A location conversion method for roads through deep learning-based semantic matching and simplified qualitative direction knowledge representation

摘要

Qualitative direction knowledge that appears in natural language descriptions of road-related locations could point to the interior of individual roads or associate multiple roads. Interpreting such descriptions to perform location conversion for roads will support intelligent road-related location services. Existing geocoding technologies could perform textual or semantic matching to transform road names to spatial locations, and research on qualitative direction reasoning could perform efficient location conversion based on semantic queries of qualitative direction knowledge between roads. However, research on geocoding lacks the consideration of matching the described internal direction knowledge of a road to a part of the road. Moreover, efficient location conversion based on semantic queries cannot scale to large road datasets due to the retrieval efficiency of a large amount of qualitative direction knowledge between roads. To accomplish this goal, this study proposes a location conversion method for roads, wherein a road ontology is designed to model the interior direction knowledge of the roads, a deep learning-based road semantic matching model is trained to match the internal direction knowledge descriptions and road segments, and a simplified qualitative direction knowledge representation between roads is performed to support rapid location conversion between roads based on efficient semantic queries. The proposed method was implemented on a road dataset of New York State. The results demonstrate that the proposed method can be effectively applied in road location conversion based on descriptions that contain qualitative direction knowledge inside individual roads or between multiple roads, which expands the scope of artificial intelligence applications.