Diseño de un sistema de recomendación basado en ganancias que usa machine learning para balancear los beneficios para el usuario y la empresa.

Abstract

This work proposes the design of a profit-aware recommender system named Multi-Aware Recommender System (MARS), where the impact on the user is subject to adjustment. The item re-ranking process is based on the profit generated for the business and other important attributes (i.e. price of the item), and it is performed using a weighted rank aggregation method. The weights are optimized iteratively using a variant of the Gradient Descent algorithm, in order to train the model and control the desired impact on the user. In this way, the model controls the impact on the user so as not to compromise the customer loyalty, while increasing the profit for the company. The weights can be unique for every user, making this approach personalized. The MARS method also controls the impact on the user by adjusting the ranking threshold TR, which determines the minimum rating that items must have to be re-ranked. Experiments showed that MARS models have the ability to control the impact on the user while increasing the profit by varying the weights associated to the considered attributes. It is important to notice that the training of the model can be parallelized; thus, the weights and results may be obtained for each user separately. By implementing MARS on 3 different data sets and various configurations, we obtained promising results in terms of the balance between impact on the user and profit generated.