Improving Customer Behaviour Prediction with the Item2Item model in Recommender Systems

Abstract

Recommender Systems are the most well-known applications in E-commerce sites. However, the trade-off between runtime and the accuracy in making recommendations is a big challenge. This work combines several traditional techniques to reduce the limitation of each single technique and exploits the Item2Item model to improve the prediction accuracy. As a case study, this paper focuses on user behaviour prediction in restaurant recommender systems and uses a public dataset including restaurant information and user sessions. Within this dataset, user behaviour can be discovered for the collaborative filtering, and restaurant information is extracted for the content-based filtering. The idea of the pre-trained word embedding in Natural Language Processing is utilized in the item-based collaborative filtering to find the similarity between restaurants based on user sessions. Experimental results have shown that the combination of these techniques makes valuable recommendations.