A method for attention knowledge tracking integrated with cognitive characterization

Through the deep integration of attention knowledge tracking method of cognitive portrayal, combined with multi-head attention mechanism and deep learning technology, the problem of failure to fully consider the multi-dimensional characteristics of learners in the existing technology is solved, and the accuracy of learners' knowledge status is achieved Evaluation and effective prediction of future performance.

CN114021722BActive Publication Date: 2024-11-08HUAZHONG NORMAL UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111278587.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-10-30
Publication Date
2024-11-08
Estimated Expiration
2041-10-30

AI Technical Summary

Technical Problem

The existing knowledge tracking methods fail to fully consider learners' learning and forgetful behavior, knowledge mastery level, test questions difficulty and test text information, making it difficult to accurately evaluate learners' knowledge status and predict future performance.

Method used

Adopt attention knowledge tracking method with deep integration of cognitive portrayal, combining big data mining, natural language processing and deep learning technology, learn students' multi-dimensional feature information through multi-head attention mechanisms, including memory, practice and language features, and dynamically evaluate knowledge Point to grasp the situation and predict future performance.

Benefits of technology

It realizes scientific, comprehensive assessment and prediction of learners' knowledge status, and improves the effect of personalized teaching and resource recommendations for students with poor academic performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114021722B_ABST
    Figure CN114021722B_ABST
Patent Text Reader

Abstract

The present invention belongs to the field of educational data mining, and provides an attention knowledge tracking method integrating cognitive characterization, including: (1) data preprocessing; (2) feature fusion; (3) performance prediction; (4) model training. The present invention proposes a set of attention knowledge tracking frameworks based on learner cognitive portraits, which respectively models the learner's memory, practice and language characteristics, and finally jointly models the three characteristics obtained above, comprehensively considering the learner's various information, so as to improve the model's performance in predicting the learner's knowledge mastery and future performance.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of educational data mining, and specifically relates to an attention knowledge tracking method integrating cognitive characterization. Technical Background

[0002] With the rise of online education, platforms such as intelligent guidance systems and massive open online courses have gradually developed, and the number of active online learners has increased, making it impossible for educators to provide personalized learning guidance based on each student's knowledge mastery. In order to build a smarter online education adaptive system and solve a series of problems in personalized teaching and learning: exercise path planning, personalized recommendation, and knowledge graph construction, knowledge tracking has become the key. Through knowledge tracking technology, the system can automatically track the learner's knowledge mastery.

[0003] Specifically, the main task of knowledge tracking is to calculate the answer records of students based on their past answers X = {x1, x2, ... x t-1}, the changes in the student's knowledge status over time (h1,h2,…h t-1 ) to conduct dynamic modeling to evaluate students’ current level of knowledge mastery t And predict the learner's response at the next moment t ; A student's record of doing a question is represented by x t =(q t ,a t ), q t is the question answered by the learner at time t, and a t Indicates that students are in q t If the answer is correct, the value is 1, otherwise it is 0.

[0004] At the same time, in the process of exploring and explaining the learning rules of learners, educational scholars have proposed the following educational theories: The practice rate in Thorndike's three laws of learning shows that when learners repeatedly practice test questions under a certain knowledge point, they will continue to learn and consolidate existing knowledge; Ebbinghaus's forgetting curve theory shows that if learners do not review, their knowledge mastery will show an exponential decay trend over time, that is, the longer the time, the more knowledge is forgotten; with the development of psychology, the item response theory was proposed and a set of mathematical models were constructed to describe the relationship between learner ability, question characteristics and learner answers. In addition, information such as the test text and the learner's mastery of the concepts or skills of the corresponding test questions has a certain degree of influence on the student's learning and answering process. In addition, based on the fact that cognitive characterization is usually divided into three parts: memory, practice and language in the fields of educational psychology and medicine, this technical solution correspondingly classifies learning and forgetting curves as memory features, learners' question records as practice features, and the text semantics of test questions as language features, and systematically constructs a cognitive portrait of learners. There is a huge challenge in how to mine the above-mentioned effective information from students' historical interaction information and model it to predict learners' future performance. On the other hand, how to comprehensively and effectively introduce cognitive information such as students' learning and forgetting, concept or skill mastery, and the relationship between test questions into the knowledge tracking task is also very difficult in terms of technology and domain.

[0005] However, traditional knowledge tracking methods mainly evaluate the changes in the learner's knowledge status dynamically based on the learner's historical answer records, but ignore the student's learning and forgetting behavior, knowledge mastery level, student ability, test difficulty, and test text information. At present, some researchers have tried to use structures such as long short-term memory networks, memory enhancement networks, and multi-head attention mechanisms to perform knowledge tracking tasks, but due to the complexity of the learner's learning process and the diversity of learning resources, most studies only consider part of the characteristics and fail to pay attention to multi-dimensional information, making it difficult to fully characterize and model test questions and learners.

