CN115062716A - Knowledge tracking method, system and storage medium fusing learning behavior characteristics - Google Patents
Knowledge tracking method, system and storage medium fusing learning behavior characteristics Download PDFInfo
- Publication number
- CN115062716A CN115062716A CN202210746234.5A CN202210746234A CN115062716A CN 115062716 A CN115062716 A CN 115062716A CN 202210746234 A CN202210746234 A CN 202210746234A CN 115062716 A CN115062716 A CN 115062716A
- Authority
- CN
- China
- Prior art keywords
- concept
- matrix
- attention
- vector
- behavior
- 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.)
- Pending
Links
- 230000006399 behavior Effects 0.000 title claims abstract description 111
- 238000000034 method Methods 0.000 title claims abstract description 66
- 239000011159 matrix material Substances 0.000 claims abstract description 146
- 239000013598 vector Substances 0.000 claims abstract description 132
- 230000007246 mechanism Effects 0.000 claims abstract description 46
- 238000004364 calculation method Methods 0.000 claims abstract description 28
- 230000008569 process Effects 0.000 claims abstract description 26
- 230000006870 function Effects 0.000 claims description 23
- 230000003993 interaction Effects 0.000 claims description 13
- 230000004913 activation Effects 0.000 claims description 12
- 238000013528 artificial neural network Methods 0.000 claims description 9
- 230000004927 fusion Effects 0.000 claims description 8
- 230000003542 behavioural effect Effects 0.000 claims description 5
- 238000004590 computer program Methods 0.000 claims description 5
- 238000005065 mining Methods 0.000 claims description 5
- 230000004044 response Effects 0.000 claims description 5
- 230000009467 reduction Effects 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 230000017105 transposition Effects 0.000 claims description 2
- 238000013507 mapping Methods 0.000 claims 1
- 230000015654 memory Effects 0.000 description 12
- 238000012549 training Methods 0.000 description 10
- 230000003287 optical effect Effects 0.000 description 4
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Tourism & Hospitality (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Biophysics (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- General Business, Economics & Management (AREA)
- Educational Technology (AREA)
- Educational Administration (AREA)
- Life Sciences & Earth Sciences (AREA)
- Economics (AREA)
- Biomedical Technology (AREA)
- Marketing (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application discloses a knowledge tracking method, a knowledge tracking system and a storage medium integrating learning behavior characteristics. The method comprises the following steps: calculating the similarity of concepts in a concept matrix by adopting a self-attention mechanism, wherein the calculation result is represented by a concept attention matrix, the similarity between the concepts contained in the problem to be predicted is calculated according to the concept attention matrix, and the calculation result is represented by the correlation weight; calculating the similarity between different concept grasping states in the concept grasping matrix by adopting a self-attention mechanism, and expressing a calculation result by using a state attention moment matrix; adopting a self-attention mechanism to mine the connection among different characteristics in the characteristic vector describing the behavior of the user in the learning process, and outputting a behavior attention vector; updating a concept master matrix by using a state attention matrix and a behavior attention vector; and predicting the answer condition of the user according to the concept mastering matrix. The invention improves the prediction precision by adopting self attention to discover the similarity between concepts.
Description
Technical Field
The present application relates to the field of knowledge tracking technologies, and in particular, to a knowledge tracking method, system and storage medium fusing learning behavior characteristics.
Background
At present, the Knowledge tracking method mainly comes from three types of models such as Bayesian Knowledge Tracking (BKT), Deep Knowledge Tracking (DKT) and Dynamic Key Value Memory network (DKMN). Bayesian knowledge tracking tracks the knowledge state of a student with the probability from an unlearned state to a learned state, the probability of forgetting a previously learned concept, the probability of not mastering a concept but guessing correctly, the probability of mastering a concept but making a mistake, and the probability values are updated in the student's problem interaction data by Hidden Markov Models (HMMs). The BKT model not only assumes that students do not forget about knowledge, i.e., that the probability of forgetting a previously learned concept is 0, but also models each concept separately. But actually, the forgetting behavior is accompanied by the beginning and the end of the learning process; the concepts are not independent from each other, but are hierarchical and interrelated. Deep knowledge tracking uses a Recurrent Neural Network (RNN) or Long-Short Term Memory Network (LSTM) to track the knowledge state of a student, represented by a hidden state vector. The dynamic key value memory network stores concepts by a key matrix and stores the mastery states of the concepts by students by a value matrix, and the DKVMN model can accurately indicate the specific knowledge states of the students on each concept by the two matrices. Although the three models achieve good results in the aspect of predicting the future answer condition of the student, the data input into the models is only exercise interaction data of the student, and the parameter updating process does not model forgetting behaviors.
However, these methods still have drawbacks. First, the impact of inter-concept similarity on the learning process is ignored. Second, the knowledge state of the student is tracked by using exercise interaction data of the student, and other abundant learning behavior characteristics of the student are ignored. Thirdly, no modeling forgetting behavior is performed in the learning process of the modeling students.
