Online news aggregation services have become the first choice to read news for many internet users. However, thousands of news articles posted on a daily basis make it impossible for users to select intriguing news articles and keep track of the latest relevant topics. The automated systems are developed to tackle information overload. Various news recommendation methods are proposed to provide personalized experiences to users from diversified backgrounds.

To capture the higher-order relations hidden in texts, we propose a general framework of personalized news recommendation which explores and exploits existing knowledge graphs. The model deploys a heuristic method to take advantage of rich knowledge crowdsourced by human editors. Furthermore, the pretrained language models grab the a...[ Read more ]

Online news aggregation services have become the first choice to read news for many internet users. However, thousands of news articles posted on a daily basis make it impossible for users to select intriguing news articles and keep track of the latest relevant topics. The automated systems are developed to tackle information overload. Various news recommendation methods are proposed to provide personalized experiences to users from diversified backgrounds.

To capture the higher-order relations hidden in texts, we propose a general framework of personalized news recommendation which explores and exploits existing knowledge graphs. The model deploys a heuristic method to take advantage of rich knowledge crowdsourced by human editors. Furthermore, the pretrained language models grab the attention of the NLP community. We demonstrated that this framework could be easily adapted to these large-scale models and exploits their representation capability. In the experiments on a real-world recommendation dataset, our model outperforms other state-of-the-art models. The further case study shows how the entity paths obtained by our model improve the recommendation quality.