To　overcome　the　shortcomings　of　current　approaches　in　feature　extraction,　specifically　the　lack　of　in-depth　feature　extraction　and　the　absence　of　high-order　feature　aggregation　for　both　users　and　items,　we　propose　a　knowledge　graph-based　dual-end　feature　aggregation　recommendation　model.　First,　initial　vectors　are　generated　using　knowledge　graph　embeddings.　Next,　a　ripple　network　and　a　knowledge　graph　attention　network　are　employed　to　extract　features　on　the　user　and　item　sides,　respectively.　Then,　features　of　various　orders　are　aggregated　and　concatenated　to　obtain　the　final　vectors.　Finally,　the　matching　probability　between　users　and　items　is　predicted.　Tests　performed　on　three　datasets　reveal　that　the　new　model　surpasses　seven　other　baseline　models　based　on　AUC,　ACC,　NDCG@K,　and　Recall@K　metrics.　©　2024　IEEE.