CN114997461A - Time-sensitive answer correctness prediction method combining learning and forgetting - Google Patents
Time-sensitive answer correctness prediction method combining learning and forgetting Download PDFInfo
- Publication number
- CN114997461A CN114997461A CN202210374206.5A CN202210374206A CN114997461A CN 114997461 A CN114997461 A CN 114997461A CN 202210374206 A CN202210374206 A CN 202210374206A CN 114997461 A CN114997461 A CN 114997461A
- Authority
- CN
- China
- Prior art keywords
- student
- answer
- formula
- time
- layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000013528 artificial neural network Methods 0.000 claims abstract description 29
- 238000012549 training Methods 0.000 claims abstract description 9
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 239000010410 layer Substances 0.000 claims description 57
- 230000015654 memory Effects 0.000 claims description 51
- 230000006870 function Effects 0.000 claims description 31
- 239000002356 single layer Substances 0.000 claims description 18
- 230000008569 process Effects 0.000 claims description 16
- 230000004913 activation Effects 0.000 claims description 15
- 239000011159 matrix material Substances 0.000 claims description 15
- 230000004044 response Effects 0.000 claims description 12
- 239000011541 reaction mixture Substances 0.000 claims description 10
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 claims description 8
- 230000003213 activating effect Effects 0.000 claims description 6
- 150000001875 compounds Chemical class 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000011835 investigation Methods 0.000 claims description 3
- 230000002123 temporal effect Effects 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 2
- 241000022852 Letis Species 0.000 claims 1
- 230000007787 long-term memory Effects 0.000 abstract description 2
- 230000006403 short-term memory Effects 0.000 abstract description 2
- 230000001149 cognitive effect Effects 0.000 description 9
- 230000008859 change Effects 0.000 description 4
- 230000001537 neural effect Effects 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 2
- 230000002452 interceptive effect Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000007087 memory ability Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000011144 upstream manufacturing Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/20—Education
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Tourism & Hospitality (AREA)
- Human Resources & Organizations (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Marketing (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Quality & Reliability (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Development Economics (AREA)
- Educational Administration (AREA)
- Educational Technology (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method for predicting the correctness of a time-sensitive answer by combining learning and forgetting, which comprises the following steps: 1, acquiring historical student answer records and performing serialization preprocessing; 2, fitting the knowledge state of the student by using a long-term and short-term memory network on continuous time and predicting the correctness of the student answering questions; and 3, training the neural network parameters to obtain a trained answer correctness prediction model for realizing the prediction of student answer correctness. The invention can realize the prediction of the correctness of student answer from end to end and can model the knowledge state of students at any time, thereby providing effective assistance for an intelligent tutoring system and teachers.
Description
Technical Field
The invention relates to the field of cognitive modeling, in particular to a time-sensitive answer correctness prediction method combining learning and forgetting.
Background
In recent years, a large amount of student exercise records are accumulated in a rapidly-developed intelligent teaching system, so that a new data-driven mode is provided for computer-aided education: and (4) cognitive modeling. The goal of cognitive modeling is to discover the knowledge level or learning ability of a student, the results of which can benefit a wide range of intelligent educational applications, such as predicting student performance and personalized course recommendations.
Given the dynamics of the learning process, many efforts are made to track changes in student knowledge levels in cognitive modeling. Existing methods can be divided into two categories: (1) traditional models represented by Bayesian Knowledge Tracking (BKT) and factorized models; (2) sequence models based on deep neural networks, such as Deep Knowledge Tracking (DKT), dynamic key-value storage network (DKVMN), etc. The deep knowledge tracking model is the first method of fitting the knowledge state of a student by using a recurrent neural network and deducing the answer performance of the current exercise according to the historical learning record of the student.
A long-standing research challenge in the field of cognitive modeling is how to naturally integrate a forgetting mechanism into the learning process of knowledge, and some researchers have incorporated a forgetting factor into student cognitive modeling to improve the accuracy of student response results and the ability to capture forgetting. Most of these methods rely on artificially designed Forgetting features (e.g., DKT + Forgetting statistics on how many times a student has made questions containing a certain knowledge point and input them as feature input models); or rely on simplified process assumptions (e.g., fixed and discrete learning intervals), greatly limiting the performance and flexibility of downstream applications, there remains a lack of a realistic cognitive modeling approach to balance the learning and forgetting processes, such that forgetting occurs at continuous times, and the student's answer performance changes as time goes by. We find that the modeling mode of the neural Hooke process is similar to the description of the memory law in cognitive psychology, and heuristically use a continuous long-time memory network in the neural Hooke to fit the learning and forgetting processes which are dependent on each other in continuous time, so that the ability of predicting the correct answer of students is improved, and meanwhile, references related to the memory abilities of the students can be provided for an intelligent tutoring system and a teacher.
