Didi Showcases AI‑Driven Intelligent Transportation Research at ACM SIGIR 2018

Recently, the premier information retrieval conference ACM SIGIR 2018 was successfully held in Ann Arbor, Michigan, USA. The Didi technology team participated deeply in the conference and hosted an Intelligent Transportation Information symposium, where they detailed Didi’s explorations and practices in the mobility field and shared experiences of industry‑academia collaboration. Didi also expressed its intention to continue open cooperation with researchers to address global transportation and environmental challenges.

Artificial Intelligence Technologies Multi‑Level Transform Transportation

ACM SIGIR, organized by the Association for Computing Machinery, is the most important academic conference in the information retrieval field. The 41st edition received 736 paper submissions, accepted 184 (including 86 long papers out of 409 submissions), and attracted nearly 800 participants.

Didi’s technical team had a paper titled “Taxi or Hitchhiking: Predicting Passenger's Preferred Service on Ride Sharing Platforms” accepted at the conference. The paper models user travel choices and proposes a recommendation system based on temporal, spatial, and behavioral features, which, according to offline simulations, significantly improves prediction accuracy and helps users plan trips more efficiently.

Didi algorithm experts explained the paper on site, attracting many domestic and international peers and scholars.

Didi’s Vice President and AI Labs head Prof. Ye Jiepeng delivered a keynote speech, describing how Didi leverages artificial intelligence to enhance user travel experience and tackle global traffic challenges. He highlighted Didi’s work in intelligent dispatch, intelligent maps, intelligent customer service, speech recognition, and smart traffic, as well as the underlying AI technologies such as speech, natural language processing, and computer vision. He also discussed AI‑enabled applications that improve user experience, support smart city traffic networks, and advance autonomous driving and new‑energy vehicles.

Prof. Ye Jiepeng, AI Labs head, discussed Didi’s AI layout and technological innovation at SIGIR.

Didi has cooperated with more than 20 cities (including Jinan, Guiyang, Shenyang, Nanjing, Wuhan, etc.), optimizing over 1,300 smart traffic lights and reducing congestion time by 10‑20% on average. Didi pledged to continue investing and collaborating widely, extending AI applications to public‑welfare domains.

Industry‑Academia Collaboration Accelerates Intelligent Transportation Innovation

Didi’s Vice President of Smart Mobility, Qu Xiaohu, shared Didi’s industry‑academia cooperation experience at the SIGIR big‑data forum. He emphasized Didi’s rich data assets, big‑data advantages, and ongoing frontier technology research, while actively partnering with academia to propose sustainable solutions.

Qu Xiaohu explained Didi’s research cooperation and talent‑cultivation mechanisms on site.

Qu noted that Didi has opened de‑identified data resources and computing infrastructure to academia, and has established collaborations with more than ten research institutions worldwide, including the University of Michigan, Stanford AI Lab, China Computer Federation, Hong Kong University of Science and Technology, and the Electrical and Electronics Engineers Association. Joint work spans AI, smart traffic, autonomous driving, economics, and operations research, with a focus on talent cultivation. Didi will soon announce a new round of thematic research programs to engage more scholars.

Prof. Pascal Van Hentenryck from the University of Michigan shared his views on the future of mobility.

Didi also hosted an Intelligent Transportation Information symposium, inviting professors from the University of Michigan, Washington University, and others to share cutting‑edge research in urban traffic governance. Seven papers were selected for presentation.

Didi highlighted two papers: “POI Semantic Model with a Deep Convolutional Structure” and “DiDi Ride Cancellation Smart Fault Determination System.” The POI semantic model uses a deep convolutional network to map input text into a semantic vector space, enabling similarity calculations that improve POI retrieval relevance and user satisfaction. The ride‑cancellation fault determination system introduces a hybrid machine‑learning‑plus‑rule algorithm that incorporates tens of thousands of features, achieving higher accuracy and recall, thereby markedly enhancing user experience.

Assistant Professor Zhang Weinan from Shanghai Jiao Tong University praised the symposium for demonstrating how information retrieval can empower new domains. Associate Professor Ban Xuegang from Washington University also commended Didi’s openness in data sharing and collaboration, noting that this approach is worth emulating by more enterprises.

•  END  •