CN116823027A - Intelligent student capability assessment method based on associated skill knowledge - Google Patents
Intelligent student capability assessment method based on associated skill knowledge Download PDFInfo
- Publication number
- CN116823027A CN116823027A CN202310656500.XA CN202310656500A CN116823027A CN 116823027 A CN116823027 A CN 116823027A CN 202310656500 A CN202310656500 A CN 202310656500A CN 116823027 A CN116823027 A CN 116823027A
- Authority
- CN
- China
- Prior art keywords
- skill
- graph
- student
- layer
- knowledge
- 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
- 238000000034 method Methods 0.000 title claims abstract description 26
- 238000013528 artificial neural network Methods 0.000 claims abstract description 28
- 230000003993 interaction Effects 0.000 claims abstract description 15
- 238000010586 diagram Methods 0.000 claims abstract description 13
- 238000012549 training Methods 0.000 claims abstract description 8
- 239000011159 matrix material Substances 0.000 claims description 37
- 230000005540 biological transmission Effects 0.000 claims description 10
- 239000013598 vector Substances 0.000 claims description 10
- 238000012546 transfer Methods 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 7
- 230000006870 function Effects 0.000 claims description 7
- 238000003475 lamination Methods 0.000 claims description 2
- 230000000306 recurrent effect Effects 0.000 claims 1
- 239000010410 layer Substances 0.000 description 49
- 238000012360 testing method Methods 0.000 description 19
- 238000002474 experimental method Methods 0.000 description 8
- 230000007246 mechanism Effects 0.000 description 7
- 238000013135 deep learning Methods 0.000 description 4
- 238000013473 artificial intelligence Methods 0.000 description 3
- 125000004122 cyclic group Chemical group 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000007418 data mining Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 238000001069 Raman spectroscopy Methods 0.000 description 1
- 230000004931 aggregating effect Effects 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000001939 inductive effect Effects 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000002356 single layer Substances 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000007474 system interaction Effects 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
- 230000017105 transposition Effects 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application provides an intelligent student capability assessment method based on associated skill knowledge, which comprises the following steps: constructing a data set; the dataset comprises: a skill diagram for representing student answer interaction information; constructing a graph-based associated skill knowledge tracking model; training the associated skill knowledge tracking model based on the dataset; based on the trained associated skill knowledge tracking model, obtaining the probability of each skill point correctly answered by the student; based on the probabilities, student competence is assessed. According to the application, the graph structure of the skills, the graph neural network and the graph-based knowledge tracking model are combined together, so that the capability of the model to perceive the relevance between the skill node and the neighbor skill node in the graph and the capability of exploring the relevance between the skills in the graph structure are improved, and the prediction accuracy of the model is improved.
Description
Technical Field
The application belongs to the technical field of artificial intelligence and information, and particularly relates to an intelligent student capability assessment method based on associated skill knowledge.
Background
The intelligent coaching system based on the artificial intelligence technology has profound influence on the online education industry, and development of online education and social progress also provide higher requirements for the intelligent and individuation level of the intelligent coaching system. Knowledge tracking is a task of modeling and predicting the knowledge state of students according to the historical answer interaction information of the students on an intelligent coaching system. Knowledge tracking can effectively predict knowledge mastering conditions of students and answering accuracy of the students, and can customize student answering paths in a personalized way, so that learning efficiency of the students is improved. The knowledge tracking plays an important role in improving the intelligent and personalized level of the intelligent coaching system, and is a key technology for realizing intelligent educational objectives by assisting the intelligent coaching system.
Deep learning has achieved great success in the field of artificial intelligence. Various knowledge tracking models based on deep learning are greatly developed, and the knowledge tracking models are excellent in tasks such as knowledge modeling and prediction. However, the following problems still generally exist in the existing depth knowledge tracking model: the historical answer information of students is simply treated as sequence data, the knowledge concepts and the graph structural characteristics of skills are ignored, and the relevance between the skill nodes in the graph and the neighbor skill nodes and the relevance between the skills on the graph structure are difficult to be effectively perceived. Therefore, the application focuses on exploring and researching the relevance among skills and proposes a correlation skill knowledge tracking model research based on deep learning.
Disclosure of Invention
In order to solve the technical problems, the application provides an intelligent student capability assessment method based on associated skill knowledge, which combines a graph structure of skills with a graph neural network and a graph-based knowledge tracking model, improves the capability of the model to perceive the relevance between a skill node and its neighbor skill node in the graph and explore the relevance between the skills in the graph structure, and improves the prediction accuracy of the model.