Disclosure of Invention
The invention aims to solve the problems in the prior art and provides a time-sensitive answer correctness prediction method combining learning and forgetting, so that the change process of the knowledge state of students under the mutual influence of learning and forgetting can be fully and truly modeled, the knowledge mastering degree of the students at any time can be obtained, the end-to-end student answer correctness prediction is realized, the student answer result prediction precision is improved, and effective assistance is provided for an intelligent tutoring system and a teacher.
The invention adopts the following technical scheme for solving the technical problems:
the invention relates to a time-sensitive answer correctness prediction method combining learning and forgetting, which is characterized by comprising the following steps of:
set students as a setSet of questions asThe knowledge concept set isWherein the student setTherein has L namesStudent, subject setThere are M questions, knowledge concept setN knowledge points are present; representing a set of students using sQ represents a question setK represents a knowledge concept setAnd set the topicsThe middle question number is 1, …, M, knowledge concept setThe number of the middle knowledge point is 1, …, N;
representing historical response records of any student s as response sequences Wherein the content of the first and second substances,for the moment of the i-th answer of student s, andnumbering the questions answered the ith time by student s,answering questions for the ith time of student sThe number of the knowledge concepts under investigation is numbered,indicates that the student s answers the question at the ith timeIn response to the above situation, ifIndicate a right answer, ifDenotes an error, i ═ 1,2, …, n s ,n s The number of times of answering questions for student s;
step 2, constructing a neural network for predicting the correctness of the knowledge state fitting-answering, comprising the following steps: a learning part represented by the continuous long-short term memory network, a forgetting part represented by the continuous long-short term memory network and an answer prediction module;
wherein, the learning part represented by the continuous long-short term memory network comprises: the system comprises a single-hot coding embedded layer, four single-layer fully-connected feedforward neural networks, two activation functions and a cell information calculation layer;
the forgetting part represented by the continuous long-short term memory network comprises: the system comprises three single-layer fully-connected feedforward neural networks, two activation functions, a memory attenuation layer and a knowledge state acquisition layer;
the answer prediction module comprises two independent hot coding embedded layers, a multilayer perceptron layer and an activation function;
step 2.1, a learning part represented by a long-short term memory network in continuous time:
step 2.1.1, the one-hot coded embedding layer utilizes formula (1) to calculate student s is inInteractive embedding when answering questions constantly
In formula (1), A is an embedded matrix to be trained, anm is the dimension of the embedding,indicates that the student is atInstant answering performanceIs encoded by one heat, andif it isDenotes s at t i No answer or question answering error exists at the knowledge point with the number of j% N at any moment, ifThen the student s is indicatedThe time instants respond correctly at knowledge points numbered j% N, where the% sign indicates the remainder and is derived from equation (2):
step 2.1.2 atAt any moment, it is set that student s answers question at ith timeKnowledge state of time isWill be provided withAndspliced into the ith input vectorThen, respectively inputting three single-layer fully-connected feedforward neural networks and correspondingly passing through sigmoid functions, thereby correspondingly outputting a first forgetting gate during the ith updateFirst input gateAnd an output gateWhen i is 1, let the initial knowledge state of student sIs a set value;
step 2.1.3, input the ith vectorInputting a fourth single-layer full-connection feedforward neural network, and outputting through a tanh activation functionCandidate memory representation of time of dayThereby using equation (3) to calculateMemory representation in temporal cell information computation layer
In the formula (3), the reaction mixture is,representThe after-attenuation memory in the time memory attenuation layer indicates that when i is 1, the method makesIs the set value;
step 2.2, forgetting part represented by long-short term memory network in continuous time:
step 2.2.1, input the ith vectorInputting the data into a fifth single-layer fully-connected feedforward neural network, and activating a function through softplus, thereby obtaining the presence of the student sForgetting factor within a time period
Step 2.2.2, input the ith vectorRespectively inputting the data into the remaining two single-layer fully-connected feedforward neural networks and correspondingly activating a function through the sigmoid so as to correspondingly obtain a second forgetting gate during the ith updateSecond input gate
Step 2.2.3, the memory attenuation layer is calculated by the formula (4)Lower memory decay limit over a period of time
