PropLLM: Propagation-Aware Scene Reconstruction for Network Fault Diagnosis
Researchers have introduced PropLLM, a novel approach for diagnosing network faults by reconstructing propagation paths. Unlike previous methods that map alerts in a single pass, PropLLM traces faults hop-by-hop, using a knowledge graph and a Temporal Causal Propagation Attention mechanism to gather evidence and guide reasoning. This method aims to resolve ambiguity in end-point symptoms and accurately pinpoint root causes. Experiments on real-world datasets show PropLLM improves diagnosis accuracy and root cause localization while significantly reducing hallucinations. AI
IMPACT Introduces a new method for network fault diagnosis using LLMs, potentially improving operational efficiency and reducing downtime.