The Research On Multi-Feature Knowledge Tracing Method Based On Temporal Convolutional Network

With the development of online education platforms,their educational goals have been upgraded from simple online education to intelligent education.They are committed to creating intelligent and personalized education platforms and optimizing the learning experience of learners.Knowledge tracking technology combines big data analysis and deep learning technology to model massive online education platform data,track changes in students’ knowledge mastery status,and provide a basis for building personalized education platforms.The knowledge tracking method based on deep learning is currently the main research direction in the field of knowledge tracking.This direction mainly uses cyclic neural networks to model the learning behavior path of students.However,there are generally long sequence dependency issues and insufficient interpretability issues in models,and in deep network structures,network degradation,gradient disappearance,or gradient explosion are prone to occur.Many researchers have made improvements to the above issues,but they can only play a mitigating role and cannot fundamentally overcome the limitations of defects.Therefore,in response to these issues,the work done in this article is as follows:(1)Aiming at the interpretability problem,this paper proposes A Deep Knowledge Tracking Model Integrating Forgetting Factors and Exercise Difficulties(FDKT-ED),which enriches the semantic information contained in the model and enhances the interpretability of the model.(2)Aiming at the problem of long sequence dependence,a Temporal Convolutional Knowledge Tracking Model Integrating Forgetting Factors and IRT(TCKT-FI)is proposed.The model uses a cyclic neural network replaced by a temporal convolutional network,and uses dilation convolution to process the deep network structure,dynamically obtaining the knowledge state of students.When dealing with long time series problems,the model has better prediction effects;When predicting the results,the forgetting factor and item response theory were comprehensively considered to further enrich students’ personality characteristics and solve the problem of poor interpretability.(3)Experiments were conducted on four widely used public datasets,and compared with three typical models in the field of knowledge tracking,namely BKT,DKT,and DKVMN,to verify the interpretability and accuracy of the FDKT-ED model;Compared with the representative DKVMN,CKT,and SAKT models,the comprehensive analysis verifies the effectiveness and accuracy of TCKT-FI model in solving the long sequence dependency problem,and has good interpretability.Finally,by comparing the calculation time,the TCKT-FI model has higher computational efficiency,significantly saving memory,and is efficient.