Disclosure of Invention
In view of at least one of the drawbacks or needs for improvement of the prior art, the present invention provides a knowledge tracking method, system and storage medium that incorporate learning behavior features to improve prediction accuracy by exploiting similarities between concepts from attention.
To achieve the above object, according to a first aspect of the present invention, there is provided a knowledge tracking method fusing learning behavior characteristics, including:
calculating the similarity of concepts in a concept matrix by adopting a self-attention mechanism, wherein the concept matrix is used for representing the related concepts of the questions in the course, the calculation result is represented by a concept attention matrix, the similarity of the concepts contained in the questions to be predicted is calculated according to the concept attention matrix, and the calculation result is represented by associated weight;
calculating the similarity between different concept mastering states in a concept mastering matrix by adopting a self-attention mechanism, wherein the concept mastering matrix is used for expressing the mastering state of the learner on the concept, and the calculation result is expressed by a state attention moment array;
adopting a self-attention mechanism to mine the connection among different characteristics in the characteristic vector describing the behavior of the user in the learning process, and outputting a behavior attention vector;
updating a concept master matrix by using a state attention matrix and a behavior attention vector;
and multiplying the associated weight by the concept mastery matrix to obtain a vector of the student concept mastery state, and inputting the vector of the student concept mastery state and the embedded vector of the exercise to be predicted into a feed-forward neural network to predict the answer condition of the user.
Further, the feature vector describing the behavior of the user in the learning process is recorded asThe method comprises exercise interaction data X and learning behavior characteristic data, wherein the exercise interaction data X is used for describing wrong answer conditions of a user at a plurality of moments before a prediction time, and the learning behavior characteristic data comprises first-class behavior characteristic dataAnd second type of behavior feature dataComprises a plurality of predefined behavior characteristic data directly obtained from the collected data of the user behavior,the system comprises a plurality of predefined behavior characteristic data which are calculated from the collected data of the user behavior and are related to the forgetting behavior of the user.
Further, the air conditioner is provided with a fan,in order to form a row vector with 3 dimensionality by three characteristics of answering times, first response time and request prompt times,the number of times of repeatedly learning the same concept, the learning time interval and the learning time interval of the same concept are combined into a row vector with a dimension of 3.
Further, the feature vectors of the behaviors in the learning process of the user are recorded asDigging by adopting self-attention mechanismThe links between the different features include:
using feed-forward self-encoder pairs with self-attention mechanismReducing the dimension to obtain a vector h after dimension reduction t Wherein the feedforward self-encoder comprises an encoder and a decoder, the encoder comprises an input layer, a self-attention layer, a full-link layer and a hidden layer, the decoder comprises a hidden layer, a full-link layer, a self-attention layer and an output layer, and the encoder and the decoder are arranged in parallelThe parts are distributed in a mirror image.
Digging h by self-attention mechanism t The relationship of internal features and the output is a behavior attention vector marked as o t 。
Further, the air conditioner is characterized in that,
softmax () represents a Softmax function and T represents transposition.
Further, the state attention moment array is denoted as C t Let the behavior attention vector be o t Using the state attention matrix C t And a behavioral attention vector o t To model the learning behavior and forgetting behavior of students and obtain a learning vector u t And a forgetting vector f t Using the learning vector u t And a forgetting vector f t Concept mastery matrix for updating t-1 timeObtaining a concept mastery matrix at time t
Further, C is t And o t After splicing, inputting the data to a full connection layer with a Tanh activation function, wherein the weight matrix of the full connection layer is w 1 Offset vector is b 1 To obtain a dimension d v Vector a of t ,
Vector a is divided into a layer of full connection layer with Tanh activation function t Conversion to learning vector u t Modeling learning behavior to obtain learning vector u t The weight matrix of the full connection layer is w u Offset vector is b u ,
Will vector a t As input of keys and values in the attention mechanism, vectorsProjection to d v After dimension is taken as the input of query in the attention mechanism, the output result of the attention mechanism is input to a full-connection layer with a Sigmoid activation function to obtain a forgetting vector f t ,
Wherein, W b Is a vectorIs projected to d v Weight term of the fully connected layer of dimension, b b Is a vectorProjection to d v Bias term, W, of the fully-connected layer of the dimension f Weight terms for the fully-connected layer of the function activated for Sigmoid, b f Softmax () represents the Softmax function, T represents the transpose,
the update formula of the concept mastery matrix is as follows:
representing concept mastery matricesThe (c) th column of (a),representing concept mastery matricesI column of (1), w t (i) Represents the associated weight w t I ranges from 1 to N.
