Design And Research Of Online Learning Effect Analysis System Based On Deep Learning

With the booming development of online education platforms,real-time evaluation of students' learning effects is an indispensable key link.At present,examination evaluation is the main method of learning effect evaluation,which virtually increases teaching costs.And according to the study of the learning effect based on the online learning platform,timely feedback(knowing the learning effect in real time)has a greater motivating effect than long-distance feedback(such as the test results).Therefore,it is necessary to introduce a function that can obtain the learning effect in real time according to the expression of students in the online learning system.This function can be unnoticed by the learner,which is more convenient and faster than the examination evaluation method.According to the research of educational technology researchers,learners' emotions are divided into six categories:interest,disgust,sadness,pleasure,sleep,and tension.Counting the number of positive emotions in real time can draw the overall learning effect.Counting the number of times that a learner has positive emotions during the entire learning process can get the individual learning effect.And verified that the learning effect of the system analysis is consistent with the experimental preset.First,emotion recognition is mainly to identify the changes in facial expression features in dynamic sequence pictures.This paper uses the length of 29 connecting lines of 44 facial feature points in the significant area of facial expression change to form a feature vector to define the expression features.In order to locate facial feature points,the traditional method is to first obtain the position information of the face area,and then use ASM(Active Shape Model)to obtain the facial feature points in each face area.For the face area in each training picture of MT CNN(Multi-task Cascaded Convolutional Networks),this paper first uses ASM to locate 33 facial feature points(the total number of feature points minus the number of feature points in the nose area minus the two mouth corner feature points that MTCNN has found).Then record their coordinates and retrain MTCNN to make it locate 33 facial feature points in each face area while detecting the face area.Compared with the traditional method,the method used in this paper can reduce the calculation steps and increase the calculation speed.Secondly,the research found that the facial features formed by the above mentioned 29 connecting lines are features adopted for basic facial expression recognition.And this feature has some redundant information.In order to extract the learner's expression features,the feature vectors that define the expression features are modified as follows:1.The value of the face area is added to it.The change of the face area indicates whether the learner is interested in the current teaching content.2.Add a vertical line from the highest point of the upper lip to line between the corners of the mouth.The length of this line indicates the learner's degree of joy-sadness.3.Since the nose area contains less emotional information,remove the 4 connecting lines of the nose area from it.This can reduce the dimensionality of features.Finally,A deep convolutional neural network with the same structure as VGG16 is designed to classify the extracted feature vectors.Because there is no public facial expression dataset for learners,this article prepares its own dataset in accordance with the organizational structure of FER 2013 data.The convolutional neural network is trained using the constructed data set and the validity of the classifier is verified using the validation set.Using the trained classifier,a deep learning-based online learning effect analysis system is designed and implemented.