Affiliation:
1. School of Information and Communication Engineering, Communication University of China, China
2. State Key Laboratory of Media Convergence and Communication, Communication University of China, China
Abstract
Personalised news recommendation comprises two crucial components: news understanding and user modelling. Previous studies have attempted to model news understanding and user interests using various internal news information and external knowledge graphs (KG). However, they have overlooked the collaborative function of the external KG and the internal information among diverse news and user behaviours, resulting in serious news cold-start problems and poor interpretability of user interests. To address these issues, this article proposes a novel approach called Relation-Aware Approach based on Multi-view News Network for News Recommendation (MNN4Rec). Specifically, MNN4Rec first constructs a Multi-view News Network (MNN), which includes candidate news and user-clicked news, and represents their exclusive multi-view information as heterogeneous nodes. Furthermore, we develop explicit and implicit news relationships and design a special sampling algorithm to search for news co-neighbours. We then use a novel dual-channel graph attention mechanism to obtain the fine-grained news understanding representation. Moreover, we construct explainable user interests by modelling the interaction of user-clicked news through the multi-headed self-attention mechanism in both semantic and relation levels. Finally, we match candidate news understanding with user interests to generate a prediction score for recommendation. Experimental results on Microsoft’s news data set MIND demonstrate that MNN4Rec outperforms existing news-recommendation methods while also mitigating the cold-start problem and enhancing the interpretability of user interests. Our code is available at https://github.com/JiangHaoPG11/MNN4Rec_code .
Subject
Library and Information Sciences,Information Systems
Reference65 articles.
1. Introduction to Recommender Systems Handbook
2. Salakhutdinov R, Mnih A. Bayesian probabilistic matrix factorization using Markov chain Monte Carlo. In: Proceedings of the international conference on machine learning, pp. 880–887. New York: ACM, https://dl.acm.org/doi/10.1145/1390156.1390267
Cited by
2 articles.
订阅此论文施引文献
订阅此论文施引文献，注册后可以免费订阅5篇论文的施引文献，订阅后可以查看论文全部施引文献