Reproduce “CMMD: Cross-Metric Multi-Dimensional Root Cause Analysis” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Shifu Yan, Caihua Shan, Wenyi Yang, Bixiong Xu, Dongsheng Li, Lili Qiu, Jie Tong, Qi Zhang

CMMD diagnoses derived-KPI anomalies by learning relationships between underlying measures and searching dimension combinations whose expected values explain the change. In a production case, it identified Search|US, where conversions fell 43.08% and accounted for 20.41% of the conversions drop. Across five production cases, mean F1 was 0.8627 versus 0.7747 for Squeeze; ground truth used recovery-ratio thresholds, not independent causal tests.

Key insights

  • Anomalies in derived KPIs are more accurately attributed by considering their underlying fundamental metrics, allowing for more precise root cause identification than single-metric analyses.
  • Graph Neural Networks can effectively model complex, unknown aggregation and calculation functions across dimensions and metrics, providing a comprehensive understanding of anomaly propagation.
  • Genetic Algorithms offer an efficient and assumption-free heuristic search method for navigating the combinatorial explosion of multi-dimensional root causes, enabling timely diagnosis in large-scale systems.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.