Abstract: Recommendation system has always been a hot issue in information retrieval research. Compared with other items such as movie recommendation and travel recommendation, news recommendation has the characteristics of large batch of news articles and strong timeliness, which has higher requirements on the algorithm. Based on the RippleNet model, this paper introduces the concept of entity entry degree, which highlights the importance of individual entities by weighting multilateral entities, thus improving the accuracy of news recommendation. The performance validation on the news dataset shows that the overall accuracy of this model is improved by 1.7% compared with other baseline models.