论文分享——Hybrid E-learning Recommendation Approach
文章目录
- 概述
- Abstract
- Conclusion
概述
论文题目为《Hybrid E-learning Recommendation Approach Based on Learners’ Influence Propagation》，跟之前分享的这篇文章来自同一作者，本文发表时间较晚，为后续工作。
同样这篇文章讲的是在线学习的推荐系统，主要 novelty 我认为是在用户间的信息传递上。方法大致为：
（1）先对每个用户建模（Learner Influence Model， LIM ），根据问卷调查确定用户的学习风格和习惯（personality），以及根据他的学习情况建立起 learning profile （学习历史），除此之外还有每个用户的水平（knowledge credibility）和值得信任的程度（learner aggregation），前者是通过学习记录计算出，后者则是通过评估用户的影响力（follower的数量、质量）得出
（2）利用 intuitionistic fuzzy logic (IFL) 模型优化 LIM 。这个方法增加了模型的多样性和可扩展性，因为它在每个用户信息中增加了一些不确定的和 intuition 的部分
（3）提出 self-organization based (SOB) recommendation approach，这一部分和作者之前那篇文章很类似，不过把算法从 learning objects 移到了用户身上，用于传播用户信息，这貌似又有点GNN的意思。在算法的收敛后会把所有用户聚成一个个的 clique ，每个clique就相当于用户的学习圈子，迭代过程中会根据用户之间互相影响的程度，让用户移动，直到熵增低于阈值即判定为稳定状态
（4）最后用 sequential pattern mining (SPM) ，决定要推荐的 learning object 和指导，策略就是根据 clique 中其他大牛的学习内容进行关联推荐，简单的说：牛逼的人学啥，我就学啥
Abstract
E-learning platforms has grown rapidly in popularity nowadays so how to recommend personalized learning resource becomes a critical problem. However, the scarcity of interpersonal information in e-learning environment makes it difficult to apply typical recommendation like collabrative filtering. The resultant extreme data sparsity poses a great challenge to the implementation of CF based recommendations. In this paper, we proposed a hybrid filtering recommendation approach SI-IFL combining learner influence model(LIM), self-organization based (SOB) and sequential pattern mining (SPM) for recommending learning objects for learners. The method works as follows: (i) LIM models learners’ profiles and are applied to acquire interpersonal information, which ease the data sparsity. (ii) SOB are used to find the optimal clique for active learners by simulating influence propogation. The algorithm reaches a stable state based on the behavior of all learners. (iii) SPM is applied to decide the final learning objects and navigational paths based on the recommended learner cliques.
Conclusion
Different from e-commerce fields, the information scarcity of e-learning hinders the application of CF recommendation. To answer the problem, this study proposes an influence user model independent from rating behaviors. It can fill the deficiency gap in the underlying data. (ii) take uncertainty and vagueness into consideration, IFL is used to optimize the LIM, which introduce more diversity and adaptability. (iii) we use self-organization theory to simulate the influence propogation so as to cluster learners into optimal cliques. And according to the experimental results, the proposed hybrid filtering approach is proved to be effective, adaptable and personalized.