Hongyi Luo, Qing Cheng, Daniel Matos, Hari Krishna Gadi, Yanfeng Zhang, Lu Liu, Yongliang Wang, Niclas Zeller, Daniel Cremers, Liqiu Meng. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

This research introduces a novel Road Network Embeddings Matching (RNEM) method for road network matching in map conflation tasks, addressing key challenges in integrating diverse map datasets. Traditional methods like Delimited Stroke Oriented (DSO) and Hootenanny face difficulties with disparities in geometric, semantic, and topological information. RNEM leverages embeddings derived from these features, significantly improving accuracy, precision, recall, and F1 score. Using pre-trained models like Bidirectional Encoder Representations from Transformers (BERT), RNEM captures semantic and topological information, while geometric embeddings are generated through resampling and normalization of polylines. Experiments on Munich datasets show that RNEM outperforms existing methods by 3.2% in accuracy. This method represents the irst approach to incorporate semantic and topological information using NLP techniques, offering a comprehensive solution for map conflation, benefiting initiatives such as the Overture Maps Foundation.