Hierarchical Code Abstraction via Unified GraphRAG on Multi-Grained Graphs
Zhongtian Bao, Wenjian Ding, Jun Wang, Yao Zhang, Zhenglu Yang, Zhe Sun, Andrzej S Cichocki
Nankai University Ludong University Juntendo University Nicolaus Copernicus University
阅读操作
确认中在文库中上传 PDF 后可生成中文音频讲解。
摘要与影响
Code summarization plays a crucial role in helping developers quickly grasp the purpose and behavior of source code. However, the majority of existing approaches concentrate solely on the function level, overlooking the richer contextual signals and hierarchical relationships that naturally exist within large software repositories. To address these shortcomings, we present a new hierarchical code summarization framework built upon GraphRAG. Our key observation is that the community structures revealed by GraphRAG inherently align with the layered organization of real-world software systems—for instance, functions forming classes, and classes composing files. Leveraging this insight, we construct a multi-level code summary graph and dynamically retrieve the most relevant subgraph to supply a large language model with structured and contextually coherent information. This design alleviates the context fragmentation commonly seen in traditional retrieval-augmented generation pipelines. Comprehensive experiments demonstrate that our framework markedly improves summary quality, especially when applied to complex or large-scale codebases.
逐年被引趋势
暂无年度引用数据
关键指标
同类平均 = 1
同领域 · 同年份 · 同类型
Google Scholar 与 OpenAlex 的被引统计范围不同,数值存在差异属正常。
AI 辅助阅读
依据:摘要
可就本文提问;依据不足时会说明。
学术脉络
学科主题
计算机 / AISoftware Engineering Research
Topic Modeling · Model-Driven Software Engineering Techniques