Coding Collaboration Process Automatically: Coding Methods Using Deep Learning Technology
Coding Collaboration Process Automatically: Coding Methods Using Deep Learning Technology
复制标题
DOI：
--
复制
发表时间：
2018
期刊：
影响因子：
--
通讯作者：
Kimihiko Ando;Chihiro Shibata;Taketoshi Inaba
中科院分区：
文献类型：
--
作者：
Kimihiko Ando;Chihiro Shibata;Taketoshi Inaba
- 在计算机支持的协作学习（CSCL）研究中，通过分析成功协作互动的机制，提取指标来识别协作过程不顺利的群体，从而获得一个指导性的支架，可以被认为是研究和教育实施中最重要的关注点。为了研究这种协作学习过程，人们尝试了不同的方法。在本文中，我们选择的口头数据分析，这种方法的优点是，它可以定量处理，同时保持定性的角度来看，与合作学习的数据相当大的规模。然而，对大规模教育数据进行编码非常耗时，有时超出了男子的能力。因此，近年来，也有人试图通过使用机器学习技术来自动化复杂的编码。在此背景下，随着CSCL系统中产生的大规模数据，我们试图利用深度学习方法实现高精度编码的自动化，这些方法来自机器学习的前沿技术。结果表明，我们的深度学习方法很有前途，优于机器学习基线。但是，可以通过构建对协作和会话上下文更敏感的编码方案和模型来提高预测精度。因此，我们提出了一种新的编码方案，可以更全面，更准确地表示在本文的最后，为下一步的研究。
— In Computer Supported Collaborative Learning (CSCL) research, gaining a guideline to carry out appropriate scaffolding by analyzing mechanism of successful collaborative interaction and extracting indicators to identify groups where collaborative process is not going well, can be considered as the most important preoccupation, both for research and for educational implementation. And to study this collaborative learning process, different approaches have been tried. In this paper, we opt for the verbal data analysis; the advantage of this method is that it enables quantitative processing while maintaining qualitative perspective, with collaborative learning data of considerable size. However, coding large scale educational data is extremely time consuming and sometimes goes beyond men’s capacity. So, in recent years, there have also been attempts to automate complex coding by using machine learning technology. In this background, with large scale data generated in our CSCL system, we have tried to implement automation of high precision coding utilizing deep learning methods, which are derived from the leading edge technology of machine learning. The results indicate that our approach with deep learning methods is promising, outperforming the machine learning baseline. But the prediction accuracy could be improved by constructing coding schemes and models more sensitive to the context of collaboration and conversation. Therefore, we propose a new coding scheme that can represent the context of learning more comprehensively and accurately at the end of this paper for the next research.