ABSTRACT This paper investigates a range of concept-based example recommendation approaches that we developed to provide example-based problem-solving support in the domain of programming. The goal of these approaches is to offer students a set of most relevant remedial examples when they have trouble solving a code comprehension problem where students examine a program code to determine its output or the final value of a variable. In this paper, we use the ideas of semantic-level similarity-based linking developed in the area of intelligent hypertext to generate examples for the given problem. To determine the best-performing approach, we explored two groups of similarity approaches for selecting examples: non-structural approaches focusing on examples that are similar to the problem in terms of concept coverage and structural approaches focusing on examples that are similar to the problem by the structure of the content. We also explored the value of personalized example recommendation based on student's knowledge levels and learning goal of the exercise. The paper presents concept-based similarity approaches that we developed, explains the data collection studies and reports the result of comparative analysis. The results of our analysis showed better ranking performance of the personalized structural variant of cosine similarity approach.

中文翻译：

---

基于概念相似性的程序示例推荐方法研究

摘要 本文研究了一系列基于概念的示例推荐方法，我们开发这些方法以在编程领域提供基于示例的问题解决支持。这些方法的目标是在学生无法解决代码理解问题时为他们提供一组最相关的补救示例，学生在这些问题中检查程序代码以确定其输出或变量的最终值。在本文中，我们使用在智能超文本领域开发的基于语义级相似性的链接的思想来生成给定问题的示例。为了确定性能最佳的方法，我们探索了两组用于选择示例的相似性方法：非结构化方法侧重于在概念覆盖方面与问题相似的例子，结构化方法侧重于内容结构上与问题相似的例子。我们还根据学生的知识水平和练习的学习目标探索了个性化示例推荐的价值。本文介绍了我们开发的基于概念的相似性方法，解释了数据收集研究并报告了比较分析的结果。我们的分析结果表明，余弦相似性方法的个性化结构变体具有更好的排名性能。练习的知识水平和学习目标。本文介绍了我们开发的基于概念的相似性方法，解释了数据收集研究并报告了比较分析的结果。我们的分析结果表明，余弦相似性方法的个性化结构变体具有更好的排名性能。练习的知识水平和学习目标。本文介绍了我们开发的基于概念的相似性方法，解释了数据收集研究并报告了比较分析的结果。我们的分析结果表明，余弦相似性方法的个性化结构变体具有更好的排名性能。