Learning performance prediction is critical to the development of personalized learning.Personalized learning aims to dynamically optimize learning progress and teaching methods based on learners’ learning abilities and learning behaviors.The development of intelligent education enables a lot of learning data to be saved.If the technology can be effectively used to mine useful information,it will help teachers to grasp the learning status of learners in time and provide personalized guidance services,so it has certain practical significance and value.At present,educational data mining is developing in many aspects,among which the research on performance prediction and knowledge tracing has received extensive attention from scholars at home and abroad.Although the relevant prediction methods have achieved certain results,there are still some shortcomings,such as lack of feature extraction,low model prediction accuracy,and poor generalization.In response to the above problems,this thesis conducts research from two aspects based on the data of learners’ learning behaviors and specific answer records.The research contents are summarized as follows:First of all,aiming at the problems of low prediction accuracy and weak generalization performance in performance prediction,this paper proposes a Performance Prediction model based on Stacking Ensemble(PPSE).According to the characteristics of the data,linear regression and BP neural network model are selected as primary learners to learn the linear relationship and nonlinear relationship between features respectively.In particular,PPSE also uses the forward stepwise regression method to select features,and concatenate the selected features with the output of the primary learner as the training set of the meta-learner to obtain the final prediction model.Two public datasets and one private dataset are used for comparative experiments to verify the effectiveness of the model structure.The experimental results show that the PPSE model has achieved better results than the baseline model,indicating that the PPSE model has better prediction accuracy and generalization performance in the task of performance prediction.In order to have detailed understanding of learner’s mastery of knowledge,this paper also proposes a Dynamic Key-Value Memory Network knowledge tracing model based on Pretraining and Multi-Head attention(DKVMN-PMH).In the pre-training process,the problem relationships,problem difficulty and knowledge difficulty are used as the constraints of the loss function to optimize the question embedding vector;At the same time,this paper combines the prediction of the previous moment,and proposes a weight calculation method based on the multi-head attention mechanism,which is used to update the knowledge state matrix by fusing learners’ correct answer times,question difficulty and knowledge difficulty features.Finally,three public datasets and a private dataset are used to conduct experiments.The experimental results show that the DKVMN-PMH model can effectively improve the prediction accuracy of knowledge tracking tasks. |