In the formula (4), the reaction mixture is,for the last period of timeThe lower limit of the internal memory attenuation is set to 1 when i is equal toIs the set value;
step 2.2.4, the memory attenuation layer is calculated by the formula (5)Memory representation c of time forgotten s (t):
Step 2.3, acquiring a hidden knowledge state:
in formula (6)To obtainMemorial representation of forgotten timeAnd is recorded as the memory representation after attenuationThe knowledge state acquisition layer calculates the position of the student s by using the formula (6)Hidden knowledge state when answering questions at any moment
In the formula (6), sigma (·) is a sigmoid activation function;
step 2.4, an answer prediction module:
step 2.4.1, orderTo solve the problemsThe two single-hot-coded embedded layers respectively use the formula (7) and the formula (8) to obtain the themeDifficulty ofAnd degree of distinction
step 2.4.2, the multilayer perceptron layer orders students to be atTemporal capability level representationThereby obtaining the question of the student s at the i +1 th answer by using the formula (9)On the prediction of correct probability of answer
In formula (9), F (-) is a multilayer perceptron;
step 2.5, assigning i +1After giving the value to i, returning to the step 2.1 for sequential execution until the historical answer sequence of the students s is completedPrediction of answer correct probability of last answer in (1)
Step 3, constructing cross entropy loss by using the formula (10)And training the knowledge state fitting-answer correctness prediction neural network to obtain a trained answer correctness prediction model for realizing the prediction of student answer correctness:
in the formula (10), the compound represented by the formula (10),for student s at t i The predicted value of the right probability of answering at a moment,for student s at t i The true value of the answer correctness at the moment, wherein,the response is shown to be wrong and the answer is wrong,indicating a right to answer.
The method for predicting the answer correctness of the time-sensitive joint learning and forgetting is characterized in that the answer prediction module in the step 2.4 is used for predicting the answer correctness according to the following process:
step 2.4.1, orderTo solve the problemsUsing the formula (11) and the formula (12) to obtain the titleDifficulty ofAnd degree of distinction
step 2.4.2, the multilayer perceptron layer utilizes formula (13) to obtain the student s is atTemporal capability level representation
step 2.4.3, the multilayer perceptron layer thus obtains the question when the student s answers at the i +1 th time by using the formula (9)On the prediction of correct probability of answer
Further, the answer prediction module in the step 2.4 is set to predict the answer correctness according to the following process:
step 2.4.1, orderTo solve the problemsUsing the formula (15) and the formula (16) respectively to obtain the titleDifficulty of (2)And degree of distinction
step 2.4.2, the multilayer perceptron layer utilizes formula (17) to obtain the student s is atTemporal capability level representation
step 2.4.3, setting the question-knowledge point matrix as Q q ={Q mn } M×N M is more than or equal to 1 and less than or equal to M, N is more than or equal to 1 and less than or equal to N, and if the problem numbered M looks at the knowledge point numbered N, Q is written mn 1 otherwise, denote Q mn =0;
The multi-layer perceptron layer obtains the question when the student s answers at the (i + 1) th time by using the formula (18)On the prediction of correct probability of answer
In the formula (18), f' (. cndot.) represents a multilayer perceptron, symbolRepresenting the multiplication of corresponding positions of the matrix.
Compared with the prior art, the invention has the beneficial results that:
1. the invention jointly models learning and forgetting by heuristically using a continuous long-short term memory network in the neural Hox process, thereby obtaining the knowledge state of the student in continuous time; the influence factors of forgetting are not only related to the current knowledge mastery degree and learning content of students, but also related to the time length, are sensitive to time factors, and can more truly and fully perform cognitive modeling on the students, so that the forgetting capacity of the students at different times can be measured, the correctness of student answers can be predicted at high accuracy, valuable references can be provided for an intelligent tutoring system, a teacher and the like to know the learning state of the learner, the students can be guided to perform targeted teaching training, and the method can be used as upstream application of self-adaptive question making and the like.
2. The invention carries out student answer expression prediction through the couplable knowledge mastery degree-question interactive function, which not only can effectively link the knowledge mastery degree and question information of students, but also can obtain the scalar comprehensive knowledge mastery degree of students or the mastery degree of students on each knowledge point, thereby enhancing the interpretability of the model, being used for visualization of knowledge state, helping intelligent tutoring systems, learners and the like to quickly know the comprehensive competence level of learners and the competence level on specific knowledge points and carrying out targeted training.
