Quick Q&A: Insights from JD JDATA Algorithm Competition
This article presents a rapid Q&A session with JD data scientists and architects, covering the benefits of algorithm contests for students, the unique advantages of the JDATA competition, scoring formulas, ways to improve results, strong feature extraction, real‑time modeling, algorithm selection, and the value of the competition’s special offer for future employment.
Guest Speakers
Li Kaidong – JD Mall Data Scientist; Zhou Mo – JD Mall Data Architect; Wang Luping – JD Mall Data Scientist and 2017 JDATA runner‑up; and the competition committee.
Q: What are the benefits for students to join algorithm competitions?
Li explains that campus datasets are small and outdated, so competitions let participants push algorithms to their limits and simulate real business scenarios, where the focus is on solving user needs. Competitions also provide exposure, networking, and attractive prizes that boost employability.
Q: What advantages does the JD JDATA competition have over similar contests?
Li highlights the competition’s high quality, realistic data, strict anti‑cheating measures, and its role as a talent pipeline for JD, recruiting top performers directly.
Q: What do the scoring formulas S1 and S2 evaluate?
S1 measures whether a user has purchased a product (higher purchase probability yields higher score); S2 measures the temporal distance between the predicted purchase date and the actual purchase date.
Q: How can participants improve their scores after submission?
Li suggests continuous idea generation, extensive data visualization to identify key features, team collaboration, model fusion, and iterative refinement.
Q: How to create strong features?
Zhou stresses deep domain knowledge; strong features arise from understanding the business, not just sophisticated statistics or models.
Q: How to judge feature importance without a model?
Zhou gives the example of “association matching” (e.g., beer and diaper co‑purchase) that can be validated through online A/B tests.
Q: Can your models handle real‑time processing?
Zhou describes JD’s real‑time computing platform using Flink for precise statistics and Spark Streaming for modeling, employing a Lambda architecture (offline batch + online incremental updates).
Q: How to choose algorithms for data‑mining competitions?
Wang categorizes methods into three groups: mathematical modeling, classic statistical machine learning, and modern representation learning, outlining their advantages, disadvantages, and suitable scenarios.
Q: Does the JDATA “Special Offer” help with JD employment?
The committee confirms that the offer is highly valuable, allowing candidates to negotiate early offers, gain longer tenure, and integrate into JD’s business faster.
Readers are encouraged to leave questions; JD will periodically invite algorithm experts to answer them.
JD Retail Technology
Official platform of JD Retail Technology, delivering insightful R&D news and a deep look into the lives and work of technologists.
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.