Sequential Recommendation Using Graph Neuron Networks

Recommendation systems (RS) are becoming increasingly popular in e-commerce, spanning areas like books, movies, music, news, and especially shopping and consumer goods. In recent years, sequential recommendation systems have gained attention as a new approach to capturing user preferences, either by analyzing the user’s entire behavior history over time or by focusing on interactions within a single session. This paper focuses on studying sequential recommender systems based on user behavior sequences with a session. Specifically, customer session data representing product selection through click behavior (number of times choosing to buy or view a product) can support effective for the e-commerce systems. Session-based recommendation system endeavors to predict the interests of users on the items they may likely click/buy next in the sequence of items which they have viewed. This article explores sequential recommender systems for analyzing user behavior within a session, with a particular focus on session-based product recommendations using graph neural networks. Experiments are conducted using RSC15 and Diginetica datasets via two metrics: accuracy measured by precision at 20 (P@20) and average reciprocal rating indicated by MRR@20. On the RSC15 dataset, the method achieves an accuracy of 71.09 (P@20) and an MRR@20 score of 30.69. Meanwhile, on the Diginetica dataset, the P@20 is 52.41 and a MRR@20 value of 18.35. These results show that the proposed approach could be used for session-based product recommendation.