Further, the concept matrix at the time t-1 is recorded as the concept matrixIs d k X N dimensional matrix, d k Is a matrixDimension of each column, N is the number of concepts, calculated using a self-attention mechanismSimilarity of concept in (1), calculating result using concept attention matrix G t It is shown that,
will exercise question q t Is noted as k t According to k t And matrix G t Calculation problem q t Similarity between concepts involved, the calculation result being given an association weight w t It is shown that,
w t =Softmax(k t ×G t )
let the concept grasping matrix at time t-1 be recorded asIs d v X N dimensional matrix, d v Is a matrixDimension of each column, and self-attention mechanism calculationThe middle concept grasps the similarity between states, and the state attention matrix C is used for calculating the result t It is shown that,
where Softmax () represents the Softmax function and T represents transpose.
According to a second aspect of the present invention, there is also provided a knowledge tracking system incorporating learning behavior features, comprising:
the concept attention module is used for calculating the similarity of concepts in a concept matrix by adopting a self-attention mechanism, the concept matrix is used for representing the related concepts of the questions in the course, the calculation result is represented by a concept attention matrix, the similarity between the concepts in the questions to be predicted is calculated according to the concept attention matrix, and the calculation result is represented by the associated weight;
the state attention module is used for calculating the similarity between different concept mastering states in a concept mastering matrix by adopting a self-attention mechanism, the concept mastering matrix is used for expressing the mastering state of the learner on the concept, and the calculation result is expressed by a state attention moment array;
the behavior attention module is used for mining the relation among different features in the feature vector describing the behavior of the user in the learning process by adopting a self-attention mechanism and outputting the behavior attention vector;
the updating module is used for updating the concept mastering matrix by utilizing the state attention matrix and the behavior attention vector;
and the prediction module is used for multiplying the association weight and the concept mastering matrix to obtain a vector of the student concept mastering state, inputting the vector of the student concept mastering state and the embedded vector of the exercise to be predicted into the feedforward neural network, and predicting the user answering situation.
According to a third aspect of the present invention, there is also provided a storage medium storing a computer program executable by a processor, the computer program, when run on the processor, causing the processor to perform the steps of any of the methods described above.
In general, compared with the prior art, the above technical solution contemplated by the present invention can achieve the following beneficial effects:
(1) the invention considers the influence of the similarity between concepts on the learning process of the students when tracking the knowledge state of the students. Existing models either do not take this into account or use other methods, the present invention can have better interpretability with a self-attention mechanism to explore similarities between concepts.
(2) According to the invention, more information influencing the learning process of the student is input into the model when the knowledge state of the student is tracked, most of the existing models only use exercise interaction data to track the knowledge state of the student, but the information contained in the data is limited, so that the knowledge state of the student cannot be accurately tracked. The invention combines the actual learning process, not only inputs exercise interaction data to the model, but also inputs abundant learning behavior characteristics to the model, thereby tracking the knowledge state of students more accurately.
(3) The invention models the forgetting behavior when tracking the knowledge state of the student, most of the existing models do not model the forgetting behavior, but the forgetting behavior is accompanied with the beginning and the end of the learning process, and the neglect of the modeling forgetting behavior is obviously unreasonable. The invention combines the reality, can change the characteristics of the attention point of the model to the input data by using the attention mechanism, and leads the model to pay more attention to the characteristics related to the forgetting behavior when modeling the forgetting behavior, thereby achieving the purpose of modeling the forgetting behavior and having better interpretability. .
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.
Fig. 1 is a schematic diagram of a knowledge tracking system incorporating learning behavior features according to an embodiment of the present application;
fig. 2 is a schematic flowchart of a knowledge tracking method incorporating learning behavior features according to an embodiment of the present application;
fig. 3 is a schematic diagram of a feed-forward self-encoder with a self-attention mechanism according to an embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
The terms "first," "second," and the like in the description and claims of the present application and in the above-described drawings are used for distinguishing between different objects and not for describing a particular order. Furthermore, the terms "include" and "have," as well as any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, article, or apparatus that comprises a list of steps or modules is not limited to the listed steps or modules but may alternatively include other steps or modules not listed or inherent to such process, method, article, or apparatus.
As shown in fig. 1, a knowledge tracking system fusing learning behavior features according to an embodiment of the present invention includes:
the concept attention module is used for calculating the similarity of concepts in a concept matrix by adopting a self-attention mechanism, the concept matrix is used for representing the related concepts of the questions in the course, the calculation result is represented by a concept attention matrix, the similarity between the concepts in the questions to be predicted is calculated according to the concept attention matrix, and the calculation result is represented by the associated weight;
the state attention module is used for calculating the similarity between different concept mastering states in a concept mastering matrix by adopting a self-attention mechanism, the concept mastering matrix is used for expressing the mastering state of the learner on the concept, and the calculation result is expressed by a state attention moment array;
the behavior attention module is used for mining the relation among different features in the feature vector describing the behavior of the user in the learning process by adopting a self-attention mechanism and outputting the behavior attention vector;
the updating module is used for updating the concept mastering matrix by utilizing the state attention matrix and the behavior attention vector;
and the prediction module is used for multiplying the association weight and the concept mastering matrix to obtain a vector of the student concept mastering state, inputting the vector of the student concept mastering state and the embedded vector of the exercise to be predicted into a feedforward neural network (full-connection neural network) and predicting the user answering condition.
