Liu Zhitong, Du Shutong, Zong Shouxin, Pan Bingyu
School of Sports Engineering, Beijing Sport University, Beijing, 100084, China.
China Sports Big Data Center, Beijing Sport University, Beijing, 100084, China.
Sci Rep. 2025 May 29;15(1):18910. doi: 10.1038/s41598-025-03531-5.
In recent years, with the spread and popularization of health knowledge, more and more people have begun to participate in fitness exercises to strengthen their bodies and prevent diseases. However, due to the lack of fitness knowledge base and the imperfection of fitness recommendation algorithm, fitness enthusiasts cannot obtain accurate fitness knowledge. Therefore, how to recommend personalized content for users according to their preferences has become a practical topic. Therefore, based on the knowledge graph technology, this paper constructs the scientific fitness knowledge graph, and proposes a model R-CKGAT that integrates collaborative knowledge embedding, user preference propagation and knowledge graph attention mechanism. Experimental results show that compared with MF, CKE and other baseline algorithms, the AUC and ACC values of the proposed algorithm in the scientific fitness data set are better than those baseline algorithms. The AUC and ACC of the model were 92.76% and 88.67% correspondingly.
近年来，随着健康知识的传播与普及，越来越多的人开始参与健身锻炼以增强体质、预防疾病。然而，由于健身知识库的匮乏以及健身推荐算法的不完善，健身爱好者无法获取准确的健身知识。因此，如何根据用户偏好为其推荐个性化内容已成为一个现实课题。为此，本文基于知识图谱技术构建科学健身知识图谱，并提出一种融合协同知识嵌入、用户偏好传播和知识图谱注意力机制的模型R-CKGAT。实验结果表明，与MF、CKE等基线算法相比，所提算法在科学健身数据集上的AUC和ACC值优于那些基线算法。该模型的AUC和ACC分别相应达到92.76%和88.67%。
