Research On Knowledge Tracking Model Based On Problem Combinatioe And Learning Ability

Knowledge tracking is a deep learning technology that models students’ knowledge mastery based on students’ past performance in answering problems,so as to obtain students’ knowledge status,so that education system can predict students’ performance in the future learning process,so as to achieve efficient personalization teaching.In recent years,many knowledge tracking models based on deep learning theory have emerged,such as deep knowledge tracking models,knowledge tracking models based on graph interaction and dynamic key-value.these models have significantly improved accuracy of knowledge tracking,but re are still some factors that need to be considered:(1)A problem is composed of two parts:problem data and problem skills.(2)There are many feature information in dataset that have not been introduced by model,such as difficulty of skills and degree of students’ mastery of skills.(3)Although some models have added forgetting factors,students’ own learning ability will also affect forgetting degree of students.This paper proposes a knowledge tracking model based on problem combination for(1)and(2).Firstly,6 relevant label data are extracted;secondly,three embedding matrices representing features of problem data,features of skills,and features of students are defined,and softmax function is used to calculate influence weight of features of problem data and skills on difficulty of problem,and carry out weighted summation,and get embedding matrix representing features of problems;then calculate corresponding predicted value,and use six mean square error loss functions to optimize embedding matrix;last,according to each answer record of student,extract corresponding student embedding,problems answered by students are embedded and related skills are embedded,and probability of students answering problems correctly is predicted through calculation methods of matrix splicing,matrix multiplication and linear transformation.Aiming at(3),this paper proposes a knowledge tracking model based on learning ability.First,extract student embedding,problem embedding obtained by multiplying problem features and difficulty of problem,and skill embedding obtained by multiplying skill features and skill difficulty;secondly,three embedding matrices are calculated by matrix connection and matrix multiplication;then,student’s learning is defined Ability and forgetting ratio,and calculate amount of learning forgetting due to student’s own learning ability and amount of learning growth that student acquires after answering problem,and n multiply se two changes with calculation result of previous step,and calculate result is passed into a linear transformation that predicts probability that student will answer problem correctly.Experiments show that two models proposed in paper have achieved better performance than benchmark model in two evaluation indicators of ACC and AUC,which verifies effectiveness of algorithm proposed in this paper.