一句话：提出 LiTracker，用 LLM 自动追踪并分析开源项目许可证演化与兼容性风险。
开源软件许可证会随项目生命周期变化，未检测的变更会带来法律风险与生态碎片化。现有自动化方法难以识别、分类许可证，推断变更动机并评估兼容性。本文提出 LiTracker，一个基于 LLM 的端到端许可证演化追踪与分析框架。它整合 SPDX 与 OSI 的 3,324 个许可证，分析 GitHub 上 3,172 个高质量项目的 1,170 万次提交和 300 万个 issue，通过领域自适应预训练与监督微调构建 License-Qwen3-8B。该模型在许可证识别和分类上取得 90.94% 与 94.12% 的 SOTA F1。实证显示 11.54% 的项目发生许可证变更，Java、C/C++、Python 生态比例最高；7.38% 变更项目存在许可证冲突，主要是 copyleft 与 permissive 之间。
The extensive adoption of open source software (OSS) makes licensing a critical determinant of project governance, reuse, and compliance. Licenses, however, may change throughout a project’s lifecycle, and undetected modifications can introduce legal risks, ecosystem fragmentation, and community trust issues. Existing automated methods remain limited in identifying and classifying licenses, inferring change motivations, and assessing compatibility, and thus fail to meet the demands of large-scale, evolving OSS ecosystems. This paper introduces LiTracker, an Large Language Model(LLM)–based framework for automated tracking and analysis of license evolution in OSS. LiTracker integrates 3,324 licenses from SPDX and OSI, and analyzes 11.7 million commits and 3 million issues from 3,172 high-quality GitHub projects. Through domain-adaptive pretraining and supervised fine-tuning, the License-Qwen3-8B model achieves state-of-the-art F1 scores of 90.94% and 94.12% for license identification and classification. Empirical results show that 11.54% of projects experience license changes, with Java, C/C++, and Python ecosystems exhibiting the highest rates. Most changes are driven by compatibility and compliance requirements, license upgrades, or business strategy shifts. LiTracker further quantifies the compatibility risks induced by these changes: 7.38% of modified projects contain license conflicts, mainly between copyleft and permissive licenses, involving Offer Source Code, Grant Patents, and Sublicense terms. In sum, LiTracker represents the first end-to-end, LLM-driven framework for semantic analysis of license evolution, providing empirical foundations for open source governance, dependency risk control, and automated compliance support.
@inproceedings{paperbot11639,
title = {Catching License Changes Before They Cost: Leveraging LLMs to Analyze License Evolution in Open Source Software},
author = {Xing Cui and Jingzheng Wu and Tianyue Luo and Wenxiang Ou and Xiang Ling},
booktitle = {ICSE},
year = {2026},
doi = {10.1145/3774748.3793628}
}