Application Of Deep Knowledge Tracing In Exercise Recommendation

Today,with the rapid development of education informatization,online education resources are constantly enriched to provide comprehensive teaching and education services for various learners.How to track the knowledge status of learners and choose appropriate resources to recommend to learners is the focus of future online education research.At present,deep knowledge tracking has achieved good application results in the application of problem recommendation,but there are two main problems: 1)Deep knowledge tracking(DKT)model based on deep learning,due to the existence of individual learners’ actual answer time Because of the difference in learning,it is impossible to collect the learning status,resulting in a low prediction accuracy.2)When recommending exercises,the existing heuristic-based exercise recommendation algorithm cannot quickly find suitable exercises to recommend,which affects the learner’s learning efficiency.In response to the above problems,this study selected an online public data set suitable for modeling the knowledge status of learners,and evaluated the optimization model.On this basis,a deep reinforcement learning recommendation algorithm was used to recommend online exercises,and Evaluate the mastery of the learners’ knowledge points.The specific work is as follows:(1)Based on the original deep knowledge tracking(DKT)modeling technology,the learning time of the learners’ knowledge points is dynamically grouped by obtaining the answer time of the learner’s twice before and after each knowledge point,and the grouping result is added to the model,A knowledge tracking model DKT-DSST that integrates learner dynamic skill learning time grouping is proposed.At the same time,in order to maximize the benefits of individual differences in learner answering time,regularization terms corresponding to the loss function of the original DKT model were introduced,and the loss function of the optimized model was reconstructed.(2)When applying the knowledge tracking model to problem recommendation,the improved DKT-DSST acts as a learner knowledge state simulator that interacts with the environment,and introduces reinforcement learning into the problem recommendation algorithm.A deep neural network-based agent is used to design the reward function for the exercises,and the strategy gradient is used to update the direction to change the strategy parameters and quickly find the appropriate exercises to improve the learner’s learning efficiency.(3)Combined with the DKT-DSST knowledge tracking model,an online answering system was designed and implemented.In the system implementation part,the full recommendation path combining the learner model and the recommendation algorithm is shown.Then design the main business function modules and data sheets of the exercise system.In the implementation part,learners are selected for online exercise recommendation.After each recommendation,the learner’s hidden knowledge level is calculated and evaluated,and the realization of the exercise recommendation function is completed.