Knowledge Graph (KG)-based recommendation systems (RS) leverage auxiliary information to address sparsity and cold-start issues. While prior work have enhanced user and item representations using multi-hop neighborhoods, they often neglect items’ inherent attractiveness to user groups. To tackle these gaps, we propose an attention-augmented KG recommendation framework that refines user and item representations through three perspectives: user preferences, item attractiveness, and higher-order KG structure. We employ a self-attention mechanism in a user preference module to model user-specific tendencies from interaction records, an attention-based item attractiveness module to highlight items’ appeal to user groups, and graph attention networks to aggregate semantic information from KG entity-relation attributes across multiple layers. Experiments on three real-world datasets show this framework outperforms baselines in click-through rate and Top-K recommendation tasks, validating our multi-perspective enhancement.