As shown in fig. 2, the knowledge tracking method with fusion learning behavior features of the embodiment of the present invention includes the steps of:
s101, calculating similarity of concepts in a concept matrix by adopting a self-attention mechanism, wherein the concept matrix is used for representing related concepts of the topics in the course, the calculation result is represented by a concept attention matrix, the similarity of the concepts contained in the problem to be predicted is calculated according to the concept attention matrix, and the calculation result is represented by associated weight.
The concept matrix is used for representing the related concepts of the subjects in the course, and the concept matrix at the time of t-1 is recorded asAnd representing the vector representation of all the related concepts of the subjects in the course obtained by the model learning at the time t-1. Before the model training, a concept matrix is initialized, and the vector representation of all the topic related concepts in the course is continuously updated by utilizing a back propagation algorithm in the model training process.
Computing matricesSimilarity between the intermediate concepts. Matrix to store conceptsAs input to a self-attention mechanism in the module, wherein d k Is a matrixThe dimension of each column, N, is the number of different concepts in the course. By means of matricesCalculating the similarity between concepts and the concept attention matrix G is used for the calculation result t Show, then according to the problem q t And matrix G t Get the associated weight w t 。
The method comprises the following steps: initializing matrices of storage concepts prior to model trainingMatrix arrayAnd continuously updating in the model training process. Using the signal as the input of the self-attention mechanism to obtain an output concept attention matrix G t . Will exercise question q t Converting into one-hot code and embedding into matrixMultiplying to obtain a dimension d k Problem embedding vector k t 。k t And G t The multiplication is converted into the associated weight w by a Softmax function t To express exercises q t Similarity between the concepts involved:
w t =Softmax(k t ×G t )
s102, calculating the similarity between different concept mastering states in a concept mastering matrix by adopting a self-attention mechanism, wherein the concept mastering matrix is used for representing the concept mastering state of the learner, and the calculation result is represented by a state attention moment array.
The concept master matrix is used for representing the master state of the learner on the concept, and the concept master matrix at the time t-1 is recorded as
Computing matricesThe concept of "Zhong" grasps the similarity between states. Matrix of mastered states of stored conceptsAs an input to a self-attention mechanism in the module, wherein d v Is a matrixThe dimension of each column, N, is the number of concepts in the course. By means of matricesCalculating similarity between user grasp states of different concepts, such as the grasp state of the user with respect to concept 1 and the grasp state of the user with respect to concept 2, and calculating a state attention matrix C for the result t And (4) showing.
The method comprises the following steps: matrix for initializing concept mastering state of storage before model trainingMatrix arrayAnd continuously updating in the model training process. It is used as self in moduleThe input of the attention mechanism is used for obtaining an output state attention matrix C t :
S103, mining the relation among different characteristics in the characteristic vector describing the behavior of the user in the learning process by adopting a self-attention mechanism, and outputting a behavior attention vector.
Receiving rich information generated in the course of student learning, inputting vectorThe self-attention mechanism enables the model to capture vectors, and comprises exercise interaction information and learning behavior characteristic informationCorrelation between internal features.
The exercise interaction data X consists of the answer conditions of the students at a plurality of moments, namely X ═ X 1 ,x 2 ,...,x t ),x t =(q t ,r t ) Is the answer condition of the student at the time t, wherein q t Questions, r, representing responses made by students t The student answers the exercise with 1, and the student answers the exercise with a mistake with 0. The learning behavior feature data represents some behavior actions in the learning process of the students, such as the number of times of answering attempts, the response time of answering for the first time, the number of times of prompting requests and the like.
Further, the learning behavior feature data includes a first type of behavior feature dataAnd second type of behavior feature dataComprises a plurality of predefined behavior characteristic data directly obtained from the collected data of user behavior,comprises a plurality of predefined behavior characteristic data which are calculated from the collected data of the user behavior and reflect the forgetting behavior of the user,the method is characterized in that key behavior characteristic data reflecting user forgetting behaviors, which types of behavior characteristic data can be defined in advance and calculated from user behavior collected dataThe method of (3) can also be predefined.
