Deep knowledge tracking method based on Bayesian neural network

Abstract

The premise of realizing personalized adaptive learning is to accurately evaluate the knowledge mastering condition of students. According to an existing knowledge tracking method based on a deep neural network, explicit encoding of knowledge in the human field is not needed, the complex relationship of student behaviors can be mined, but the effect is poor and over-fitting is easy to occur. The invention provides a deep knowledge tracking method based on a Bayesian neural network, a model employs an embeded+LSTM+Dense structure, parameters are replaced by a normal step-by-step mode from a point value, and the parameter adjustment is introduced into a Bayesian back propagation method. An embedding layer is used for training a student behavior vector with task guidance so as to improve theaccuracy of the model, an LSTM layer can effectively process the long dependence of data, and a Dense layer adjusts the dimension of output data; carrying out model parameter distribution, and clearlyrepresenting uncertainty of a prediction result and randomness of model optimization; bayesian prior knowledge accelerates model convergence, effectively prevents overfitting, and enhances generalization ability.

Description

technical field [0001] The deep knowledge tracing based on Bayesian neural network is a knowledge tracing method, and the invention relates to an online evaluation technology. Background technique [0002] In the context of big data, indicators such as students' learning habits, learning progress, and learning status are effectively quantified and collected. It is an important task of Intelligent Tutoring Systems (ITS) to conduct data analysis on students' learning behaviors, provide timely feedback on students' learning effects, recommend reasonable learning paths and learning resources with appropriate difficulty, and realize efficient knowledge imparting and learning. research topic. The premise of realizing personalized adaptive learning is to have an accurate assessment of students' knowledge mastery. The relationship between student behavior data is complex. The traditional method based on feature engineering to mine the learning behavior relationship and evaluate th...

AI Technical Summary

Problems solved by technology

The data volume of students' learning behavior data is huge and complex, and there are a large number of potential relationships in the data. Not only are there connections between knowledge points, but there are also connections between students' behaviors. The order in which students answer questions and each knowledge point The right or wrong order will affect the final result and performance. It is extremely difficult for traditional feature engineering modeling methods to model student behavior data with rich features.

(2) The existing deep neural network knowledge tracking model adopts the traditional Recurrent Neural Network (RNN) and uses the time truncated backpropagation algorithm, which is prone to overfitting problems

Images