A critical road segment identification method using two-stage feature learning with dynamic and static road segment embedding

Recently, Journal of Geo-information Science published an online research achievement led by Professor Sheng Wu from the Digital China Research Institute (Fujian) at Fuzhou University, along with his graduate student Weiyi Wu. Their research is dedicated to solving a long-standing problem: traditional methods often overlook the road segments that are crucial for local areas in low-traffic regions, which impedes effective traffic management in large-scale road networks.

In terms of research methodology, the researchers reconceptualized urban road networks as a traffic corpus, treating road segments as "words" and travel routes as "sentences". By leveraging natural language processing techniques, they first utilized Word2Vec to generate static embeddings for road segments, thereby capturing the inherent properties of roads, such as connectivity and geographical location information. Subsequently, they employed the deep contextualized ELMo model to infer dynamic embeddings for road segments, which can reflect real-time traffic changes. Meanwhile, through attention pooling and differentiable clustering techniques, they integrated the static and dynamic embeddings of road segments, enabling the adaptive identification of road segments that are crucial for both the overall urban traffic flow and local area requirements.

To verify the effectiveness of this method, they conducted a study within the area enclosed by the third ring road in Fuzhou City, using mobile positioning data. The experimental results show that this new method can effectively identify critical road segments in large-scale road networks and relatively critical segments in local areas. It offers a more accurate and practical approach for identifying critical road segments, which is of great significance for urban traffic management and optimization.

For more details, please refer to the original article:

A Critical Road Segment Identification Method Using Two-Stage Feature Learning with Dynamic and Static Road Segment Embedding.

https://www.sciengine.com/JGIS/doi/10.12082/dqxxkx.2025.240483（If you want to see the English version of the full text, please click on the “iFLYTEK Translation” in the article page.）

---

---