An interactive address matching method based on a graph attention mechanism

Modernizing and standardizing place names and addresses is a key challenge in the development of smart cities. This paper proposes a solution to address matching challenges, such as incomplete descriptions, reversed word order, and the diverse descriptions often found in Chinese addresses. Leveraging the hierarchical structure of Chinese addresses, this study introduces the interactive address matching graph attention model (IAMGAM). In the IAMGAM, an attention-based feature interaction method (AFIM) is employed. To reflect the hierarchical nature of address elements, a directed graph is used to model the address data, and the model is trained and tested using a graph attention mechanism. Experiments demonstrate that the IAMGAM achieves an accuracy and F1-score of 99.61%. Compared with the existing address matching methods, the IAMGAM improves the accuracy by 0.66% to 2.57%, and the F1-score by 0.68% to 2.55%, outperforming baseline models. Additionally, ablation experiments confirm the effectiveness of each component within the model. Furthermore, when fine-tuned using ChatGLM2-6B, the results show that the IAMGAM still outperforms ChatGLM2-6B. IAMGAM demonstrates excellent performance in Chinese address matching tasks, and the Large Language Model (LLM)-based methods, such as ChatGLM2-6B, show great potential for future development in this area. • The employed GAM effectively captures complex relations between address elements. • The GAM network extracts rich features of the addresses. • The proposed IAMGAM outperforms other methods in address-matching tasks.