Context: Software developers often ask questions on Technical Q&A forums like Stack Overflow (SO) to seek solutions to their programming-related problems (e.g., errors and unexpected behavior of code). Problem: Many questions miss required code snippets due to the lack of readily available code, time constraints, employer restrictions, confidentiality concerns, or uncertainty about what code to share. Unfortunately, missing but required code snippets prevent questions from getting prompt and appropriate solutions. Objective: We plan to introduce GENCNIPPET, a tool designed to integrate with SO's question submission system. GENCNIPPET will generate relevant code examples (when required) to support questions for their timely solutions. Methodology: We first downloaded the SO April 2024 data dump, which contains 1.94 million questions related to Python that have code snippets and 1.43 million questions related to Java. Then, we filter these questions to identify those that genuinely require code snippets using a state-of-the-art machine learning model. Next, we select questions with positive scores to ensure high-quality data. Our plan is to fine-tune Llama-3 models (e.g., Llama-3-8B), using 80% of the selected questions for training and 10% for validation. The primary reasons for choosing Llama models are their open-source accessibility and robust fine-tuning capabilities, which are essential for deploying a freely accessible tool. GENCNIPPET will be integrated with the SO question submission system as a browser plugin. It will communicate with the fine-tuned model to generate code snippets tailored to the target questions. The effectiveness of the generated code examples will be assessed using automatic evaluation against ground truth, user perspectives, and live (wild) testing in real-world scenarios.

翻译：背景：软件开发人员经常在Stack Overflow（SO）等技术问答论坛上提问，以寻求解决其编程相关问题（例如代码错误和意外行为）的方案。问题：由于缺乏现成代码、时间限制、雇主限制、保密担忧或不确定应分享哪些代码，许多问题缺失必需的代码片段。遗憾的是，缺失但必需的代码片段会阻碍问题获得及时且适当的解决方案。目标：我们计划引入GENCNIPPET，这是一个设计用于集成到SO问题提交系统中的工具。GENCNIPPET将在需要时生成相关代码示例，以支持问题获得及时解决。方法：我们首先下载了SO 2024年4月的数据转储，其中包含194万个附带代码片段的Python相关问题和143万个Java相关问题。接着，我们使用最先进的机器学习模型筛选这些问题，以识别真正需要代码片段的问题。然后，我们选择评分积极的问题以确保数据质量。我们的计划是使用所选问题的80%进行训练、10%进行验证，对Llama-3模型（例如Llama-3-8B）进行微调。选择Llama模型的主要原因是其开源可访问性和强大的微调能力，这对于部署一个可免费访问的工具至关重要。GENCNIPPET将作为浏览器插件与SO问题提交系统集成。它将与微调后的模型通信，生成针对目标问题定制的代码片段。生成代码示例的有效性将通过基于真实数据的自动评估、用户视角以及在真实场景中的现场（实际环境）测试来进行评估。