[0006] Specifically, the Deep Knowledge Tracing (DKT) model based on the long short-term memory network represents the learner's mastery of all knowledge points with a hidden state vector, which makes it difficult to accurately memorize sequences of more than hundreds of time steps and cannot explain the student's mastery level of each knowledge point; the Dynamic Key-Value Memory Networks for Knowledge Tracing (DKVMN) model based on the memory enhancement network solves the above problem by allowing the network to retain multiple hidden state vectors and read and write these vectors separately, but does not take into account that different students have different speeds in mastering knowledge points; and the knowledge tracing models based on the multi-head attention mechanism, such as the self-attention knowledge tracing model (ASelf-Attentive modelfor Knowledge Tracing, SAKT) and the relation-aware self-attention knowledge tracing model (Relation-Aware Self-Attention for Knowledge Tracing, SAKT) Knowledge Tracking (RKT) benefits from the superiority of the self-attention mechanism and is significantly better than previous research works in terms of the interpretability of results and performance prediction. However, students' learning process is often complex and changeable, and learning resources are also rich and diverse. Most of the current knowledge tracking studies based on the attention mechanism have failed to fully consider and characterize the characteristic information of students and test questions, such as students' abilities, forgetting and learning behaviors, mastery of concepts or skills, difficulty of test questions, and the relationship between test questions.

[0007] Therefore, exploring an attention knowledge tracking method that deeply integrates cognitive characterization, using popular big data mining, multi-head attention mechanism and deep learning technology to comprehensively analyze learners' online historical learning trajectories, accurately evaluate learners' current knowledge status and predict their future performance, has important research significance and application value. Summary of the invention

[0008] The purpose of the present invention is to address the shortcomings of existing learner knowledge diagnosis methods and provide an attention knowledge tracking method that deeply integrates cognitive characterization. It comprehensively utilizes big data mining technology, natural language processing and deep learning methods, and comprehensively considers the relationship between test questions, test question difficulty, student ability, learning and forgetting behavior, and learner skill mastery information to jointly model the learner's knowledge status. It uses a multi-head attention mechanism to learn the above-mentioned feature information, scientifically and comprehensively diagnose and evaluate the changes in learners' knowledge points over time, and predict learners' future performance, so as to provide personalized teaching and resource recommendations for students with unsatisfactory academic performance, and make targeted improvements.

[0009] To achieve the object of the invention, the following technical solutions are adopted in the present invention.

[0010] An attention knowledge tracking method integrating cognitive characterization includes: (1) data preprocessing; (2) feature fusion; (3) performance prediction; and (4) model training.

[0011] The specific data preprocessing in step (1) is as follows: corresponding data records are collected and processed into the data format required by the knowledge tracking model on this basis, that is, a data model is constructed to provide data preparation for subsequent feature fusion.

[0012] (1-1) Data collection: Collect the historical learning records of primary school mathematics learners on the "Learning Big Data" platform, including original data such as user names, question numbers answered by students, answering situations, knowledge points, time information, and test question texts.

[0013] (1-2) Constructing a data model: First, considering the need to protect student privacy, the user names of students in the original data are mapped with digital ids; secondly, the original data is grouped according to student ids and a learner question-solving sequence set arranged in ascending order of time is established; then the data is cleaned, deleting test questions with less than 3 learner practices and student data with less than 3 question-solving records; further feature extraction is performed on the cleaned data, including counting the number of repeated practices of a learner's series of practices belonging to the same knowledge point, obtaining learner ability and test question difficulty based on the Rasch model, and using the Bert model to extract text feature vectors of test question texts and constructing an adjacency matrix of test question relationships based on these vectors; finally, the processed data is saved to the corresponding dataset file.

[0014] The specific feature fusion in step (2) is as follows: Comprehensive and systematic features regarding student cognitive characterization are extracted and constructed from three aspects: the memory dimension, the practice dimension, and the language dimension, deeply exploring the learning behavior patterns of learners and the dynamic change process of their mastery of knowledge points.

