Part 4/12:
The team first established a baseline using matrix factorization techniques like ALS (Alternating Least Squares), which are widely used in recommendation systems — similar to Netflix's movie suggestions. These models predict user preferences based on past interactions, effectively ranking products on the likelihood of purchase.
The ALS baseline achieved a precision at K (top recommendations) of around 2.26% at the third position, with Desile Gain (a measure of rank ordering efficacy) of 67.1%. These results, while decent, highlighted the need for models capable of capturing complex sequential relationships inherent to retail purchase behavior.