In order to achieve the above purpose, the application provides an intelligent student capability assessment method based on associated skill knowledge, which comprises the following steps:
constructing a data set; the data set comprises a skill diagram for representing student answer interaction information;
constructing a graph-based associated skill knowledge tracking model;
training the associated skill knowledge tracking model by utilizing the data set;
based on the trained associated skill knowledge tracking model, obtaining the probability of each skill point correctly answered by the student;
based on the probabilities, student competence is assessed.
Optionally, constructing the dataset comprises:
obtaining answer interaction information of students;
constructing a skill graph G (V, E) based on the answer interaction information, wherein V is a vertex set of the skill graph, and E is a set of edges in the skill graph;
based on the skill graph, the dataset is constructed.
Optionally, the associated skill knowledge tracking model includes: the system comprises a coding layer, a picture scroll layer, a message transmission layer and a skill picture feature memory and update layer which are connected in sequence;
the coding layer is used for coding the input data;
the graph volume lamination layer is used for extracting structural features of the nodes and the neighbor nodes;
the message transfer layer is used for transferring answer information to the associated skill point of the current answer skill point, namely the neighbor node;
the skill diagram feature memorizing and updating layer is used for memorizing and updating the skill diagram node features after the message passing layer, memorizing the student history answer information and modeling the student knowledge state.
Optionally, in the encoding layer, encoding the input data includes:
encoding the skill graph node characteristics; coding answer information of students; an index of skills in the skill graph is encoded.
Optionally, the graph roll layer adopts a graph roll neural network;
the graph convolution neural network is as follows:
wherein ,represents the sum of the adjacency matrix A and the identity matrix I, A represents the adjacency matrix,/and->Representation->D represents the degree matrix of the adjacency matrix A, W l The representation is a training weight matrix of layer I, sigma represents the activation function, H l Representing a layer i activation matrix.
Optionally, the message transmission layer is used for performing splicing operation on the coded answer information code, the index code of the skills in the skill graph and the skill point characteristics after passing through the graph convolution layer;
the message transfer layer is also used for transferring the answer information to the associated skill point of the current answer skill point.
Optionally, in the skill graph feature memorizing and updating layer, a GRU (generalized graph unit) cyclic neural network is adopted to memorize and update the skill graph node features after the message passing through the message passing layer.
Optionally, the probability that the student correctly answers each skill point is:
wherein ,Wout General weight matrix for representing node characteristics, b i The bias term representing the node i is represented,probability of representing correct answering skill point i, +.>And (5) representing a knowledge state feature vector of the skill point i at the time t+1.
Compared with the prior art, the application has the following advantages and technical effects:
the application provides a graph-based associated skill knowledge tracking model. The application explores the knowledge concept and the graph structure characteristic of the skills, constructs the graph structure of the skills according to the historical answer sequence information of the students, and defines the associated skill points of the skills in the graph, namely the 1-hop neighbor skill nodes of the skill points. The graph structure of skills is combined with the graph neural network and the graph-based knowledge tracking model, the relevance capability between the skill node and the neighbor skill node in the graph and the relevance capability between the explored skills in the graph structure are improved, and the prediction accuracy of the model is improved.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the application. In the drawings:
FIG. 1 is a schematic diagram of a graph-based associative skill knowledge tracking model in accordance with an embodiment of the present application;
FIG. 2 is a graph showing test AUC results over six data sets for a model in accordance with an embodiment of the present application;
FIG. 3 is a schematic diagram of a part of skill indexes and names in the dataset ASSISTMEnts2015 according to an embodiment of the present application;
fig. 4 is a diagram illustrating the relationship intention of skill points in the data set ASSISTments2015 in the graph structure according to the embodiment of the present application.
Detailed Description
It should be noted that, without conflict, the embodiments of the present application and features of the embodiments may be combined with each other. The application will be described in detail below with reference to the drawings in connection with embodiments.
It should be noted that the steps illustrated in the flowcharts of the figures may be performed in a computer system such as a set of computer executable instructions, and that although a logical order is illustrated in the flowcharts, in some cases the steps illustrated or described may be performed in an order other than that illustrated herein.
The application provides an intelligent student capability assessment method based on associated skill knowledge, which comprises the following steps:
constructing a data set; the dataset comprises: a skill diagram for representing student answer interaction information;
constructing a graph-based associated skill knowledge tracking model;
training the associated skill knowledge tracking model based on the dataset;
based on the trained associated skill knowledge tracking model, obtaining the probability of each skill point correctly answered by the student;
based on the probabilities, student competence is assessed.
