论文分享：E-Learning personalization based on hybrid recommendation strategy and learning style identification

E-Learning personalization based on hybrid recommendation strategy and learning style identification

Background

我刚看到这篇文章的时候吓了一跳，好家伙，这个刊（Computers & Education）是100%的1区期刊，而且这篇文章的引用量好像贼高（~540）。看了一眼作者信息，这是哪里人？一查，塞尔维亚。。。。

整体上来讲，这篇文章介绍了他们所设计的系统——Protus——中的推荐算法部分。文章称，整个Protus系统可以根据用户不同的背景、偏好、学习目的以及其他因素，向用户推荐合适的学习资料。按照这个描述来看，这个系统是很符合“因材施教”的原则的。

Prior knowledge

这篇文章在一开始介绍了一个很重要的概念：不同的人有不同的学习风格（learning style）（这应该很符合因材施教的理论）。在学习中，集中精力是学好的一个必要条件；文章注意到了这个现象，因此希望能借助学习风格理论，判断出不同学生的学习风格，从而以不同的方式令学生集中注意，达到教好学生的目的。

做法

整个Protus系统分为好几个大块（如上图）。与我们目前正在做的工作非常类似，Protus中储存了所有的学习资料（包括教学资料以及题目），也储存了所有用户的信息（包括个人信息以及历史记录）。Protus会在用户使用系统的过程中不断地更新用户画像。当然，文章的重点还是放在了推荐系统上面。

为了进行用户建模，文章使用Aprioriall算法，对用户的历史记录（这里可以是任何学习资源的历史记录）进行分析。显然，学习风格类似的用户，生成的历史记录也是类似的（例如，React的官方Tutorial提供了两种不同的学习方式：要么边写代码边学，要么直接看文档。如果有两个人选择了相同的方式，那么他们的学习风格大概率也是相同的）。在使用这个Aprioriall算法进行完建模之后，就可以使用协同过滤算法来推荐用户想要的学习资源了。

---

Abstract

It is important to design personalized online learning systems which can overcome the disadvantages of traditional classroom learning, which cannot nicely fit students' learning styles and preferences. In this paper, we designed a e-learning system Protus, which can provide learning materials based on different students' learning styles and current knowledge level. By analyzing users' activity history using the AproriAll algorithm, this recommendation system can successfully illustrate users to various learning styles, and thus can do recommendation based on collaborative filtering. Experiments have been done on actual learners, and the result shows a significant improvement over those who did not use this system.

Conclusion

In this paper, we stated the importance for e-learning systems to consider students' preferences and learning styles. To implement this functionality, we described Protus, a e-learning system to detect students' learning styles based on AproriAll algorithm and recommends learning materials to students by applying collaborative filtering. By applying this system to actual learners, we have seen a result which indicates this system will have noticeable improvements in various aspects, including but not limited to outcome test score, learning efficiency, and learning speed. Although currently this system is designed for tutoring only programming languages such as Java, it can be hopefully used in other educational fields.