Further, three characteristics of number of times of trying to answer, first response time and number of times of requesting and prompting are combined into a row vector with dimension 3The same number of repeated learning of the same concept, the learning time interval and the learning time interval of the same concept are combined into a line vector with the dimension of 3
Interacting the exercises with data x t The one-hot coding is converted into the one-hot coding,to solve x t Problem of sparseness, by combining it with an embedding matrixMultiplying to obtain a dimension d v (Vector)Exercise interaction data x representing students at time t t . Will vector(Vector)Vector e t Composition matrixMatrix arrayThe first line of (1) represents the student's problem interaction data x at time t t Each of the remaining rows represents a learning behavior feature, e.g., the second row represents the number of attempts and the values in that row are the same. From this it can be seen thatThere are a number of repetitive features, so embodiments of the present invention preferably employ a Feed-Forward auto-encoder With Self-Attention mechanism (FFA-SA) pair matrixAnd (5) reducing the dimension, and mining the relation of the internal features of the reduced vector by adopting a self-attention mechanism to output the vector as a behavior attention vector.
Fig. 3 shows a specific structure of the FFA-SA. Unlike conventional feed forward self encoders, the embodiments of the present invention introduce a self attention layer.
The FFA-SA is divided into an encoder and a decoder, wherein the encoder consists of an input layer, a self-attention layer, a full connection layer and a hidden layer; the decoder is composed of a hidden layer, a full-link layer, a self-attention layer and an output layer. The two portions are arranged in mirror image.
Matrix in an encoderAs input, a self-attention mechanism is used to discover the connections between features within the input data:
matrix S of outputs e The method comprises the following steps of inputting the relation between the characteristic interiors into a full connection layer, and performing first dimension reduction and coding:
wherein the weight matrix of the full connection layerOffset vectorF is to be e Inputting the data into a hidden layer to obtain a final dimension reduction and coding result:
Vector h in decoder t As input, it is used as a full connection layer to carry out the first time of dimension raising and decoding:
wherein the weight matrix of the full connection layerOffset vectorF is to be d Input to the self-attention layer for discovering associations inside featuresComprises the following steps:
matrix S of outputs d And (3) containing the relation among the characteristic interiors, and performing final dimension increasing and encoding on the input and output layers:
matrix arrayAnd matrixThe closer the vector is, the better the self-encoder FFA-SA training is represented, and the output vector h of the hidden state is in the continuous training process of the FFA-SA t And matrixThe closer the expressed property information. Digging vector h with self-attention mechanism t The association of internal features, the output of which is the behavioral attention vector o t ,o t Not only contains the exercise mutual information of students but also contains the learning behavior characteristic information of students:
and S104, updating the concept grasping matrix by using the state attention matrix and the behavior attention vector.
And initializing a concept mastering matrix before model training, continuously updating the concept mastering matrix in the model training process, and updating the concept mastering matrix at the previous moment in the adjacent moments to obtain the concept mastering matrix at the next moment in the adjacent moments. For example, the concept understanding matrix at time t-2 is updated to obtain the concept understanding matrix at time t-1, and the concept understanding matrix at time t-1 is updated to obtain the concept understanding matrix at time t.
Only updateTo obtainFor illustration, the update principle at other times is the same, and is not described herein again.
Further, a state attention matrix C is utilized t And a behavioral attention vector o t To model the learning behavior and forgetting behavior of students and obtain a learning vector u t And a forgetting vector f t Using the learning vector u t And a forgetting vector f t Concept mastery matrix for updating t-1 timeObtaining a concept mastery matrix at time t
The method comprises the following steps: output result C of state attention module and behavior attention module t And o t Splicing and inputting the data into a full connection layer with a Tanh activation function, wherein the weight matrix of the full connection layer is w 1 Offset vector is b 1 To obtain a dimension d v Vector a of t Is a vector o t And matrix C t Is shown in summary.
Modeling learning behaviors: vector a is divided into a layer of full connection layer with Tanh activation function t Conversion to learning vector u t The weight matrix of the full connection layer is w u Offset vector is b u ,
Modeling forgetting behaviors: will vector a t As input of keys and values in the attention mechanism, vectorsProjection to d v After dimension is taken as the input of query in the attention mechanism, the output result of the attention mechanism is input to a full-connection layer with a Sigmoid activation function to obtain a forgetting vector f t :
Wherein, W b Is a vectorIs projected to d v Weight term of the fully connected layer of dimension, b b Is a vectorProjection to d v Bias term, W, of the fully-connected layer of the dimension f Weight terms for the fully-connected layer of the Sigmoid activation function, b f For the offset terms of the fully-connected layer of the Sigmoid-enabled function, Softmax () represents the Softmax function and T represents the transpose.
wherein the content of the first and second substances,representing concept mastery matricesThe (c) th column of (a),representing concept mastery matricesI column of (1), w t (i) Represents the associated weight w t I ranges from 1 to N, i is updated concept by concept.
And S105, multiplying the associated weight by the concept grasping matrix to obtain a vector of the student concept grasping state, inputting the vector of the student concept grasping state and the embedded vector of the exercise to be predicted into a feedforward neural network (full-connection neural network), and predicting the answer condition of the user.