Constructing the data set includes:
obtaining answer interaction information of students;
constructing a skill graph G (V, E) based on the answer interaction information, wherein V is a vertex set of the skill graph, and E is a set of edges in the skill graph;
based on the skill graph, the dataset is constructed.
The constructed associated skill knowledge tracking model comprises the following steps: the system comprises a coding layer, a picture scroll layer, a message transmission layer and a skill picture feature memory and update layer which are connected in sequence;
an encoding layer for encoding input data;
the graph convolution layer is used for extracting structural features of the nodes and the neighbor nodes;
the message transmission layer is used for transmitting the answer information to the associated skill point of the current answer skill point, namely the neighbor node;
and the skill diagram feature memorizing and updating layer is used for memorizing and updating the skill diagram node features after the message transmission layer, memorizing the student history answer information and modeling the student knowledge state.
In the encoding layer, encoding the input data includes:
encoding the skill graph node characteristics; coding answer information of students; an index of skills in the skill graph is encoded.
The embodiment mainly aims at the common problems of the existing knowledge tracking model: the knowledge tracking model based on the graph can excavate the association relation of skills on the graph structure to a certain extent, but the knowledge tracking model based on the graph only simply simulates the transmission process of answer information on the graph structure, so that the relevance between the skill node and the neighbor skill node in the graph is difficult to be effectively perceived. In the embodiment, the correlation among skills of the multi-skill composition questions, the graph structure characteristics of the skills and the correlation among the skills in the composition graph structure are explored deeply, a correlation skill knowledge tracking model based on deep learning is provided, and an intelligent student capability assessment method based on the correlation skill knowledge is constructed by using the model.
The method proposed by the embodiment can be applied to an intelligent coaching system, and the embodiment uses a depth knowledge tracking model combining the Graph structural features of the exploration skills with a Graph neural network (Graph Neural Network, GNN), creatively combines the Graph structures of the skills with the Graph neural network and a Graph-based knowledge tracking model, and proposes a Graph-based associated skill knowledge tracking model (Graph-based Associated Skills Knowledge Tracing, GASKT) and is used for constructing an intelligent student capability assessment method. The intelligent student capability assessment method can assess the mastering degree of the student on each skill point in the system according to the historical interaction record of the student user, so that the intelligent coaching system can conduct personalized teaching and guidance on the student.
And (3) data collection:
firstly, the intelligent student capability assessment system collects interaction data of student users and removes user sensitive information and other irrelevant information. After system processing, the student answer interaction information at the time t can use the tuple x t ＝(q t ,a t ) Representation, q t Representing the answering skill points of students, a t And giving correct and wrong information to students. Building a skill graph G (V, E) for all student answer records, where v= { V 1 ,v 2 ,…,v N The set of vertices of the skill graph, is a collection of edges in the skill graph. The neighbor node of node i is denoted N i . The history answer sequence from student to t time is X t ＝(x 1 ,…,x t ) In order to realize intelligent assessment of student ability, it is necessary to predict the correct answering skill point q of students at time t+1 t+1 Probability P (a) t+1 ＝1|q t+1 ,X t G (V, E)) or predicting student mastery of overall skills (P) 1 ,…,P N ). Therefore, the embodiment provides a graph-based associated skill knowledge tracking model.
Model structure:
structure of graph-based associative skill knowledge tracking model: coding layer, graph convolution layer, message transfer layer, skill graph feature memory and update, as shown in FIG. 1.
Coding layer:
firstly, coding student answer information at t moment and skill characteristics in the graph:
skill graph node feature encoding:
at time t, for each node feature matrix in the skill graph G (V, E), it is represented as d represents the encoding dimension.
And (3) coding answer information of students:
at time t, student answer information tuple is x t ＝(q t ,a t), wherein qt Representing skill points of the answer, a t Represents q t Whether or not the answer is correct, answer pair q t ，a t =1, otherwise, a t =0. In order to express which skill the student answers to and the wrong answer information, the model carries out one-hot coding with the dimension of 2N on the student answer information: x is x t ＝δ(q t +a t ×N)∈{0,1} 2N Delta represents one-hot coding;
setting an answer information embedding matrixObtaining answering information of student at t time and embedding code x' t ：
Skill index encoding:
encoding the skill index in the graph, wherein the skill index embedding matrix is as followsE c (i) Representation skill index embeddingMatrix E c Is the i-th row vector of (c).
