MVDE-MSA: A Multi-view Disentangled Enhancement Framework with Matching Scenario Awareness for Robust Address Matching

Standardized address data plays a pivotal role in urban governance applications such as emergency response systems, yet existing address matching methods face significant challenges due to inconsistent data collection standards across departments. These challenges manifest as colloquial expressions, missing elements, and high textual similarity among geographically proximate addresses, collectively degrading matching performance. To address these limitations, we propose MVDE-MSA, a robust address matching framework incorporating Multi-View Disentangled Enhancement and Matching Scenario Awareness. The framework features two key innovations: a Multi-View Disentanglement-Enhanced Encoding module that explicitly preserves complementary information across different address representations, and a Matching Scenario-Aware Dynamic Weight Allocation network that adaptively adjusts view-specific similarity weights based on scenario characteristics. Comprehensive evaluations on real-world datasets reveal that MVDE-MSA achieves a 6.0% improvement in Recall@1 over state-of-the-art methods, proving particularly effective in handling ambiguous and incomplete address data.