[0015] (2-1) Problem definition. Assume there are H students, N concepts, and M different questions in the dataset, N << M. Represent the questions answered by the learner, the concepts to which the questions belong, and the answering responses as a tuple where represents the question attempted by student h at time t, is the concept index, represents the answer, and then given the historical learning records of student h from time 1 to t-1 the goal is to predict the probability that student h correctly answers a question belonging to concept at the next time t, that is the probability, that is

[0016] (2-2) Question, concept and interaction embedding. Under the setting of step (2-1), the input of the embedding layer is the question sequence Q = {q1, q2, …, q t}、Concept sequence C={c1,c2,…,c t}, reaction sequence R = {r1, r2, ..., r t}, the output is question embedding, concept embedding, question-response interaction embedding and concept-response interaction embedding; question embedding: using the question embedding matrix E Q ∈R M×d , so that the problem set Set Q Each exercise in q i Embedded into the i-th row, that is Where d represents the dimension of embedding into high-dimensional space; Concept embedding: define a concept embedding matrix E C ∈R N×d , used to obtain the concept set Set C Each concept in c i The embedding representation of

[0017] Question-response interaction embedding: By comprehensively considering the difficulty of the test questions, the students’ response information and the order of interaction, the answer situation r t Expanded to a d-dimensional vector r t =[r t ,r t ,…,r t ]∈R d , get a single question-response interaction embedding Specifically expressed as:

[0018]

[0019]

[0020] Among them, δ qt Represents the difficulty of the problem, P∈R l×d is the position embedding matrix containing the interaction position information, Indicates splicing;

[0021] Concept-Response Interaction Embedding: Training a concept change embedding matrix E D ∈R N×d , used to perceive and obtain the contextual representation of different questions under the same concept. Get a single concept-reaction interaction embedding As shown below:

[0022]

[0023]

[0024] The exercise-response interaction embedding sequence is represented as in The concept-response interaction embedding sequence is represented as in

[0025] (2-3) Cognitive characterization feature representation. Three feature fusers are constructed in the model to obtain students’ memory, practice and language features. The memory feature fuser uses the question embedding sequence Question-Reaction Interaction Embedding Sequence Number of exercises N = {n1, n2, ..., n t-1} and the time interval between questions ΔT = {Δt1, Δt2, …, Δt t-1} as input and output the student memory-related features M = {m1,m2,…,m t}. The practical feature fuser uses concept embedding And concept-reaction embedding As input, it outputs the student practice features P = {p1, p2, ..., p t}. The language feature fusion is embedded in the sequence of questions Question-Reaction Interaction Embedding Sequence And the question relationship adjacency matrix R = {R t,1 ,R t,2 ,…,R t,t-1} as input and output the student language features L = {l1,l2,…,l t};

[0026] In the above steps (2-3), the specific method of obtaining the student memory-related features M, obtaining the student practice-related features P, and obtaining the student language-related features L is as follows:

[0027] Let α ij are the attention weights learned using the general scaled dot-product attention mechanism:

[0028]

[0029] Then, the value is weighted and summed according to the obtained weight coefficient to obtain the final representation of the i-th element, o i ∈R d :

[0030]

[0031] Among them, q i , k j and v j Represent query, key and value respectively, W Q ∈R d×d , W K ∈R 2d×d and W V ∈R 2d×d They are the projection matrices for query, key and value respectively;

[0032] Among them, the practical feature fusion adopts concept embedding and concept-reaction interaction embedding As input, the concept embedding sequence As the query, concept-response interaction embedding sequence in the attention mechanism As the key and value matrix, only the general attention mechanism is used to obtain and output the students' practice-related features, which reflects the students' knowledge mastery;

[0033]

[0034] Finally, the practice-related feature fuser obtains and outputs the student practice-related features P = {p1, p2, ..., p t};

[0035] By incorporating students' abilities, learning behaviors, and forgetting patterns, the memory feature fuser uses a modified memory-aware attention mechanism to embed the question sequence Question-Reaction Interaction Embedding Sequence As input, the question is embedded in the sequence As the query, question-response interaction embedding sequence in the attention mechanism As the key and value matrix, the lg(N) function is used to model the students' learning behavior, and the exponential decay function e -ΔT To model the forgetting rules of students, the memory-related attention weight score obtained by adjusting the attention weight score is:

[0036]

[0037] Among them, θ h represents the ability of the hth student, which reflects the student's learning ability to a certain extent, and also reflects the student's forgetting rate; λ1 represents a trainable parameter of the model. Then according to the obtained weight score β i The question-response interaction embeddings are weighted summed to obtain the final representation of memory-related features, m i∈R d :

[0038]

[0039] Finally, the memory-related feature fusion unit obtains and outputs the student memory-related features M = {m1, m2, ..., m t};

[0040] The language feature fuser uses a modified exercise relation-aware attention mechanism to adopt question embedding sequence Question-Reaction Interaction Embedding Sequence As input, the question is embedded in the sequence As the query, question-response interaction embedding sequence in the attention mechanism As key and value, the attention weight is modified by constructing the exercise relationship adjacency matrix R based on the test text vector extracted by the Bert model to obtain the required language-related attention weight score:

[0041] γ i =λ2*Softmax(R i )+(1-λ2)α i

[0042] Among them, λ2 is also a trainable parameter of the model; based on the weight score γ i The question-response interaction embeddings are weighted summed to obtain the final representation of language-related features, l i ∈R d :

[0043]

[0044] Finally, the language-related feature fusion unit obtains and outputs the student language-related features L = {l1,l2,…,l t}.