Graph convolution layer:
existing graph neural networks are mostly based on a messaging mechanism (MessagePassingMechanism): and circularly iterating and aggregating the characteristics of the neighbor nodes to update the characteristic representation of the current node. Such a graph neural network is also called a messaging neural network (MessagePassingNeuralNetworks, MPNN), and its general framework is described as follows:
wherein ,representing a set of neighbor nodes of node i, M t As a message function (MessageFunction), U t Updating the function (updateFunctions) for node characteristics, e ij Representing the characteristics of edge (i, j).
The industry typically divides graph neural networks into two categories, spectral-based and Spatial-based. In this embodiment, three graph neural networks are used to construct a graph convolutional layer based on a graph-based associative skill knowledge tracking model, and as shown in table 1, the graph neural network convolutional layer is implemented by using torch-geometry.
TABLE 1 graphic neural network model employed by the graphic convolution layer and classification and implementation thereof
Graph convolutional neural network (GraphConvolutionalNetwork, GCN):
the Layer-wise (Layer-wise) propagation rule of the graph roll-up neural network is as follows:
wherein ,the purpose is to consider the case where a node points to itself, i.e. edges (I, I), I N Is an identity matrix (identity matrix) of dimension n×n. />W (l) Is the training weight matrix of the first layer. σ is an activation function, and ReLU (x) =max (0, x) or the like can be selected. H l Is a layer I activation matrix initialized to H 0 And c, X is a feature input matrix.
Graph attention network (GraphAttentionNetwork, GAT):
the graph Attention network uses Self-Attention mechanism (Self-Attention) to calculate the influence weight factors of the nodes and the neighbors thereof, and GAT has the advantages of being independent of the graph structure known in advance, not needing expensive matrix operation and the like, and can be well applied to inductive learning tasks.
The node characteristic matrix for graph G (V, E) isGAT employs a mechanism of attentionLet the weight matrix be->For any two nodes i and j in the graph G, the attention coefficients between them are calculated as:
e ij ＝a(Wh i ,Wh j ) (5)
GAT considers graph structure on the basis of self-attention, and adopts mask attention mechanism (mask attention), only calculates the attention between node and its neighbor node:
combining equation (5) and equation (6), and the attention mechanism a may be a single layer feedforward neural network, consisting of weight vectorsParameterization can yield:
wherein [ (i ] represents a splicing operation), and T represents a matrix transposition operation. GAT considers the topology (neighbors) of the node, and the feature vector updated by node i is:
for GAT employing a multi-head K-attention mechanism, the feature vectors updated by node i are:
graph homogeneous network (GraphIsomorphismNetwork, GIN):
graph isomorphism networks explored the learning and expression capabilities of the graph neural network on isomorphism graphs relative to the Wei Sifei ler-raman test (WeisfeilerLehmanTest, WLTest), and GIN explored the impact of an aggregator-SUM (MEAM, MAX) on isomorphism graph resolving capabilities in an aggregation operation. GIN can be described by the formula:
where e is optionally 0, the mlp may be a multi-layered linear connected layer, and l represents the layer i iteration.
At time t, the node characteristic matrix in the skill graph G (V, E) is as followsLearning and sensing the structural features of the skill graph by using a graph neural network:
where GNNConv represents a graph neural network convolutional layer, GNNConv may be GCNConv, GATConv, GINConv in a graph-based associative skill knowledge tracking model. A is an adjacency matrix representation of the skill graph G (V, E) structure. The graph convolution layer can aggregate the 1-hop neighbor skill node information of the current skill node by utilizing a message transmission mechanism of the graph neural network, can capture the substructure information of the skills in the skill graph, and senses the relevance between the current skill and the associated skill.
Message passing layer:
coding answer information of student at time t in coding layerSkill index code E c Skill point feature H 'after passing through graph convolution layer' t-1 And (3) performing splicing operation:
the message transmission layer needs to answer the question informationTo the current answering skill point q t Is->The relevant skill points are affected by the answer information of the current time t, so that the model can learn about the current skill point q of students t The grasping degree of the skill and how the answer information affects the dynamic process of other skill characteristics, and the calculation formula is as follows:
wherein ,fS Representing a multi-layer perceptron network, f N Representing a neighbor information transfer function:
wherein A is the adjacency matrix representation of the skill graph G (V, E), f to ，f from Are multi-layer perceptron networks, i.e. the model not only considers the forward transmission q of information answering information t I, also consider the back propagation of information i→q t 。
Graph feature memory and update:
the associated skill knowledge tracking model based on the graph adopts a GRU cyclic neural network to memorize and update the skill graph node characteristics after the message passing layer:
the GRU cyclic neural network can memorize and learn answer information of students at each moment, improves dynamic prediction capability of a model on knowledge state and knowledge mastering of the students, and the GRU can be described by the following formula:
z t ＝σ(W z x t +U z h t-1 +b z ) (16)
r t ＝σ(W r x t +U r h t-1 +b r ) (17)
wherein ,xt Is the input of the model at the moment tVector, z t ，r t Respectively, an update gate (UpdateGate) vector, and a reset gate (ResetGate) vector. W and b represent the weight matrix and bias vector of the corresponding gate, respectively, and a indicates hadamard product (hadamard product).
