This experiment demonstrates how item-to-item collaborative filtering can generate product suggestions for incomplete shopping carts.
This experiment demonstrates how the Matchbox Recommender can be leveraged to deliver product suggestions based on the current contents of an online shopping cart. Sales transaction data are simulated for a collection of products such that some items are more likely to be bought together (e.g. phones and phone chargers). A portion of these transactions, as well as randomly-generated negative examples, are used to train a Matchbox Recommender. The trained recommender is then used to identify top suggestions for each product. For each transaction in the test set, we identify the most frequent suggestions for all products in the shopping cart *except* the last product added. The suggestion "accuracy" is assessed by calculating how often the last product added to the shopping cart is among the suggested products.