融合知识图谱和轻量级图卷积网络推荐系统的研究

Research on the fusion of knowledge graph and lightweight graph convolutional network recommendation system

基于协同过滤的算法是推荐系统中最重要的方法,由于冷启动和数据稀疏性的特点,限制了其推荐性能。为了应对以上问题,提出了知识图谱和轻量级图卷积网络推荐系统相结合的模型,该模型通过将知识图谱中的各个实体(项目)进行多次迭代嵌入传播以获取更多的高阶邻域信息,通过轻量聚合器进行聚合,进而预测用户和项目之间的评分。最后,在3个真实的数据集上MovieLens-20M、Last.FM和Book-Crossing的实验结果表明,该模型与其他基准模型相比可以得到较好的性能。

The algorithm based on collaborative filtering is the most important method in the recommendation system.However,the cold start and data sparsity characteristics limit its recommendation performance.We propose a model that combines a knowledge graph and a lightweight graph convolutional network recommendation system to address the aforementioned issues.The model embeds and propagates multiple items in the knowledge graph to obtain more highorder neighborhood information.It aggregates through a lightweight aggregator to predict the score between users and items.Finally,the experimental findings of MovieLens-20M,Last.FM and Book-Crossing on three real datasets show that compared with other benchmark models,this model can achieve better performance.