Classic recommendation systems are very good at predicting user preferences but they do not take into account the fact that some products are limited by their inventory. In this thesis, we presented some classical approaches to modeling user preferences using matrix factorization. We presented a dynamic and a static recommendation system that optimizes inventory consumption. We also proposed a hybrid model that can combine the predictions of a recommendation system which models preference, and a recommendation system that optimizes product consumption. In addition to the model, we also presented a way to test the performance of systems using a product-shopping simulation. The recommended systems and simulation have been combined into the library PyRec. The performance of the implemented models was tested on the MovieLense 1M dataset and private commercial data. We show that with the proposed approach we can improve the stock consumption of classical systems.
|