Model output and optimization:
the model uses feature vectors of skill node i to predict the probability of a student correctly answering skill point i:
wherein ,Wout Is a node characteristic general weight matrix, b i Is the bias term for node i. Loss optimization of the graph-based associative skill knowledge tracking model is consistent with DKT, i.e., minimizes negative log likelihood loss (NLL) of answer information for an observed sequence of student answers.
Student competence assessment
Through the knowledge tracking model, the probability of each skill point being correctly answered by the student can be finally obtained. It may be provided that when the probability of a student correctly answering a certain skill point is less than 0.5, the student is judged not to grasp the skill point, when the probability of correctly answering is greater than 0.5 and less than 0.9, the student is judged to basically grasp the skill point, and when the probability of correctly answering is greater than 0.9, the student is judged to have been proficient in grasping the skill point. Through the assessment of student's ability, intelligent tutoring system can provide individualized tutoring for the student, for example reduces the skill point exercise that has been mastered, increases the skill point exercise that does not master to improve intelligent tutoring system's teaching efficiency.
The present embodiment further details the experimental data set and experimental setup of the graph-based associative skill knowledge tracking model:
experimental data set:
the adjacency matrix A for constructing the skill graph G (V, E) only depends on the skill answer path of students, does not need a question field, and has relatively more knowledge tracking data sets without the question field. Thus, this embodiment employs the most comprehensive benchmark dataset in the current knowledge tracking field to evaluate the performance of a graph-based associative skill knowledge tracking model, collecting a total of the most widely used 6 datasets:
(1)Algebra05：
algebra05 was collected from KDCup 2010 educational data mining challenges. The present embodiment preprocesses the data set Algebra05 into a "skillbuilder" data set similar to ASSIST 09. For a multi-skill question, the interactive records of the learner can be independently formed into an answer record by taking each skill as a skill label.
(2) ASSISTMEnts series data set
The ASSISTMENTS dataset is the best known benchmark data collected by the ASSISTMENTS online tutorial platform. There are a total of four ASSISTments data sets: 1ASSISTMEnts2009 (ASSIST 09), the experiment is carried out by adopting an updated data set with a Skill-builder format; 2ASSISTMEnts2012 (ASSIST 12); 3ASSISTMEnts2015 (ASSIST 15); 4ASSISTMENTS2017 (ASSIST 17), ASSIST17 is The most recent ASSISTMENTS dataset provided by The educational data mining competition The2017ASSISTMENTS DataMiningComposition.
(3)EdNet
The EdNet is a data set of all student system interactions collected by the multi-platform AI intelligent coaching system Santa in 2 years, and the student users of Santa in korea can be served to students through Android, iOS and Web platforms up to 78 ten thousand. The experiment is carried out by adopting an Ednet-KT1 format data set, and consists of question answering logs of students, namely, answer interaction information of each student in a Santa system is recorded in a log file which the student belongs to, and the record is collected according to a question answering sequence format since 4 months in 2017.
Reference model
To test the performance of the graph-based associative skill knowledge tracking model and demonstrate the improvement of the model herein to the existing knowledge tracking model, the present embodiment compares the graph-based associative skill knowledge tracking model to the most advanced knowledge tracking model as follows:
BKT: bayesian knowledge tracking model
BKT uses a Hidden Markov Model (HMM) to model the potential knowledge state of students as a set of binary variables. The experiment uses pyBKT to implement the BKT model and set the parameters seed=42, num_bits=1.
DKT, depth knowledge tracking model:
the experiment uses LSTM networks of "Tanh" activation functions to implement DKT.
DKVMN, dynamic key value memory network knowledge tracking model:
DKVMN is a variant of memory enhanced neural network (MANN) that uses a static matrix called Key (Key) to store knowledge concepts and a dynamic matrix called Value (Value) to store and update student's knowledge concepts's state of mastery.
