CN-116090632-B - Depth knowledge tracking method and device based on multitasking

CN116090632BCN 116090632 BCN116090632 BCN 116090632BCN-116090632-B

Abstract

The invention discloses a depth knowledge tracking method and device based on multiple tasks, wherein the method comprises the steps of obtaining the question characterization, the knowledge point characterization and the answer correct and error result of the current time and all previous times, inputting the question characterization, the knowledge point characterization and the answer correct and error result to a knowledge tracking model after training for processing, and obtaining the knowledge point answer correct and error result of the next time in response to the processing of the knowledge tracking model, wherein the knowledge tracking model comprises a first auxiliary task component for predicting the knowledge point corresponding to the current time t question and a basic knowledge tracking task component for predicting the answer correct and error result of the next time t+1 knowledge point based on a specified knowledge tracking network. According to the scheme, on the basis of the original basic task of predicting whether the student is correct or incorrect in answering, knowledge points input by the student at the current moment are predicted, and the historical overall answering accuracy of the student is predicted at each time step, so that the effect of evaluating the answering is improved.

Inventors

LIU ZITAO
LIU QIONGQIONG
HUANG SHUYAN
CHEN JIAHAO
LUO WEIQI

Assignees

北京乐柏信息咨询有限公司

Dates

Publication Date: 20260508
Application Date: 20230106

Claims (13)

1. A depth knowledge tracking method based on multitasking, comprising: Acquiring the current time Characterization of topics at all previous moments Knowledge point characterization And correct and incorrect results of the answer ; Inputting the topic representation The knowledge point representation And the correct and incorrect result of the answer Until the trained knowledge tracking model is processed, and Obtaining a next time in response to the processing of the knowledge tracking model Knowledge points answer the correct and incorrect results; Wherein the knowledge tracking model is based on a specified knowledge tracking network, and comprises a method for predicting the current moment First auxiliary task component for question corresponding knowledge point and method for predicting next time The first auxiliary task component and the basic knowledge tracking task component participate in training the knowledge tracking model together; the knowledge tracking model further comprises a second auxiliary task component for predicting the historical overall response accuracy of the student, and the second auxiliary task component, the first auxiliary task component and the basic knowledge tracking task component participate in training of the knowledge tracking model together; wherein the knowledge state required by the predictions of the second auxiliary task component and the basic knowledge tracking task component is obtained Comprising: inputting the answer correct and error result And the knowledge point characterization To the answer coding unit, outputting the relation answer coding characterization ; Computing the answer code characterizations Characterization of relationship And the knowledge point characterization Is a joint characterization of (2) Inputting the joint representation Outputting knowledge state to the knowledge tracking network Wherein the knowledge state Tracking the knowledge state of the last moment in the network by the specified knowledge And the combined characterization And (5) determining.
2. The method of claim 1, wherein the first auxiliary task component comprises a topic encoding unit, a relationship learning network unit, and a topic tagging predictor, and wherein the process of predicting a current topic correspondence knowledge point by the first auxiliary task component comprises: Inputting the topic representation And the knowledge point characterization After the topic code unit is reached, the relation representation is output through the relation learning network unit ; Inputting the relationship representation Outputting knowledge point prediction characterization to the topic labeling predictor ; Obtaining a first loss function 。
3. The multi-task based depth knowledge tracking method as claimed in claim 2, wherein the basic knowledge tracking task component includes a knowledge tracking predictor, and wherein the process of predicting correct and incorrect results of knowledge points at a next time by the basic knowledge tracking task component includes: Inputting the knowledge state Outputting the knowledge point answer correct and incorrect result prediction characterization at the next moment to the knowledge tracking predictor ; Obtaining a third loss function 。
4. The multi-tasking based depth knowledge tracking method of claim 3 wherein the second auxiliary task component comprises an answer encoding unit and a personalized prior knowledge predictor and wherein the process of predicting the historical overall answer accuracy of a student by the second auxiliary task component comprises: obtaining the accuracy rate of the historical overall response in a specified time range The history overall answer accuracy The ratio of the correct number of historical answers to the total number of historical answers is given; Inputting the knowledge state Outputting the historical overall answer accuracy predictive characterization of the students to the personalized priori knowledge predictor ; Obtaining a second loss function 。
5. The multitasking-based depth knowledge tracking method of claim 4 in which the knowledge tracking model is model trained by the first auxiliary task component, the second auxiliary task component and the primary knowledge tracking task component in combination.
6. The method of claim 5, wherein the model trained objective loss function is Wherein, the method comprises the steps of, And Is a super parameter for adjusting the weights of the first and second loss functions.
7. A multitasking based depth knowledge tracking method as claimed in claim 3 wherein said knowledge tracking model is model trained by said first auxiliary task component in conjunction with said primary knowledge tracking task component.
8. The method of claim 7, wherein the model trained objective loss function is Wherein, the method comprises the steps of, Is a super parameter for adjusting the weight of the first loss function.
9. The method of claim 6 or 8, wherein the extremum of the objective loss function is calculated to obtain the parameter optimum of the model training process.
10. The multi-task based depth knowledge tracking method of claim 1, wherein the specification based knowledge tracking network comprises a recurrent neural network, a memory network, a graph network, and/or an attention network.
11. A multitasking based depth knowledge tracking apparatus comprising: a first module capable of acquiring the current time Characterization of topics at all previous moments Knowledge point characterization And correct and incorrect results of the answer ; A second module capable of inputting the topic representation The knowledge point representation And the correct and incorrect result of the answer Until the trained knowledge tracking model is processed, and A third module capable of obtaining a next time in response to the processing of the knowledge tracking model Knowledge point mastery degree of (2); Wherein the knowledge tracking model is based on a specified knowledge tracking network, and comprises a method for predicting the current moment First auxiliary task component for question corresponding knowledge point and method for predicting next time The first auxiliary task component and the basic knowledge tracking task component participate in training the knowledge tracking model together; the knowledge tracking model further comprises a second auxiliary task component for predicting the historical overall response accuracy of the student, and the second auxiliary task component, the first auxiliary task component and the basic knowledge tracking task component participate in training of the knowledge tracking model together; wherein the knowledge state required by the predictions of the second auxiliary task component and the basic knowledge tracking task component is obtained Comprising: inputting the answer correct and error result And the knowledge point characterization To the answer coding unit, outputting the relation answer coding characterization ; Computing the answer code characterizations Characterization of relationship And the knowledge point characterization Is a joint characterization of (2) Inputting the joint representation Outputting knowledge state to the knowledge tracking network Wherein the knowledge state Tracking the knowledge state of the last moment in the network by the specified knowledge And the combined characterization And (5) determining.
12. An electronic device, comprising: processor, and A memory arranged to store computer executable instructions which, when executed, cause the processor to perform the multitasking depth knowledge tracking method of any one of claims 1 to 10.
13. A computer readable storage medium storing one or more programs, which when executed by an electronic device comprising a plurality of application programs, cause the electronic device to perform the multitasking depth knowledge tracking method of any one of claims 1-10.

