| In recent years,the popularity of online education platforms has promoted the research of adaptive learning systems,in which exercise score prediction is one of the most critical modules in the adaptive learning process.Problem score prediction includes design research on problem characteristics,student abilities and interaction functions.A large number of existing studies have proved the feasibility of problem score prediction based on knowledge tracking,but there are also some challenges: for example,knowledge tracking is difficult to capture multiple knowledge points.The changes of students’ abilities in the exercises,the lack of attention to the structural information of knowledge points in the prediction of exercise scores,and the interactive functions based on cognitive diagnosis rely heavily on manual design.In response to the above three challenges,this article starts with two aspects of work,trying to use the attention mechanism and graph neural network methods to model the learning process of students,and combine the structural information of the knowledge points to generate exercise features,and finally based on neurocognitive diagnosis The problem score is predicted in the matching function of the framework.The main work of this paper includes:1.Exercise Score Prediction based on Neural Attentive Knowledge Tracing.In order to pay attention to the changes of students’ abilities,we divide the students’ ability characteristics into long-term characteristics and short-term characteristics,and use the attention mechanism to improve the knowledge tracking model based on long and short term memory networks-NAKTM,which uses the short-term characteristics of students captured by LSTM Together with the long-term characteristics captured by the attention mechanism,it represents the comprehensive ability of the student.At the same time,with the help of neural network pooling operation,the NAKTM model uses as many knowledge points involved in the exercises as possible to generate exercise features.Finally,the model design uses a bilinear model to match the characteristics of the exercises and the characteristics of the students’ ability,and predict the students’ scores on the candidate exercises at the next moment.2.Prediction of Exercise Score Based on Graph Neural Network and Knowledge Structure.In order to improve the predictive effect of the cognitive diagnosis model,this paper proposes a new cognitive diagnosis framework—a neurocognitive diagnosis framework based on knowledge tracking enhancement.In this framework,knowledge tracking analyzes the student’s answering process and integrates students’ abilities in The latent space is expressed in the form of feature vectors,and then combined with the structural information of the exercise knowledge points,input into the neurocognitive diagnosis function based on the DNN model,and finally output the prediction result.In this paper,the DKT model is first tested to prove the validity of the framework,and then based on the gated neural network,the knowledge tracking model is improved,and a new model is proposed,it is named GKT-CD. |
