T Towards a Perspective of Hybrid Approaches and Methodologies in Recommender Systems

Recommender Systems apply machine learning and data mining techniques to filter undetected information and can predict whether a user of a system would like a given resource based on his/her interests and preferences. To date a number of recommendation algorithms have been proposed, where Collaborative Filtering (CF) and Content-Based Filtering (CBF) are the two most famous and adopted recommendation techniques. CF Recommender Systems recommend items by identifying other users with similar taste and use their opinions for recommendation. CF Recommender Systems suffer from problems and challenges such as scalability, first rater (new item), data sparsity and cold-start problems. On the other hand, CBF Recommender Systems recommend items based on the content information of the items and match these items with interest and preferences of a user and therefore suffer from an overspecialization problem. In order to generate accurate and good recommendations, Hybrid Recommender Systems combine CF and CBF Recommender Systems to avoid the above aforementioned problems and challenges. This paper initially discusses Recommender Systems in general, then presents an overview of the state-of-the-art research in the area of Hybrid Recommender Systems, specifically from the perspective of types, applications, architectures and algorithms and finally discusses relevant open issues of Hybrid Recommender Systems.