The prediction module is based on the matrixTo predict the answer situation of the student at the moment t.
The method comprises the following steps: associate a weight w t And matrixMultiplying to obtain a vector n t The output of the concept mastery state of the student is as follows:
considering that there will be some difference between the problems, such as different difficulty coefficients, the vector n is divided t And problem embedding vector k t Splicing is carried out, and the vector obtained in this way contains the studentsThe concept grasping state comprises exercise information, the vector is input into a full-connection layer with Tanh activation function, and the weight matrix of the full-connection layer is w 2 Offset vector is b 2 To obtain a vector i t :
Finally, an output layer with a Sigmoid activation function is utilized, and the weight matrix of the output layer is w 3 Offset vector is b 3 I is to t As input, for predicting student's problem q t The performance condition of (2):
simulation tests are carried out in different data sets, the knowledge tracking method fusing learning behavior characteristics is marked as ILB-KT and compared with SAKT, DKT and DKEMN, and as shown in Table 1, the prediction accuracy of the method disclosed by the invention is higher than that of the prior art.
TABLE 1
The implementation principle of the system is the same as that of the method, and details are not described here.
The present application also provides a storage medium storing a computer program executable by a processor, the computer program, when executed on the processor, causing the processor to perform any of the above-described steps of the knowledge tracking method with fusion of learning behavior features. The computer-readable storage medium may include, but is not limited to, any type of disk including floppy disks, optical disks, DVD, CD-ROMs, microdrive, and magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic or optical cards, nanosystems (including molecular memory ICs), or any type of media or device suitable for storing instructions and/or data.
It should be noted that for simplicity of description, the above-mentioned embodiments of the method are described as a series of acts, but those skilled in the art should understand that the present application is not limited by the described order of acts, as some steps may be performed in other orders or simultaneously according to the present application. Further, those skilled in the art should also appreciate that the embodiments described in the specification are preferred embodiments and that the acts and modules referred to are not necessarily required in this application.
In the foregoing embodiments, the descriptions of the respective embodiments have respective emphasis, and for parts that are not described in detail in a certain embodiment, reference may be made to related descriptions of other embodiments.
In the several embodiments provided in the present application, it should be understood that the disclosed system may be implemented in other ways. For example, the above-described system embodiments are merely illustrative, and for example, the division of the modules is merely a logical division, and in actual implementation, there may be other divisions, for example, multiple modules or components may be combined or integrated into another system, or some features may be omitted, or not implemented. In addition, the shown or discussed coupling or direct coupling or communication connection between each other may be through some service interfaces, indirect coupling or communication connection of systems or modules, and may be in electrical or other forms.
The modules described as separate parts may or may not be physically separate, and parts displayed as modules may or may not be physical modules, may be located in one place, or may be distributed on a plurality of network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment.
In addition, functional modules in the embodiments of the present application may be integrated into one processing module, or each of the modules may exist alone physically, or two or more modules are integrated into one module. The integrated module can be realized in a hardware mode, and can also be realized in a software functional module mode.
The integrated module, if implemented in the form of a software functional module and sold or used as a stand-alone product, may be stored in a computer readable memory. Based on such understanding, the technical solution of the present application may be substantially implemented or a part of or all or part of the technical solution contributing to the prior art may be embodied in the form of a software product stored in a memory, and including several instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method described in the embodiments of the present application. And the aforementioned memory comprises: various media capable of storing program codes, such as a usb disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a removable hard disk, a magnetic disk, or an optical disk.
Those skilled in the art will appreciate that all or part of the steps in the methods of the above embodiments may be implemented by a program, which is stored in a computer-readable memory, and the memory may include: flash disks, Read-Only memories (ROMs), Random Access Memories (RAMs), magnetic or optical disks, and the like.
The above description is only an exemplary embodiment of the present disclosure, and the scope of the present disclosure should not be limited thereby. That is, all equivalent changes and modifications made in accordance with the teachings of the present disclosure are intended to be included within the scope of the present disclosure. Embodiments of the present disclosure will be readily apparent to those skilled in the art from consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
It will be understood by those skilled in the art that the foregoing is only an exemplary embodiment of the present invention, and is not intended to limit the invention to the particular forms disclosed, since various modifications, substitutions and improvements within the spirit and scope of the invention are possible and within the scope of the appended claims.
