Leveraging Large Language Models for Log Anomaly Detection: Case Studies and Comparative Analysis

School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu, Sichuan, 610000, China

^* Corresponding author: 2023310101020@std.uestc.edu.cn

Abstract

As computer systems become more complex, log anomaly detection has become more crucial and challenging. However, the conventional methods’ dependence on log parsing places significant limitations on them. Many problems still beset the widely used deep- learning techniques. Large Language Models (LLMs) offer an innovative approach to these restrictions via their advanced contextual reasoning and natural language comprehension skills. The enormous potential of LLMs to improve performance from an abundance of aspects throughout the various log anomaly detection stages will serve as the primary focus of this study. Several frameworks are examined in the case study, applying LLMs to improve the frameworks’ performance during the anomaly detection stage as well as to construct the preprocessed data. The study will additionally explore how some LLM-related technologies, for example, Retrieval- Augmented Generation (RAG), integrate into this field. In the meantime, the benefits and drawbacks of these frameworks are also discussed. The study will demonstrate the tremendously vast potential of LLM in log anomaly detection through the comparative analysis. Finally, this paper discusses the current challenges and limitations of LLM-based approaches and proposes directions for future development.