SAKT, self-attention knowledge tracking model:
SAKT constructs a Query (Query) and a Key (Key) by using the history answer correct information of students, and constructs a Value (Value) by using an answer skill sequence.
GKT, graph-based depth knowledge tracking model:
the GKT models and predicts the knowledge state of students by constructing a graph structure of skills and then simply simulating the process of transmitting answer information to neighbor nodes.
The graph-based correlation skill knowledge tracking model optimal hyper-parameters are shown in table 2. For all models, the experiment was optimized using Adam optimizer, setting learning_rate=0.001, beta1=0.9, beta2=0.999, epsilon=1 e-8. Batch size (BatchSize) was set to 32. In order to utilize the student answer sequences as much as possible, the constructed skill graph G (V, E) adjacency matrix A is more perfect so as to find more answer paths of students, the student answer sequences selected in the experiment are longer, the minimum sequence length of the student answers in the data set is set to be 5, and the maximum sequence length is set to be 200, namely 5< sequence length <200. In order to compare the performance advantages and disadvantages of the graph-based associated skill knowledge tracking model and the graph-based knowledge tracking model (GKT), the experimental setting of the graph-based associated skill knowledge tracking model and the graph-based knowledge tracking model (GKT) are consistent, a training set and a verification set are set, and the ratio of the testing set to the verification set is 6:2:2.
Table 2 graph-based correlation skill knowledge tracking model optimal hyper-parameter table
Results and analysis
In this embodiment, the area under the curve (AreaUnderCurve, AUC) of the most common receiver operation feature (ReceiverOperatingCharacteristic, ROC) in the knowledge tracking field is used as an evaluation index of the model performance, the test AUC results of all knowledge tracking models in six real world data sets are shown in fig. 2, and the test AUC results of the top three of the performance ranks are marked in fig. 2, wherein the deeper the blue color of the unit cell, the better the performance is.
The following conclusions were drawn:
the graph-based associative skill knowledge tracking model (GASKT) presented in this embodiment performs better than the existing most advanced methods on six data sets (Algebra 05, ASSIST09, ASSIST12, ASSIST15, ASSIST17, edNet), and achieves test AUC results of 79.31%, 84.74%, 84.70%, 83.27%, 78.65%, 86.73%, respectively. The test AUC results of the graph-based associative skill knowledge tracking model (GASKT) were better than the best performing baseline models 0% (DKT 79.31%), 1.16% (SAKT 83.58%), 1.45% (DKT 83.25%), 1.52% (SAKT 81.75%), 0.61% (GKT 78.04%), 3.44% (GKT 83.29%), respectively, on average, by 1.36% over the six data sets.
On dataset ASSIST09, DKT and GASKT performed equally well, both 79.31%. SAKT achieved second best test AUC results on data sets Algebra05, ASSIT09, ASSIT15, 78.98%, 83.58%, 81.75%, respectively. Furthermore, DKT gave the second best test AUC results on dataset ASSIST12, 83.25%. GKT gave the second best test AUC results in dataset ASSIT17, edNet, 78.04%, 83.29%, respectively. SAKT obtained third best test AUC results on data sets ASSIST12, ASSIT17, 82.48%, 77.20%, respectively; the DKT obtains the third best test AUC results on the data sets ASSIST15 and Ednet, which are 80.34 percent and 82.14 percent respectively; GKT gave the third best test AUC results in data sets Algebra05, ASSIT09, 78.75%, 82.57%, respectively.
For non-graph-based methods, SAKT performed most excellent on datasets ASSIT09, ASSIT15, ASSIT17, achieving test AUC results of 83.58%, 81.75%, 77.20%, respectively. DKT performed most excellent on data sets Algebra05, ASSIT12, edNet, achieving test AUC results of 79.31%, 83.25%, 82.14%, respectively. Of these, BKT and DKVMN perform poorly.
For graph-based knowledge tracking methods: GKT and GASKT, graph-based correlation skill knowledge tracking model (GASKT) test AUC results (79.31%, 84.74%, 84.70%, 83.27%, 78.65%, 86.73%) on all six datasets (Algebra 05, ASSIST09, ASSIST12, ASSIST15, ASSIST17, edNet) were better than GKT (78.75%, 82.57%, 81.14%, 78.58%, 78.04%, 83.29%), with average elevations of 2.51% over GKT0.56%, 2.17%, 3.56%, 4.69% and 0.61%, 3.44%, respectively. Experimental results show that the graph-based associative skill knowledge tracking model presented herein is an effective graph-based depth knowledge tracking model.