[0045] The performance prediction in step (3) specifically includes: feature fusion, using the relative position attention mechanism to construct a prediction model, diagnosing the students' knowledge mastery and predicting the learners' performance.

[0046] (3-1) Feature fusion: The prediction stage uses the relative position attention mechanism for prediction, so the exercise embedding sequence and the concept embedding sequence are concatenated and linearly transformed, and used as the query matrix of the prediction model. Similarly, the student memory, practice and language related features output by the memory, practice and language feature aggregator mentioned above are concatenated and linearly transformed to obtain the final embedding representation, which is used as the key and value matrix. The specific calculation is as follows:

[0047]

[0048]

[0049]

[0050]

[0051]

[0052]

[0053] clip(x,k)=max(-k,min(k,x))

[0054] Among them, q i ,k i ,v i ∈R d , A vector representing the relative position information between elements, where k is the maximum distance between the set elements;

[0055] The output o of the model is then passed to the feedforward layer, which is calculated as follows:

[0056] F = ReLU(oW (1) +b (1) )W (2) +b (2)

[0057] Among them, W (1) ,W (2) ∈R d×d is the weight matrix, b (1) , b (2) ∈R d is the bias vector; in addition to the above modeling structure, residual connection, layer normalization and random inactivation are added after the self-attention layer and the feedforward layer;

[0058] (3-2) Diagnostic results of learners’ knowledge mastery h t This is the final result F output by the above feedforward layer.

[0059] (3-3) Learner performance prediction: The probability that the learner correctly answers the test question at time t for:

[0060]

[0061] The step (4) of model training aims to minimize the negative log-likelihood of the student's response sequence. and the true label r of the learner’s response t The cross entropy loss between them is used to learn the parameters.

[0062]

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] 1. The present invention proposes a set of attention knowledge tracking framework based on learner cognitive portraits. This framework models the learner's memory, practice and language characteristics respectively, and finally performs joint modeling based on the above three characteristics, comprehensively considering various aspects of learner information, so as to improve the model's performance in predicting learners' knowledge mastery and future performance.

[0065] 2. Based on the cognitive characterization attention knowledge tracking framework, a method for modeling learners' memory-related characteristics is proposed. The learner's learning behavior is modeled with lg(x), and it is assumed that the improvement is obvious before 10 exercises, and after more than 10 exercises, it is assumed that the student has mastered the exercise; the learner's forgetting law is modeled with e -ΔT The degree of knowledge mastery will show an exponential decay trend with the passage of time. That is, the longer the time, the more knowledge is forgotten. Taking into account the different abilities of different students, the memory-related characteristics of learners are modeled through a modified attention mechanism that can perceive memory.

[0066] 3. Based on the cognitive characterization attention knowledge tracking framework, a method for modeling learners' practice characteristics is proposed. The difficulty of the corresponding test questions measured by the item response theory and the students' answer sequence are used to perform joint modeling based on the attention mechanism to obtain learners' practice-related characteristics.

[0067] 4. Based on cognitive characterization and attention knowledge tracking, a method for modeling learners' language characteristics is proposed. By constructing a question relationship adjacency matrix with highly relevant text semantics, a customized relationship-aware attention mechanism is used to mine learners' language-related characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 Schematic diagram of the framework of the attention knowledge tracking method integrating cognitive characterization in the present invention.

[0069] Figure 2 Model diagram of the attention knowledge tracking method integrating cognitive characterization in the present invention.

[0070] Figure 3 Comparison chart of multi-dimensional feature systems. DETAILED DESCRIPTION

[0071] The embodiment of the present invention discloses a cognitive profile-based attention knowledge tracing method (Cognitive Profiles Attentive Knowledge Tracing, CPKT). Specifically, firstly, learners' historical practice records of elementary school mathematics subjects are collected from the "Learning Big Data" platform, and then the raw data obtained above are cleaned and processed to a certain extent, and the processed data is used to model a dynamic change process of the learner's knowledge state over time, and three feature fusers are designed, including a memory feature fuser, a practice feature fuser, and a language feature fuser to respectively obtain the learner's memory, practice, and language features, and then the relative position attention mechanism is used during the training process to update the learner's knowledge state and predict the probability of the learner correctly answering the question at the next moment.

