Cycling Map Inference Using Global Positioning System Trajectories: A Case Study in the Online Food Delivery Business

Abstract

The online food delivery (OFD) business is booming in China. Owing to the timeliness requirements, delivery personnel in OFD platforms usually use electric bicycles to make deliveries. However, the accuracy and the coverage rate of existing cycling maps are relatively low, as is evidenced by a considerable amount of cycling global positioning system (GPS) trajectories that cannot be matched to existing maps, thus the efficiency of delivery is affected. Although there has been a proliferation of studies on driving or walking map inference using GPS trajectories, to the authors’ knowledge, none of them systematically investigate the cycling scenario. Our study addresses this gap. We work with Meituan—the largest OFD platform in China—and use the GPS trajectories reported by delivery personnel to infer the underlying cycling map. We first adapt three popular map inference algorithms, namely, k-means clustering, kernel density estimation, and trace merging. We also propose a new approach that infers the cycling network. We perform an initial inference of the underlying road network through an iterative process and apply a series of map refinement techniques to further improve the appearance of the inferred road network. The result shows that our algorithm reaches an F-score of 0.41, whereas the best existing algorithm we adapt reaches an F-score of 0.39. We also consider a special case that uses the driving map information in the area. In this case, a map-matching step is included and the overall F-score further increases from 0.41 to 0.70.