Graph Model based Recommendation Architecture for E-commerce Applications

Abstract

It is very challenging to provide relevant data to users almost instantaneously due to a large amount of data present in an application. The role of recommendations is to provide relevant data to users considering relationships among data and users. Graph models are enriched in relationships; therefore, we propose an architecture for recommendations based on a graph model in e-commerce. The proposed architecture consists of two phases: offline phase for graph creation and recommendation phase for results generation. In the offline phase, different data sources are unified into a recommendation graph which is utilised by different recommendation algorithms to generate results. We also design algorithms for content-based and collaborative recommendations based on the generated graph. We implement a prototype of the proposed architecture in e-commerce and analyse and compare its performance with the relational model. We also verify the improved performance of the proposed graph model asymptotically. The graph model outperformed the relational model for content-based and collaborative recommendations. Thus, our architecture can be used in various applications for recommendations.