Optimization And Application Of Knowledge Tracing Model Based On Multiple Behavioral Features

In recent years,efficient and intelligent online education has gradually demonstrated certain advantages that traditional offline education cannot replace.Against this background,knowledge tracing,as one of the most important aspects of data mining in online education,aims to track students’ knowledge status and predict their mastery of knowledge based on their answering performance.As a current research focus in the field of intelligent education,deep knowledge tracing models have received much attention.However,many scholars have overlooked the impact of various behavioral characteristics of students during the learning process.Although some researchers have used decision trees to model behavioral characteristics,this algorithm is prone to falling into local optima traps.In addition,traditional modeling methods cannot simultaneously consider the differences in different student behavioral characteristics and analyze the learning behavior patterns of the same student in a time series.To address these two issues,this thesis proposes two methods and applies them to a personalized exercise recommendation system.The specific research content is as follows.1.Propose an improved knowledge tracing method based on DKVMN-GADT for multiple behavioral characteristics.To address the problem of decision trees being prone to falling into local optima,this method introduces a two-layer CART decision tree based on a genetic algorithm.The first layer optimizes the upper layer of CART by using a genetic algorithm to find the optimal sequence of behavioral characteristics.The second layer optimizes the lower layer of CART based on the first layer by using a genetic algorithm to select the optimal feature weights as decision tree splitting points.The decision tree response and the student’s answer situation are then cross-calculated.Finally,using DKVMN as the underlying model,the student’s future answering situation is predicted.This method continuously finds the optimal features through genetic operations,which improves the classification accuracy of the decision tree and effectively alleviates the problem of local optima.Compared with classic knowledge tracing models(DKT,DKVMN,DKVMN-DT)on three public datasets(ASSISTments2009,ASSISTments2012,and Algebra2005-2006),this method shows an improvement in the AUC index.2.Propose a knowledge tracing method that combines time-series behavior features with learning ability time-forgetting features.In response to the impact of students’ timeseries behavior features on the prediction of knowledge tracing models during the learning process,this method introduces a time convolutional network to preprocess the time-series behavior features.The time convolutional network mainly consists of causal convolution,dilated convolution,and residual connections.At the same time,considering the impact of the time interval between answering questions on the learning ability and forgetting of different students,the method uses the preprocessed results to calculate the learning ability and then introduces an enhanced time-forgetting effect mechanism based on the learning ability to strengthen the impact of long time intervals on students’ knowledge state changes.This makes the knowledge tracing model more in line with the time-forgetting law of students’ learning.Finally,the proposed model was compared with representative models(DKT,DKVMN,DKVMN-DT)on two public datasets(ASSISTments2012,Algebra2005-2006)in comparative experiments,and the results show that the AUC and ACC indicators were improved to varying degrees.3.A prototype of a personalized exercise recommendation system based on a multibehavior feature knowledge tracing algorithm has been implemented.The system mainly uses an improved knowledge tracing algorithm to track student learning progress,predict student answering results,facilitate teachers’ understanding of students’ mastery of different knowledge points,and recommend suitable exercises for students to improve their learning initiative and achieve personalized learning goals.