Intent-Guided Diffusion with Adaptive Intent Fusion for User Cold-Start Recommendation

User cold-start recommendation aims to provide personalized recommendations for users with extremely limited interaction data and remains a fundamental challenge in recommender systems. Existing approaches mainly rely on auxiliary information, meta-learning, or sparsity simulation to compensate for insufficient behavioral signals. However, these methods still depend heavily on the limited observations available for cold-start users and rarely exploit transferable preference patterns embedded in historical interaction data. To address this issue, we propose an Intent-Guided Diffusion Model (IGDM) for user cold-start recommendation. The proposed framework consists of three components: an intent extraction module, an adaptive intent fusion module, and an intent-guided diffusion module. Specifically, latent preference prototypes are first extracted from historical user interaction sequences to capture transferable preference patterns. An adaptive intent fusion mechanism then dynamically integrates the limited behavioral representation of a cold-start user with these prototypes according to their relevance, producing a more informative fused preference representation. Finally, the fused representation is introduced as a conditional signal to guide the diffusion-based denoising process, allowing the sparse and potentially noisy user representation to be progressively refined toward a more reliable preference representation. Extensive experiments on three real-world datasets, namely MovieLens-1 M, Last.FM, and Yelp, demonstrate that IGDM consistently outperforms multiple representative baselines in terms of Recall, NDCG, and MAP, with more pronounced improvements under highly sparse interaction settings. These results demonstrate the effectiveness of combining transferable preference priors with diffusion-based representation refinement for user cold-start recommendation.