Research On Deep Knowledge Tracing Method Based On Learner Ability

In recent years,with the popularity of the internet,more and more students have chosen to use online education platforms for learning.These platforms can collect and analyze students’ interaction data to evaluate their learning progress and mastery of knowledge,and adjust teaching content and learning paths based on their performance to help learners better grasp knowledge.In the process of online education,students’ knowledge levels are constantly changing,and learners’ mastery of knowledge is dynamically changing.Therefore,modeling learners’ learning processes and dynamically updating their knowledge level evaluation is particularly important.In order to more accurately evaluate learners’ learning levels,knowledge tracking tasks have emerged.Through analyzing students’ historical learning information,it dynamically tracks students’ knowledge levels and predicts their future learning performance.Students’ future learning conditions are crucial for teachers,who can use the predicted results to teach students according to their aptitudes,adopt different educational methods for students in different situations,and develop more accurate teaching plans.Although the current methods have good performance,there are still some problems.For example,the models cannot model individual differences between learners well,cannot take into account learners’ ability levels,and cannot dynamically update learners’ knowledge states.This paper focuses on student interaction records,analyzes them based on learners’ abilities,and comprehensively considers the effects of individual differences in learners’ abilities and learning behavior characteristics on their learning processes.It proposes two different knowledge tracking models and an improved particle swarm algorithm for optimizing the parameters of the knowledge tracking model.The specific research content is as follows:(1)A deep knowledge tracking model CA-DKTA based on individual differences in students’ abilities and attention mechanism is proposed.Starting from the interaction process between learners and exercises,it considers the continuous changes in individuals’ abilities during the learning process and the differences between different individuals,and introduces attention mechanism in the prediction process to calculate the correlation weights between exercises and knowledge points,further explaining students’ mastery of knowledge.The model was experimented on three real datasets and achieved good results,visualizing changes in students’ knowledge states and ability levels.(2)Students’ learning abilities are also reflected in their learning behaviors to some extent,such as their memory of knowledge and dynamic learning states.This paper proposes a knowledge tracking model that combines students’ forgetting factors and unique learning behavior characteristics.Forgetting behavior is modeled using learning interval time and repetition learning times,and individual behavior differences are combined with students’ abilities to reflect learners’ positive or negative learning states.(3)To further optimize the proposed model in this paper,an optimization strategy with dimensionality reduction and clustering ideas is proposed.Combining the particle swarm algorithm in natural computation methods,the high-dimensional decision variables of the population are mapped to a low-dimensional space through the idea of locally embedded linear dimensionality reduction,and K-means clustering is used to achieve information exchange between local particles,improving the efficiency of the algorithm while ensuring good optimization ability.This algorithm is applied to design a two-layer structure optimization method to optimize the batch＿size and learning rate of the model.