用代码语法树构建图谱，比大模型生成更准更快地理解代码架构。
Reliable Graph-RAG for Codebases: AST-Derived Graphs vs LLM-Extracted Knowledge Graphs
- 用 Tree-sitter 和双向遍历生成确定性代码图谱，替代大模型提取知识图谱。
- 在 Shopizer 上，确定性图谱正确率最高，大模型图谱有377个文件遗漏。
- 构建速度快、成本低，适合需要精准多跳推理的代码理解场景。
软件工程中的检索增强生成常依赖向量相似性搜索，但难以处理控制器→服务→仓库等多跳架构推理。本文在 Java 代码库（Shopizer，另含 ThingsBoard 与 OpenMRS Core）上对比三种检索流程：(A) 仅向量的无图 RAG，(B) 大模型生成的知识图谱 RAG（LLM-KB），(C) 基于语法树（AST）的确定性知识图谱 RAG（DKB），使用 Tree-sitter 与双向遍历构建。每库测试 15 个架构与代码追踪查询，评估索引时间、查询延迟、语料覆盖率、成本及答案正确性。DKB 索引仅需数秒，而 LLM-KB 耗时更长；在 Shopizer 上，LLM-KB 有 377 个文件被跳过或遗漏，嵌入块覆盖率与图规模低于 DKB。端到端成本方面，DKB 相对基线略高但远低于 LLM-KB，尤其在代码库扩大时差距显著。查询延迟上，无图与 DKB 接近，而 LLM-KB 更慢且波动大。在 Shopizer 测评中，DKB 正确率最高，LLM-KB 次之，向量基线在上游架构查询中表现最差，且幻觉风险最高。总体而言，确定性 AST 图谱在覆盖可靠性与多跳定位上优于大模型提取图谱，且索引成本大幅降低。
原文摘要 · Abstract (English)
Retrieval-Augmented Generation for software engineering often relies on vector similarity search, which captures topical similarity but can fail on multi-hop architectural reasoning such as controller to service to repository chains, interface-driven wiring, and inheritance. This paper benchmarks three retrieval pipelines on Java codebases (Shopizer, with additional runs on ThingsBoard and OpenMRS Core): (A) vector-only No-Graph RAG, (B) an LLM-generated knowledge graph RAG (LLM-KB), and (C) a deterministic AST-derived knowledge graph RAG (DKB) built with Tree-sitter and bidirectional traversal. Using 15 architecture and code-tracing queries per repository, we measure indexing time, query latency, corpus coverage, cost, and answer correctness. DKB builds its graph in seconds, while LLM-KB requires much longer graph generation. LLM-KB also shows indexing incompleteness: on Shopizer, 377 files are skipped or missed, reducing embedded chunk coverage and graph size compared to DKB. End-to-end cost is modest for DKB relative to the vector-only baseline but much higher for LLM-KB, especially as repository scale increases. Query latency is similar for No-Graph and DKB, while LLM-KB is slower and more variable. On the Shopizer question suite, DKB achieves the highest correctness, LLM-KB is close behind, and the vector-only baseline performs worst on upstream architectural queries and has the highest hallucination risk. Overall, deterministic AST-derived graphs provide more reliable coverage and multi-hop grounding than LLM-extracted graphs at substantially lower indexing cost.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。