Meituan Builds AI on IA: From Enhanced Customer Experience to Differentiated Cloud Services

In July 2017, Meituan and Intel announced their deep collaboration in Artificial Intelligence (AI), focusing on building Deep Learning capabilities over Intel Architecture (IA). What is intriguing here is Meituan started their AI journey on GPU but eventually transitioned to Intel Xeon Phi as an alternative. Before we delve into why Meituan pivoted to IA, I want to give you some background on Meituan and their AI application.

Who is Meituan?

The $13 billion-dollar deal of Amazon buying Wholefoods marks the mainstream adoption in the US market the trend of O2O: Online to Offline. But O2O concept actually took off even earlier in China, due to the sheer number of people connected on mobile.

Meituan, the world’s largest online and on-demand delivery platform, emerged as a winner in this O2O wave. If you have not heard about Meituan, you can think of it as China’s Groupon, Yelp and Booking combined. With more than 180M active users per month and 600M total registered users as of June 20161, Meituan processes 10M daily orders and deliveries. An anticipated new funding round could potentially raise Meituan’s valuation to $25 billion to $30 billion, making it one of the most valuable private companies in the world.

Deep Learning in Meituan

Meituan leverages AI and deep learning to deliver an optimized user experience. It uses deep-learning-powered natural language processing to perform text analysis, semantic matching and search engine ranking to return the most relevant vendors and services when users conduct a search. It is worth mentioning that although semantic matching is a common technology used in search engines, Meituan’s O2O business model adds further complexity, because a search phrase may mean totally different things depending on the user’s intent and state (for example, physical location). For example, a search phrase of “Great Wall” potentially could mean tour services to The Great Wall if the user is visiting Beijing, or restaurants whose names contain “Great Wall” if the user has mostly been searching for local places to eat, or movie theaters playing the movie “The Great Wall” if the user regularly search for movies on Saturday night (38% on Rotten Tomatoes, ouch!). Meituan’s creative data scientists considered the O2O scenario when building their deep learning models, using the customer’s browsing and purchase history to guide their semantic match model optimization.

In addition, Meituan also leverages computer vision to enhance text and object recognition, image classification and image quality ranking to present the most attractive pictures of a vendor or service to a customer based on his or her sensitivity and preference with image properties such as color, resolution, and composition. Furthermore, Meituan has extended its AI expertise to its cloud service offerings, launching AI-as-a-Service in Meituan Open Services (MOS).

Why AI on IA?

Meituan has a long and collaborative relationship with Intel and that has extended to deep learning. It recently deployed Intel® Xeon® Scalable processors and Intel® Xeon® Phi to run its AI workloads for both internal applications and AIaaS in MOS.

Why has Meituan opted to build AI on IA? Here are the three top reasons:

Trust in Intel’s AI Portfolio and Strategy

Meituan trusts Intel’s capability to innovate and deliver the best AI solution on IA, and this trust roots in Intel’s strong brand, track record to deliver best-performing silicon and innovative solutions, and the long collaboration between the two companies.

Intel has the full stack of end-to-end AI products for building AI solutions at scale, from edge to cloud. Meituan started with Xeon Phi Knights Landing to build out its deep learning cluster, which can smoothly upgrade to the upcoming Knight Mill platform with more deep learning optimization. Down the road, the Intel Nervana hardware will deliver unparalleled deep learning performance and efficiency that dwarf general-purpose GPUs.

Beyond silicon, Intel is committed to optimizing the most popular analytics, machine & deep learning frameworks including Tensorflow, Caffe, Torch, and Spark-based MLlib and BigDL, so that they run best on IA.

AI Made Easy

With Intel AI solution, Meituan could deploy AI today using existing, familiar infrastructure, and streamline the management of their cloud infrastructure supporting their AI as well as other workloads. Furthermore, there is no additional burden to train developers, and they can be more productive when focusing more on the problems they want to solve with deep learning, instead of learning a new programming framework and language.

Teamwork makes the dream work

Intel does not just sell processors. Intel helps our customer solve problems, deliver amazing user experience and create differentiated services to grow in new markets. The teamwork started with joint development effort between Intel and Meituan. Intel provided hardware, software and strong technical support to help Meituan develop and enhance MOS public cloud platform through innovations, and assisted MOS to create the differentiated AIaaS offering. With strong application software expertise, Meituan provides software development and timely feedback of its development and testing results, ensuring that both sides work together smoothly2. This win-win strategic cooperation is accelerating the evolution of AI and creating the next-gen, intelligent Internet in Meituan and at large.

Intel understands how using AI can help its customers differentiate their business and maintain their competitive advantage. AI is what Intel does. To learn more about AI technology, applications, and Intel AI solutions, check out