作为信息检索的重要应用场景，推荐系统已被应用在电子商务、信息流等数字化场景以实现用户与信息的高效匹配。为提升推荐效果和算法可解释性，最新研究聚焦在知识增强的推荐研究：这类方法引入知识图谱为代表的场景外知识来丰富待推荐项联系，构造包含用户、待推荐项、实体的异质信息网络，再基于图嵌入或路径的方法学习用户和待推荐项表示，最后用待推荐项间关系建模历史交互对用户需求影响。但已有研究仍面临以下挑战：1.异质信息融合不足，表示学习时忽略了场景中内容特征；2.建模用户需求时并未考虑其动态性；3.引入大规模知识信息严重影响了算法效率。针对这些挑战，本项目拟开展动态高效的知识融合推荐方法研究，主要包括：融合外部知识和场景内容特征的表示学习研究、基于待推荐项知识关联和用户交互历史的动态需求建模、针对知识融合推荐方法的算法优化。我们认为，该研究符合推荐领域的发展方向，有助于改进推荐效果并提升海量用户的体验。

As a key application scenario of information retrieval, recommender systems have been applied to various digital scenarios, such as e-commerce and information flow, to achieve efficient matching of users and information. To improve the effectiveness and interpretability of recommender systems, recent studies focus on knowledge-enhanced recommendation algorithms. This type of algorithm introduces external knowledge, such as knowledge graph, to enrich the connections between items and then construct a heterogeneous information network including users, items, and entities. User and item embeddings are learned based on graph embedding methods or path-based methods, and item associations are applied to model the impact of user historical interaction on current demand. However, existing studies still faces the following challenges: 1. An insufficient fusion of heterogeneous information, which ignores the content features in recommendation scenarios when conduct embedding learning; 2. The time factor is not considered when modeling the dynamic demands of users; 3. The introduction of large-scale knowledge information seriously affects the efficiency of the algorithms. In response to these challenges, this project intends to carry out a study on dynamic and efficient knowledge fusion recommendation algorithms, which mainly includes: representation learning that integrates external knowledge and content features in recommendation scenario, user dynamic demand modeling based on item associations and user interaction history, and algorithm optimization of knowledge enhanced recommendation algorithms. We believe that this work will contribute to the study of recommendation systems, improve the recommendation performance, and serve mass users better.