🤖 AI Summary
To address the low efficiency of code understanding and file localization in large-scale software repositories, this paper proposes a graph-augmented hybrid retrieval framework. First, it leverages large language models (LLMs) to extract semantic summaries and generate fine-grained vector embeddings, while integrating static analysis to construct a knowledge graph encoding syntactic–semantic relationships—including inheritance, method calls, and references. Second, it introduces a graph-aware retrieval expansion mechanism that jointly optimizes semantic similarity matching and subgraph traversal, supporting LLM-generated constrained natural language queries and interpretable reasoning. Experimental results demonstrate significant improvements in accuracy and robustness for problem-driven file retrieval across multiple open-source projects. The approach achieves a substantial leap in automated code understanding capability and establishes a scalable, knowledge-infused infrastructure for intelligent development toolchains.
📝 Abstract
We present a repository decomposition system that converts large software repositories into a vectorized knowledge graph which mirrors project architectural and semantic structure, capturing semantic relationships and allowing a significant level of automatization of further repository development. The graph encodes syntactic relations such as containment, implementation, references, calls, and inheritance, and augments nodes with LLM-derived summaries and vector embeddings. A hybrid retrieval pipeline combines semantic retrieval with graph-aware expansion, and an LLM-based assistant formulates constrained, read-only graph requests and produces human-oriented explanations.