University of Linz research explores LLMs for software dependency upgrades

A master's thesis from the University of Linz investigates the use of Large Language Models (LLMs) to assist in software dependency upgrades. The research addresses the difficulty developers face when identifying changes between library versions.

To support this, a tool named SemanticAnalyzer was developed. It generates structured semantic analysis reports by utilizing documentation sources and abstract syntax tree (AST) diffs of Android and Java libraries. The study found that using documentation-based prompts with cloud-based models produced the most reliable reports with minimal incorrect claims, whereas local models faced limitations with larger inputs.