Graph-based skill relevance
In the embodiment, the feature similarity is used for measuring and calculating the relevance between any node i and other nodes, and the cosine similarity is a method for effectively calculating the feature similarity. Feature h for node i and node j i and hj The similarity sim (i, j) is calculated as follows:
fig. 3 is a partial skills index and name in dataset assistants 2015. FIG. 4 is a graph showing the t-SNE descent after cosine similarity is calculated for skill point features in the dataset ASSISTMEnts2015Visualization after dimension. For each skill node i, the experiment only selects a node set with stronger relevance to the skill node iAnd performing wire connection visualization. The connecting line between the two points indicates that the skill point i and the skill point j have relevance, and the thickness of the line indicates the strength of the relevance.
The graph structure has very clear and definite expression capability, can convey information which is difficult to transfer by other data structures, and can effectively discover the relevance among elements in the graph. For example, as can be seen intuitively from the lower right hand corner in fig. 4, there is a correlation between skills 45 (equallionsolungtwoorfeworsteps) and skills 41 (additionalndsubsectionpositvidedemals), but the correlation is not strong and the lines are thin. The correlation between skills 45 (equationsolvengtwoorfeworsteps) and skills 34 (DividiltyRules), and between skills 45 (equationsolvengtwoorforsteps) and skills 38 (AlgebraicSolving) is strong, and the lines are thick. As another example, as seen from the lower right corner of fig. 4, skills 34 (DividilityRules) have 5 skills with strong relevance, which are all its neighbor nodes, namely skills 10 (Congrequence), skills 33 (scientific Notification), skills 38 (AlgebraicSolving), skills 43 (AdditionalSubstctionFraction), skills 45 (EqualitionSolngTwoFewerSteps), respectively.
In summary, the graph-based associative skill knowledge tracking model can effectively explore the associative relationship between skills in the graph structure.
The embodiment provides a graph-based associated skill knowledge tracking model. The embodiment explores the knowledge concept and the graph structure characteristics of the skills, constructs the graph structure of the skills according to the historical answer sequence information of the students, and defines the associated skill points of the skills in the graph, namely the 1-hop neighbor skill nodes of the skill points. The graph structure of skills is combined with the graph neural network and the graph-based knowledge tracking model, the relevance capability between the skill node and the neighbor skill node in the graph and the relevance capability between the explored skills in the graph structure are improved, and the prediction accuracy of the model is improved.
The present example conducted extensive, systematic experiments to test the performance of the model. The test AUC results of the graph-based associative skill knowledge tracking model on the six selected public data sets (Algebra 05, ASSIST09, ASSIST12, ASSIST15, ASSIST17, edNet) were improved by 1.36% on average over the best performing baseline model. The embodiment also deeply explores detailed processes of modeling and predicting the knowledge state of the student, and discusses tasks such as relevance among knowledge concepts and skills, condition influence among exercise of problems, knowledge concept discovery and the like.
The present application is not limited to the above-mentioned embodiments, and any changes or substitutions that can be easily understood by those skilled in the art within the technical scope of the present application are intended to be included in the scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.
Claims (8)
1. An intelligent student capability assessment method based on associated skill knowledge is characterized by comprising the following steps:
constructing a data set; the data set comprises a skill diagram for representing student answer interaction information;
constructing a graph-based associated skill knowledge tracking model;
training the associated skill knowledge tracking model by utilizing the data set;
based on the trained associated skill knowledge tracking model, obtaining the probability of each skill point correctly answered by the student;
based on the probabilities, student competence is assessed.
2. The intelligent student competency assessment method based on associated skill knowledge of claim 1, wherein constructing the data set comprises:
obtaining answer interaction information of students;
constructing a skill graph G (V, E) based on the answer interaction information, wherein V is a vertex set of the skill graph, and E is a set of edges in the skill graph;
based on the skill graph, the dataset is constructed.
3. The intelligent student competency assessment method based on associated skill knowledge of claim 1, wherein the associated skill knowledge tracking model comprises: the system comprises a coding layer, a picture scroll layer, a message transmission layer and a skill picture feature memory and update layer which are connected in sequence;
the coding layer is used for coding the input data;
the graph volume lamination layer is used for extracting structural features of the nodes and the neighbor nodes;
the message transfer layer is used for transferring answer information to the associated skill point of the current answer skill point, namely the neighbor node;
the skill diagram feature memorizing and updating layer is used for memorizing and updating the skill diagram node features after the message passing layer, memorizing the student history answer information and modeling the student knowledge state.