[0072] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0073] like Figure 1 As shown in the figure, an attention knowledge tracking method integrating cognitive characterization can effectively evaluate the specific changes of learners' knowledge status as they continue to learn and predict learners' future performance. It mainly includes the following steps:

[0074] (1) Data preprocessing

[0075] (1-1) Data collection: The learners’ historical learning records in mathematics (at least one semester) are collected from the “Learning Big Data” platform, as shown in Tables 1 and 2, including original data such as user name, question number answered by the students, answer status, knowledge points, time information, and test text.

[0076] Table 1 Description of key fields of student test record

[0077]

[0078]

[0079] Table 2 Key fields of the exercise text

[0080]

[0081] (1-2) Construct a data model: First, considering the need to protect students' privacy, map the usernames of students in the original data to digital IDs, and also perform numerical mapping on questions and knowledge points; second, group the original data according to student IDs and establish a set of learners' question-solving sequences arranged in ascending order of time; then clean the data, delete the test questions with less than 3 learners' practices and the student data with less than 3 question-solving records; further preprocess the cleaned data, including counting the time taken by learners to solve corresponding questions, the number of repeated practices of the same knowledge point for the series of practices, the specific calculation method is shown in Table 3, obtaining learners' abilities and question difficulties based on the simple and interpretable Rasch model, extracting the text feature vectors of test question texts using the Bert model, and constructing an adjacency matrix of question relationships based on these vectors; then considering the storage size of the computer and the calculation efficiency of the model, divide the processed questions, concepts, responses, response times, and repeated practice number sequences into subsequences of fixed length S = {s1, s2, …, s l}, where l is the maximum length that the model can process, usually set to 200; when the sequence length t is less than l, repeatedly fill 0 at the end of the question and concept sequences, and repeatedly add -1 at the end of the response sequence; when t is greater than l, divide the sequence into t / / n + 1 subsequences. If the length of the last subsequence is less than 3, then it is t / / n. All these subsequences are used as inputs to the model; finally, save the processed data into the corresponding dataset files.

[0082] Table 3 Calculation method of memory features

[0083]

[0084] (2) Feature fusion

[0085] Extract and construct a systematic and comprehensive feature system of students' cognition from the three dimensions of memory, practice, and language.

[0086] (2-1) Problem definition

[0087] Suppose there are H students, N concepts, and M different questions in the dataset, N << M. Represent the questions answered by learners, the concepts to which the questions belong, and the response as a tuple where represents the question attempted by student h at time t, is the concept index, represents the answer. Then, given the historical learning records of student h from time 1 to t - 1 The goal is to predict the probability that student h correctly answers the question belonging to concept at the next time t, that is

[0088] (2-2) Problem, concept, and interaction embedding

[0089] Under the setting of step (2-1), the input of the embedding layer of this embodiment is the question sequence Q = {q1, q2, ..., q t}、Concept sequence C={c1,c2,…,c t}, reaction sequence R = {r1, r2, ..., r t}, the output is question embedding, concept embedding, question-response interaction embedding, and concept-response interaction embedding. Question embedding: Using the question embedding matrix E Q ∈R M×d Set the problem set Q Each exercise in q i Embedded into the i-th row, that is Where d represents the dimension of embedding into high-dimensional space; Concept embedding: define a concept embedding matrix E C ∈R N×d Get concept set Set C Each concept in c i The embedding representation of

[0090] Question-Response Interaction Embedding: Answering the question r t Expanded to a d-dimensional vector r t =[r t ,r t ,…,r t ]∈R d , get a single question-response interaction embedding As shown below:

[0091]

[0092]

[0093] Among them, δ qt Represents the difficulty of the learned problem, P∈R l×d is the position embedding matrix containing the interaction position information.

[0094] Concept-response interaction embedding: Embedding matrix E using concept changes D ∈R N×d Perceive and obtain contextual representations of different questions under the same concept, and then obtain a single concept-response embedding Specifically expressed as:

[0095]

[0096]

[0097] The exercise-response interaction embedding sequence is represented as in The concept-response interaction embedding sequence is represented as in

[0098] (2-3) Cognitive Characterization Feature Representation

[0099] like Figure 2 As shown in the figure, three feature fusers are constructed in the CPKT model to obtain students' memory, practice and language features.

[0100] (2-3-1) Memory feature fusion module

[0101] This module dynamically evaluates learners' knowledge status based on their learning behaviors and forgetting patterns, and takes into account the different abilities of different students to achieve personalized modeling.

[0102] First, extract the learning features, that is, the number of repeated exercises n: the number of repetitions of interactively corresponding knowledge points in the sequence; for example: as shown in Table 3, the number of knowledge point repetitions is counted from the beginning of the sequence, and the number of repetitions is accumulated by 1 when encountering a test question containing the same knowledge point as T2.

