Multisource point-of-interest matching method based on multi-attribute feature similarity

Multisource Point-of-Interest (POI) matching refers to the process of recognizing and accurately associating the same real-world location in different data sources and is a core technology in multi-source geospatial data fusion. Aimed at eliminating the current difficulty in efficiently capturing the complex semantic information of texts in Chinese POI matching, leading to weak semantic constraints, a POI matching method based on multi-attribute feature similarity was proposed. The method fully combines the characteristics of Chinese characters and considers the semantic features of non-spatial attributes; calculates the similarity from five dimensions: name, sound, address, category, and contact information; and adopts a machine-learning algorithm of Support Vector Classification (SVC) to transform the POI matching problem into a binary classification problem. Finally, the matching method was validated using the public POI dataset of Wuhou District, Chengdu City, in AMap and Tencent Map. The experimental results show that the proposed POI matching method has high matching accuracy and reliable applicability, and precision, recall, and F1 score reach high levels among similar methods.