Abstract: Aiming at the problem of low accuracy and low recommendation efficiency in  the traditional algorithms during the recommendation process of institutional documents, we proposed a text recommendation algorithm based on knowledge embedding. By transforming knowledge in the  knowledge graph  into feature vectors and combining them with neural network models, the accuracy and stability of the recommendation system were effectively improved when dealing with massive and diverse data. Experimental results show that the proposed algorithm exhibits better recommendation accuracy and stability than the traditional methods in the face of cold start and diverse user interests. It provides a more efficient and reliable solution to the data sparsity problem and personalisation requirements in large-scale recommendation systems, which helps to improve the user experience and enhance the overall performance of the recommendation system.

Key words: knowledge graph, word embedding, graph embedding, neural network