The recommendation system is of great significance for screening effective information and improving the efficiency of information acquisition. Traditional recommendation systems face problems such as sparse data and cold starts. Based on the combination of external scoring and item connotation knowledge, an e-product recommendation model RKGCF based on cyclic knowledge graph and collaborative filtering is proposed. After fully considering the correlation between items, users, and ratings, Top-K recommendations are made using collaborative filtering based on items and users. To reveal the semantics between entities and relationships and understand user interests, external data and user preference data of items are added to the knowledge graph to extract the dependency relationship between entities and construct interactive in-formation between users and items. Multiple sets of different negative samples are used to train the model, and the real e-commerce data are used for testing. The experimental results demonstrate that the model has significantly improved the accuracy of the recommendation effect.