Claims (10)
1. A knowledge tracking method fusing learning behavior characteristics is characterized by comprising the following steps:
calculating the similarity of concepts in a concept matrix by adopting a self-attention mechanism, wherein the concept matrix is used for representing the related concepts of the questions in the course, the calculation result is represented by a concept attention matrix, the similarity of the concepts contained in the questions to be predicted is calculated according to the concept attention matrix, and the calculation result is represented by associated weight;
calculating the similarity between different concept mastering states in a concept mastering matrix by adopting a self-attention mechanism, wherein the concept mastering matrix is used for expressing the mastering state of the learner on the concept, and the calculation result is expressed by a state attention moment array;
adopting a self-attention mechanism to mine the connection among different characteristics in the characteristic vector describing the behavior of the user in the learning process, and outputting a behavior attention vector;
updating a concept master matrix by using a state attention matrix and a behavior attention vector;
and multiplying the association weight by the concept mastering matrix to obtain a vector of the student concept mastering state, and inputting the vector of the student concept mastering state and the embedded vector of the exercise to be predicted into a feed-forward neural network to predict the user answer condition.
2. The method for knowledge tracking with fusion learning behavior features as claimed in claim 1, wherein the feature vector describing the behavior of the user in the learning process is recorded as The method comprises exercise interaction data X and learning behavior characteristic data, wherein the exercise interaction data X is used for describing wrong answer conditions of a user at a plurality of moments before a prediction time, and the learning behavior characteristic data comprises first-class behavior characteristic dataAnd second type of behavior feature data Comprises a plurality of predefined behavior characteristic data directly obtained from the collected data of user behavior,the system comprises a plurality of predefined behavior characteristic data which are calculated from the collected data of the user behavior and are related to the forgetting behavior of the user.
3. The knowledge tracking method with fusion learning behavior feature of claim 2,in order to form a row vector with 3 dimensionality by three characteristics of answering times, first response time and request prompt times,the number of times of repeatedly learning the same concept, the learning time interval and the learning time interval of the same concept are combined into a row vector with a dimension of 3.
4. The method for knowledge tracking with fusion learning behavior features as claimed in claim 1, wherein the feature vector of the behavior is recorded asDigging by adopting self-attention mechanismThe links between the different features include:
using feed-forward self-encoder pairs with self-attention mechanismReducing the dimension to obtain a vector h after dimension reduction t The feedforward self-encoder comprises an encoder and a decoder, the encoder comprises an input layer, a self-attention layer, a full-connection layer and a hidden layer, the decoder comprises a hidden layer, a full-connection layer, a self-attention layer and an output layer, and the encoder and the decoder are distributed in a mirror image mode.
Digging h by self-attention mechanism t The relationship of internal features and the output is a behavior attention vector marked as o t 。
6. The converged learning line of claim 1 or 2A method of tracking knowledge of a feature, characterized by mapping a state attention moment array as C t Denote the behavioral attention vector as o t Using the state attention matrix C t And a behavioral attention vector o t To model the learning behavior and forgetting behavior of students and obtain a learning vector u t And a forgetting vector f t Using the learning vector u t And a forgetting vector f t Concept mastery matrix for updating t-1 timeObtaining a concept mastery matrix at time t
7. The method of knowledge tracking with fusion learning behavior features of claim 6, characterized in that C is used t And o t After splicing, inputting the data into a full connection layer with a Tanh activation function, wherein the weight matrix of the full connection layer is w 1 Offset vector is b 1 To obtain a dimension d v Vector a of t ,
Vector a is divided by a full connection layer with Tanh activation function t Conversion to learning vector u t Modeling learning behavior to obtain learning vector u t The weight matrix of the full connection layer is w u Offset vector is b u ,
Will vector a t As inputs for keys and values in the attention mechanism, vectorsProjection to d v After dimension is taken as the input of query in the attention mechanism, the output result of the attention mechanism is input to a full-connection layer with a Sigmoid activation function to obtain a forgetting vector f t ,
Wherein, W b Is a vectorProjection to d v Weight term of the fully connected layer of dimension, b b Is a vectorProjection to d v Bias term, W, of the fully-connected layer of the dimension f Weight terms for the fully-connected layer of the function activated for Sigmoid, b f Softmax () represents the Softmax function, T represents the transpose,
the update formula of the concept mastery matrix is as follows:
8. The method for tracking knowledge of fusion learning behavior characteristics according to claim 1, wherein the concept matrix at the time t-1 is recorded as a concept matrix Is d k X N dimensional matrix, d k Is a matrixDimension of each column, N is the number of concepts, calculated using a self-attention mechanismSimilarity of concept in (1), calculating result using concept attention matrix G t It is shown that,
will exercise question q t Is noted as k t According to k t And matrix G t Calculation problem q t Similarity between concepts involved, the calculation result being given an association weight w t It is shown that,
w t =Softmax(k t ×G t )
let the concept grasping matrix at time t-1 be recorded as Is d v X N dimensional matrix, d v Is a matrixDimension of each column, and self-attention mechanism calculationThe middle concept grasps the similarity between states, and the state attention matrix C is used for calculating the result t It is shown that,
where Softmax () represents the Softmax function and T represents transpose.
