A Survey and Taxonomy of Sequential Recommender Systems for E-commerce Product Recommendation

E-commerce recommendation systems facilitate customers’ purchase decision by recommending products or services of interest (e.g., Amazon). Designing a recommender system tailored toward an individual customer’s need is crucial for retailers to increase revenue and retain customers’ loyalty. As users’ interests and preferences change with time, the time stamp of a user interaction (click, view or purchase event) is an important characteristic to learn sequential patterns from these user interactions and, hence, understand users’ long- and short-term preferences to predict the next item(s) for recommendation. This paper presents a taxonomy of sequential recommendation systems (SRecSys) with a focus on e-commerce product recommendation as an application and classifies SRecSys under three main categories as: (i) traditional approaches (sequence similarity, frequent pattern mining and sequential pattern mining), (ii) factorization and latent representation (matrix factorization and Markov models) and (iii) neural network-based approaches (deep neural networks, advanced models). This classification contributes towards enhancing the understanding of existing SRecSys in the literature with the application domain of e-commerce product recommendation and provides current status of the solutions available alongwith future research directions. Furthermore, a classification of surveyed systems according to eight important key features supported by the techniques along with their limitations is also presented. A comparative performance analysis of the presented SRecSys based on experiments performed on e-commerce data sets (Amazon and Online Retail) showed that integrating sequential purchase patterns into the recommendation process and modeling users’ sequential behavior improves the quality of recommendations