3. The invention models the dynamic change of the knowledge state of the student through the continuous long-term and short-term memory network, and the modeling mode can combine the learning and forgetting processes, so that the change of the knowledge state is close to the real change process, and the prediction precision is further improved in the future performance prediction of the student.
4. Experiments show that compared with other advanced algorithms, the invention has stable performance of predicting the answer on different sequence lengths (namely the number of the answers made by each student) and shows good robustness.
Drawings
FIG. 1 is a diagram of a model framework corresponding to the method of the present invention.
Detailed Description
In this embodiment, referring to fig. 1, a method for predicting correctness of a time-sensitive answer by combining learning and forgetting is performed according to the following steps:
set students as a setThe topic set isThe knowledge concept set isWherein the student setIn which there are L students and the question setThere are M questions, knowledge concept setsN knowledge points are present; representing a set of students using sQ represents a question setOne problem in (1), k represents a set of knowledge conceptsAnd set the topicsMiddle problemsNumbered 1, …, M, set of knowledge conceptsThe number of the middle knowledge point is 1, …, N;
representing historical response records of any student s as response sequences Wherein, the first and the second end of the pipe are connected with each other,for the moment of the i-th answer of student s, andnumbering the questions answered the ith time for student s,answering questions for the ith time of student sThe number of the knowledge concepts under investigation is numbered,indicating that student s answered the question at the ith timeIn response to the above situation, ifIndicate a right answer, ifDenotes an error, i ═ 1,2, …, n s ,n s The number of times of answering questions for student s; due to the fact thatThe middle school students have different answering lengths, the maximum length is set to be ML, answering records exceed the ML and are cut into new sequences, and the shortage is filled with 0. Three real datasets assisment 12, assisment 17, and slepemapy. cz are used in this embodiment, and ML is set to 100. The example uses 5-fold cross training, the experimental results are averaged over 5 trains, 20% of the data set is used as the test set, 10% is used as the validation set, and 70% is used as the training set.
Step 2, constructing a neural network for predicting the correctness of the knowledge state fitting-answering, comprising the following steps: a learning part represented by the continuous long-short term memory network, a forgetting part represented by the continuous long-short term memory network and an answer prediction module;
wherein, the learning part represented by the continuous long-short term memory network comprises: the system comprises a single-hot coding embedded layer, four single-layer fully-connected feedforward neural networks, two activation functions and a cell information calculation layer;
the forgetting part represented by the continuous long-short term memory network comprises: the system comprises three single-layer fully-connected feedforward neural networks, two activation functions, a memory attenuation layer and a knowledge state acquisition layer;
the answer prediction module comprises two independent hot coding embedded layers, a multilayer perceptron layer and an activation function;
step 2.1, a learning part represented by a long-short term memory network in continuous time:
step 2.1.1, calculating student s is in the student s by using the single-hot coding embedding layer according to the formula (1)Interactive embedding during answer at all times
In formula (1), A is an embedded matrix to be trained, andm is the embedding dimension, in this example, setIndicates that the student is atInstant answering performanceIs encoded by one heat, and if it isDenotes s at t i No answer or question answering error exists at the knowledge point with the number of j% N at any moment, ifThen the student s is indicatedThe time instants are correctly answered at knowledge points numbered j% N, where the% sign indicates the remainder is taken,obtained by using the formula (2):
step 2.1.2 atAt any moment, it is set that student s answers question at ith timeKnowledge state of time isWill be provided withAndspliced into the ith input vectorThen, respectively inputting three single-layer fully-connected feedforward neural networks and correspondingly passing through sigmoid functions, thereby correspondingly outputting a first forgetting gate during the ith updateFirst input gateAnd an output gateWhen i is 1, let the initial knowledge state of student sIs the set value. In this example, d is set to 64; when the value of i is 1, the reaction condition is shown,
step 2.1.3, input the ith vectorInputting a fourth single-layer fully-connected feedforward neural network, and outputting through a tanh activation functionCandidate memory representation of time of dayThereby using equation (3) to calculateMemory representation in temporal cell information computation layer
In the formula (3), the reaction mixture is,to representThe memory after attenuation in the time memory attenuation layer indicates that when i is equal to 1, the control unit commandsIs the set value. In this embodiment, when i is set to 1,
step 2.2, forgetting part represented by long-short term memory network in continuous time:
step 2.2.1, input the ith vectorInputting the data into a fifth single-layer fully-connected feedforward neural network, and activating a function through softplus, thereby obtaining the presence of the student sForgetting factor within a time period
Step 2.2.2, input the ith vectorRespectively inputting the data into the remaining two single-layer fully-connected feedforward neural networks and correspondingly activating a function through the sigmoid so as to correspondingly obtain a second forgetting gate during the ith updateSecond input gate
Step 2.2.3, the memory attenuation layer is calculated by formula (4)Lower memory decay limit over a period of time
In the formula (4), the reaction mixture is,for the last period of timeThe lower limit of the internal memory attenuation is set to 1 when i is equal toIs the set value. In this embodiment, when i is set to 1,
step 2.2.4, noteThe memory attenuation layer is calculated by formula (5)Memory representation c of time forgotten s (t):
Step 2.3, acquiring a hidden knowledge state:
in formula (6)To obtainMemorial representation of forgotten timeAnd is recorded as the memory representation after attenuationThe knowledge state acquisition layer calculates the position of the student s by using the formula (6)Hidden knowledge state when answering questions at any moment
In equation (6), σ (·) is a sigmoid activation function.
