[论文解读] Constructing a personalized learning path using genetic algorithms approach

Lumbardh Elshani, Krenare Pireva|arXiv (Cornell University)|Apr 22, 2021

Online Learning and Analytics被引用 7

一句话总结

本文提出了一种基于遗传算法（GA）的框架，通过优化课程概念的难度、持续时间、评分以及概念间的关系，生成个性化的学习路径。该方法优于传统的固定路径学习，其性能最高可比传统方法提升8.34%，其中锦标赛选择与循环交叉的组合效果最佳。

ABSTRACT

A substantial disadvantage of traditional learning is that all students follow the same learning sequence, but not all of them have the same background of knowledge, the same preferences, the same learning goals, and the same needs. Traditional teaching resources, such as textbooks, in most cases pursue students to follow fixed sequences during the learning process, thus impairing their performance. Learning sequencing is an important research issue as part of the learning process because no fixed learning paths will be appropriate for all learners. For this reason, many research papers are focused on the development of mechanisms to offer personalization on learning paths, considering the learner needs, interests, behaviors, and abilities. In most cases, these researchers are totally focused on the student's preferences, ignoring the level of difficulty and the relation degree that exists between various concepts in a course. This research paper presents the possibility of constructing personalized learning paths using genetic algorithm-based model, encountering the level of difficulty and relation degree of the constituent concepts of a course. The experimental results shows that the genetic algorithm is suitable to generate optimal learning paths based on learning object difficulty level, duration, rating, and relation degree between each learning object as elementary parts of the sequence of the learning path. From these results compared to the quality of the traditional learning path, we observed that even the quality of the weakest learning path generated by our GA approach is in a favor compared to quality of the traditional learning path, with a difference of 3.59\\%, while the highest solution generated in the end resulted 8.34\\% in favor of our proposal compared to the traditional learning paths.

研究动机与目标

• 为解决传统教育中“一刀切”学习路径的局限性，实现基于学习者个体需求的个性化学习。
• 开发一种混合遗传算法模型，整合课程概念的难度、持续时间、评分以及概念间的依赖关系。
• 评估并比较不同种群初始化方法、选择策略及交叉技术对学习路径优化的影响。
• 证明基于GA的学习路径在质量与适应性方面可持续优于传统固定顺序学习路径。
• 提供一个灵活且可扩展的个性化学习路径生成框架，可进一步集成额外的学习者属性。

提出的方法

• 将学习路径建模为染色体，其中每个基因代表一个学习对象（概念），其顺序定义了学习路径。
• 设计适应度函数，通过考虑难度等级、持续时间、评分及概念间关联程度，以最大化学习路径质量。
• 评估锦标赛选择与轮盘选择在父代选择中的表现，结果显示锦标赛选择收敛性更优。
• 比较循环交叉与PMX（部分映射交叉）在重组中的表现，循环交义在搜索空间探索方面略胜一筹。
• 种群初始化方法包括随机初始化与模拟退火，后者生成的初始解质量更高。
• 算法演化100代，记录适应度值以评估收敛性与优化性能。

实验结果

研究问题

• RQ1 遗传算法能否有效生成优于传统固定顺序学习路径的个性化学习路径？
• RQ2 不同选择策略（锦标赛选择 vs. 轮盘选择）如何影响生成学习路径的质量与收敛性？
• RQ3 在优化学习路径适应度方面，循环交叉与PMX交叉的相对表现如何？
• RQ4 种群初始化方法的选择（随机初始化 vs. 模拟退火）如何影响初始解质量与整体优化结果？
• RQ5 引入概念难度、持续时间、评分及关联依赖关系在多大程度上提升了学习路径的个性化与有效性？

主要发现

• 表现最佳的配置——锦标赛选择配合随机初始化与循环交叉——生成的学习路径适应度值达到45.04，显著优于传统路径。
• 即使GA方法生成的最差解也比传统学习路径高出3.59%，表明其改进具有持续性。
• GA模型生成的最高质量解相比传统学习路径在适应度上提升了8.34%，证实其优越性。
• 锦标赛选择在早期收敛与最终解质量方面均优于轮盘选择，尽管两者在代际间均呈现改善趋势。
• 循环交叉在适应度收敛方面略胜PMX交叉，尽管差异微小，表明两者在路径生成中均具可行性。
• 基于模拟退火的初始化方法生成的初始种群质量高于随机初始化，加速了向最优解的收敛。