Research On Knowledge Tracking Model Integrating Forgetting Factors And Learning Ability

With the rapid development of big data technology and the gradual popularization of applications in various fields,data-driven intelligent education services,such as intelligent tutoring systems,make personalized learning possible.However,how to accurately and comprehensively model learners through learners ’ learning process data is a difficult problem for personalized learning technology.Knowledge tracking is a technology based on the sequence of learners ’ answer behavior to model and predict the state of learners ’ knowledge mastery,which solves the problem of learners ’ personalized modeling to a certain extent.Through knowledge tracking,teachers can grasp students ’ learning status in time,further arrange exercises for students in a targeted manner,teach students in accordance with their aptitude,and improve learning efficiency.Knowledge tracking has achieved rich results in the long-term research accumulation,but there are three main problems in the existing knowledge tracking model : First,the existing mainstream knowledge tracking model updates the knowledge mastery of learners ’ answering exercises,but many ignore learners ’ forgetting behavior;second,the evaluation of learners ’learning ability is not perfect.The existing methods of predicting learners ’ future answers from the perspective of ability are not reasonable enough.Third,there is a lack of methods to effectively integrate knowledge attribution and ability attribution into an overall knowledge tracking framework.Aiming at these problems,this thesis constructs a knowledge tracking model that integrates forgetting factors and learning ability.The main research work is as follows:(1)Combined with educational psychology,the model considers three factors that affect knowledge forgetting : the number of times learners learn knowledge points,the number of answers and the mastery of knowledge points.Based on the above information,the forgetting behavior is modeled to fit the change of learners ’ knowledge mastery caused by forgetting behavior.(2)The representation method of learners ’ learning ability is proposed.Based on the interaction records of learners and exercises,this thesis considers that learners with similar behaviors have similar learning abilities,constructs a learner-exercise interaction heterogeneous graph,and uses the relationship graph convolutional network to embed the nodes of learners and exercises respectively.The representation vector of learners ’ learning ability is finally obtained.(3)A method of predicting learners ’ answers from the perspective of ability is proposed.In this thesis,the multi-layer fully connected layer is used to simulate the complex interaction process of learners ’ questions,and to predict learners ’ answers from the perspective of ability more accurately.On the ASSISTments2009-2010 and AICFE-math datasets,the model prediction effect is improved compared with other baseline models under the same evaluation index.In this thesis,the learner ’s learning ability embedding representation method based on the relationship graph convolutional network algorithm solves the problem of learner ’s learning ability representation.On this basis,the forgetting effect in the learner ’s learning process is considered,and the knowledge ability double tracking model is constructed to solve the existing knowledge tracking model.In the model,the knowledge module does not consider forgetting and the ability module construction is not perfect,which leads to the problem of low prediction accuracy and weak interpretability.