[0103] Secondly, extract the forgetting feature, that is, the sequence time interval Δt: the time interval between two directly adjacent interactions in the sequence; for example: as shown in Table 3, the first interaction time interval defaults to 0, and the interval time of each subsequent exercise is calculated as: Δt n =T n -T n-1 (e.g., T4-T3).

[0104] Finally, the input of the memory feature fusion is the question embedding sequence Question-Reaction Interaction Embedding Sequence Number of exercises N = {n1, n2, ..., n t-1} and the time interval between questions ΔT = {Δt1, Δt2, …, Δt t-1}, where the question is embedded in the sequence As the query, question-response interaction embedding sequence in the multi-head attention mechanism It serves as the key and value matrix, and outputs the student memory-related features obtained using the modified memory-aware attention mechanism.

[0105] Specifically, the memory feature fuser first calculates the different degrees of influence of past question j on future question i, that is, the weight score α ij .

[0106]

[0107] Where d is the embedding size, They are the projection matrices for query and key respectively.

[0108] Then based on the number of exercises N = {n1, n2, ..., n t-1} and the time interval between questions ΔT = {Δt1, Δt2, …, Δt t-1}, the logarithmic function lg(N) is used to model the students’ learning behavior, and the exponential decay function e -ΔT To model the student forgetting pattern to modify the influence weight Finally, the memory-related attention weight score of student h is:

[0109]

[0110] Among them, θ h represents the ability of the hth student, which reflects the student's learning ability to a certain extent, and also reflects the student's forgetting rate; λ1 represents a trainable parameter of the model; and then according to the obtained weight score Perform weighted summation on the value matrix to obtain the final representation of memory-related features, m i ∈R d :

[0111]

[0112] in, is the projection matrix for value. Finally, the memory-related feature fuser obtains and outputs the student memory-related features M = {m1, m2, …, m t}.

[0113] (2-3-2) Practical Feature Fusion Module

[0114] The practice-related feature fuser uses concept embedding and concept-reaction interaction embedding As input, the concept embedding sequence As the query, concept-response interaction embedding sequence in the multi-head attention mechanism As the key and value matrix, the general attention mechanism is used to obtain and output the student's practice-related features, which mainly reflect the student's current level of knowledge mastery. The specific calculation is as follows:

[0115]

[0116]

[0117]

[0118] in and are the linear projection matrices for query, key and value, respectively, which are the training parameters of the model. Finally, the practice-related feature fuser obtains and outputs the student practice-related features P = {p1, p2, ..., p t}.

[0119] (2-3-3) Language feature fusion module

[0120] In the process of extracting learner language features, this module uses the Bert model to obtain the text embedding vector E of the corresponding test question. t-1 ,E t , and then use the cosine similarity formula to calculate the similarity between exercises and construct the question relationship adjacency matrix R. The construction process of the exercise relationship adjacency matrix R is as follows:

[0121]

[0122]

[0123] Among them, K represents the threshold of whether the test questions are relevant, and its size is set to 0.1*M.

[0124] Finally, the language feature fuser embeds the sequence with the question Question-Reaction Interaction Embedding Sequence And the question relationship adjacency matrix R = {R t,1 ,R t,2 ,…,R t,t-1} as input, where the question is embedded in the sequence As the query, question-response interaction embedding sequence in the attention mechanism Then, as the key and value matrix, the modified relation-aware attention mechanism is used to obtain and output the language features of the students.

[0125] As with the memory fusion, first calculate the general attention weight score Reuse the practice relation R i Adjust the weights to get the student language-related attention score γ required for this module i :

[0126]

[0127]

[0128] Among them, λ2 is an adjustable parameter of the model. Based on the weight score Perform weighted summation on the value matrix to obtain the final representation of language-related features, li ∈R d :

[0129]

[0130] Likewise, is the parameter to be trained for the model. Finally, the language-related feature aggregator obtains and outputs the student’s language-related features L = {l1,l2,…,l t}.

[0131] Finally, we compare the previous research work and summarize the multi-dimensional feature system constructed by the CPKT model, such as Figure 3 shown.

[0132] (3) Performance prediction

[0133] The prediction and application of this embodiment specifically include: feature fusion, using the relative position attention mechanism to build a prediction model, diagnosing students' knowledge mastery, and predicting learners' performance.

[0134] (3-1) Feature Fusion

[0135] The prediction stage uses the relative position attention mechanism to perform predictions, concatenate and linearly transform the exercise embedding sequence and the concept embedding sequence as the query matrix of the prediction module. The student memory, practice and language related features output by the memory, practice and language feature fuser are concatenated and linearly transformed to obtain the final embedding representation F pred , and then take the features from time 1 to time t-1 As key and value matrix:

[0136]

[0137]

[0138]

[0139] in, is the training parameter of the model, 0∈R d is the zero vector, Indicates splicing.

