数据来源于华南师范大学砺儒云平台 2019 － 2020 年第二学期开设的 《翻转课堂与混合学习》课程中师生之间的论坛交互

社会网络分析法开展实证研

1/3SURYLGHVWHFKQLTXHVWKDWDXWRPDWHWKHDQDO\VLVof language, which allows researchers to establish a better understanding of language and of the roles that language potentially plays in various aspects of RXUOLYHV

1/3WRROVDQGWHFKQLTXHVDUHRIWHQJXLGHGby theories, models, and algorithms developed in the āHOGRIFRPSXWDWLRQDOOLQJXLVWLFVEXWWKHSULPDU\SXUSRVHRI1/3WRROVLVWKHDXWRPDWHGLQWHUSUHWDWLRQof human language input.

7H[WDQGdiscourse analysis provides one means to understand FRPSOH[SURFHVVHVDVVRFLDWHGZLWKWKHXVHRIODQJXDJH'LVFRXUVHDQDO\VWVV\VWHPDWLFDOO\H[DPLQHVWUXFWXUHVDQGSDWWHUQVZLWKLQZULWWHQWH[WDQGVSRNHQGLVFRXUVHand their relations to behaviours, psychological pro-cesses, cognition, and social interactions. Indeed, WH[WDQGGLVFRXUVHDQDO\VLVKDVSURYLGHGDZHDOWKRIinformation about language.Chapter 8: Natural Language Processing and Learning AnalyticsDanielle S. McNamara1, Laura K. Allen1, Scott A. Crossley2, Mihai Dascalu3, Cecile A. Perret4/DQJXDJHLVRIFHQWUDOLPSRUWDQFHWRWKHāHOGRIHGXFDWLRQEHFDXVHLWLVDFRQGXLWIRUFRP-PXQLFDWLQJDQGXQGHUVWDQGLQJLQIRUPDWLRQ7KHUHIRUHUHVHDUFKHUVLQWKHāHOGRIOHDUQLQJDQDO\WLFVFDQEHQHāWIURPPHWKRGVGHYHORSHGWRDQDO\]HODQJXDJHERWKDFFXUDWHO\DQGHIāFLHQWO\1DWXUDOODQJXDJHSURFHVVLQJ 1/3 WHFKQLTXHVFDQSURYLGHVXFKDQDYHQXH1/3techniques are used to provide computational analyses of different aspects of language as they relate to particular tasks. In this chapter, the authors discuss multiple, available 1/3WRROVWKDWFDQEHKDUQHVVHGWRXQGHUVWDQGGLVFRXUVHDVZHOODVVRPHDSSOLFDWLRQVRIthese tools for education. A primary focus of these tools is the automated interpretation of human language input in order to drive interactions between humans and computers, RUKXPDQ×FRPSXWHULQWHUDFWLRQ7KXVWKHWRROVPHDVXUHDYDULHW\RIOLQJXLVWLFIHDWXUHVLPSRUWDQWIRUXQGHUVWDQGLQJWH[WLQFOXGLQJFRKHUHQFHV\QWDFWLFFRPSOH[LW\OH[LFDOGL-versity, and semantic similarity. The authors conclude the chapter with a discussion of FRPSXWHUEDVHGOHDUQLQJHQYLURQPHQWVWKDWKDYHHPSOR\HG1/3WRROV LH,76022&Vand CSCL) and how such tools can be employed in future research. Keywords:1DWXUDOODQJXDJHSURFHVVLQJ 1/3 ODQJXDJHFRPSXWDWLRQDOOLQJXLVWLFVOLQ-guistic features, automated writing evaluation, intelligent tutoring systems, CSCL, MOOCsABSTRACT1Psychology Department, Arizona State University, USA2Applied Linguistics and ESL Department, Georgia State University, USA3Computer Science Department, University Politehnica of Bucharest, Romania4Institute for the Science of Teaching and Learning, Arizona State University, USA

In statistical modelling, there are generally four types of data considered: categorical, ordinal, interval, and ratio.