Step 2.4, an answer prediction module:
step 2.4.1, orderTo solve the problemsThe two single-hot-coded embedded layers respectively use the formula (7) and the formula (8) to obtain the themeDifficulty ofAnd degree of distinction
step 2.4.2, the multilayer perceptron layer makes students s inTemporal capability level representationThus, the question when the student s answers the (i + 1) th time is obtained by the formula (9)On the prediction of correct probability of answer
In the formula (9), F (-) is a multilayer perceptron, and in this embodiment, F (-) is a three-layer fully-connected neural network;
step 2.5, after assigning the value of i +1 to i, returning to the step 2.1 to execute in sequence until the historical answer sequence of the students s is completedPrediction of answer correct probability of last answer in (1)
Step 3, constructing cross entropy loss by using the formula (10)And training the knowledge state fitting-answer correctness prediction neural network to obtain a trained answer correctness prediction model for realizing the prediction of student answer correctness. In this example implementation, an Adam optimizer is used:
in the formula (10), the compound represented by the formula (10),for student s at t i The predicted value of the correct probability of answering at the moment,for student s at t i And (3) the true value of the answer correctness at the moment, wherein 0 represents wrong answer and 1 represents right answer.
In specific implementation, the answer prediction module in step 2.4 may also predict the correctness of the answer according to the following process:
step 2.4.1, orderTo solve the problemsUsing the formula (11) and the formula (12) to obtain the titleDifficulty ofAnd degree of distinction
step 2.4.2, the multilayer perceptron layer utilizes formula (13) to obtain the student s is atTemporal capability level representation
Step 2.4.3, the multilayer perceptron layer thus obtains the question when the student s answers at the i +1 th time by using the formula (9)On the prediction of correct probability of answer
In specific implementation, the answer prediction module in step 2.4 may also predict the correctness of the answer according to the following process:
step 2.4.1, orderTo solve the problemsUsing the formula (15) and the formula (16) respectively to obtain the titleDifficulty ofAnd degree of distinction
step 2.4.2, the multilayer perceptron layer utilizes formula (17) to obtain the student s is atTemporal capability level representation
step 2.4.3, setting the question-knowledge point matrix as Q q ={Q mn } M×N M is more than or equal to 1 and less than or equal to M, N is more than or equal to 1 and less than or equal to N, and if the problem numbered M looks at the knowledge point numbered N, Q is written mn If not, Q is noted mn 0; the multilayer perceptron layer obtains the question when the student s answers at the (i + 1) th time by using the formula (18)On the prediction of correct probability of answer
In equation (18), f' (. cndot.) represents a multilayer perceptron, and the sign ° represents multiplication of corresponding positions of the matrix. In this embodiment, f' (. cndot.) is a three-layer fully-connected neural network.
Examples
In order to verify the effectiveness of the method, three public data sets ASSIST (American data set) 12, ASSIST (American data set) 17 and Slepemapy (American data set) which are widely used in the field of education are selected. For these three data sets, their maximum sequence length is set to 100, the student sequences that exceed this length are truncated into several pieces, and the deficiencies are complemented by 0; at the same time, to ensure that each sequence has sufficient data for training, sequences below 5 interactions are removed.
This example uses Accuracy (ACC) and area under ROC curve (AUC) as evaluation criteria.
