OPTIMIZING RECOMMENDER SYSTEMS: INTEGRATING COLLABORATIVE FILTERING WITH BAYESIAN AND GAUSSIAN TECHNIQUES

Abstract

One frequently utilized method in recommendation systems is collaborative filtering (CF). CF’s strength is that it generates recommendations based on past users’ ratings or purchase behavior instead of requiring thorough user information for customer profiling. Fine-tuning the hyperparameters of CF algorithms remains a difficult task even with developments in modeling consumers and products/services. This work explores an alternative approach for this aim by means of Bayesian optimization using Gaussian processes during hyperparameter altering. This method reduces the time and effort usually needed for manual tuning by autonomously adjusting hyperparameters for two basic and simple CF algorithms on three popular datasets, yielding competitive results: Netflix Prize, Movielens 1M, and Movielens 10M. Therefore, it could enable practitioners to improve the performance of their recommendation systems whilst greatly shortening the time and effort spent on tuning their systems.