Description

Depth knowledge tracking method and device based on multitasking Technical Field The present disclosure relates to the field of computer software technologies, and in particular, to a depth knowledge tracking method, device, electronic apparatus, and storage medium based on multitasking. Background In the educational scene, how to timely pay attention to learning and mastering abilities of all students is important to design of teaching schemes and rapid improvement of the students' abilities. The knowledge tracking is a method for automatically judging the learning ability of the corresponding students according to the historic learning conditions of the students by using a computer algorithm, and can quickly evaluate the learning conditions of all the students in real time. The depth knowledge tracking method based on multitasking mainly relies on the history question making records of students to infer the future question performance of the students, and can acquire the mastering condition of the students on each knowledge point. Most of the current methods replace the answer records of students on the topics with the answer records on the corresponding knowledge points of the topics, further predict the answer conditions of the students on the future knowledge points by using a machine learning/deep learning method, predict the answer conditions of the students on the future knowledge points by using additional information such as the topic text, the answer time of the students, the number of attempts and the like or obtain the topic difficulty and the like through learning, and model the problems directly by using a small number of methods. The current depth knowledge tracking method based on multitasking mainly comprises a method based on a psychology statistical model, a method based on traditional machine learning and a method based on depth learning. However, these methods are complicated to model, resulting in poor model effect or too large model to be deployed in real scenes. Therefore, how to improve the evaluation effect of future question performance of students in a mode of adding a small amount of parameters on the basis of not introducing additional information is a technical problem to be solved. Disclosure of Invention An object of embodiments of the present disclosure is to provide a depth knowledge tracking method, apparatus, electronic device, and storage medium based on multitasking, in view of the above-mentioned problems. In order to solve the above technical problems, the embodiments of the present specification are implemented as follows: in a first aspect, a depth knowledge tracking method based on multitasking is provided, including: Acquiring a question token [ q 0,…,qt ], a knowledge point token [ c 0,…,ct ] and a correct and incorrect answer result [ r 0,…,rt ] at the current time t and at all the previous times; Inputting the question characterization [ q 0,…,qt ], the knowledge point characterization [ c 0,…,ct ] and the correct and incorrect answer result [ r 0,…,rt ] to a trained knowledge tracking model for processing, and Responding to the processing of the knowledge tracking model, and obtaining a correct and incorrect result of a knowledge point at the next time t+1; the knowledge tracking model is based on a specified knowledge tracking network and comprises a first auxiliary task component for predicting a knowledge point corresponding to a topic at the current moment t and a basic knowledge tracking task component for predicting a correct and incorrect result of a knowledge point at the next moment t+1, wherein the first auxiliary task component and the basic knowledge tracking task component participate in training of the knowledge tracking model together. Optionally, the first auxiliary task component comprises a topic coding unit, a relation learning network unit and a topic labeling predictor, and the process for predicting the current topic corresponding knowledge point by the first auxiliary task component comprises the following steps: Inputting the topic representation [ q 0,…,qt ] and the knowledge point representation [ c 0,…,ct ] to the topic coding unit, and then outputting a relationship representation z t through the relationship learning network unit; Inputting the relation representation z t to the topic labelling predictor, outputting knowledge point prediction representation Obtaining a first loss function Optionally, the knowledge tracking model further comprises a second auxiliary task component for predicting the historical overall response accuracy of the student, and the second auxiliary task component participates in training of the knowledge tracking model together with the first auxiliary task component and the basic knowledge tracking task component. Optionally, obtaining the knowledge state h t required by the second auxiliary task component and the primary knowledge tracking task component prediction includes: Inputting the answer corre