Skip navigation

DSpace JSPUI

DSpace preserves and enables easy and open access to all types of digital content including text, images, moving images, mpegs and data sets

Learn More
DSpace logo
English
中文
  • Browse
    • Communities
      & Collections
    • Publication Year
    • Author
    • Title
    • Subject
  • Search TDR
  • Rights Q&A
  • Help
    • My Page
    • Receive email
      updates
    • Edit Profile
  1. NTU Theses and Dissertations Repository
  2. 管理學院
  3. 資訊管理學系
Please use this identifier to cite or link to this item: http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/94666
Title: 基於對比學習實現之自監督式系統日誌異常檢測
Framework for Self-Supervised Log Anomaly Detection Based on Contrastive Learning
Authors: 王佑豪
Yu-Hao Wang
Advisor: 莊裕澤
Yuh-Jzer Joung
Keyword: 系統日誌,異常檢測,非監督式學習,對比學習,時間序列分析,
System Log,Anomaly Detection,Unsupervised Learning,Contrastive Learning,Time Series Analysis,
Publication Year : 2024
Degree: 碩士
Abstract: 本研究針對系統日誌異常檢測問題提出了一個創新的二階段訓練框架,先利用日誌的模板(Template)資訊訓練一個基於 Contrastive Learning 的 Transformer 模型,將日誌直接轉換成一個特徵向量的資料點,我們會保證資料點能夠保留事件模板的資訊,並套用在下游的正常日誌之建模任務上,使得我們可以對異常日誌進行檢測。根據本論文的架構設計,我們可以在非監督式的異常檢測任務上獲得平均 94.84 的 F1 Score,比先前的 State-of-the-Art 模型 LogBERT 的表現(89.93)還要更高。文獻中的非監督式學習模型會將日誌轉換為模板後,使用一個 Embedding 層來將模板對應到特徵向量,而本研究會在推論時會直接使用日誌內容轉換為特徵向量,我們會在展示與其他模型的表現比較後,接著討論這個架構設計的合理性。其次,我們將提出一個方法,在推論時使用簡單的嵌入層替換掉 Transformer 模型,這樣的取代能夠大幅增加推論速度,使該論文架構能夠成為線上日誌異常檢測的優秀選擇。
This study proposes an innovative two-stage training framework for the problem of system log anomaly detection. First, we use the template information of the logs to train a Transformer embedder based on Contrastive Learning, directly converting the logs into feature vector data points. We ensure that the data points retain the event template information and apply them to the downstream task of modeling normal logs by training a simple LSTM network, allowing us to detect anomalous logs. According to the framework design in this paper, we achieve an average F1 Score of 94.84 in unsupervised anomaly detection tasks, which is higher than the previous state-of-the-art model LogBERT's performance (89.93). We will discuss our intuition and rationale in depth in the experiment part. Moreover, we propose a method to get rid of the embedder at inference time, making it a competitive option for online log anomaly detection system.
URI: http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/94666
DOI: 10.6342/NTU202402929
Fulltext Rights: 同意授權(全球公開)
metadata.dc.date.embargo-lift: 2029-07-31
Appears in Collections:資訊管理學系

Files in This Item:
File SizeFormat 
ntu-112-2.pdf
  Until 2029-07-31
2.08 MBAdobe PDF
Show full item record


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

社群連結
聯絡資訊
10617臺北市大安區羅斯福路四段1號
No.1 Sec.4, Roosevelt Rd., Taipei, Taiwan, R.O.C. 106
Tel: (02)33662353
Email: ntuetds@ntu.edu.tw
意見箱
相關連結
館藏目錄
國內圖書館整合查詢 MetaCat
臺大學術典藏 NTU Scholars
臺大圖書館數位典藏館
本站聲明
© NTU Library All Rights Reserved