一句话：用 LLM 智能体自动构建并迭代修正 GitHub 仓库的层次分类体系，端到端完成仓库归类。
GitHub 开源生态缺乏系统的层次化分类：主流机制 GitHub Topics 扁平、不一致且仅覆盖 67% 的仓库。本文提出 ATLAS，首个能自动构建软件仓库层次分类体系并端到端完成归类的框架。它把 LLM 的全局知识与真实仓库分布相结合，由 Designer Agent 提出划分维度、Classifier Agent 分配仓库，并用自我纠正精化循环依据分类失败、通过逐级升级的策略修订维度。作者在 54,387 个 GitHub 仓库上，与覆盖四类范式的六个基线对比，并在两个下游任务和三个模型族上评估。在 2,001 仓库的分层基准上，ATLAS 的分类体系质量 F 值（TQF）达 83.13%，超出最佳基线 15 个百分点（全量 54k 语料上约 73.0%，差距源自 Path Granularity 对长路径的全有全无评分，而非分类准确率更低），是唯一同时具备高结构质量与高实用性的方法。下游任务中，该分类体系支持软件替代品发现达到 P@1 = 85.71%，超过人工整理列表（62.34%），并在仓库检索上取得最高 P@1。分类体系还揭示出扁平标签或相似度方法难以发现的结构性生态趋势：从库向 AI/ML 应用的转移仅在层次化、基于类型的归类下才可见，该类项目已占新近被社区采纳项目的 61%。
The open-source ecosystem on GitHub lacks a systematic hierarchical taxonomy of software repositories. GitHub Topics, the dominant organizational mechanism, is flat, inconsistent, and covers only 67% of repositories. We present ATLAS, the first framework that automatically constructs a hierarchical taxonomy for software repositories and classifies repositories into it end-to-end. By combining LLM global knowledge with real repository distributions, ATLAS proposes meaningful splitting dimensions and iteratively corrects those that fail to accommodate real repositories. A Designer Agent proposes splitting dimensions while a Classifier Agent assigns repositories; a self-corrective refinement loop uses classification failures to drive dimension revision through escalating strategies. We evaluate ATLAS on 54,387 GitHub repositories against six baselines spanning four paradigm families, two downstream software engineering tasks, and three model families. On a stratified 2,001-repository benchmark, ATLAS achieves a Taxonomy Quality F-score (TQF) of 83.13%, outperforming the best baseline by 15 percentage points (on the full 54k corpus the approximate TQF is 73.0%, a gap driven by Path Granularity's all-or-nothing scoring on longer paths rather than lower classification accuracy; see Section 4.5). It is the only method to simultaneously achieve high structural quality and high practical applicability. On downstream tasks, the ATLAS taxonomy enables software alternative discovery with P@1 = 85.71%, surpassing even human-curated lists (62.34%), and achieves the highest P@1 for repository retrieval. The taxonomy further reveals structural ecosystem trends that are difficult to obtain from flat tags or similarity methods: the shift from libraries to AI/ML applications (now 61% of newly community-adopted projects) becomes visible only through hierarchical, type-based categorization. An interactive taxonomy explorer is available at https://atlas-taxonomy.netlify.app/.
@inproceedings{abs-2606-21597,
title = {ATLAS: Agentic Taxonomy of LArge-Scale Software Ecosystems},
author = {Junyi Lu and Mengyao Lyu and Jiahui Wu and Lei Yu and Chengwei Liu and Fengjun Zhang and Li Yang and Chun Zuo and Yang Liu},
booktitle = {ASE},
year = {2026},
doi = {10.1145/3832783.3834361}
}