In the embodiment, five methods are selected for effect comparison with the method of the invention, the selected methods are DKT, DKT _ V, DKT + Forgetting, AKT and HawkesKT respectively, CT-NCM is the method of the invention, CT-NCM _ IRT and CT-NCM _ NCD are two expanding methods of step 2.4 indicated by the right 2 and the right 3, and the experimental results are shown in Table 1.
TABLE 1 Experimental results of student answer prediction on three data sets by the method of the present invention and other comparative algorithms
From table 1, it can be seen that the CT-NCM and its two variants all achieve excellent results on three public large data sets, and the CT-NCM achieves optimal results on three data sets, and experiments prove that the invention has high accuracy and reliability in predicting student answer correctness.
Claims (3)
1. A time-sensitive answer correctness prediction method combining learning and forgetting is characterized by comprising the following steps:
step 1, obtaining student historical answer records and carrying out serialization preprocessing:
set students as a setTopic collectionIs composed ofThe knowledge concept set isWherein the student setIn which there are L students and the question setThere are M questions, knowledge concept setsN knowledge points are present; representing a set of students using sQ represents a question setK represents a knowledge concept setAnd set the topicsThe middle question number is 1, …, M, knowledge concept setThe number of the middle knowledge point is 1, …, N;
representing historical response records of any student s as response sequences Wherein the content of the first and second substances,for the moment of the i-th answer of the student s, an Numbering the questions answered the ith time for student s,answering questions for the ith time of student sThe number of the knowledge concepts under investigation is numbered,indicates that the student s answers the question at the ith timeIn response to the above situation, ifIndicate a right answer, ifDenotes an error, i ═ 1,2, …, n s ,n s The number of times of answering questions for student s;
step 2, constructing a neural network for predicting the correctness of the knowledge state fitting-answering, comprising the following steps: a learning part represented by the continuous long-short term memory network, a forgetting part represented by the continuous long-short term memory network and an answer prediction module;
wherein, the learning part represented by the continuous long-short term memory network comprises: the system comprises a single-hot coding embedded layer, four single-layer fully-connected feedforward neural networks, two activation functions and a cell information calculation layer;
the forgetting part represented by the continuous long-short term memory network comprises: the system comprises three single-layer fully-connected feedforward neural networks, two activation functions, a memory attenuation layer and a knowledge state acquisition layer;
the answer prediction module comprises two independent hot coding embedded layers, a multilayer perceptron layer and an activation function;
step 2.1, a learning part represented by a long-short term memory network in continuous time:
step 2.1.1, the one-hot coded embedding layer utilizes the formula (1) to calculate the position of the student sInteractive embedding during answer at all times
In formula (1), A is an embedded matrix to be trained, anm is the dimension of the embedding,indicates that the student is atInstant answering performanceIs encoded by one heat, andif it isDenotes s at t i No answer or question answering error exists at the knowledge point with the number of j% N at any moment, ifThen the student s is indicatedThe time instants are correct at knowledge points numbered j% N, where the% sign indicates the remainder is taken and is derived from equation (2):
step 2.1.2 atAt any moment, the question of student s answering at the ith time is setKnowledge state of time isWill be provided withAndspliced into the ith input vectorThen respectively inputting three single-layer materialsConnecting with feedforward neural network and correspondingly passing through sigmoid function, thereby correspondingly outputting the first forgetting gate during ith updateFirst input gateAnd an output gateWhen i is 1, let the initial knowledge state of student sIs the set value;
step 2.1.3, input the ith vectorInputting a fourth single-layer fully-connected feedforward neural network, and outputting through a tanh activation functionCandidate memory representation of time of dayThereby using equation (3) to calculateMemory representation in temporal cell information computation layer
In the formula (3), the reaction mixture is,to representThe after-attenuation memory in the time memory attenuation layer indicates that when i is 1, the method makesIs the set value;
step 2.2, forgetting part represented by long-short term memory network in continuous time:
step 2.2.1, input the ith vectorInputting the data into a fifth single-layer full-connection feed-forward neural network, and activating a function through softplus, thereby obtaining the position of the student sForgetting factor within a time period