While both methods intend to inform the design of intervention systems, the former does so by building software based on theory GHYHORSHGGXULQJWKHUHYLHZRIH[SODQDWRU\PRGHOVE\H[SHUWVZKLOHWKHODWWHUGRHVVRXVLQJGDWDFROOHFWHGIURPKLVWRULFDOORJāOHV LQWKLVFDVHFOLFNVWUHDPGDWD The largest methodological difference between the two modelling approaches is in how they address the issue RIJHQHUDOL]DELOLW\,

largely with DIRFXVRQSUHGLFWLQJVWXGHQWVXFFHVVDVRSHUDWLRQDOL]HGE\DFDGHPLFDFKLHYHPHQW,

本研究弥补了中文反思写作指导与反馈相关研究领域的空白，并以计算机支持的反思写作切入反思性教学领域，探究了信息技术和相关工具对学习者反思写作技能与反思能力的支持效度，具有开拓中文反思写作实践路径的独特研究价值

智能反思写作反馈系统提供在线反思写作与指导、文本分析、写作反馈功能

经历、情感、信念、视角、困难、自我批判、学习收获、计划等八类

Gibbs[9]认为个体从实践中反思习得新知，提出包含描述、情感、评价、分析、结论、行动计划六阶段的反思循环模型

结果表明：①只采用学生的部分学习过程数据预测学习成绩时，学生的行为模式与学习成绩的映射关系存在多种模式，试图构建学生在线学习行为与学生成绩一一对应预测关系的预测模型难以获得较好的预测结果。②本研究中的预测模型最高可预测正确所有混合课程中74.7% 的学生，不同学习成绩等级学生的预测准确率差异较大，成绩为A和B的学生其被预测准确率

因此整个预测模型表示的是学生的在线学习行为指标与其成绩等级之间的唯一映射关

通常评价多分类问题的指标包括权重准确率（weightedaccuracy）、权重查全率（weighted recall）和权重F1值（weighted F1 measure）评价。

问题1：如何解释混合课程中学生成绩预测模型？问题2：在混合课程中构建学生成绩预测模型能为教师提供哪些帮助

本研究主要分析以下评估指标：准确率 = （TP+FP）/（TP+FP+TN+FN）；灵敏度 = 真正率（TPR）= TP/（TP+FN）；特异度 = 1 - 假正率（FPR）= FP/（FP+TN）；召回率 = TP/（FP+FN）。

主要聚焦在多场景学习过程中通过倾向性指标和行为表现指标预测大学生学习绩效是否大于、等于或小于80分，是一个典型的二分类问题。

多场景行为表现指标有较强的预测能力，深度神经网络模型预测准确率最高（82%）但耗时最多。

EDM and LAK both reflect the emergence of data-intensive approaches to education. In sectors such as government, health care, and industry, data mining and analytics have become increasingly prominent for gaining insight into organizational activities. Drawing value from data in order to guide planning, interventions, and decision-making is an important and fundamental shift in how education systems function. LAK and EDM share the goals of improving education by improving assessment, how problems in education are understood, and how interventions are planned and selected. Extensive use by administrators, educators, and learners of the data produced during the educational process raises the need for research-based models and strategies. Both communities have the goal of improving the quality of analysis of large-scale educational data, to support both basic research and practice in education.

EDM as follows: “Educational Data Mining is an emerging discipline, concerned with developing methods for exploring the unique types of data that come from educational settings, and using those methods to better understand students, and the settings which they learn in.”

Growing interest in data and analytics in education, teaching, and learning raises the priority for increased, high-quality research into the models, methods, technologies, and impact of analytics.

KEY DISTINCTIONS BETWEEN COMMUNITIES

LAK and EDM share the goals of improving education by improving assessment, how problems in education are understood, and how interventions are planned and selected.

“Educational Data Mining is an emerging discipline, concerned with developing methods for exploring the unique types of data that come from educational settings, and using those methods to better understand students, and the settings which they learn in.”

urrent forms of data commonly used in learning analytics work include activity data (traces of what students did) and artifact data (things that students created). O

Size here refers to two distinct characteristics.

here are three critical characteristics that distinguish learning analytics from prior educational research: the data the work is based on, the kinds of analyses employed

Learning data are not “natural” because they are produced as students interact with designed environments. The data thus represent aspects of what students do in response to that specific environment.

here are three critical characteristics that distinguish learning analytics from prior educational research: the data the work is based on, the kinds of analyses employed

characterized by their large quantity and relatively small grain‐siz

his chapter describes three characteristics of Learning Analytics work that distinguish it from prior educational research to give readers a concise overview of what makes learning analytics a unique and especially promising technology to improve teaching and learning.

