Next point-of-interest (POI) recommendation has attracted increasing attention with the rapid development of location-based social networks. It aims to predict the next location a user is likely to visit based on their historical check-in behavior. Recent research indicates that diffusion models are gaining traction for their effectiveness in sequential recommendation tasks. However, these methods often overlook key spatiotemporal factors such as POI transition dynamics and target visit time. We propose DiffSTRec (Diffusion-based Spatiotemporal Recommendation), a diffusion-based framework that integrates a POI transition graph and multi-period time encoding to better capture POI semantics and users’ temporal preference dynamics. By incorporating noisy target representations and visiting time into the preference extraction module, we dynamically estimate user preference signals to guide the denoising process at each step. This design allows the model to better capture latent and multifaceted user intent. Experiments show that DiffSTRec consistently outperforms competitive baselines on three real-world datasets.