METoNR: A meta explanation triplet oriented news recommendation model

作者:

Highlights:

摘要

Personalized news recommendation is an important task for online news platforms to target user interests and alleviate information overload. Most existing methods leverage news contents to make recommendation due to the importance of contents to distinguish pieces of news. Although Heterogeneous Graph (HG) shows great potential in organizing and exploiting varieties of side information to boost recommendation performance in general recommender systems, and at the same time there exist rich side information in real news recommendation scenario, existing methods lack attention to utilizing HG to exploit multiple kinds of side information to enhance news recommendation accuracy. In addition, they pay no attention to providing understandable explanations to improve user satisfaction. To this end, we propose a meta explanation triplet oriented news recommendation model. Specifically, we first construct an HG to extract high-order relatedness knowledge between users and news from various side information. Then, a dedicated neural network is designed to leverage rich side information and contents in a joint way to make news recommendation. Finally, we provide user-centered and news-centered recommendation explanations for users based on meta explanation triplets. Extensive experiments on two benchmark real-world datasets show that our model could improve news recommendation performance compared with state-of-the-art methods and provide effective explanations.

论文关键词:Knowledge extraction,News recommendation,Meta explanation triplet,Supervised multi-task learning,Recommendation explanation

论文评审过程:Received 23 August 2021, Revised 29 October 2021, Accepted 6 December 2021, Available online 14 December 2021, Version of Record 20 December 2021.

论文官网地址:https://doi.org/10.1016/j.knosys.2021.107922