Map-matching, a critical process for aligning location records with actual routes in road networks, faces unique challenges when applied to Bluetooth (BT) data. Unlike GPS trajectories, BT readings from roadside stations introduce significant uncertainties due to their wide detection range and station-centered nature. This study identifies and addresses three primary types of spatiotemporal uncertainty in BT data: misdetection, disordering, and duplication. To overcome limitations in existing approaches, we propose CLMM, a novel map-matching method that learns uncertainty-aware representations through contrastive learning. CLMM utilizes a Seq2Seq module to accomplish the map-matching tasks while leveraging multiple tailored augmentation operators to generate diverse positive samples for each uncertainty type. Guided by contrastive loss, the model iteratively adapts to data uncertainty, resulting in more accurate representations and map-matching results. Comprehensive experiments validate our model’s effectiveness and efficiency, advancing map-matching techniques for BT data and improving foundations for real-world urban applications.