Knowledge graph-based recommendation systems face challenges like sparse supervision, long-tail distribution, and data noise, limiting accuracy and generalization. To address these issues, we propose a novel recommendation framework, Multi-Level Contrastive Learning Enhanced Recommendation (MLCLR). MLCLR integrates self-supervised contrastive learning into knowledge graph-based recommendation, aiming to improve representation learning by capturing both structural and semantic information at multiple levels. Specifically, MLCLR employs a multi-level contrastive learning mechanism that constructs both global and local views to enhance user-item embeddings. Additionally, we introduce two data augmentation strategies that mitigate the effects of noise and sparsity, leading to more robust recommendations. Extensive experiments on three public datasets (MovieLens-1M, Last.FM, and Book-Crossing) demonstrate that MLCLR significantly outperforms state-of-the-art baselines in terms of recommendation accuracy and robustness, validating its effectiveness in addressing long-tail and sparse supervision issues.