The important characteristic of discourse analytics is extensive use of natural language processing techniques (Kao & Poteet, 2007) for extracting quantitative measures from written text.

rly 2011, a small group of educational researchers hosted The First International Learning Analytics (LAK’11) Conference in Banff, Canada. A goal of this first gathering was to define and scope the emergent research focusing on understanding student learning through the use of machine learning, data mining and data visualisation methods

One of the most highly cited examples of an early learning analytics tool is designed to aid instructors provide feedback to students based on their predicted success

In short, the learning design of a course needs to be factored into the development of any predictive model.

Learning analytics draws on theories and methods from machine learning and data science, education, cognitive psychology, statistics, computer science, neuroscience, and social and learning sciences to name but a few

“measurement, collection, analysis and reporting of data about learners and their contexts, for purposes of understanding and optimizing learning and the environments in which it occurs”

，需要将学习特征数据与场景信息相结合: 1) 有效融合线上学习与线下学习行为数据，通过提炼、整合场景信息，利用行为特征提取模型实现自下而上的特征表达提取，特征分类存储与索引，建立学习行为特征数据库; 2) 通过自上向下的任务分解( 用户需求理解) ，建立学习行为特征的提取与表达、任务相关性映射的特征计算模型，分析影响隐性学习行为的关联因素，建立表达学习行为模式分析的综合模型; 3) 通过对教育大数据的属性分层，应用数据挖掘技术，发掘学习者行为背后的学习准备、学习动机、归因和自我效能感的关联，洞察影响学习的诸多因素; 4) 对学习过程节点进行模式分类，实现对学生学习的过程性评估和学习轨迹呈现。

教育数据挖掘则是运用统计学的数据挖掘算法解答教育问题。

可以看到，让学习分析技术回归“以人为本”，需要考虑包括用户界面、对工作实践的影响、用户权力和控制的转移，以及融入数据模型的价值观，这也是未来学习分析技术升级转型的发展方向。

是教育场景的复杂性与多样性，难以抽象出通用模式对数据作通用性解读。学习分析所做的数据挖掘，需要将数据结合具体教育场景进行解读

）微 观 化 的 服 务 层

（3）可 视 化 的 分 析 结 果

）。不 同 来 源 的 海 量数 据 为 提 供 自 动 化 的 学 习 支 持 和 针 对 性 的 学 习 服 务 提 供 了可 能 性 ，同 时 其 最 大 的 挑 战 是 如 何 将 不 同 来 源 的 多 样 性 数 据加 以 整 合 ，并 将 这 些 多 来 源 的 数 据 导 入 到 同 一 个 分 析 框 架 中（常 需 要 采 用 第 三 方 分 析 软 件 ）进 行 分 析 和 运 算 ，从 而 提 供 有关 于 学 生 学 习 情 况 的 可 视 化 分 析 结 果 。

，学 习 分 析 技 术 是 测量 、收 集 、分 析 和 报 告 有 关 学 生 的 学 习 行 为 以 及 学 习 环 境 的 数 据 ，用 以 理 解 和 优 化 学 习 及 其 产 生 的 环 境 的

可 视 化 的 分 析 结 果

（2）多 重 角 度 的 分 析 技 术

（1）复 合 化 的 数 据 采 集 。

正 式 和 非 正 式 学 习 过 程 中

一 类 是 学 生 数 据 （learners off-put data），即 学 生 在 学 习 过程 中 由 移 动 终 端 、社 会 性 软 件 和 学 习 管 理 系 统 所 记 录 的 数据 ；另 一 类 是 智 能 化 数 据 （intelligent data），即 可 以 通 过 语 义 分析 以 及 连 接 技 术 来 处 理 的 源 自 课 程 、学 期 考 试 和 其 他 来 源 的数 据 ，这 类 数 据 与 学 习 者 的 学 习 过 程 间 接 相 关 。

有 关 学 生 的 学 习 行 为 以 及 学 习 环 境 的 数 据