9. A knowledge tracking system that fuses learning behavior features, comprising:
the concept attention module is used for calculating the similarity of concepts in a concept matrix by adopting a self-attention mechanism, the concept matrix is used for representing the related concepts of the questions in the course, the calculation result is represented by a concept attention matrix, the similarity between the concepts in the questions to be predicted is calculated according to the concept attention matrix, and the calculation result is represented by the associated weight;
the state attention module is used for calculating the similarity between different concept mastering states in a concept mastering matrix by adopting a self-attention mechanism, the concept mastering matrix is used for expressing the mastering state of the learner on the concept, and the calculation result is expressed by a state attention moment array;
the behavior attention module is used for mining the relation among different features in the feature vector describing the behavior of the user in the learning process by adopting a self-attention mechanism and outputting the behavior attention vector;
the updating module is used for updating the concept mastering matrix by utilizing the state attention matrix and the behavior attention vector;
and the prediction module is used for multiplying the association weight and the concept mastering matrix to obtain a vector of the student concept mastering state, inputting the vector of the student concept mastering state and the embedded vector of the exercise to be predicted into the feedforward neural network, and predicting the user answering situation.
10. A storage medium, characterized in that it stores a computer program which, when run on a processor, causes the processor to carry out the steps of the method according to any one of claims 1 to 8.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210746234.5A CN115062716A (en) | 2022-06-29 | 2022-06-29 | Knowledge tracking method, system and storage medium fusing learning behavior characteristics |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210746234.5A CN115062716A (en) | 2022-06-29 | 2022-06-29 | Knowledge tracking method, system and storage medium fusing learning behavior characteristics |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115062716A true CN115062716A (en) | 2022-09-16 |
Family
ID=83203830
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210746234.5A Pending CN115062716A (en) | 2022-06-29 | 2022-06-29 | Knowledge tracking method, system and storage medium fusing learning behavior characteristics |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115062716A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115545160A (en) * | 2022-09-26 | 2022-12-30 | 长江大学 | Knowledge tracking method and system based on multi-learning behavior cooperation |
CN116127048A (en) * | 2023-04-04 | 2023-05-16 | 江西师范大学 | Sequential self-attention knowledge tracking model integrating exercises and learning behavior characterization |
-
2022
- 2022-06-29 CN CN202210746234.5A patent/CN115062716A/en active Pending
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115545160A (en) * | 2022-09-26 | 2022-12-30 | 长江大学 | Knowledge tracking method and system based on multi-learning behavior cooperation |
CN116127048A (en) * | 2023-04-04 | 2023-05-16 | 江西师范大学 | Sequential self-attention knowledge tracking model integrating exercises and learning behavior characterization |
CN116127048B (en) * | 2023-04-04 | 2023-06-27 | 江西师范大学 | Sequential self-attention knowledge tracking model integrating exercises and learning behavior characterization |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110379225B (en) | System and method for interactive language acquisition | |
Trask | Grokking deep learning | |
Li et al. | Dialogue learning with human-in-the-loop | |
CN110428010B (en) | Knowledge tracking method | |
CN115062716A (en) | Knowledge tracking method, system and storage medium fusing learning behavior characteristics | |
CN111582694A (en) | Learning evaluation method and device | |
CN113610235B (en) | Adaptive learning support device and method based on depth knowledge tracking | |
CN112115246A (en) | Content recommendation method and device based on conversation, computer equipment and storage medium | |
CN115545160B (en) | Knowledge tracking method and system for multi-learning behavior collaboration | |
CN114818691A (en) | Article content evaluation method, device, equipment and medium | |
CN114299349B (en) | Crowdsourcing image learning method based on multi-expert system and knowledge distillation | |
CN112069827B (en) | Data-to-text generation method based on fine-grained subject modeling | |
CN116136870A (en) | Intelligent social conversation method and conversation system based on enhanced entity representation | |
CN115544158A (en) | Multi-knowledge-point dynamic knowledge tracking method applied to intelligent education system | |
CN115329959A (en) | Learning target recommendation method based on double-flow knowledge embedded network | |
CN114861754A (en) | Knowledge tracking method and system based on external attention mechanism | |
CN114385801A (en) | Knowledge tracking method and system based on hierarchical refinement LSTM network | |
CN115238169A (en) | Mu course interpretable recommendation method, terminal device and storage medium | |
Ferlitsch | Deep Learning Patterns and Practices | |
CN112949628A (en) | Track data enhancement and track identification method based on embedding-mixing | |
LU503730B1 (en) | Sequence recommendation method and system based on the coupling relationship between item attributes and time sequence patterns | |
CN114971066A (en) | Knowledge tracking method and system integrating forgetting factor and learning ability | |
Yue et al. | Augmenting interpretable knowledge tracing by ability attribute and attention mechanism | |
CN114943276B (en) | Depth knowledge tracking method based on tree-type attention mechanism | |
Pineda | The mode of computing |
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 |