Research On Knowledge Tracing And Exercise Recommendation For Intelligent Tutoring

As information technology and education continue to integrate and develop,intelligent tutoring technology are being applied widely in the field of education.The intelligent tutoring technology can provide personalized learning guidance based on each student’s unique learning situation,helping students better master knowledge.Building accurate student models is the key to achieving intelligent tutoring and personalized educational services.At the same time,in the field of intelligent tutoring,exercise recommendation is also very important.Exercise recommendation can analyze students’ learning data,predict their knowledge level and learning preferences,and recommend exercises suitable for their learning level and preferences,helping students better consolidate knowledge and improve learning efficiency.Therefore,based on the above research background,this article carries out a series of studies:Firstlym in response to the insufficient modeling of difficulty characteristics of exercises,and the lack of refinement of students’ ability characteristics and knowledge status in existing student modeling methods.A difficulty-aware convolutional knowledge tracing model is proposed,which extracts several types of effective information to model the difficulty characteristics of exercises.Two matrices are established based on the students’ historical learning records,to respectively extract the micro changes in the students’ knowledge status and ability level.A psychological measurement model is introduced to enhance the interpretability of the model.The experiments show that this model effectively enhances the difficulty characteristics of exercises,and outperforms mainstream methods on two datasets with different data distributions.Secondly,for the field of programming intelligent tutoring,a exercise recommendation model based on deep reinforcement learning and strategy selection is proposed,which mines the implicit relationships between knowledge points using the knowledge graph in programming and uses the knowledge graph embedding method to enhance exercise representation.At the same time,the concept of current learning strategy is introduced to divide the student’s learning strategies and a strategy selection network is used to learn the dynamic changes of strategies,which improved the performance of recommendation.Finally,in the exercise recommendation model,the aforementioned knowledge tracing model was used as a student simulator and three types of rewards were set: exercise sequence reward,knowledge sequence reward,and participation reward.By simulating the student’s answer situation,the participation reward value of the student was kept within a reasonable range.The experiments show that compared to other baseline models,this recommendation model provides better recommendation strategies in various recommendation system indicators on two online judge system datasets.