[0140] (3-2) Prediction model construction

[0141] Pass the input data to the relative position attention layer for prediction:

[0142]

[0143]

[0144]

[0145]

[0146]

[0147] clip(x,k)=max(-k,min(k,x))

[0148] in, A vector representing the relative position information between elements, where k is the maximum distance between the set elements.

[0149] The output of the relative position attention layer is then fed into the feed-forward layer, and the final output is:

[0150]

[0151] in, is the weight matrix, and is the bias vector. In addition to the above modeling structure, residual connections, layer normalization and random inactivation are added after the self-attention layer and the feed-forward layer.

[0152] (3-3) Knowledge status diagnosis and performance prediction

[0153] (3-3-1) Knowledge status diagnosis: Diagnostic results of learners’ knowledge mastery t This is the final result F output by the above feedforward layer.

[0154] (3-3-2) Learner performance prediction: After diagnosing the learner’s knowledge mastery, predict the learner’s future performance, that is, the probability that the learner will correctly answer the test question at time t for:

[0155]

[0156] (4) Model training: The training goal is to minimize the negative log-likelihood of the student response sequence.

[0157] (4-1) Loss function: By minimizing the probability of predicting that the learner can answer the question correctly and the true label r of the learner’s response t The cross entropy loss between them is used to learn the parameters.

[0158]

[0159] (4-2) Optimization function: Update the parameters in the model by using the Adam optimization algorithm.

[0160] Parts of the present invention that are not described in detail belong to the well-known technology in the art.

[0161] It should be noted that the above-mentioned examples of the present invention are intended to explain the technical features of the present invention in detail. Without departing from the present invention, several improvements and modifications made are also protected by the present invention, so the protection scope of the present invention should be based on the content defined by the claims of this application.

