AbnormalLog: A Deep Anomaly Detection Method for Log Sequence Data
Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: Log Data Analysis, Anomaly Detection, Complex Sequential Data analysis, Log parsing
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Abstract: Anomaly detection for computer log sequence data plays a very important role in various industries. Log data is complex time series with plenty of text information, which is difficult to process due to both its non-structural characteristics and temporal correlation. Existing log anomaly detection schemes do not utilize all available data information such as the semantic and parameter information, nor do they consider weighting of data based on time. The AbnormalLog algorithm proposed in this paper implements semantic parsing technique to expand current detection schemes by analyzing template and parameter information of the log data. AbnormalLog is comprised of four functional modules: Log Parsing, Semantic Embedding, Parameter Anomaly Detection and Template Anomaly Detection. We compare the proposed method to three most commonly used log anomaly detection methods in industry. The results demonstrate that AbnormalLog is superior to the other algorithms with respect to common model evaluation criteria.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 184
Loading