An Efficient Cold Start Solution for Recommender Systems Based on Machine Learning and User Interests

Recommender Systems are used to provide suggestions to users based on their interests. One of the well-known problems in recommender systems is the cold start problem, which concentrates on providing recommendations when the user data is not sufficient. Although many solutions have been proposed in the literature, the majority did not concentrate on using both the hidden user motifs. We proposed previously a user-interest based cold start solution, however, finding similar users needed optimization. In this work, we propose the use of machine learning techniques to find patterns associating user profile information and user extracted interests. This would further improve the accuracy provided suggestions for new users. Experimental work showed that our solution is efficient in terms of training time, classification time, and accuracy. In details, using Bayesian Classifier Chain classifier proved to be the fastest training and classification time, while all classifiers proved to be efficient in term of accuracy.