In this paper, we base our research on a fictional collective programming scenario: A group of physically-distributed programmers try to collaboratively solve a programming problem in a web-based development environment, through a continually executing loop, consisting of three concurrent activities: exploration, integration, and feedback. In exploration, a programmer is freely to submit a sequence of gradually improved solutions until achieving a correct one. All programmers’ latest submissions are integrated into a collective artifact through integration. And through feedback, any programmer who hasn’t achieved a correct solution is continuously pushed with personalized feedback information from the collective artifact, to help the programmer improve her/his submission. In order to facilitate the realization of this fictional scenario, we narrow the target problems to those introductory programming problems, design a genetic algorithm to integrate a set of syntax-correct programs into a collective program dependence graph (CPDG), and propose an automatic feedback generation method based on the CPDG and a programmer’s latest submission. The key idea is to generate feedback from mutual inspiration: Any programmer’s submission (even not correct) may possess information that could provide inspiration for others. We evaluate the proposed approach through a set of simulated experiments, as well as a set of real experiments. The results show that our approach has a precision of 90% and a recall of 80% in randomly generated data sets on average, and a precision of 69% and a recall of 77% in real student submissions on average.

中文翻译：

---

走向虚构的集体编程场景：一种基于 EIF 循环的方法

在本文中，我们的研究基于虚构的集体编程场景：一组物理分布的程序员尝试在基于 Web 的开发环境中协作解决编程问题，通过一个持续执行的循环，由三个并发活动组成：探索、整合和反馈。在探索中，程序员可以自由地提交一系列逐渐改进的解决方案，直到获得正确的解决方案。所有程序员的最新提交通过集成集成到一个集体工件中。并且通过反馈，任何没有得到正确解决方案的程序员都会不断收到来自集体工件的个性化反馈信息，以帮助程序员改进她/他的提交。为了便于实现这个虚构的场景，我们将目标问题缩小到那些介绍性的编程问题，设计了一种遗传算法将一组语法正确的程序集成到一个集体程序依赖图（CPDG）中，并提出了一种基于 CPDG 和程序员最新提交的自动反馈生成方法. 关键思想是从相互启发中产生反馈：任何程序员的提交（即使是不正确的）都可能包含可以为他人提供灵感的信息。我们通过一组模拟实验以及一组真实实验来评估所提出的方法。结果表明，我们的方法在随机生成的数据集中平均准确率为 90%，召回率为 80%，在真实学生提交的平均准确率为 69% 和召回率为 77%。