Abstract: In the recommendation model combined with the knowledge graph, the item vector obtained by user history behavior and knowledge graph embedding may lose some information, which makes the item vector representation inaccurate. In addition, most recommendation model cannot fully model the feature interaction between user and item. To solve the above problems, a recommender model based on knowledge graphs sharing information (ISRS) is proposed. In the knowledge graph module, the training of entity vector had to consider the current triad (head， relation， tail), which was modeling the relationship between head vector and relation vector and then sharing information with the item vector. The feature interactions of user and item were extracted by DeepFM layers, and the low-order and high-order interactions were modeled. The experiments demonstrate that ISRS model performs better in CTR prediction and Top-K recommendation, compared with other state-of-the-art methods.