| Recently,the "quality crisis" of online learning with "high dropout rate and low completion rate" is becoming increasingly severe.Existing research have showed that low level of emotional engagement in learning is one of the significant reasons interpreting why the learning performance and achievement is limited adopting online learning methods.Therefore,how to recognize participants’ level of emotional engagement in online learning environments with an efficient way and then to conduct early warning to the non-engagement behavior is the key to alleviate the bad learning achievement rooted by learners’ low level of emotional engagement.However,traditional methods for measuring emotional engagement,such as self-reporting and manual observation,mostly are time-consuming,labor-intensive as well as subjective,and lack the time cues which is required to understand the interaction between emotional engagement and learning.Ground on this consideration,the present study uses educational technology theories and deep learning methods,based on the online learning environment,and attempts to solve the problem of identifying emotional engagement in online learning environment from multiple perspectives.The main contents of the thesis are as follows:(1)Building an online learning emotional engagement database.Collecting online learning emotional engagement data in natural scenes is one of the most challenging tasks.With regard to the fuzzy problem of the classification of emotional engagement in online learning environments,this article clarifies the emotion states that are highly related to learner’ learning,and based on the task learning method,establishes the corresponding relationship between engagement and facial expression,and divides four emotional engagement categories,including no engagement at all,low engagement,medium engagement,and high engagement.This study collected a large number of learning videos related to learners in an unconstrained online learning environment,and finally selected 32904 images containing four types of emotional engagement.(2)Design a deep learning-based online learning emotional engagement recognition method.First,in order to ensure that the image is not affected by irrelevant backgrounds,this study uses YOLOv4 to perform face detection.Second,an improvement model is proposed to deal with issues of the huge amount of VGG16 network parameters and time-consuming training.At the same time,in the model training process,the deep deterministic information bottleneck method(Deep Deterministic Information Bottleneck,DIB)is used to make up for the deficiency of the traditional loss function in order to obtain a more compact feature expression,to reduce generalization errors,to improve the generalization of the model,and to achieve an more accurate recognition of emotional engagement in complex online learning scenarios.Last,the proposed method in this study is verified by comparing it with traditional machine learning and other deep learning methods. |
