Bifrost: Empowering Pretrained Language Model with Fallibility Representation for Log-Based Fault Diagnosis

2026-07-25

AI Summary

Existing pre-trained language models struggle to effectively capture the multi-level structural information in system logs that reflects failures, thereby limiting log-based diagnostic performance. To address this challenge, this work proposes Bifrost, a novel approach that introduces the concept of “falsifiability-aware representations” and integrates domain knowledge from site reliability engineers. Bifrost employs a self-supervised contrastive learning strategy to jointly model both semantic and structural characteristics of logs. Evaluated on three public systems and an industrial ML-as-a-Service (MLaaS) platform, Bifrost significantly enhances failure diagnosis: it improves anomaly detection F1 by 9.83% on average, increases root cause localization Hit Rate@k by 18.28%, and boosts fault identification Macro-F1 by 20.88%.

Abstract

Log-based fault diagnosis is crucial for runtime debugging and maintenance. Existing fault diagnosis methods use language models pre-trained on natural language (PLMs) for log representation. However, system faults are reflected in the multi-level structure of system logs. PLMs pre-trained on natural language struggle to comprehensively capture multi-level fault information, failing to meet the requirements of fault diagnosis. We refer to this information as fallibility representations. To address this problem, we propose a novel log representation learning method, Bifrost. It draws inspiration from the log analysis experience of Site Reliability Engineers and meticulously designs strategies based on self-supervised contrastive learning to learn the fallibility representations of logs. Across three public systems and one industrial ML-as-a-Service system, the log representations produced by Bifrost outperform existing PLMs by average margins of 9.83% in F1 for anomaly detection, 18.28% in HR@k for root cause localization, and 20.88% in Macro-F1 for fault identification.

Problem

• log-based fault diagnosis
• pretrained language models
• fallibility representation
• multi-level log structure
• fault information

Innovation

• fallibility representation
• log-based fault diagnosis
• self-supervised contrastive learning
• pretrained language model
• root cause localization