Reproduce “Knowledge Graph Based Repository-Level Code Generation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Mihir Athale, Vishal Vaddina
This research introduces a knowledge graph-based framework for repository-level code generation that grounds Large Language Models in structured codebase context. The approach achieved a 36.36% pass@1 score on the EvoCodeBench dataset, demonstrating improved contextual accuracy and dependency alignment compared to existing methods.
Key insights
- Providing rich, structured repository context through Knowledge Graphs significantly enhances the quality of code generated by Large Language Models.
- A hybrid retrieval system combining full-text, semantic, and graph-based searches effectively identifies and leverages complex dependencies and architectural patterns within codebases.
- Grounding LLMs with relevant sub-graphs from a Knowledge Graph leads to the generation of more consistent, dependency-aware, and functionally correct code outputs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.