Claims

1. A method for tracking attention knowledge by integrating cognitive characterization, Features The method comprises the following steps: (1) Data preprocessing: Collect relevant data records and process them into the data format required by the knowledge tracking model, that is, build a data model to provide data preparation for subsequent feature fusion; (1-1) Data collection: Collect learners' mathematics learning history records on the "Learning Big Data" platform, for at least one semester, including user name, question number answered by the student, answer status, knowledge points, time information, and original question text data; (1-2) Constructing a data model: First, map the student user names in the original data with digital IDs; second, group the original data by student ID and establish a sequence of learners' questions in ascending time order; then clean the data and delete questions that are practiced by less than 3 learners and student data with less than 3 question records; (2) Feature fusion: Extract and construct comprehensive and systematic features of students’ cognitive characterization from the three dimensions of memory, practice, and language, and deeply explore the dynamic changes in learners’ learning behavior patterns and their knowledge mastery; (2-1) Problem Definition: Suppose there are H students, N concepts, and M different questions in the dataset, where N << M. Represent the questions answered by the learner, the concepts to which the questions belong, and the response as a tuple where represents the question attempted by student h at time t, is the concept index, represents the answer. Then, given the historical learning records of student h from time 1 to t-1 the goal is to predict the probability that student h will correctly answer a question belonging to concept at the next time step t, i.e., ​ (2-2) Question, concept and interaction embedding: Under the setting of step (2-1), the input of the embedding layer is the question sequence Q = {q 1 ,q 2 ,…,q t }、Concept sequence C={c 1 ,c 2 ,…,c t }, reaction sequence R = {r 1 ,r 2 ,…,r t }, the output is question embedding, concept embedding, question-response interaction embedding and concept-response interaction embedding; question embedding: using the question embedding matrix E Q ∈R M×d Set the problem set Q Each exercise in q i Embedded into the i-th row, that is Where d represents the dimension of embedding into high-dimensional space; Concept embedding: using the concept embedding matrix E C ∈R N×d Get concept set Set C Each concept in c i The embedding representation of Question-Response Interaction Embedding: Answering the question r t Expanded to a d-dimensional vector r t =[r t ,r t ,…,r t ]∈R d , get a single question-response interaction embedding As shown below: Among them, δ qt Represents the difficulty of the problem, P∈R l×d is the position embedding matrix containing the interaction position information, Indicates splicing; Concept-response interaction embedding: Embedding matrix E using concept changes D ∈R N×d Perceive and obtain contextual representations of different questions under the same concept to obtain a single concept-response embedding Specifically expressed as: The exercise-response interaction embedding sequence is represented as in The concept-response interaction embedding sequence is represented as in (2-3) Cognitive characterization feature representation: Three feature fusers are constructed in the model to obtain students’ memory, practice, and language features. The memory feature fuser uses the problem embedding sequence Question-Reaction Interaction Embedding Sequence Number of exercises N = {n 1 ,n 2 ,…,n t-1 } and the time interval between questions ΔT = {Δt 1 ,Δt 2 ,…,Δt t-1 } as input and output the student memory-related features M = {m 1 ,m 2 ,…,m t }, Practice Feature Fusion Using Concept Embedding And concept-reaction embedding As input, it outputs the student practice-related features P = {p 1 ,p 2 ,…,p t }, the language feature fuser embeds the sequence with the question Question-Reaction Interaction Embedding Sequence And the problem relation adjacency matrix R = {R t,1 ,R t,2 ,…,R t,t-1 } as input, and outputs the student language related features L = {l 1 ,l 2 ,…,l t }; (3) Performance prediction: feature fusion, using the relative position attention mechanism to build a prediction model to diagnose students’ knowledge mastery and predict learners’ performance; (3-1) Feature fusion: In the prediction stage, the relative position attention mechanism is used for prediction. The exercise embedding sequence and the concept embedding sequence are concatenated and linearly transformed, and used as the query matrix of the prediction model. Similarly, the student memory, practice and language related features output by the memory, practice and language feature aggregator are concatenated and linearly transformed to obtain the final embedding representation, which is used as the key and value matrix. The specific calculation is as follows: clip(x,k)=max(-k,min(k,x)) Among them, q i ,k i ,v i ∈R d , A vector representing the relative position information between elements, where k is the maximum distance between the set elements; Finally, the output o of the model is passed to the feedforward layer, which is calculated as follows: F=ReLU(oW (1) +b (1) )W (2) +b (2) Among them, W (1) ,W (2) ∈R d×d is the weight matrix, b (1) , b (2) ∈R d is the bias vector; in addition to the above modeling structure, residual connection, layer normalization and random inactivation are added after the self-attention layer and the feedforward layer; (3-2) Diagnostic results of learners’ knowledge mastery h t This is the final result F output by the above feedforward layer; (3-3) Learner performance prediction: The probability that the learner correctly answers the question at time t for: (4) Model training: The goal of the training is to minimize the negative log-likelihood of the student’s response sequence by minimizing the probability of predicting that the learner will answer the question correctly. and the true label r of the learner’s response t The cross entropy loss between them is used to learn the parameters, 2. The method for tracking attention knowledge integrating cognitive characterization according to claim 1, Features: The specific method of obtaining the student memory-related feature M, the student practice-related feature P, and the student language-related feature L in step (2-3) is: Let α ij are the attention weights learned using the general scaled dot-product attention mechanism: Then, the value matrix is ​​weighted and summed according to the obtained weight score to obtain the final representation of the i-th element, o∈R d : Among them, q i , k j and v j Represent query, key and value respectively, W Q ∈R d×d , W K ∈R 2d×d and W V ∈R 2d×d They are the projection matrices for query, key and value respectively; Practical Feature Fusion Using Concept Embedding and concept-reaction interaction embedding As input, the concept embedding sequence As the query, concept-response interaction embedding sequence in the attention mechanism Then, as the key and value matrix, only the general attention mechanism is used to obtain and output the students' practice-related features; Finally, the practice-related feature fusion device obtains and outputs the student practice-related features P = {p 1 ,p 2 ,…,p t }; The memory feature fuser uses a modified memory-aware attention mechanism to embed the question sequence Question-Reaction Interaction Embedding Sequence As input, the question is embedded in the sequence As the query, question-response interaction embedding sequence in the attention mechanism As the key and value matrix, the logarithmic function lg(N) is used to model the students' learning behavior, and the exponential decay function e -ΔT To model the forgetting rules of students, the memory-related attention weight score obtained by adjusting the attention weight score is: Among them, θ h represents the ability of the hth student, λ 1 Represents a trainable parameter of the model; then according to the obtained weight score β i The question-response interaction embeddings are weighted summed to obtain the final representation of memory-related features, m i ∈R d : Finally, the memory-related feature fusion unit obtains and outputs the student memory-related feature M = {m 1 ,m 2 ,…,m t }; The language feature fuser uses a modified exercise relation-aware attention mechanism to adopt question embedding sequence Question-Reaction Interaction Embedding Sequence As input, the question is embedded in the sequence As the query, question-response interaction embedding sequence in the attention mechanism As the key and value matrix, the attention weight is modified by constructing the exercise relationship adjacency matrix R based on the test text vector extracted by the Bert model to obtain the required language-related attention weight score: c i =λ 2 *Softmax(R i )+(1-l 2 )a i Among them, λ 2 It is also a trainable parameter of the model; based on the weight score γ i The question-response interaction embeddings are weighted summed to obtain the final representation of language-related features, l i ∈R d : Finally, the language-related feature fusion unit obtains and outputs the student language-related features L = {l 1 ,l 2 ,…,l t }.