Didi AI Labs Showcases Oral Papers and AI Innovations at KDD 2019
Didi AI Labs presented over ten technical reports at KDD 2019, including three oral papers on dialogue summary generation, deep value-network dispatching, and environment reconstruction with hidden confounders, plus invited talks, posters, demos of AI-driven mobility solutions, earning recognition for breakthroughs in ETA, path planning, and demand prediction.
From August 4‑8, 2019, the 25th ACM SIGKDD Conference (KDD 2019) was held in Anchorage, Alaska, gathering more than 3,100 researchers and industry professionals. Didi participated with over ten technical reports, including three oral papers that highlight the company’s recent AI research.
Oral Paper 1: Automatic Dialogue Summary Generation for Customer Service – Didi’s AI Labs introduced a Leader‑Writer network that generates concise, logical, and accurate summaries of customer service tickets using deep learning and a key‑point sequence guidance.
Oral Paper 2: A Deep Value‑Network Based Approach for Multi‑Driver Order Dispatching – The paper presents a novel deep reinforcement learning method combined with a semi‑Markov decision process to optimize long‑term spatial‑temporal objectives for ride‑sharing dispatch, achieving more accurate value estimation.
Oral Paper 3: Environment Reconstruction with Hidden Confounders for Reinforcement‑Learning‑Based Recommendation (DEMER) – In collaboration with Nanjing University’s LAMDA Lab, Didi proposed a reconstruction framework that treats invisible confounders as hidden policies, modeling the environment, recommender, and user as three interacting agents. The approach leverages imitation learning and GANs to recreate realistic environments, validated on Didi’s driver activity recommendation platform with significant offline and online gains.
In addition to the oral papers, Didi’s senior leaders delivered invited talks. Didi Vice President and AI Labs head Ye Jiepeng presented “Transforming Transportation: A Data‑Driven Approach,” discussing AI applications across the entire ride‑sharing workflow, including reinforcement learning for dispatch, behavior‑economics‑driven gamification, and large‑scale data analytics. Didi Vice President and CTO Qu XiaoHu gave a talk titled “Smart Mobility: Redefine Transportation with AI,” covering intelligent dispatch, map services, and demand‑supply prediction.
The conference also featured a poster session and on‑site demonstrations of Didi’s AI‑driven products such as intelligent order dispatch, AI‑powered customer service, AIoT solutions, and the newly announced “Group Goose Plan – Open Mobility Platform.” The Chinese Association for Artificial Intelligence recognized Didi’s breakthroughs in ETA, path planning, intelligent dispatch, and demand prediction as internationally leading.
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
