【Meituan Technology Salon】Sponsored by the Meituan technical team and the Meituan Association for Science and Technology, each salon invites technical experts from Meituan and other Internet companies to share practical experience from the frontline, covering all major technical fields.
From September 2020, the Meituan Technology Salon will create a series of academic activities, including top conference paper sharing, academic hotspot discussions, etc., inviting industry and academia to discuss cutting-edge issues.
Activity time : 14:00-16:30, September 5, 2020
Event address : online event
Event registration : poke me to sign up
| Introduction
Data mining technology is widely used in the Internet industry. KDD, as the top conference in the field of data mining, is the focus of attention in academia and industry. This technical salon will introduce the results published by Meituan in KDD2020, covering distribution range optimization, ranking learning, and the experience sharing of Meituan in this KDD Cup competition. Hope to exchange and learn with the industry's technical colleagues.
| Schedule
| Share Introduction
1. Delivery Scope: A New Way of Restaurant Retrieval For On-demand Food Delivery Service
Ding Xuetao, master of Tsinghua University, is currently responsible for the research on the network planning of Meituan Daojiao's delivery, the pricing of delivery fees and the direction of supply and demand balance.
brief introduction
The scope of delivery is one of the core operating tools in food delivery. It not only determines the merchant supply that users can trade, but also determines the scope of the merchant’s service, which in turn affects the merchant’s turnover and also affects the efficiency and experience of delivery performance . How to pass the three-party appeal of the distribution scope Balance has become an important issue. This paper proposes a method based on computational geometry + machine learning + combinatorial optimization to determine the shape of each merchant, and on the basis of ensuring a reasonable range of services, to achieve a better solution for scale and delivery experience.
2. KDD Cup 2020 Winning Solution and Its Applications in Meituan Search Ads
Huang Jianqiang, master of Peking University, is currently engaged in the optimization of advertising quality estimation model in the search advertising algorithm group of the advertising platform department of Meituan Dadian.
brief introduction
The KDD Cup is a top international event in the field of data mining research hosted by SIGKDD, and is currently the most influential event in the field of data mining. Deviation optimization, automated map learning, and multi-modal matching are three of the five questions in this year’s KDD Cup. The search advertising algorithm team of the Meituan advertising platform has accumulated these three questions this year based on the technologies in these three fields. Achieved two championships in one season. This speech will share the solutions to these three competition questions and our advertising business applications in these three areas.
3. KDD Cup 2020 2nd Place Solution and Its Applications in Meituan-Dianping Search
Zuo Kai, Master of Southeast University, is currently engaged in multi-modal search related work in the Visual Application Group of the Search and NLP Department of Meituan AI Platform.
brief introduction
The KDD Cup is a top international event in the field of data mining research hosted by SIGKDD, and is currently the most influential event in the field of data mining. The review search vision team achieved the second place in the multi-modal matching task based on the accumulation of technology in the multi-modal search. This speech mainly shares our solutions and the application of multi-modal search in the search business of Meituan Dianping.
4. Learning-To-Rank with Context-Aware Position Debiasing
Xiao Keyi, master of National Taiwan University, is currently engaged in the sorting of result pages in the core sorting group of the AI platform search and NLP department of Meituan.
brief introduction
The location information in the sorting scene has a great influence on the user's click behavior, and one consequence is that the user click data collected online exists in Bias. This paper analyzes how bias causes the offline evaluation to be inconsistent with the online effect, and proposes some accurate evaluation schemes. For biased training data, the training method of debias is given, and significant results have also been achieved on the search ranking line. .
5. Ranking with Deep Multi-Objective Learning
Xuezhi Cao, Ph.D. from Shanghai Jiaotong University, is currently responsible for the construction of product knowledge graph and its application in search scenarios in the Search and NLP Department of the AI Platform of Meituan.
brief introduction
The existing ranking learning methods mainly use one of the following three training methods: single-point training, paired training and list training. Different training methods often have their own advantages in different scenarios and different evaluation indicators. In this article, we combine single-point training and list training, and propose a multi-target training method for the deep ranking model, which complements the information mined by different training methods to improve the training effect of the ranking model.
| Thanks
Event organizer : Meituan technical team, Meituan Science and Technology Association
Propaganda partners : event line, academic headlines, school online
|How to sign up
"Meituan Technology Salon Issue 54: Meituan Data Mining Technology Practice-KDD Special" registration please click to read the original text .
Scan the QR code below to add Meimei WeChat (MTDPtech03), and reply to the keyword: 0905 or KDD into the group, you can join the active WeChat group and communicate with the lecturer.