Step 2.2.2, input the ith vectorRespectively inputting the data into the remaining two single-layer fully-connected feedforward neural networks and correspondingly activating a function through the sigmoid so as to correspondingly obtain a second forgetting gate during the ith updateSecond input gate
Step 2.2.3, the memory attenuation layer is calculated by the formula (4)Lower memory decay limit over a period of time
In the formula (4), the reaction mixture is,for a previous period of timeThe lower limit of memory decay, when i equals 1, letIs the set value;
step 2.2.4, the memory attenuation layer is calculated by the formula (5)Memory representation c of time forgotten s (t):
Step 2.3, acquiring the hidden knowledge state:
in formula (6)To obtainMemorial representation of forgotten timeAnd is recorded as the memory representation after attenuationThe knowledge state acquisition layer calculates the position of the student s by using the formula (6)Hidden knowledge state when answering questions at any moment
In the formula (6), sigma (·) is a sigmoid activation function;
step 2.4, an answer prediction module:
step 2.4.1, orderTo solve the problemsThe two single-hot-coded embedded layers respectively use the formula (7) and the formula (8) to obtain the themeDifficulty ofAnd degree of distinction
step 2.4.2, the multilayer perceptron layer makes students s inTemporal capability level representationThereby obtaining the question of the student s at the i +1 th answer by using the formula (9)On the prediction of correct probability of answer
In the formula (9), F (·) is a multilayer perceptron;
step 2.5, after assigning the value of i +1 to i, returning to the step 2.1 to execute in sequence until the historical answer sequence of the students s is completedPrediction of answer correct probability of last answer in (1)
Step 3, constructing cross entropy loss by using formula (10)And training the neural network for predicting the correctness of the knowledge state fitting-answer to obtain a trained answer correctness prediction model for realizing the prediction of the correctness of the student answer:
in the formula (10), the reaction mixture is,for student s at t i The predicted value of the correct probability of answering at the moment,for student s at t i The true value of the answer correctness at the moment, wherein,the response is shown to be wrong and the answer is wrong,indicating a right to answer.
2. The method for predicting the correctness of jointly learning and forgetting time-sensitive answers of claim 1, wherein the answer predicting module in the step 2.4 is configured to predict the correctness of answers according to the following process:
step 2.4.1, orderTo solve the problemsUsing the formula (11) and the formula (12) to obtain the titleDifficulty of (2)And degree of distinction
step 2.4.2, the multilayer perceptron layer utilizes formula (13) to obtain the student s is atTemporal capability level representation
step 2.4.3, the multilayer perceptron layer thus obtains the question when the student s answers at the i +1 th time by using the formula (9)On the prediction of correct probability of answer
3. The method for predicting the correctness of jointly learning and forgetting time-sensitive answers of claim 1, wherein the answer predicting module in the step 2.4 is configured to predict the correctness of answers according to the following process:
step 2.4.1, orderTo solve the problemsUsing the formula (15) and the formula (16) respectively to obtain the titleDifficulty ofAnd degree of distinction
step 2.4.2, the multilayer perceptron layer utilizes formula (17) to obtain the student s is atTemporal capability level representation
step 2.4.3, setting the question-knowledge point matrix as Q q ={Q mn } M×N M is more than or equal to 1 and less than or equal to M, N is more than or equal to 1 and less than or equal to N, and if the problem numbered M looks at the knowledge point numbered N, Q is written mn If not, Q is noted mn =0;
The multilayer perceptron layer obtains the question when the student s answers at the (i + 1) th time by using the formula (18)On the prediction of correct probability of answer
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210374206.5A CN114997461B (en) | 2022-04-11 | 2022-04-11 | Time-sensitive answer correctness prediction method combining learning and forgetting |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210374206.5A CN114997461B (en) | 2022-04-11 | 2022-04-11 | Time-sensitive answer correctness prediction method combining learning and forgetting |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114997461A true CN114997461A (en) | 2022-09-02 |
CN114997461B CN114997461B (en) | 2024-05-28 |
Family
ID=83023373
