We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.CY

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computers and Society

Title: Moving Metric Detection and Alerting System at eBay

Abstract: At eBay, there are thousands of product health metrics for different domain teams to monitor. We built a two-phase alerting system to notify users with actionable alerts based on anomaly detection and alert retrieval. In the first phase, we developed an efficient anomaly detection algorithm, called Moving Metric Detector (MMD), to identify potential alerts among metrics with distribution agnostic criteria. In the second alert retrieval phase, we built additional logic with feedbacks to select valid actionable alerts with point-wise ranking model and business rules. Compared with other trend and seasonality decomposition methods, our decomposer is faster and better to detect anomalies in unsupervised cases. Our two-phase approach dramatically improves alert precision and avoids alert spamming in eBay production.
Comments: The work is oral presented on the AAAI-20 Workshop on Cloud Intelligence, 2020
Subjects: Computers and Society (cs.CY); Machine Learning (cs.LG); Social and Information Networks (cs.SI); Machine Learning (stat.ML)
Cite as: arXiv:2004.02360 [cs.CY]
  (or arXiv:2004.02360v1 [cs.CY] for this version)

Submission history

From: Zezhong Zhang [view email]
[v1] Mon, 6 Apr 2020 00:28:39 GMT (427kb,D)

Link back to: arXiv, form interface, contact.