AI and Big Data in Didi’s Mapping Services – Insights from WGDC 2018

Didi’s ride‑hailing app first shows green recommended pick‑up points; the accuracy and timeliness of map data, route planning, and estimated time of arrival (ETA) are fundamental for smooth travel.

Leveraging massive real‑time travel data, Didi integrates machine‑learning and deep‑learning algorithms into its map system to enhance these services, and a range of AI technologies underpins its products and features.

At the WGDC 2018 conference in Beijing, Zhang Xian, General Manager of Didi Maps, detailed the AI techniques behind Didi’s mapping platform.

Didi Maps provides core services such as ETA, route planning, pick‑up points, and the “you may want to go” recommendation, and supports higher‑level business systems including fleet dispatch, demand‑supply forecasting, car‑pooling, and intelligent order allocation. Machine‑learning and deep‑learning are extensively used to improve accuracy, processing capability, and data update efficiency.

Specific algorithmic examples include a new ETA model built on massive real‑time data that overcomes traditional shortcomings, a demand‑supply prediction model using billions of orders and driver locations to forecast regional demand, and an intelligent order‑allocation system that applies reinforcement learning to continuously refine matching probabilities and optimize overall traffic efficiency.

When the system detects a passenger’s pick‑up location, it can predict the destination within 2 ms and recommend the most likely target, achieving about 90 % prediction accuracy for the “you may want to go” feature.

Through big‑data algorithms, Didi recommends nearby optimal pick‑up points, reducing driver‑passenger communication and improving trip efficiency. As of 2017, the platform hosts over 30 million recommended points, with 75 % of users directly using this feature, resulting in an average distance of less than 25 m to a point.

Didi employs nearly 9,000 staff, almost half of whom are big‑data scientists or engineers. Within the map division, 40 % come from the mapping industry, 30 % specialize in machine learning, and another 30 % have a computer‑science background.

WGDC is a leading GIS conference that promotes innovation in spatial big‑data. During the summit, Didi announced the “Didi Maps Open Topics” program, offering de‑identified data, computing resources, and funding to universities and research institutions to accelerate high‑quality research and its industrial application.

The first three open topics are: “Intelligent Traffic Path Computation”, “Multi‑Sensor Indoor Positioning”, and “Large‑Scale Scene 3D Reconstruction”. Interested parties are invited to contact [email protected] for more information.