Personalized Learning Resource Recommendation in Real-practice

With the advent of portable devices such as tablets and e-readers (amazon kindle, ipad, Google Chromebook, etc.), reading online content for educational, learning, training or recreational purposes has become a very popular activity. Compared to printed material, readers of digital content are offered several levels of interactivity. For example, digital content allows more interactive and collaborative learning, users may read additional or supplementary online content related to a specific part of the e-text that they have difficulty understanding or wish to explore more; they can add annotations; zoom-in on a picture, or play a video embedded in the content. Despite these advantages, printed media still provides other benefits that cannot be matched by digital. Some of the advantages of printed material include: 10-30% faster reading rate, lack of distractions, no device compatibility or Internet connection issues, cost effectiveness and, most importantly, the fact that print is still the medium preferred by the majority of students. Instead of eliminating these benefits, we believe that learning should be based on print and enhanced by the use of technology rather than replaced by it.

This presentation focuses on introducing a learning system, namely METIS, which leverages the benefits of reading of both printed and digital content and provide further enhancements to the reading experiences. The system architecture, along with machine learning services and algorithms will be discussed and introduced as well. The developed system has been deployed and piloted in Silicon Valley local schools and universities with thousands of participants in real class.

Outline:

1. Background of Hybrid Learning
   1. Digital v.s. Physical Content
   2. Hybrid Content
2. Hybrid Learning System
   1. System Architecture
   2. Content Creator View & Functions
   3. Reader’s View & Functions
3. Machine Learning Technologies
   1. User profiling
   2. Personalized Content Recommendation
   3. Learning Graph Generation
   4. Illustration Image Recommendation
   5. EEG signal for attention detection
4. Challenges
   1. Data sparsity
   2. “Too long to process” issue
   3. Semantic topic discovery and representation
5. System deployment and Discussion
   1. Performance evaluation in real system
   2. Feedbacks

参考译文：

随着各种便携式设备如平板电脑、Kindle、Chromebook的广泛使用，在线教育、培训和娱乐等内容越来越流行。和传统的纸质材料相比，电子读物更具互动性，可以在线传播，学习更多的在线相关内容，遇到理解障碍可以在线查询，拓展知识视野，还可以阅读过程中添加注释，缩放，或者播放阅读内容里的嵌入视频多媒体资源。

然而，纸质读物也有自身优势，例如阅读速度更快，不会分心，不需要考虑和电子设备的兼容，成本低，最重要的是，绝大多数学生需要纸质读物。排除这些优势，我们相信在电子技术的辅助下，纸质读物仍然是学习的主体，而且也不该被电子设备完全替代。

本演讲将重点介绍METIS学习系统，该系统将电子和纸质读物内容的优势结合起来，为未来的高效阅读体验提供混合式的增强手段。我会在演讲中介绍并分析该系统中的机器学习和算法技术架构，这套系统已经在硅谷当地的中学和大学获得广泛应用。

演讲提纲：

1. 混合学习的背景

• 电子和纸质阅读资源对比
• 混合阅读资源

2. 混合学习系统

• 系统架构
• 内容创建者-系统功能
• 读者-系统功能

3. 机器学习技术

• 用户建模
• 个性化内容推荐
• 学习路径自动规划
• 图像推荐实例
• 机遇脑电波信号的注意力检测系统

4. 系统挑战

• 数据稀疏性
• 查询太长不能处理问题
• 语义主题发现和表征

5. 系统部署和讨论

• 系统性能评估
• 客户效果反馈