Recommender systems: an overview

Аннотация

In this article, we look at the key advances in collaborative filtering recommender systems, focusing on the evolution from research focused solely on algorithms to research focused on the broad set of issues surrounding user experience with the recommender. The Internet provides a huge amount of heterogeneous information, and its number is increasing rapidly every year. Related to this is the concept of information overload, when a person is exposed to an excessive amount of information and is not able to effectively process and use information. Recommender systems are trying to eliminate this problem. However, the difficulty of evaluating and selecting relevant information is constantly increasing. Therefore, new recommendation systems using various evaluation strategies are constantly emerging. There are also studies that demonstrate the effectiveness of their use in solving problems with information retrieval and recommendations. Unlike traditional models (such as collaborative filtering), deep learning provides a better understanding of user requirements, item characteristics, and the historical interaction between them. Earlier methods use models such as matrix factorization, SVD (Singular Value decomposition), to recommend them. These systems only work with user ratings. Gradually, recommendation systems were created that also used available information about items.