<p>Knowledge graphs (KGs) play a pivotal role by providing auxiliary information in addressing data sparsity and cold-start issues in recommender systems. Nevertheless, existing KG-based recommendation models still encounter two critical challenges. The first is an overemphasis on capturing popular intents, leading to an underutilization of niche intents that could improve recommendation effectiveness. The second is the neglect of active learning on the knowledge-side, coupled with the failure to enhance the quality of collaborative signals and knowledge information effectively. To tackle these challenges, this paper proposes the multi-side contrastive learning with intent-enhanced (MCLI) framework. Specifically, we construct an intent-enhanced KG to overcome the limitation of focusing solely on popular intents while capturing niche intents valuable for recommendations. Next, we encode task-relevant information from dual-side graph refining using a local feature aggregator, and perform dual-view contrastive learning between the interaction-side and knowledge-side to facilitate active learning of knowledge-side information. Additionally, we preserve the collaborative knowledge side of the global structure and conduct global-view contrastive learning between it and the dual sides to ensure that the model retains all relevant information. Unlike the state-of-the-art baseline KGIC, which emphasizes local and non-local graph contrast, MCLI integrates an enhanced KG that captures niche intents and employs dual-view contrastive learning to distill task-relevant information from both the interaction and knowledge sides. On the Last.FM dataset, MCLI achieves substantial improvements in CTR prediction, with AUC and F1 gains of 3.48% and 5.29%, respectively.</p>