Research On Knowledge Tracing Model Based On Student Behavior State Sequence

In current research,the continuous innovation of knowledge tracking technology is creating new opportunities for the mining and analysis of educational big data in the future.At this stage,a new environment has emerged,that is,web-based education or learning management system(LMS).LMS is a digital framework for managing and simplifying online learning.Its main purpose is to manage learners’ learning behavior,monitor students’ participation,and track their progress in the whole system learning.This paper investigates the publicly available dataset of Students’ Academic Performance Dataset collected from LMS using the Experience API(x API).The role of x API in the process of collecting big data in education is to monitor students’ behavior through the education process,in order to evaluate characteristics that may affect students’ academic performance.This paper focuses on researching methods for improving knowledge tracking models from different perspectives to improve the accuracy of academic performance prediction,exploring the potential relationship between student behavior state feature sequences and academic performance,so as to provide the model with higher quality prediction performance and more accurate fitting results.In order to achieve the knowledge tracking task of this article,the following research work was mainly completed:First,the original dataset is preprocessed,the data is cleaned and encoded,the number of key feature values is verified,and the corresponding specific key feature attributes are extracted,the feature contribution score of the full dimension is calculated,and the student behavior state characteristics that best fit the research in this paper are found,and the correlation between them is analyzed using mathematical statistics tools.The main purpose of this work is to supplement the missing sequence sample information of the model as much as possible,so that the prediction results of the knowledge tracking task are as accurate as possible and have reference value.Secondly,GM-HMM,a hidden Markov knowledge tracking model optimized by the Gauss Newton’s method,is proposed.Considering that students’ academic achievements under different behaviors are different,the modeling of behavior state feature sequence and academic achievements is used to predict students’ final mastery of major course knowledge.Finally,the application of Gaussian exponential function fitting in education big data is discussed,and GM-GF is proposed,which makes the Gaussian exponential model optimized by Gauss Newton iteration possible in the solution of knowledge tracking tasks.On this basis,we further completed the GM-GF grouping comparison experiment of univariate features and bivariate multi features,explored the comparison experiment of multivariate single Gaussian GM-GF model and multi Gaussian mixture model,and further discussed and verified the Goodness of fit of GM-GF model.All group comparison experiments have shown that the model proposed in this article outperforms other conventional prediction models in terms of fitting results and goodness of fit.