Intelligent learning state tracking method and system based on multi-task framework and application

Abstract

The invention belongs to the technical field of personalized learning, and discloses an intelligent learning state tracking method, system and application based on a multi-task framework, and the method comprises the steps: collecting external learning behavior features, learning resource features and potential capability features of a learner, and carrying out the preprocessing operation, and obtaining learning features containing prior information; then constructing a plurality of stacked convolutional neural networks to carry out deep representation learning on the learning features, controlling the forgetting condition of learners, and constructing deep learning features; performing deep and shallow feature fusion, introducing a bidirectional recurrent neural network, and creating an intelligent learning state tracking model based on long time sequence dependence; and finally, quantifying and predicting the learning state of the learner, and constructing a loss function to perform multi-task training. The prediction performance of the knowledge tracking model in the aspect of predicting the learning state of the learner can be improved, the prediction field and the education application field of the knowledge tracking model are expanded, and the development of personalized education and intelligent education is promoted.

Description

technical field [0001] The invention belongs to the technical field of personalized learning, and in particular relates to an intelligent learning state tracking method, system and application based on a multi-task framework. Background technique [0002] At present, with the development of artificial intelligence technology in the field of educational applications, a series of online learning platforms, such as MOOC, Edx, Coursera, etc., have sprung up, enabling the field of education to better realize individualized teaching. At the beginning of 2020, the online teaching mode reached an unprecedented level. However, the teaching mode of teachers still stays in the traditional offline teaching in the past. The learning data of learners on the online learning platform has not been effectively mined, resulting in the inability of learners to receive personalized services. , teachers are also unable to provide individualized instruction. [0003] The intelligent learning stat...

AI Technical Summary

Problems solved by technology

At the beginning of 2020, the online teaching mode reached an unprecedented level. However, the teaching mode of teachers still stays in the traditional offline teaching in the past. The learning data of learners on the online learning platform has not been effectively mined, resulting in the inability of learners to receive personalized services. , teachers are also unable to provide individualized instruction

The core technology of the Bayesian-based intelligent learning state tracking method is the hidden Markov model, which uses statistical learning methods and machine learning methods to update the knowledge state of learners. It has strong interpretability in the field of education, and the model is simple It is easy to understand, but it also has some shortcomings: 1) It relies too much on the understanding of educational experts; 2) It is well applied on small data sets, so when the amount of learner interaction data is large, the model prediction effect is not good; 3) In the Insufficient representation of the learner's learning state

The intelligent learning state tracking method based on deep learning can better adapt to large data sets, and has natural advantages in representing the learning state of learners, but it also has shortcomings: 1) The interpretability in the field of education is weak ; 2) Failure to take into account the impact of mistakes and guesswork

[0005] (1) The traditional intelligent learning state tracking method ignores other learning factors that affect the learner's learning state in the learning process, such as answering time and potential ability, and does not fully characterize the learner's learning process, resulting in ineffective subsequent predictions. precise;

[0006] (2) The traditional intelligent learning state tracking method only predicts the learner's answering response, but does not predict the learner's answering time, potential ability, etc., nor does it predict the learner's answering response, answering time, and potential ability at the same time ;

[0007] (3) The traditional intelligent learning state tracking method is not stable enough when predicting the learner's learning state, but the change process of the learner's learning state should be stable;

[0011] (3) How to solve the problem of fluctuating prediction results of traditional intelligent learning state tracking methods

Images