4. The intelligent student competence assessment method based on associated skill knowledge as set forth in claim 3, wherein in the encoding layer, encoding input data comprises:
encoding the skill graph node characteristics; coding answer information of students; an index of skills in the skill graph is encoded.
5. A method of intelligent student competence assessment based on associative skill knowledge as claimed in claim 3, wherein the atlas employs a atlas neural network;
the graph convolution neural network is as follows:
wherein ,represents the sum of the adjacency matrix A and the identity matrix I, A represents the adjacency matrix,/and->Representation->D represents the degree matrix of the adjacency matrix A, W l The representation is a training weight matrix of layer I, sigma represents the activation function, H l Representing a layer i activation matrix.
6. The intelligent student ability assessment method based on associated skill knowledge according to claim 4, wherein the message transfer layer is used for performing a splicing operation on the coded answer information codes, the index codes of the skills in the skill graph and the skill point characteristics after passing through the graph convolution layer;
the message transfer layer is also used for transferring the answer information to the associated skill point of the current answer skill point.
7. The intelligent student competence assessment method based on associated skill knowledge as set forth in claim 1, wherein in the skill graph feature memorizing and updating layer, a GRU recurrent neural network is adopted to memorize and update skill graph node features after passing through a message passing layer.
8. The intelligent student competence assessment method based on associated skill knowledge as claimed in claim 1, wherein the probability that a student correctly answers each skill point is:
wherein ,Wout Representing node characteristics in generalWeight matrix, b i The bias term representing the node i is represented,probability of representing correct answering skill point i, +.>And (5) representing a knowledge state feature vector of the skill point i at the time t+1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310656500.XA CN116823027A (en) | 2023-06-05 | 2023-06-05 | Intelligent student capability assessment method based on associated skill knowledge |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310656500.XA CN116823027A (en) | 2023-06-05 | 2023-06-05 | Intelligent student capability assessment method based on associated skill knowledge |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116823027A true CN116823027A (en) | 2023-09-29 |
Family
ID=88123271
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310656500.XA Pending CN116823027A (en) | 2023-06-05 | 2023-06-05 | Intelligent student capability assessment method based on associated skill knowledge |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116823027A (en) |
-
2023
- 2023-06-05 CN CN202310656500.XA patent/CN116823027A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113033808B (en) | Deep embedded knowledge tracking method based on problem difficulty and student capability | |
| CN113344053B (en) | Knowledge tracking method based on examination question different composition representation and learner embedding | |
| CN111814982A (en) | Multi-hop question-answer oriented dynamic reasoning network and method | |
| CN115510814B (en) | Chapter-level complex problem generation method based on dual planning | |
| CN113361791A (en) | Student score prediction method based on graph convolution | |
| CN114021722A (en) | Attention knowledge tracking method integrating cognitive portrayal | |
| CN115544158A (en) | Multi-knowledge-point dynamic knowledge tracking method applied to intelligent education system | |
| CN113378581B (en) | Knowledge tracking method and system based on multivariate concept attention model | |
| CN113283488B (en) | Learning behavior-based cognitive diagnosis method and system | |
| CN114077659A (en) | Knowledge graph question-answering method and system based on neighbor interaction network | |
| CN116611517A (en) | Knowledge tracking method integrating graph embedding and attention | |
| CN115795015A (en) | Comprehensive knowledge tracking method for enhancing test question difficulty | |
| CN116823027A (en) | Intelligent student capability assessment method based on associated skill knowledge | |
| CN112785039B (en) | Prediction method and related device for answer score rate of test questions | |
| CN115423546A (en) | Course recommendation method based on multi-graph neural network | |
| CN114742292A (en) | Knowledge tracking process-oriented two-state co-evolution method for predicting future performance of students | |
| CN114240539A (en) | Commodity recommendation method based on Tucker decomposition and knowledge graph | |
| CN115525742A (en) | Knowledge graph question-answering method based on representation learning | |
| Yang et al. | Skill-Oriented Hierarchical Structure for Deep Knowledge Tracing | |
| CN114943276B (en) | Depth knowledge tracking method based on tree-type attention mechanism | |
| CN116151242B (en) | Intelligent problem recommendation method, system and storage medium for programming learning scene | |
| CN114861916B (en) | Knowledge association path-fused cognitive tracking method | |
| Li et al. | Graph Knowledge Tracking Interaction Model Combining Classification and Regression Tree | |
| CN117911206A (en) | Dynamic assessment student knowledge level method based on double-attention mechanism | |
| CN118095531A (en) | Attention knowledge tracking method based on multi-scale space-time convolution |
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 |