Research On Confused State Recognition Based On BiLSTM-SVM Model

With the development of science and technology and the progress of the times,all walks of life have changed dramatically compared to the past,and people's lifestyles and working methods have also undergone tremendous changes.Today,with the development of Internet technology,traditional classroom education is also facing huge challenges.In recent years,online course education has become increasingly popular,people can learn by video at home with only a computer,which is very convenient.However,it is undeniable that although online course education has many advantages,it also has many shortcomings compared with traditional education methods,and the lack of immediate feedback is one of the main problems of online course education.Especially this year,due to the impact of COVID-19,students need to attend classes online,and it is more difficult for teachers to master the students 'course learning situation,so it is urgent to find an effective way for teachers to obtain the instant feedback from students in class.At present,electroencephalogram(EEG)has been successfully applied in many research fields,such as disease detection,physiology,engineering application fields,etc,but there are few studies on the state of brain confusion based on EEG.Therefore,in order to solve the above problem,this paper uses the EEG signals of students when watching online courses as a data set,combines the powerful data feature learning capabilities of deep learning methods with the good feature-based classification performance of support vector machine(SVM),and designs and builds the confused state recognition framework based on BiLSTM-SVM model,to classify and predict the state of confusion of students.Through a series of parameter adjustments,the model performs very well on the data set,which verifies the effectiveness of the model in this paper,and provides a feasible idea for solving the problem of instant feedback in online course education.The main research work is as follows:(1)Introduced the basic structure and related theoretical knowledge of recurrent neural network(RNN)and its variant long short-term memory(LSTM)network in detail,designed and built a confused state recognition framework based on BiLSTM-SVM model.The model is mainly divided into 5 parts,namely the input layer,batch normalization(BN)layer,BiLSTM layer,SVM classification layer and output layer;then the role of the BN layer,BiLSTM layer,SVM classifier in the model is elaborated in detail,in which the BN layer is On the basis of batch normalization,it speeds up the network training rate and also plays a role in preventing overfitting to a certain extent;BiLSTM layer performs a back propagation on the basis of unidirectional LSTM,which can extracts contextual features better;SVM classifier has better classification performance for low-dimensional features,and can make better classification predictions for features learned by BiLSTM layer.(2)Build a model for experiment and analysis.In order to make the model proposed in this paper obtain better results on the data,a series of parameter optimizations are performed to obtain a set of optimal parameters.Experiments show that the BiLSTM-SVM model obtained a higher classification result in this confused state recognition task,which was 3.37%higher than the current highest accuracy rate of 73.30%on this data set,and performed better.Then the BiLSTM model was constructed on the same platform for experiment.By optimizing the parameters of the BiLSTM model,the optimal results of the BiLSTM model were obtained,and the results of the two models were compared and analyzed,indicating the superiority of the BiLSTM-SVM model in the task of confused state recognition.