Institutional Repository of Vilnius University
Automatinis klaidų aptikimas programinės įrangos kodo peržiūrose /
Abstract
dc:descriptionThis thesis analyzes methods of integrating the outputs of rulebased SA tools with LLMs for automated flaw detection in Java code reviews. PMD was chosen as the static analysis tool, and Qwen2.5Coder Instruct as the base model. The work addresses the high falsepositive rate of standalone static analyzers and the cost and latency of manual reviews. A mixed approach is proposed to improve detection effectiveness in which PMD warnings are fed into the LLM’s commentgeneration pipeline. Two main methods were implemented and evaluated. The first method fine tunes the LLM by augmenting real code review comments with PMD warnings and filtering out lowquality examples in the training dataset. The second method uses a base model with CoT, USC, and fewshot prompting techniques. Both methods outperform using the LLM alone, yet they still miss more complex issues that demand deeper contextual understanding.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Klinkovas, Meinardas,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD239031215&prefLang=en_US