Study And Achieve Emotion Recognition For E-learning

Education has always been an area that people attach importance to.It is not only related to the growth of a person,but also to the development of a nation.E-learning has become one of the hot topics in recent years,but the current E-learning still lacks the humanized and personalized interaction methods for different users.How to improve the human-computer interaction performance of E-learning has naturally become the focus of research.Aiming at how to improve the interactive performance of E-learning,this paper proposes a new framework of adaptive E-learning,which can adjust the content displayed by the platform according to two different interactive ways,and then improve its self-adaptability.One way is to acquire face information through the user’s camera,recognize the emotion based on face,and adjust the teaching content adaptively according to the result of face emotion recognition.The second way is to conduct a guided human-machine dialogue through the user’s answers to the test questions,and adjust the teaching content adaptively based on the reply content of the multi-round dialogue.Aiming at the framework of E-learning designed in this paper,we propose one algorithms to recognize static emotion and two algorithms to recognize dynamic emotion respectively.To solve the problem of static emotion recognition,a convolution neural network combined with soft-max layer is proposed to recognize facial expressions.Because increasing the number of network layers and reducing the size of convolution filter can make the network extract details better,this paper takes these two optimization methods as the primary method to improve the accuracy of the algorithm.Five groups of static emotion recognition algorithms with different convolution layers are compared.In the experiment,the number of convolution layers is set to 6,8,12 and 13 respectively,and the recognition accuracy of VGG16 convolution neural network based on transfer learning is compared.Thirteen layers of convolutional neural network showed the best recognition effect,with 91.6%accuracy,which was slightly higher than that of transfer learning VGG16neural network with 89.6%accuracy.To solve the problem of dynamic emotion recognition,this paper designs two deep learning algorithms which are suitable for dynamic emotion recognition.one is a hybrid network which is combining convolution neural network with bidirectional long and short memory neural network,the other is 3D convolution neural network.Firstly,an emotion recognition algorithm is designed in the form of convolution neural network and bidirectional long-short memory neural network.The convergence effect and the accuracy of emotion recognition are verified by three groups of comparative experiments,and the accuracy of dynamic emotion recognition is 62.3%.Then,another dynamic emotion recognition algorithm is designed based on the idea of 3D convolution neural network.It has two functions:feature extraction and feature modeling in time series,and achieves 70%accuracy in identifying dynamic emotions.Compared with 61.87%accuracy of the best dynamic emotion recognition algorithm in EmotiW2018^[1],the proposed algorithm achieves nearly 10%improvement.Finally,the paper gives a summary and outlook of the full text,combs the design principles and experience of emotional recognition,and summarizes the methods of further network optimization.The research work of this paper provides some useful methods and effective measures for the research of human-computer interaction method of E-learning.