Enhancing Recommender Systems: A New Approach Using Collaborative Filtering with Bayesian Optimization and Gaussian Processes

Tóm tắt

Recommendation systems frequently make use of collaborative filtering (CF). CF derives its strength from the fact that in order to profile its users, it does not require a lot of knowledge about them, but it depends on earlier users’ ratings in which they were choosing products to recommend. Although there has been progress in modeling users as well as items, calibrating CF algorithms’ hyperparameters will
continue to be a difficult task. This paper has come up with a new way of doing this by the use of Bayesian optimization with Gaussian processes throughout hyperparameter tuning. The method in question automatically works on hyperparameters to make two fundamental and straightforward CF algorithms have solid results against three famous datasets (Netflix Prize, MovieLens 1M, MovieLens 10M), and two other datasets (Douban and Jester dataset 2). This solution not only exhibits solid performance in controlled experiments but also provides a simplified approach for practitioners, minimizing time intensive manual adjustments while ensuring low overhead, therefore necessitating relatively low computational and devel opmental resources. This results in expedited deployment and simplified integration into practical systems, hence enhancing the dependability and scalability of recommendation engines.