Research On Knowledge Tracking And Learning Resource Recommendation Model For Personalized Learning

Personalized learning aims to enable learners to obtain appropriate learning styles and learning resources according to their learning ability,cognitive level,knowledge status,etc.,which is one of the goals pursued by teaching activities for a long time.In recent years,with the deepening of the process of education informatization,various online learning platforms have sprung up,providing learners with an open learning environment and rich learning resources,and at the same time generating massive learning data,which provides strong support for assessing learners’ cognitive level and providing appropriate learning resources.Therefore,mining and analyzing learner learning data,and then exploring personalized learning is an important task of education informatization.Therefore,this paper systematically conducts research on knowledge tracking and learning resource recommendation methods for personalized learning.Knowledge tracking(KT)aims to model the changing knowledge status of students with learning time based on students’ historical answer records,and then predict students’ answering performance,which is the core module supporting the smart education system.The learning answer record is the embodiment of students’ real learning level,and quantifies students’ intrinsic and tacit knowledge point mastery status from these external and explicit learning performances through preset models,and predicts their performance in unknown exercises.Specifically:(1)In view of the problem of sparse input data in the existing model,this paper proposes to use Q-matrix(Q-matrix)to mine the mapping relationship between knowledge points and questions,and map one or more knowledge points contained in the question to the knowledge space,which not only considers the correlation between the question and the knowledge point,but also alleviates the problem of data sparseness.(2)Aiming at the problem that the existing model cannot capture the relationship between the test questions,a method of integrating the attention mechanism is proposed,and a test question similarity calculation method based on the difficulty of the test question and the similarity of knowledge points is given,in which the difficulty of the question depicts the difficulty of the test question,even if two questions with the same knowledge point may have a better degree of differentiation for students.The similarity of knowledge points describes the degree of overlap between two questions at the level of knowledge points,and there are differences in students’ mastery of different knowledge points,so it is necessary to quantify the similarity of knowledge points measured by different questions.(3)Aiming at the problem of poor interpretability of existing models,the traditional psychometric item response theory(IRT)is proposed to make the model parameters have better explanatory performance.In order to solve the common learning forgetting characteristics in the real learning environment,the Ebbinghaus learning forgetting curve is introduced to dynamically quantify the degree of change of students’ knowledge status with the increase of learning time.Learning resource recommendation is another important aspect of personalized learning,large-scale open online courses(MOOC)platform brings together a large number of high-quality teaching resources,these various,dazzling resources for the majority of learners to achieve independent learning at the same time,but also caused "information overload" and "information trek" and other problems.Aiming at the problems that the existing MOOC recommendation methods do not make full use of the implicit information contained in MOOC videos,are easy to form the "cocoon effect",and it is difficult to capture the dynamically changing learning needs and interests of learners,this paper proposes a dynamic MOOC recommendation model integrating video subtitle information,with BERT(bidirectional encoder representations from transformers,BERT)as the encoder By integrating the multi-head attention mechanism,the semantic information of the subtitle text of the MOOC video is deeply explored,the network based on LSTM(long short-term memory,LSTM)architecture is used to dynamically capture the changing knowledge preference state of learners with learning,the attention mechanism is introduced to mine the personality information and common information between MOOC videos,and finally the MOOC video with the recall probability Top N is recommended based on the learner’s knowledge preference state.Finally,the experimental dataset collected in the real learning scenario verifies the effectiveness of the model.