Addressing cold start in recommender systems with neural networks: a literature survey

Abstract

Filtering information on the Internet and recommending the right choices is more than important for Internet users and various businesses that offer products and services. Although recommender systems do this work efficiently, problems such as Cold Start often appear when new users or items enter the system. The traditional methods of recommender systems, collaborative filtering and content–based techniques, do not offer an optimized solution to this problem. The integration of neural networks in recommender systems offers a new approach to solving cold start. Whether using the feature of extracting hidden data, or using deep learning algorithms with more layers, the accuracy of recommendations and predictions has increased significantly. We have analyzed 40 papers that approached solving the cold start problem using neural networks. We have researched how neural networks are integrated into recommender systems, what they are used for, which neural network algorithms have shown to be more efficient in solving the cold start problem, and which algorithms have increased the accuracy of the recommendation. We aim to answer these questions with other subquestions related to types of cold start such as item or user cold start and warm, partial, or strict cold start.