An Efficient Collaborative Filtering for Recommendation Systems Using Differential Machine Learning

Utilizing deep learning, in contrast to standard recommendation models, is capable of successfully capturing non-linear and non-trivial user-item connections and codifying extremely complex abstractions as data representations in higher layers. Differential machine learning as a type of supervised learning trains models based on the twin networks that consider inputs and labels and differentials of labels to inputs. This paper proposes an efficient collaborative filtering method for recommendation systems using differential machine learning, called Differential Collaborative Filtering (DiffCF). DiffCF integrates the cost of derivatives and errors in values, which increases accuracy remarkably. The experiments are performed to train recommender system models on various datasets with different feature dimensions, while a significant reduction in errors and regularization penalties is obtained in comparison to standard machine learning.