Research On Enchanced Deep Knowledge Tracing Based On Recurrent Neural Network

In recent years,with the development of intelligent education,many online learning platforms have emerged.Online educational platforms break the limits of space and time.These platforms provide open learning resources and realize the sharing of educational resource for students.Thus,more and more students choose e-learning.In addition,the student learning records accumulated a large amount of educational data resources,which contains abundant information and value.Making use of the educational data resource can realize knowledge tracing or other educational research tasks.Knowledge tracing task is achieved by tracing the evolution of each student knowledge state via a series of learning activities.It aims to identify students’ knowledge proficiency and learning ability,so as to make personalized learning programs and improve learning efficiency for students.Therefore,how to realize knowledge tracing task by analyzing and studying educational data resources has become a research hotspot in the field of educational data mining.At present,deep knowledge tracing is a prevalent model which by encoding student learning records and using recurrent neural network to trace student’s knowledge state.However,this method only focuses on the usage of exercise results and ignores the impact of other features.In addition,the method also ignores student features impact on student performance.Therefore,this paper based on enhanced deep knowledge tracing model.First,in view of the traditional models only chosen exercise features,we propose to incorporate both student features and exercise features into the knowledge tracing model.Second,we propose to integrate attention mechanism and multiple features based on the knowledge tracing models in view of traditional model neglect the effect weight of each feature on student performance.The main work and innovation points of this paper are as follows:（1）This paper proposes a novel framework –Multiple Features Enhanced Deep Knowledge Tracing.First,we propose integrate student features and exercise features based on deep knowledge tracing.Second,to comprehensively analyze the impact of each feature and the relationship between features on student performance,a principal component analysis method is employed to automatically handle multiple features and learn their representation.Finally,a recurrent neural network is used to trace knowledge states of students by encoding their learning activities for their performance prediction.The experimental results verify the effectiveness of our method and improve the accuracy of prediction.（2）Based on the point（1）,we propose a novel framework--Multiple Features Fusion Attention Mechanism Enhanced Deep Knowledge Tracing.First,this paper proposes based on deep knowledge tracing model combined with multiple features.Then,a recurrent neural network is used to trace the knowledge states of students overtime by processing their learning sequences.Finally,a fusion attention mechanism based on the original recurrent neural network architecture is used for knowledge tracing task.In this way,student knowledge state can be traced more accurately.The experimental results show that our method can enhance the interpretability and improve the accuracy of student performance prediction.Extensive experiments on a real-world dataset show that our approaches proposed in this paper could trace student knowledge state and predict their performance accurately in the future.