Probabilistic Approach for Recommendation Systems

In this article, we propose a new probabilistic approach for product recommendations using deep learning framework, combining information from historical observations, similar users and prior knowledge. The deep learning approach is using autoregressive recurrent networks to model the recommendations probabilistically from a Bernoulli distribution. If prior information exists we implement a Pseudo-Bayesian approach, where we obtain posterior samples assuming Bernoulli likelihood on the sampled data from the deep learning model. The proposed approach allows for a very flexible modeling of product recommendations and quantifying uncertainty in predictions. Simulations and experiments were conducted to demonstrate the applicability and performance of the model. Comparisons made to related recommendation models revealed more accurate predictions among the proposed models.