用BERTopic自动提炼社区问答，生成简洁实用的安卓API文档
Automating API Documentation with LLMs: A BERTopic Approach
- 从360万条问答中提取主流话题，用抽取式摘要生成内容
- 用户测试显示新文档显著提升开发效率与满意度
- 适合需要快速理解API用法的开发者和文档团队
开发者依赖API文档，但官方资料常冗长复杂或不完整。许多人转向社区论坛如Stack Overflow获取实用信息。本文提出自动化提炼非正式来源的方法，聚焦安卓API。利用BERTopic从360万条Stack Overflow帖子中提取主流话题，并采用抽取式摘要技术生成包含代码片段的简洁总结。通过30名安卓开发者的用户研究评估摘要的连贯性、相关性、信息量与满意度，结果显示生产力明显提升。将正式API知识与社区生成内容结合，可增强文档可用性，使API资源更易获取且可操作。
原文摘要 · Abstract (English)
Developers rely on API documentation, but official sources are often lengthy, complex, or incomplete. Many turn to community-driven forums like Stack Overflow for practical insights. We propose automating the summarization of informal sources, focusing on Android APIs. Using BERTopic, we extracted prevalent topics from 3.6 million Stack Overflow posts and applied extractive summarization techniques to generate concise summaries, including code snippets. A user study with 30 Android developers assessed the summaries for coherence, relevance, informativeness, and satisfaction, showing improved productivity. Integrating formal API knowledge with community-generated content enhances documentation, making API resources more accessible and actionable work.
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。