A Utility-Based Recommendation Approach for E-Commerce Websites Based on Bayesian Networks

Abstract

Although utility-based recommendation in E-Commerce can provide much better recommendation accuracy, there are still no effective approaches to build the utility function of each user. In order to overcome this problem, an approach based on Bayesian networks is proposed. Firstly, based on the common user utility function of a specific commodity which has already been constructed by domain experts, a prior Bayesian network can be established. Secondly, the prior Bayesian network is modified based on the current user’s implicit feedback, so that his utility function can be represented by means of Bayesian networks. Finally, according to his utility function, objects the current user may like are recommended to him. Compared with other approaches, this approach may acquire utility functions more approximately and automatically. Furthermore, it could extend the range of applications for which utility-based recommendation would be more useful.