A Survey on Graph-Based Collaborative Filtering Techniques in Recommender Systems

Abstract

Recommender Systems (RS) are software tools which can be used in making useful predictions of items to users. RS has been an important research area since the mid-1990s, and there are a lot of RS tools built since then to improve user satisfaction. While building the RS tools, researchers face a lot of problems in the form of data sparsity, information overload, cold-start, scalability, lack of resources and time. These factors may reduce the accuracy of predictions. To overcome these problems, researchers model the rating data as graphs. Through graphs, we can explore the transitive associations and hidden information in our dataset. Particularly, this work focuses only on the graph-based collaborative filtering (GBCF) techniques introduced in the recommendation systems. We have studied and analyzed various GBCF articles published in the most popular online digital libraries during the last two decades. These approaches have been categorized into three broader categories: user-item, user-user, and item-item based on the model of the graph, which is used for the recommendation purpose. This survey provides an understanding of how graph-based CF has helped researchers and developers built a more efficient RS model in terms of different RS goals, such as accuracy.

Keywords: Cold-start, Scalability, Data sparsity, Graph-based collaborative filtering, Information overload, Recommender systems.

View PDF