Free

Using LLMs to Detect Anomalies Present in User Behavioral Logs

Abstract

User behavioral logs may be used to detect anomalies present in them (i.e. a shift in purchases trend in an e-commerce store). This has been traditionally handled with rules based systems or statistics based machine learning modules. This thesis investigates Large Language Models (LLMs) as a highly performant alternative to the traditional approaches. Do they reduce the required data annotation and training efforts?

The study involves the creation of a small prototype that illustrates the approach.

Starting Point

Guan, W., Cao, J., Qian, S., Gao, J., & Ouyang, C. (2024). Logllm: Log-based anomaly detection using large language models. arXiv preprint arXiv:2411.08561.
Alnegheimish, S., Nguyen, L., Berti-Equille, L., & Veeramachaneni, K. (2024). Large language models can be zero-shot anomaly detectors for time series?. arXiv preprint arXiv:2405.14755.
Fariha, A., Gharavian, V., Makrehchi, M., Rahnamayan, S., Alwidian, S., & Azim, A. (2024, August). Log anomaly detection by leveraging LLM-based parsing and embedding with attention mechanism. In 2024 IEEE Canadian Conference on Electrical and Computer Engineering (CCECE) (pp. 859-863). IEEE.

Interested in this topic?

If you are a master student at the University of Tirana and would like to work on this thesis topic, please get in touch.

Contact Me