华东师范大学学报(教育科学版) ›› 2022, Vol. 40 ›› Issue (9): 32-44.doi: 10.16382/j.cnki.1000-5560.2022.09.004

• 理论前沿与人才战略 • 上一篇    下一篇

在线教育中的学习情感计算研究——基于多源数据融合视角

翟雪松1, 许家奇1, 王永固2   

  1. 1. 浙江大学教育学院,杭州 310058
    2. 浙江工业大学教育科学与技术学院,杭州 310023
  • 出版日期:2022-09-01 发布日期:2022-08-24
  • 基金资助:
    2019年度国家社会科学基金重大项目“人工智能促进未来教育发展研究”(19ZDA364)

Research on Learning Affective Computing in Online Education: From the Perspective of Multi-source Data Fusion

Xuesong Zhai1, Jiaqi Xu1, Yonggu Wang2   

  1. 1. College of Education, Zhejiang University, Hangzhou 310058, China
    2. College of Educational Science and Technology, Zhejiang University of Technology, Hangzhou 310023, China
  • Online:2022-09-01 Published:2022-08-24

摘要:

学习情感是影响学习绩效、学习感知及高阶思维能力的重要因素。现有学习情感计算研究主要基于重量级生理反馈技术的小样本分析,缺少大规模在线开放课程环境下的实践研究。这一方面是由于在线课程环境下,学习情感计算的数据来源较为有限,多为单一的面部表情数据;另一方面,学习者在线学习场景下常处于监督不足的状态,学习者身体姿态的随意性较大,因此极有可能影响到面部特征的提取。然而,本研究认为在线学习者的姿态也具有情感特征,同样是情感信息的关键来源。因此,尝试将学习者的姿态数据融合到面部表情数据中,构建多源数据融合的深度学习情感计算模型,弥补学习者姿态变化带来的面部识别缺陷,同时进行多源情感数据的协同分析,实现数据的交叉印证和相互补偿。研究得出:通过训练构建的包含7878张在线学习者面部表情和姿态图像的数据集,利用卷积神经网络和决策融合的方法将学习者面部姿态数据融入表情数据中,学习情感识别准确率较单一的面部表情识别提高了3%,是在线学习情感计算的有效方法。本研究在理论上为多源数据融合在学习者情感计算的有效性提供模型基础,在实践上,为在线教育环境下的学习情感计算提供了有效的技术路径。

关键词: 情感计算, 在线教育, 学习情感, 多源数据融合, 人工智能, 深度学习

Abstract:

Learning affection is an essential factor affecting learning performance, perception, and higher-order thinking ability. Existing research on learning affective computing is mainly based on a small sample analysis of heavyweight physiological feedback technology. There is a lack of learning affective computing research in online courses. In the online course environment, the data sources for learning affective computing are relatively limited, mainly based on a single facial expression data. On the other hand, learners are often under insufficient supervision in online learning scenarios. The body posture is more random, so it is very likely to affect the extraction of facial features. However, this study believes that the pose of online learners also has affective characteristics and is also a key source of affective information. Therefore, we try to fuse the learner’s posture data into the facial expression data, build a multi-source data fusion deep learning affective calculation model, and make up for the facial recognition defects caused by the learner’s posture change. Also, we perform collaborative analysis multi-source affective data to realize data cross verification and mutual compensation. The research concludes that it is an effective method for online learning affective calculation to build a dataset containing 7,878 facial expressions and posture images of online learners constructed through training, and use the convolutional neural networks and decision fusion methods to integrate learner posture data into facial expression data.The accuracy of affections recognition increases 3%, compared with a single facial expression recognition. In theory, this research provides a model basis for the effectiveness of multi-source data fusion in learners’ affective computing. In practice, it provides an effective technical path to learning affective computing in an online education environment.

Key words: affective computing, online education, learning affection, multi-source data fusion, artificial intelligence, deep learning