Recommender systems are of increasing importance owing to the growth of social networks and the complexity of user behavior, and cater to the personalized needs of users. To improve recommendation performance, several methods have emerged and made a combination of knowledge graphs and recommender systems. However, the majority of approaches faces issues like overlooking spatiotemporal features and lacking dynamic modeling. The former restricts the flexibility of recommendations, while the latter renders recommendations unable to adapt to the changing interests of users. To overcome these limitations, a novel dynamic preference recommendation model based on spatiotemporal knowledge graphs (DRSKG), which captures preferences dynamically, is proposed in this paper. Constructed by knowledge graphs, the model integrates spatiotemporal features and takes into account the dynamic preferences of users across various temporal, spatial, and situational contexts. Therefore, DRSKG not only describes the spatiotemporal characteristics of user behaviors more accurately but also models the evolution of dynamic preferences in spatiotemporal changes. Massive experiments demonstrate that the proposed model exhibits significant recommendation enhancement compared with the traditional one, achieving up to 7% and 5% improvements in terms of Precision and Recall metrics, respectively.

由于社交网络的增长和用户行为的复杂性，推荐系统变得越来越重要，并满足了用户的个性化需求。为了提高推荐性能，出现了几种方法，并将知识图谱和推荐系统相结合。然而，大多数方法都面临着诸如忽略时空特征和缺乏动态建模等问题。前者限制了推荐的灵活性，而后者则使推荐无法适应用户不断变化的兴趣。为了克服这些限制，该文提出了一种基于时空知识图谱（DRSKG）的新型动态偏好推荐模型，该模型可以动态捕获偏好。该模型由知识图谱构建，集成了时空特征，并考虑了用户在各种时间、空间和情境环境中的动态偏好。因此，DRSKG 不仅更准确地描述了用户行为的时空特征，而且对动态偏好在时空变化中的演变进行了建模。大量实验表明，与传统模型相比，所提出的模型表现出显著的推荐增强，在 Accuracy 和 Recall 指标方面分别提高了 7% 和 5%。