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210374206.5A Active CN114997461B (en) | 2022-04-11 | 2022-04-11 | Time-sensitive answer correctness prediction method combining learning and forgetting |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114997461B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116166998A (en) * | 2023-04-25 | 2023-05-26 | 合肥师范学院 | Student performance prediction method combining global and local features |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20110022125A (en) * | 2009-08-27 | 2011-03-07 | 김정민 | Repeat learning method by applying estimation and management of memory maintainment deadline |
JP2017134184A (en) * | 2016-01-26 | 2017-08-03 | 株式会社ウォーカー | Learning support system having continuous evaluation function of learner and teaching material |
CN112800323A (en) * | 2021-01-13 | 2021-05-14 | 中国科学技术大学 | Intelligent teaching system based on deep learning |
CN113033808A (en) * | 2021-03-08 | 2021-06-25 | 西北大学 | Deep embedded knowledge tracking method based on exercise difficulty and student ability |
CN113793239A (en) * | 2021-08-13 | 2021-12-14 | 华南理工大学 | Personalized knowledge tracking method and system fusing learning behavior characteristics |
CN113947262A (en) * | 2021-11-25 | 2022-01-18 | 陕西师范大学 | Knowledge tracking method based on different composition learning fusion learning participation state |
-
2022
- 2022-04-11 CN CN202210374206.5A patent/CN114997461B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20110022125A (en) * | 2009-08-27 | 2011-03-07 | 김정민 | Repeat learning method by applying estimation and management of memory maintainment deadline |
JP2017134184A (en) * | 2016-01-26 | 2017-08-03 | 株式会社ウォーカー | Learning support system having continuous evaluation function of learner and teaching material |
CN112800323A (en) * | 2021-01-13 | 2021-05-14 | 中国科学技术大学 | Intelligent teaching system based on deep learning |
CN113033808A (en) * | 2021-03-08 | 2021-06-25 | 西北大学 | Deep embedded knowledge tracking method based on exercise difficulty and student ability |
CN113793239A (en) * | 2021-08-13 | 2021-12-14 | 华南理工大学 | Personalized knowledge tracking method and system fusing learning behavior characteristics |
CN113947262A (en) * | 2021-11-25 | 2022-01-18 | 陕西师范大学 | Knowledge tracking method based on different composition learning fusion learning participation state |
Non-Patent Citations (1)
Title |
---|
徐墨客;吴文峻;周萱;蒲彦均;: "多知识点知识追踪模型与可视化研究", 电化教育研究, no. 10, 21 September 2018 (2018-09-21) * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116166998A (en) * | 2023-04-25 | 2023-05-26 | 合肥师范学院 | Student performance prediction method combining global and local features |
CN116166998B (en) * | 2023-04-25 | 2023-07-07 | 合肥师范学院 | Student performance prediction method combining global and local features |
Also Published As
Publication number | Publication date |
---|---|
CN114997461B (en) | 2024-05-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110428010B (en) | Knowledge tracking method | |
CN113033808A (en) | Deep embedded knowledge tracking method based on exercise difficulty and student ability | |
CN112800323B (en) | Intelligent teaching system based on deep learning | |
CN112116092A (en) | Interpretable knowledge level tracking method, system and storage medium | |
CN114595923A (en) | Group teaching recommendation system based on deep reinforcement learning | |
Roskos et al. | Learning the art of instructional conversation: The influence of self-assessment on teachers' instructional discourse in a reading clinic | |
CN113610235A (en) | Adaptive learning support device and method based on deep knowledge tracking | |
CN111461442A (en) | Knowledge tracking method and system based on federal learning | |
Pedamallu et al. | A system dynamics model for improving primary education enrollment in a developing country | |
CN114911975A (en) | Knowledge tracking method based on graph attention network | |
Millán et al. | Adaptive Bayesian networks for multilevel student modelling | |
Singh et al. | Implementation and evaluation of intelligence incorporated tutoring system | |
CN115455186A (en) | Learning situation analysis method based on multiple models | |
CN114429212A (en) | Intelligent learning knowledge ability tracking method, electronic device and storage medium | |
CN114997461A (en) | Time-sensitive answer correctness prediction method combining learning and forgetting | |
CN113378581B (en) | Knowledge tracking method and system based on multivariate concept attention model | |
CN113704235A (en) | Depth knowledge tracking model based on self-attention mechanism | |
CN113591988A (en) | Knowledge cognitive structure analysis method, system, computer equipment, medium and terminal | |
CN116402134A (en) | Knowledge tracking method and system based on behavior perception | |
CN114676903A (en) | Online prediction method and system based on time perception and cognitive diagnosis | |
Arafiyah et al. | MONITORING LEARNERS’PERFORMANCE BY MODELING LEARNING PROGRESS USING MACHINE LEARNING | |
CN114117033A (en) | Knowledge tracking method and system | |
Nafiseh et al. | Evaluation based on personalization using optimized firt and mas framework in engineering education in e-learning environment | |
Mu | Gated recurrent unit framework for ideological and political teaching system in colleges | |
Collins et al. | Using Prior Data to Inform Model Parameters in the Predictive Performance Equation. |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant |