A learner cognitive tracking method and system for learning process evaluation

Abstract

The present invention belongs to the field of intelligent learning evaluation technology and discloses a learner cognitive tracking method and system for learning process evaluation. The method collects data and extracts multidimensional features based on the test knowledge cognition tensor TKC to generate the original four-tuple of answer features; sets hyperparameters to process the original four-tuple into a data format of equal length, encodes the answer features of different dimensions separately and fuses and embeds them to obtain a hierarchical embedding representation; constructs a cognitive tracking unit based on four gate structures: Slip Gate, Guess Gate, Level Gate, and Output Gate, and builds a deep cognitive tracking model with a fusion gate structure; tracks and outputs the learner's cognitive level of each knowledge point at different times, and predicts the learner's future answer performance on the knowledge point. The present invention promotes personalized learning for learners and helps model the learner's cognitive level change process in both objective and subjective senses.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent learning evaluation, and in particular relates to a learner cognition tracking method and system for learning process evaluation. Background Art

[0002] The continuous development of information technologies such as artificial intelligence, big data, and intelligent tutoring systems (ITS) has invisibly driven the shift from traditional education to new models. While the vast amount of teaching resources brings significant convenience to learners, it can also mean that the resources provided by tutoring systems may not be tailored to meet learners' specific needs. Therefore, implementing new teaching models, exemplified by personalized instruction, has become a major challenge in the field of intelligent education. These challenges include understanding learners' knowledge based on their past learning activities, predicting their future performance, rationally planning their learning paths, efficiently recommending learning resources, helping them tailor their learning plans, addressing gaps, and establishing a closed-loop learning-practice-testing system to truly achieve individualized teaching and personalized learning.

[0003] Learner cognitive tracking technology is a key technology for personalized instruction. This supervised sequential learning task aims to create a model of how a learner's knowledge state and cognitive structure evolve over time. Using the learner's historical learning history as input, the model assesses the learner's cognitive level—that is, how their knowledge mastery changes over time—in order to predict their future performance. This technology then explores the cognitive information underlying the learner's scores, analyzes their strengths and weaknesses, and ultimately provides targeted, personalized learning guidance.

[0004] Currently, researchers in the field have proposed two main types of learner cognitive tracking models: interpretable learner cognitive tracking based on probabilistic graphical models and high-precision learner cognitive tracking based on deep learning. Interpretable learner cognitive tracking methods based on probabilistic graphical models, such as Bayesian Knowledge Tracing (KT), are a special case of the Hidden Markov Model (HMM). They typically treat the learner's historical responses as an observation sequence and the learner's mastery of knowledge points as a state sequence, taking into account factors such as guesswork, errors, and forgetting. BKT models each knowledge point independently, assuming that the learner's response is a binary observation variable: correct or incorrect; the learner's knowledge state is a binary variable: mastered or not mastered. Learners can transition from the "not mastered" state to the "mastered" state by learning knowledge points, without forgetting during the learning process. These methods are simple and easy to understand. They rely on experts to label the relationship between exercises and knowledge points, effectively assessing each learner's knowledge level based on expert definitions, and have good interpretability. However, they also have certain drawbacks, such as ignoring the impact of exercise order on learning, significantly increasing labor costs, and the simple model architecture cannot simulate the highly complex learning process of the human brain. With the development of artificial intelligence and deep learning technologies, high-precision learner cognitive tracking methods based on deep learning, such as Deep Knowledge Tracing, have also rapidly developed. These methods use recurrent neural networks to model learners' responses, simulate changes in their knowledge state during the learning process, discover correlations between exercises, and predict future performance based on learners' historical response sequences. High-precision learner cognitive tracking methods based on deep learning can directly learn feature hierarchies from data using flexible functions and use historical information for prediction, achieving good prediction results for sequential problems. However, the operations within the model are black-box; for example, feature extraction and filtering are invisible, and the extracted features are difficult to interpret. Furthermore, both interpretable learner cognitive tracking based on probabilistic graphical models and high-precision learner cognitive tracking based on deep learning use only learners' interaction sequences as model input, ignoring the influence of learners' behavioral characteristics and other information collected by online platforms on the model. Both methods focus solely on the learners' outcome evaluation and assume that learners' responses are binary, with only two states (0, 1)—correct or incorrect. Consequently, they fail to explore the fine-grained changes in learners' cognitive states during the learning process.

[0005] How to effectively utilize the rich information contained in learning resources, and when extracting and embedding features of knowledge points, take into account that the learner's mastery of each knowledge point is not an extreme "either 0 or 1", so as to build a more accurate and detailed learner cognitive tracking model; how to capture the potential impact of objective factors such as learning resources on learners while modeling cognitive levels based on the subjective factors of learners' learning behavior; how to effectively combine process-oriented learner learning evaluation with traditional result-oriented learner learning evaluation, has become a difficult problem in learner cognitive tracking.

[0006] Through the above analysis, the problems and defects of the existing technology are as follows:

[0007] (1) The current interpretable learner cognitive tracking method based on probabilistic graphical models ignores the impact of exercise order on learning, increases labor costs, and the simple model architecture cannot simulate the highly complex learning process of the human brain; the operations within the high-precision learner cognitive tracking method based on deep learning are black box, such as feature extraction and screening are invisible, and the extracted features are difficult to explain.

[0008] (2) Existing learner cognitive tracking methods only focus on the outcome evaluation of learners and assume that learners’ responses are binary, with only two states: correct or incorrect. They are unable to mine the fine-grained cognitive state change information of learners during the learning process; they only use the learners’ interaction sequence as the input of the model, ignoring the influence of learners’ behavioral characteristics and other information collected by the online platform on the model.

[0009] (3) Traditional learner cognitive tracking methods simplify the information contained in learning resources, using only the knowledge points and correctness involved in the test questions as model inputs, ignoring other features in the learning process, resulting in low resource utilization of the model; the perspective of modeling learner cognitive states is relatively single, and the impact of objective factors such as learning resources on learners in the learning process is not considered, resulting in low model accuracy.

[0010] (4) Traditional learner cognitive tracking methods usually only use the learner's final binary answer (correct answer: 1; incorrect answer: 0) as a way to evaluate the learner's learning behavior. It is unable to effectively combine it with fine-grained process evaluation, resulting in poor interpretability of the model. Summary of the Invention

[0011] In view of the problems existing in the prior art, the present invention provides a learner cognition tracking method and system for learning process evaluation.

[0012] The present invention is implemented as follows: a learner cognitive tracking method for learning process evaluation, the learner cognitive tracking method for learning process evaluation includes: systematically collecting learner data and performing multi-dimensional feature extraction based on the test knowledge cognitive tensor TKC to generate an original quadruple of answer features; setting hyperparameters, processing the original quadruple into a data format of equal length, encoding answer features of different dimensions separately and fusing and embedding them to obtain a hierarchical embedded representation; constructing a cognitive tracking unit based on four gate structures of Slip Gate, Guess Gate, Level Gate and Output Gate, and building a deep cognitive tracking model with a fusion gate structure; tracking and outputting the learner's cognitive level of each knowledge point at different times based on cognitive state information and knowledge point parameter information, predicting the learner's future answer performance on the knowledge point, and making personalized learner learning resource recommendations.

[0013] Furthermore, the learner cognitive tracking method for learning process evaluation includes the following steps:

[0014] Step 1: Define and construct the "practice-knowledge-answer-cognitive hierarchy" quadruple: Collect learners' learning resources and answer data, and extract features for each learner based on the test knowledge and cognition tensor TKC introduced in Bloom's taxonomy of educational objectives in cognitive domains to generate the original quadruple of answer features;

[0015] Step 2: Data preprocessing and multi-dimensional feature encoding, fusion, and embedding: Hyperparameters are set based on the length of the original quadruple of learner response features, and data is processed into a uniform length format. The learner response features from different dimensions contained in the original quadruple are encoded and fused to obtain the final hierarchical embedding representation of each knowledge point.

[0016] Step 3: Construct a cognitive tracking model based on four gate structures: The model input is obtained by encapsulating the embedded representation in step 2, and a cognitive tracking unit is constructed based on the four gate structures of Slip Gate, Guess Gate, Level Gate, and Output Gate, thereby building a deep cognitive tracking model that integrates the gate structure.

[0017] Step 4: Track and output the learner's cognitive level at different times and predict future learning performance: Based on the learner's cognitive state information and knowledge point parameter information during the tracking process, dynamically output the learner's cognitive level at each knowledge point at different times and predict the learner's future answer performance.

[0018] Furthermore, the definition and construction of the "Practice - Knowledge - Answer - Cognitive Level" quadruple in step one includes:

[0019] (1) Construct a matrix of learning resource collection and learner historical responses;

[0020] S={s1,s2,…,s M};

[0021] E={e1,e2,…,e N};

[0022] K={k1,k2,…,k L};

[0023]

[0024] Among them, S is the learner set, M is the number of learners; E is the test question set, N is the number of test questions; K is the knowledge point set, L is the number of knowledge points; R is the learner's historical answer matrix, r mn =0 means that learner m answered question n incorrectly, r mn =1 means that learner m answered question n correctly.

[0025] (2) Bloom's taxonomy of educational objectives in the cognitive domain is introduced to divide learners' mastery of knowledge points into seven levels, and a specific and unified definition of the examination level of each knowledge point in the cognitive level interview questions is obtained, thereby constructing the test knowledge cognition tensor TKC that integrates Bloom's taxonomy of educational objectives in the cognitive domain;

[0026]

[0027] The seven cognitive levels are: not mastered 0, knowing 1, understanding 2, application 3, analysis 4, synthesis 5 and evaluation 6; 0≤c nl ≤6 means that question n tests knowledge points at levels l to c.

[0028] (3) Based on the learner's historical answer matrix R and the test knowledge cognition tensor TKC that incorporates Bloom's cognitive domain education objective taxonomy, the learner's learning data is feature extracted, and the learner's answer record is processed into an answer feature quadruple R containing four types of information: practice, knowledge, answer, and cognitive level. s ;

[0029] R s =(e s ,(k s ,r s ,tkc s ));

[0030] Among them, e s is the interactive record of learner s in practice, k s is the ID sequence of all knowledge points answered by learner s in chronological order, r s k sThe sequence of learner s’ answers corresponding to the knowledge point ID, tkc s k s The TKC level sequence of the knowledge point corresponding to the knowledge point ID is examined.

[0031] Furthermore, the data preprocessing and multi-dimensional feature encoding, fusion and embedding in step 2 include:

[0032] (1) Set the hyperparameter MAX_STEP to process the original answer feature quadruple into a data format of equal length;

[0033]

[0034] Among them, k s The ID sequence of the knowledge points answered by learner s, r s The real situation sequence for learner s's answer, tkc s is the cognitive level sequence of the knowledge points that learner s answers; 0 is the zero vector, when R s e in s When it is an integer multiple of MAX_STEP, the length of the zero vector is 0, otherwise the length of the zero vector is MAX_STEP minus e s The remainder of dividing by MAX_STEP, C(·) represents data filling processing; the filled k s ,r s ,tkc s Reshape them into answer matrices with a fixed column width of MAX_STEP to obtain a model input format of equal length.

[0035] (2) For the learner's answer feature quadruple R s The feature information IDk from different dimensions of the knowledge point contained in s 、Right or Wrong s and inspection level tkc s , first encode them separately and then embed and concatenate them to obtain the final hierarchical embedding representation of each knowledge point;

[0036]

[0037]

[0038] Among them, e ID is the traditional one-hot encoding of the ID of knowledge point k, and the dimension of the 0 vector is the number of knowledge points L contained in the dataset, e r To encode the answer situation that incorporates the knowledge point ID feature, e tkc is the traditional one-hot encoding of the cognitive level of the knowledge point being examined, e kIt is a multi-feature encoding vector that finally integrates knowledge point ID information, answer information, and cognitive level information.

[0039] Furthermore, the cognitive tracking model constructed based on the four gate structures in step three includes:

[0040] (1) Encapsulate the embedding representation in step 2 to obtain the final model input x that integrates the knowledge point answer information and hierarchical information t ; Before proceeding to the next step of calculation, t Extract answer features r t and hierarchical features tkc t , define k t is the cognitive state vector of the learner at time t;

[0041] r t =x t ⊙A;

[0042] tkc t =x t ⊙B;

[0043] Among them, A and B are extraction parameter matrices.

[0044] (2) Based on the four gate structures, a cognitive tracking unit is constructed to build a deep cognitive tracking model that integrates the gate structure; corresponding to the different levels of knowledge points and learners, the model sets up Slip Gate, Guess Gate, Level Gate and Output Gate respectively.

[0045] (3) Slip Gate and Guess Gate focus on the knowledge point level. Each knowledge point has a slip parameter and a guess parameter, which represent the probability that the learner learns the knowledge point but fails to reach the specified cognitive level, or the probability that the learner does not learn the knowledge point but explicitly reaches the specified cognitive level.

[0046] s t =σ(U s r t +V s k t-1 +b s );

[0047] g t =σ(U g r t +V g k t-1 +b g );

[0048] Level Gate focuses on the learner level. When a learner correctly answers a knowledge point at a certain level, the learner's cognitive state of that knowledge point is no lower than the level at which the knowledge point was tested. Otherwise, the learner is temporarily considered to be at the lowest level of Bloom's Cognitive Domain Taxonomy of Educational Objectives.

[0049] l t =σ(a t ⊙k t-1 );

[0050] Based on the traditional cognitive diagnosis method, the information from different levels obtained from Slip Gate, Guess Gate and Level Gate is integrated and updated to obtain the latent information about the learner's cognitive state;

[0051]

[0052] Output Gate responds based on the response information t , hierarchical information tkc t And the hidden information c t , determine the learner's final cognitive state k t ;

[0053] o t =σ(U o r t +V o k t-1 +b o );

[0054] k t =o t ⊙tanh(c t +U tkc tkc t );

[0055] Among them, U s 、V s 、b s , U g 、V g 、b g , U o 、V o 、b o , U tkc is the model training parameter matrix, a t is the learner's actual answer to the knowledge point, t The value range is {0, 1}; ⊙ represents element-by-element multiplication, σ represents the sigmoid activation function, and tanh represents the tanh activation function.

[0056] Furthermore, in step 4, tracking and outputting the learner's cognitive level at different times and predicting their future learning performance include:

[0057] (1) According to the learner’s cognitive state information k obtained during the tracking process t , build a classification function to output the learner's specific cognitive level at different knowledge points at each moment;

[0058] ct=V×k t ×M;

[0059]

[0060] Among them, level n is the maximum value of the cognitive level examined in the dataset, V and M are respectively the dimensions [L,1] and [L,level n ] is a two-dimensional matrix, cl is the output cognitive level of the learner.

[0061] (2) Add a fully connected layer for mapping, matching the learner's cognitive state with the knowledge point answer one by one, and obtain the learner's future answer performance on each knowledge point;

[0062] y t =σ(Wk t +b);

[0063] Among them, W is the weight coefficient matrix of the fully connected layer, and b is the bias coefficient matrix of the fully connected layer, which is obtained by training.

[0064] (3) Calculate the loss function of the model, which is the output prediction value y t and the true label r t The cross entropy between

[0065] loss CLKT =-∑(r t logy t +(1-r t )log(1-y t ));

[0066] (4) Update all parameter matrices in the model according to the loss function value and the gradient descent rule.

[0067] Another object of the present invention is to provide a learner cognition tracking system for learning process evaluation using the learner cognition tracking method for learning process evaluation. The learner cognition tracking system for learning process evaluation comprises:

[0068] The multi-dimensional feature data extraction module is used to collect learners' learning resources and answer data. Based on the test knowledge cognition tensor TKC introduced in Bloom's taxonomy of educational objectives in the cognitive domain, it extracts features from learners' learning data and processes learners' answer records into original answer feature quadruples containing four types of information: practice, knowledge, answer, and cognitive level.

[0069] The padding data processing module is used to set hyperparameters based on the length of the original quadruple of the learner's answer feature, and process the original quadruple of the answer feature into a data format of equal length;

[0070] The multi-dimensional feature embedding and fusion module is used to encode the feature information ID, correct or incorrect answer, and assessment level from different dimensions of the knowledge point contained in the learner's answer feature quadruple, and then embed and splice them to obtain the final hierarchical embedding representation of each knowledge point;

[0071] The deep cognitive tracking module is used to encapsulate the answer embedding representation to obtain the required model input. It constructs a cognitive tracking unit based on four gate structures: SlipGate, Guess Gate, Level Gate, and Output Gate. This constructs a deep cognitive tracking model that integrates gate structures. It models and tracks the learner's cognitive state of different knowledge points during the learning process, as well as the parameter information of the knowledge points themselves.

[0072] The cognitive level output module is used to dynamically output the learner's cognitive level at each knowledge point at different moments based on the learner's cognitive state information during the tracking process;

[0073] The future performance prediction module is used to predict the learner's future performance on the knowledge point based on the learner's cognitive state information and knowledge point parameter information during the tracking process;

[0074] The personalized resource recommendation module is used to recommend personalized learning resources to learners based on the learner's cognitive level information, knowledge point parameter information and test question knowledge cognition tensor TKC information during the tracking process.

[0075] Another object of the present invention is to provide a computer device, which includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes the steps of the learner cognitive tracking method for learning process evaluation.

[0076] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to execute the steps of the learner cognitive tracking method for learning process evaluation.

[0077] Another object of the present invention is to provide an information data processing terminal, which is used to implement the learner cognitive tracking system for learning process evaluation.

[0078] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0079] First, in view of the technical problems existing in the above-mentioned prior art and the difficulty of solving these problems, we closely combine the technical solutions to be protected by the present invention and the results and data during the research and development process, and analyze in detail and in depth how the technical solutions of the present invention solve the technical problems and some creative technical effects brought about by solving the problems. The specific description is as follows:

[0080] The learner cognitive tracking method for learning process evaluation provided by the present invention first defines and constructs the "exercise-knowledge-answer-cognitive level" quadruple, collects learners' learning resources and answer data, and extracts features for each learner based on the test knowledge cognition tensor TKC introduced by Bloom's cognitive domain education objective taxonomy to generate an original quadruple of answer features; then performs data preprocessing and multi-dimensional feature encoding, fusion and embedding, sets hyperparameters based on the length of the original quadruple of learner answer features, and processes it into a data format of equal length; encodes and fuses learners' answer features from different dimensions contained in the original quadruple to obtain the final hierarchical embedding representation of each knowledge point; then constructs a cognitive tracking model based on four gate structures, obtains the required model input by encapsulating the embedded representation, and The four gate structures of Gate construct a cognitive tracking unit, thereby building a deep cognitive tracking model that integrates gate structures; finally, it tracks and outputs the learner's cognitive level at different times and predicts their future learning performance. According to the learner's cognitive state information and knowledge point parameter information during the tracking process, it dynamically outputs the learner's cognitive level at each knowledge point at different times, and predicts the learner's future answer performance. The present invention is conducive to comprehensive and process-oriented modeling of learners' learning behavior and cognitive level, and is conducive to accurate and effective process-oriented evaluation and mining of learners' overall knowledge mastery and specific cognitive level, thereby promoting learners' personalized learning and providing new ideas for tracking and predicting learners' knowledge mastery and cognitive level in online learning platforms.

[0081] The present invention constructs a learner cognitive tracking unit based on four gate structures: Slip Gate, Guess Gate, Level Gate, and Output Gate. It designs a learner cognitive tracking model that can integrate the influence of learning resources. Modeling the learner's cognitive level helps model the learner's cognitive level change process in both objective and subjective senses, simulates and tracks the learner's cognitive state and the parameter information of the knowledge point itself during the learning process, and effectively restores the changes in the learner's cognitive state in actual scenarios. The present invention has higher accuracy in evaluating and tracking the learner's overall knowledge mastery and specific cognitive levels during the learning process, which is superior to traditional learner cognitive tracking methods. It can provide learners with more reliable self-examination information and teachers with more reasonable tutoring information. In addition, the present invention has more fully explored the rich features contained in the knowledge points. In the data preprocessing stage, it not only focuses on the answer status of the knowledge points, but also introduces professional knowledge in the field of education, taking into account the test knowledge cognition tensor TKC that integrates Bloom's cognitive domain education objective classification of the knowledge points being examined, and uses a customized multidimensional feature extraction and embedding method to perform a comprehensive and fine-grained embedding representation of the knowledge points based on information from different dimensions.

[0082] Second, considering the technical solution as a whole or from the perspective of the product, the technical effects and advantages of the technical solution to be protected by the present invention are described in detail as follows:

[0083] The present invention effectively extracts and embeds multidimensional features of knowledge points, and takes into account that the learner's cognitive state at different knowledge points may be affected by objective factors such as learning resources. The parameter information of knowledge points and learners is also modeled using a cognitive tracking unit based on four gate structures, which improves the accuracy of the model and is closer to reality.

[0084] The present invention uses the test knowledge cognitive tensor TKC that integrates Bloom's cognitive domain education objective taxonomy to define cognitive hierarchical information. While updating the learner's overall cognitive level at every moment, it also outputs the learner's hierarchical changes in specific knowledge points in a fine-grained manner, avoiding the extreme evaluation of learners' knowledge mastery based solely on the results of "correct answer" and "incorrect answer", making full use of professional knowledge in the field of education, and improving the effectiveness and interpretability of the model.

[0085] The learner cognitive tracking method for learning process evaluation provided by the present invention can explore the learner's cognitive level on knowledge points, fully consider the relationship between the subjective factors and objective factors affecting the learner, and thus track and evaluate the learner's overall knowledge mastery and specific cognitive level in a process-oriented manner, provide learners with effective evaluation and tracking information, help learners adjust subsequent learning plans, check for omissions and fill in the gaps, and improve learning efficiency.

[0086] Third, as auxiliary evidence for the inventiveness of the claims of the present invention, it is also reflected in the following important aspects:

[0087] (1) The expected benefits and commercial value of the technical solution of the present invention after transformation are: using big data and artificial intelligence to support intelligent learning evaluation and realize the educational concept of teaching students in accordance with their aptitude. It can be widely used in smart education systems, intelligent guidance systems, intelligent teaching assistants, adaptive learning systems and other fields, and has huge commercial value.

[0088] (2) The technical solution of the present invention fills the technical gap in the industry at home and abroad: the present invention integrates the rich information contained in learners and knowledge points with technology by constructing a learner cognitive tracking model oriented to the learning process, and designs a learner cognitive tracking unit based on four gate structures: Slip Gate, Guess Gate, Level Gate and Output Gate. It can obtain a more explanatory evaluation and analysis of the learner's cognitive level, filling the gap that the industry technology only evaluates learners' learning results from the two perspectives of "correct answer" and "wrong answer", and cannot conduct multi-dimensional, comprehensive and process evaluation and tracking of learners' learning behavior and cognitive status, thereby effectively modeling the changes in learners' overall knowledge mastery and specific cognitive level during the learning process, and providing comprehensive and accurate guidance information for learners' personalized learning.

[0089] (3) The technical solution of the present invention solves a technical problem that people have always wanted to solve but have never been able to solve successfully: when mining and tracking learners' cognitive levels, people have always wanted to achieve a fine-grained, process-based learning outcome and cognitive level evaluation method. Learners' learning of knowledge points is not limited to the final result evaluation of "right answer" and "wrong answer", and this cognitive state is not only affected by the learners' own subjective factors, but also related to the objective factors contained in the learning resources, and will change as learning continues to deepen or forget. The present invention is based on the mining of dynamic knowledge mastery and cognitive status, and overcomes the technical difficulties of process evaluation and tracking cognitive levels.

[0090] (4) The technical solution of the present invention overcomes technical bias: Traditional learner cognitive tracking methods simply assume that learners’ learning status for knowledge points is limited to two outcomes: “correct answer” and “incorrect answer”. This bias greatly hinders the application of learner cognitive tracking. Based on professional knowledge in the field of education, the present invention constructs a learner cognitive tracking method for learning process evaluation, which overcomes traditional technical bias and greatly enhances the application value of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0092] Figure 1 This is a flow chart of a learner cognition tracking method for learning process evaluation provided by an embodiment of the present invention;

[0093] Figure 2 1 is a schematic diagram of a learner cognitive tracking method for learning process evaluation provided by an embodiment of the present invention;

[0094] Figure 3 This is a TKC tensor heat map of test question knowledge cognition provided by an embodiment of the present invention;

[0095] Figure 4 is a comparison chart of experimental results provided by an embodiment of the present invention;

[0096] Figure 5 This is a radar chart showing an example of the learner cognitive level mining results provided by an embodiment of the present invention;

[0097] Figure 6 This is an example line chart of the learner's cognitive level tracking results provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0098] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0099] In view of the problems existing in the prior art, the present invention provides a learner cognitive tracking method and system for learning process evaluation. The present invention is described in detail below with reference to the accompanying drawings.

[0100] like Figure 1 As shown, the learner cognition tracking method for learning process evaluation provided by the embodiment of the present invention includes the following steps:

[0101] S101, the system collects the learner's learning resources and answer data and performs multi-dimensional feature extraction based on the test question knowledge cognition tensor TKC to generate the original answer feature quadruple;

[0102] S102: Set hyperparameters, process the original quadruple into a data format of equal length, encode the answer features of different dimensions separately, and fuse and embed them to obtain a hierarchical embedding representation;

[0103] S103, constructs a cognitive tracking unit based on four gate structures: Slip Gate, Guess Gate, Level Gate, and Output Gate, and builds a deep cognitive tracking model that integrates gate structures;

[0104] S104, tracks and outputs the learner's cognitive level of each knowledge point at different times, predicts the learner's future answer performance on the knowledge point and makes personalized learning resource recommendations.

[0105] As a preferred embodiment, Figure 2 As shown, the learner cognition tracking method for learning process evaluation provided by the embodiment of the present invention specifically includes the following steps:

[0106] (1) Define and construct the “practice-knowledge-answer-cognitive level” quadruple: collect learners’ learning resources and answer data, extract features for each learner based on the test knowledge and cognition tensor TKC introduced in Bloom’s taxonomy of educational objectives in cognitive domains, and generate the original quadruple of answer features;

[0107] (2) Data preprocessing and multi-dimensional feature encoding, fusion, and embedding: Hyperparameters are set based on the length of the original quadruple of learner response features to process them into a data format of equal length; learner response features from different dimensions contained in the original quadruple are encoded and fused to obtain the final hierarchical embedding representation of each knowledge point;

[0108] (3) Constructing a cognitive tracking model based on four gate structures: The required model input is obtained by encapsulating the embedded representation in step 2, and a cognitive tracking unit is constructed based on the four gate structures of Slip Gate, Guess Gate, Level Gate, and Output Gate, thereby building a deep cognitive tracking model that integrates the gate structure;

[0109] (4) Tracking and outputting the learner's cognitive level at different times and predicting their future learning performance: Based on the learner's cognitive state information and knowledge point parameter information during the tracking process, the learner's cognitive level at each knowledge point at different times is dynamically output, and the learner's future answer performance is predicted. The symbols and their meanings that appear in the embodiments of the present invention are shown in Table 1.

[0110] Table 1 Symbol description

[0111]

[0112]

[0113] The definition and construction of the "exercise-knowledge-answer-cognitive level" quadruple in step (1) provided in the embodiment of the present invention includes:

[0114] (1.1) Construct a matrix of learning resource collection and learner historical responses;

[0115] S={s1,s2,…,s M}

[0116] E={e1,e2,…,e N}

[0117] K={k1,k2,…,k L}

[0118]

[0119] Among them, S is the learner set, M is the number of learners; E is the test question set, N is the number of test questions; K is the knowledge point set, L is the number of knowledge points; R is the learner's historical answer matrix, r mn =0 means that learner m answered question n incorrectly, r mn =1 means that learner m answered question n correctly.

[0120] (1.1.1) The dataset information collected from the 2021 undergraduate course "C Language Programming and Design" at the Department of Artificial Intelligence Education, Central China Normal University is shown in Table 2.

[0121] Table 2 Dataset information

[0122] Number of learners 51 Number of test questions 124 Number of knowledge points 17 Learner's longest interaction record 124 Learner's shortest interaction record 10 Total interaction records 6165

[0123] (1.1.2) Preprocess the dataset. First, remove learners with fewer than 50 interaction records. Then, use zero-padding to address missing interaction records. Finally, arrange all learner interaction records in chronological order. This yields a two-dimensional matrix R (with dimensions [50, 124]) representing the responses of the 50 learners to the 124 test questions.

[0124] (1.1.3) This dataset contains a total of 17 knowledge points. The detailed description of the knowledge points is shown in Table 3.

[0125] Table 3 Knowledge point description

[0126] Knowledge Points describe <![CDATA[k1]]> program <![CDATA[k2]]> function <![CDATA[k3]]> Notes <![CDATA[k4]]> algorithm <![CDATA[k5]]> Data Structure <![CDATA[k6]]> Identifier <![CDATA[k7]]> Data Type <![CDATA[k8]]> Constants and variables <![CDATA[k9]]> Arithmetic operators <![CDATA[k 10 ]]> Input statement <![CDATA[k 11 ]]> Output Statement <![CDATA[k 12 ]]> expression <![CDATA[k 13 ]]> Logical operators <![CDATA[k 14 ]]> Judgment statement <![CDATA[k 15 ]]> Loop Statement <![CDATA[k 16 ]]> pointer <![CDATA[k 17 ]]> Array

[0127] (1.2) Introducing Bloom’s Taxonomy of Educational Objectives in the Cognitive Domain, the learners’ mastery of knowledge points is divided into seven levels, and a specific and unified definition of the examination level of each knowledge point in the cognitive level interview questions is obtained, thereby constructing the test knowledge cognition tensor TKC that integrates Bloom’s Taxonomy of Educational Objectives in the Cognitive Domain;

[0128]

[0129] The seven cognitive levels are: not mastered (0), known (1), understood (2), applied (3), analyzed (4), synthesized (5), and evaluated (6); 0≤c nl ≤6 means that question n tests knowledge points at levels l to c.

[0130] like Figure 3 As shown in Figure 1, based on Bloom's taxonomy of educational objectives in cognitive domains, educational experts labeled the knowledge points and their levels tested in each test question, constructing different test-knowledge-cognition tensors (TKCs). Considering the classroom's teaching objectives for first-year undergraduates new to programming, the highest level of knowledge points tested in this dataset is Level 3, the application level.

[0131] (1.3) Based on the learner's historical answer matrix R and the test knowledge cognition tensor TKC that incorporates Bloom's cognitive domain education objective taxonomy, the learner's learning data is feature extracted and the learner's answer record is processed into an answer feature quadruple R containing four types of information: practice, knowledge, answer, and cognitive level. s ;

[0132] R s =(e s ,(k s ,r s ,tkc s ))

[0133] Among them, e s is the interactive record of learner s in practice, k s is the ID sequence of all knowledge points answered by learner s in chronological order, r s k s The sequence of learner s’ answers corresponding to the knowledge point ID, tkc s k s The TKC level sequence of the knowledge point corresponding to the knowledge point ID in the test. In other words, when focusing on the characteristics of the knowledge point, not only the learner's answer to each knowledge point is considered, but also the TKC level information of the knowledge point is incorporated based on the parameters of the knowledge point itself.

[0134] In this dataset, examples of learners' answer feature quadruples are as follows:

[0135] R s =(1,(5,0,3))

[0136] The data preprocessing and multi-dimensional feature encoding, fusion and embedding in step (2) provided in the embodiment of the present invention include:

[0137] (2.1) Considering that the lengths of the answer feature quads of different learners may be inconsistent, the hyperparameter MAX_STEP is set to process the original answer feature quads into a data format of equal length to facilitate subsequent input into the model;

[0138]

[0139] Among them, k s The ID sequence of the knowledge points answered by learner s, r s The real situation sequence for learner s's answer, tkc s is the cognitive level sequence of the knowledge points that learner s answers; 0 is the zero vector, when R s e in s When it is an integer multiple of MAX_STEP, the length of the zero vector is 0, otherwise the length of the zero vector is MAX_STEP minus e s The remainder of dividing by MAX_STEP, C(·) represents data filling processing; the filled k s ,r s ,tkc s Reshape them into answer matrices with MAX_STEP as fixed column width to obtain the model input format of equal length;

[0140] (2.2) For the learner's answer feature quadruple R s The feature information k from different dimensions of knowledge points contained in s (ID), r s (correct answer) and tkc s (Inspection level), first encode them separately and then embed and splice them to obtain the final hierarchical embedding representation of each knowledge point;

[0141]

[0142]

[0143] Among them, e ID is the traditional one-hot encoding of the ID of knowledge point k, the dimension of the 0 vector is the number of knowledge points L contained in the dataset (L = 17 in this dataset), e r To encode the answer situation that incorporates the knowledge point ID feature, e tkc is the traditional one-hot encoding of the cognitive level of the knowledge point being examined, e k It is a multi-feature encoding vector that ultimately integrates knowledge point ID information, answer information, and cognitive level information.

[0144] The cognitive tracking model constructed based on four gate structures in step (3) provided by the embodiment of the present invention includes:

[0145] (3.1) Use the DataLoader method under the pytorch framework to encapsulate the input data and obtain the input data x that can be used by the deep learning algorithm t ;

[0146] (3.2)x t It is an encoding vector that combines the knowledge point answer information and hierarchical information. Before proceeding to the next step of calculation, first start from x t Extract answer features r t and hierarchical features tkc t , define k t is the cognitive state vector of the learner at time t;

[0147] r t =x t ⊙A

[0148] tkc t =x t ⊙B

[0149] Among them, A and B are extraction parameter matrices, whose values ​​can be set manually or trained together with the model;

[0150] (3.3) Constructing cognitive tracking units based on four gate structures, and building a deep cognitive tracking model that integrates gate structures; corresponding to different levels of knowledge points and learners, the model sets up Slip Gate, Guess Gate, Level Gate, and Output Gate respectively;

[0151] (3.4) Slip Gate and Guess Gate focus on the knowledge point level. Each knowledge point has a slip parameter and a guess parameter, which are the probability that the learner has learned the knowledge point but failed to reach the specified cognitive level, or the probability that the learner has not learned the knowledge point but has explicitly reached the specified cognitive level.

[0152] s t =σ(U s r t +V s k t-1 +b s )

[0153] g t =σ(U g r t +V g k t-1 +b g )

[0154] Level Gate focuses on the learner level. When a learner correctly answers a knowledge point at a certain level, their cognitive state of that knowledge point should be no lower than the level at which the knowledge point was tested. Otherwise, they are temporarily considered to be at the lowest level of Bloom's taxonomy of educational objectives in the cognitive domain.

[0155] l t =σ(a t ⊙k t-1 )

[0156] Based on the traditional cognitive diagnosis method, the information from different levels obtained from Slip Gate, Guess Gate and Level Gate is integrated and updated to obtain the latent information about the learner's cognitive state;

[0157]

[0158] Output Gate responds based on the response information t , hierarchical information tkc t And the hidden information c t , determine the learner's final cognitive state k t ;

[0159] o t =σ(U o r t +V o k t-1 +b o )

[0160] k t =o t ⊙tanh(c t +U tkc tkc t )

[0161] Among them, U s 、V s 、b s , U g 、V g 、b g , U o 、V o 、b o , U tkc is the model trainable parameter matrix, a t is the learner's actual answer to the knowledge point, which can be obtained by r t The calculated value range is {0,1}, ⊙ represents element-by-element multiplication, σ represents the sigmoid activation function, and tanh represents the tanh activation function.

[0162] The tracking and outputting of the learner's cognitive level at different times and the prediction of their future learning performance in step (4) provided by the embodiment of the present invention include:

[0163] (4.1) According to the learner’s cognitive state information k obtained during the tracking process t , build a classification function to output the learner's specific cognitive level at different knowledge points at each moment;

[0164] ct=V×k t ×M

[0165]

[0166] Among them, level n is the maximum value of the cognitive level examined in the dataset, V and M are respectively the dimensions [L,1] and [L,level n ] two-dimensional matrix (in this data set, L = 17, level n =3), its value can be set manually, and cl is the output learner cognitive level.

[0167] (4.2) In order to map the learner’s cognitive state to the knowledge point answer, a fully connected layer is added for mapping, and the learner’s future answer performance on each knowledge point is obtained;

[0168] y t =σ(Wk t +b)

[0169] Among them, W is the weight coefficient matrix of the fully connected layer, and b is the bias coefficient matrix of the fully connected layer, which can be obtained by training.

[0170] (4.3) Calculate the loss function of the model, which is the output prediction value y t and the true label r t The cross entropy between

[0171] loss CLKT =-∑(r t logy t +(1-r t )log(1-y t ))

[0172] (4.4) Update all parameter matrices in the model according to the loss function value obtained in step (4.3) and the gradient descent rule.

[0173] The experimental parameter settings of the embodiment of the present invention are shown in Table 4.

[0174] Table 4 Experimental parameter settings

[0175] parameter value MAX_STEP 10 BATCH_SIZE 64 LEARNING RATE 0.001 EPOCH 200 / 500 / 800 INPUT_SIZE 34 LEVEL_SIZE 3 HIDDEN_SIZE 68 OUTPUT_SIZE 17

[0176] The learner cognitive tracking system for learning process evaluation provided by the embodiment of the present invention includes:

[0177] The multi-dimensional feature data extraction module is used to collect learners' learning resources and answer data. Based on the test knowledge cognition tensor TKC introduced in Bloom's taxonomy of educational objectives in the cognitive domain, it extracts features from learners' learning data and processes learners' answer records into original answer feature quadruples containing four types of information: practice, knowledge, answer, and cognitive level.

[0178] The padded data processing module is used to set hyperparameters based on the length of the original quadruple of learner response features, processing the original quadruple of response features into a data format of equal length to facilitate subsequent input into the model, thereby solving the problem of inconsistent length of quadruple of response features of different learners;

[0179] The multi-dimensional feature embedding and fusion module is used to encode the feature information from different dimensions of the knowledge point (ID, correct or incorrect answer, and assessment level) contained in the learner's answer feature quadruple, and then embed and splice it to obtain the final hierarchical embedding representation of each knowledge point;

[0180] The deep cognitive tracking module is used to encapsulate the answer embedding representation to obtain the required model input. It constructs a cognitive tracking unit based on four gate structures: SlipGate, Guess Gate, Level Gate, and Output Gate. This constructs a deep cognitive tracking model that integrates gate structures. It models and tracks the learner's cognitive state of different knowledge points during the learning process, as well as the parameter information of the knowledge points themselves.

[0181] The cognitive level output module is used to dynamically output the learner's cognitive level at each knowledge point at different moments based on the learner's cognitive state information during the tracking process;

[0182] The future performance prediction module is used to predict the learner's future performance on the knowledge point based on the learner's cognitive state information and knowledge point parameter information during the tracking process;

[0183] The personalized resource recommendation module is used to recommend personalized learning resources to learners based on the learner's cognitive level information, knowledge point parameter information and test question knowledge cognition tensor TKC information during the tracking process.

[0184] The learner cognitive tracking method for learning process evaluation provided by the application embodiment of the present invention is applied to a computer device, which includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes the steps of the learner cognitive tracking method for learning process evaluation.

[0185] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware portion can be implemented using dedicated logic; the software portion can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those skilled in the art will understand that the above-mentioned devices and methods can be implemented using computer-executable instructions and / or contained in processor control code, for example, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, or can be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0186] In order to illustrate the positive effects of the learner cognitive tracking method provided by the embodiment of the present invention, the learner cognitive tracking method for learning process evaluation proposed in the present invention is used to model and track the changes in the learner's cognitive level during the learning process, so as to predict the learner's future answering performance. In addition, the experimental results of the method of the present invention are compared with those of the traditional learner cognitive tracking method, using AUC, F1SCORE, RECALL and PRECISION as evaluation indicators. The comparison results are shown in Table 5.

[0187] Table 5 Experimental results

[0188] method AUC F1SCORE RECALL PRECISION Traditional learner cognitive tracking method 97.72％ 93.10％ 93.88％ 92.33％ This method 97.92％ 93.38％ 98.37％ 94.95％

[0189] Figure 4The figure is a comparative diagram of the experimental results provided by the embodiment of the present invention. The experimental results show that, compared with the traditional learner cognitive tracking method, the learner cognitive tracking method for learning process evaluation proposed by the present invention has an AUC, F1Score, Recall and Precision improved by 0.20%, 0.28%, 4.49% and 2.62% respectively. This shows that the present invention can better explore the learner's cognitive level at each knowledge point during the learning process, track the changes in the learner's cognitive level at different stages, and has a higher accuracy in predicting the learner's answer performance, which is more effective than the traditional learner cognitive tracking method.

[0190] like Figures 5 and 6 The experimental results shown in the figure indicate that the test question knowledge cognition tensor TKC and the multidimensional feature extraction and embedding method constructed by the learner cognitive tracking method for learning process evaluation provided by the embodiment of the present invention are effective for multi-feature information mining of learners and knowledge points, can fully focus on the additional features of knowledge points, and perform fine-grained embedding representation of knowledge points based on information of different dimensions. In addition, the cognitive tracking unit is constructed based on four gate structures: Slip Gate, Guess Gate, Level Gate, and Output Gate. A learner cognitive tracking model that can consider the influence of knowledge point parameters is designed. Modeling the learner's cognitive level helps to model the learner's cognitive level change process in both objective and subjective senses, simulate and track the learner's cognitive state and the parameter information of the knowledge point itself during the learning process. While updating the learner's overall cognitive level at each moment, the learner's hierarchical changes on specific knowledge points are mined and output in a fine-grained manner, thereby effectively predicting their answer performance on the corresponding knowledge points, and providing a more reasonable and explainable learning evaluation for the learner's learning process and learning results.

[0191] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with this technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered by the scope of protection of the present invention.

Claims

1. A learner cognitive tracking method for learning process evaluation, characterized by: include: The answer features of different dimensions are encoded separately and then fused and embedded to obtain a hierarchical embedding representation; Track and output the learner's cognitive level of each knowledge point at different times based on cognitive state information, predict the learner's future answer performance on the knowledge point, and make personalized learning resource recommendations for learners; The learner cognitive tracking method for learning process evaluation includes the following steps: Step 1: Define and construct the "exercise-knowledge-answer-cognitive hierarchy" quadruple: Collect learners' learning resources and answer data. Based on the test knowledge and cognitive tensor (TKC) based on Bloom's taxonomy of educational objectives in cognitive domains, extract features for each learner and generate the original quadruple of answer features. Step 2: Data preprocessing and multi-dimensional feature encoding, fusion, and embedding: Hyperparameters are set based on the length of the original quadruple of learner response features, and data is processed into a uniform length format. The learner response features from different dimensions contained in the original quadruple are encoded and fused to obtain the final hierarchical embedding representation of each knowledge point. Step 3: Construct a cognitive tracking model based on four gate structures: The model input is obtained by encapsulating the embedded representation in step 2, and a cognitive tracking unit is constructed based on the four gate structures of Slip Gate, Guess Gate, Level Gate, and Output Gate, thereby building a deep cognitive tracking model that integrates the gate structure. Step 4: Track and output the learner's cognitive level at different times and predict future learning performance: Based on the learner's cognitive state information and knowledge point parameter information during the tracking process, dynamically output the learner's cognitive level at each knowledge point at different times and predict the learner's future answer performance; The cognitive tracking model constructed based on the four gate structures in step 3 includes: (1) Encapsulate the embedding representation in step 2 to obtain the final model input x that integrates the knowledge point answer information and hierarchical information t ; Before proceeding to the next step of calculation, t Extract answer features r t and hierarchical features tkc t , define k t is the cognitive state vector of the learner at time t; r t =x t ⊙A; tkc t =x t ⊙B; Among them, A and B are extraction parameter matrices; (2) Based on the four gate structures, a cognitive tracking unit is constructed to build a deep cognitive tracking model that integrates the gate structure. Corresponding to the different levels of knowledge points and learners, the model sets up Slip Gate, Guess Gate, Level Gate, and Output Gate respectively. (3) Slip Gate and Guess Gate focus on the knowledge point level. Each knowledge point has a slip parameter and a guess parameter, which represent the probability that the learner learns the knowledge point but fails to reach the specified cognitive level, or the probability that the learner does not learn the knowledge point but explicitly reaches the specified cognitive level. s t =σ(U s r t +V s k t-1 +b s ); g t =σ(U g r t +V g k t-1 +b g ); Level Gate focuses on the learner level. When a learner correctly answers a knowledge point at a certain level, the learner's cognitive state of that knowledge point is no lower than the level at which the knowledge point was tested. Otherwise, the learner is temporarily considered to be at the lowest level of Bloom's Cognitive Domain Taxonomy of Educational Objectives. l t =σ(a t ⊙k t-1 ); Based on the traditional cognitive diagnosis method, the information from different levels obtained from Slip Gate, Guess Gate and Level Gate is integrated and updated to obtain the latent information about the learner's cognitive state; Output Gate responds based on the response information t , hierarchical information tkc t And the hidden information c t , determine the learner's final cognitive state k t ; the t =σ(U o r t +V o k t-1 +b o ); k t =o t ⊙tanh(c t +U tkc tkc t ); Among them, U s 、V s 、b s , U g 、V g 、b g , U o 、V o 、b o , U tkc is the model training parameter matrix, a t is the learner's actual answer to the knowledge point, t The value range is {0, 1}; ⊙ represents element-by-element multiplication, σ represents the sigmoid activation function, and tanh represents the tanh activation function.

2. The learner cognitive tracking method for learning process evaluation according to claim 1, characterized in that: The definition and construction of the "Practice - Knowledge - Answer - Cognitive Level" quadruple in step 1 includes: (1) Construct a matrix of learning resource collection and learner historical responses; S={s1,s2,…,s M }; E={e1,e2,…,e N }; K={k1,k2,…,k L }; Among them, S is the learner set, M is the number of learners; E is the test question set, N is the number of test questions; K is the knowledge point set, L is the number of knowledge points; R is the learner's historical answer matrix, r mn =0 means that learner m answered question n incorrectly, r mn =1 means that learner m answered question n correctly; (2) Bloom's taxonomy of educational objectives in the cognitive domain is introduced to divide learners' mastery of knowledge points into seven levels, and a specific and unified definition of the examination level of each knowledge point in the cognitive level interview questions is obtained, thereby constructing the test knowledge cognition tensor TKC that integrates Bloom's taxonomy of educational objectives in the cognitive domain; The seven cognitive levels are: not mastered 0, knowing 1, understanding 2, application 3, analysis 4, synthesis 5 and evaluation 6; 0≤c nl ≤6 means that question n tests knowledge points at levels l to c; (3) Based on the learner's historical answer matrix R and the test knowledge cognition tensor TKC that incorporates Bloom's cognitive domain education objective taxonomy, the learner's learning data is feature extracted, and the learner's answer record is processed into an answer feature quadruple R containing four types of information: practice, knowledge, answer, and cognitive level. s ; R s =(e s ,(to s ,r s ,tkc s )); Among them, e s is the interactive record of learner s in practice, k s is the ID sequence of all knowledge points answered by learner s in chronological order, r s k s The sequence of learner s’ answers corresponding to the knowledge point ID, tkc s k s The TKC level sequence of the knowledge point corresponding to the knowledge point ID is examined.

3. The learner cognitive tracking method for learning process evaluation according to claim 1, characterized in that: The data preprocessing and multi-dimensional feature encoding, fusion and embedding in step 2 include: (1) Set the hyperparameter MAX_STEP to process the original answer feature quadruple into a data format of equal length; Among them, k s The ID sequence of the knowledge points answered by learner s, r s The real situation sequence for learner s's answer, tkc s is the cognitive level sequence of the knowledge points that learner s answers; 0 is the zero vector, when R s e in s When it is an integer multiple of MAX_STEP, the length of the zero vector is 0, otherwise the length of the zero vector is MAX_STEP minus e s The remainder of dividing by MAX_STEP, C(·) represents data filling processing; the filled k s ,r s ,tkc s Reshape them into answer matrices with MAX_STEP as fixed column width to obtain the model input format of equal length; (2) For the learner's answer feature quadruple R s The feature information ID k from different dimensions of the knowledge point contained in s 、Right or Wrong s and inspection level tkc s , encoded separately and then embedded and spliced ​​to obtain the final hierarchical embedding representation of each knowledge point; Among them, e ID is the traditional one-hot encoding of the ID of knowledge point k, and the dimension of the 0 vector is the number of knowledge points L contained in the dataset, e r To encode the answer situation that incorporates the knowledge point ID feature, e tkc is the traditional one-hot encoding of the cognitive level of the knowledge point being examined, e k It is a multi-feature encoding vector that finally integrates knowledge point ID information, answer information, and cognitive level information.

4. The learner cognitive tracking method for learning process evaluation according to claim 1, characterized in that: Step 4 involves tracking and outputting the learner's cognitive level at different times and predicting their future learning performance. (1) According to the learner’s cognitive state information k obtained during the tracking process t , build a classification function to output the learner's specific cognitive level at different knowledge points at each moment; ct=V×k t ×M; Among them, level n is the maximum value of the cognitive level examined in the dataset, V and M are respectively the dimensions [L,1] and [L,level n ] is a two-dimensional matrix, cl is the output learner cognitive level; (2) Add a fully connected layer for mapping, matching the learner's cognitive state with the knowledge point answer one by one, and obtain the learner's future answer performance on each knowledge point; y t σ(Wk t +b) Among them, W is the weight coefficient matrix of the fully connected layer, and b is the bias coefficient matrix of the fully connected layer, which is obtained by training; (3) Calculate the loss function of the model, which is the output prediction value y t and the true label r t The cross entropy between loss CLKT =-∑(r t logy t +(1-r t )log(1-y t )); (4) Update all parameter matrices in the model according to the loss function value and the gradient descent rule.

5. A learner cognition tracking system for learning process evaluation using the learner cognition tracking method for learning process evaluation according to any one of claims 1 to 4, characterized in that: The learner cognitive tracking system for learning process evaluation includes: The multi-dimensional feature data extraction module is used to collect learners' learning resources and answer data. Based on the test knowledge cognition tensor TKC introduced in Bloom's taxonomy of educational objectives in the cognitive domain, it extracts features from learners' learning data and processes learners' answer records into original answer feature quadruples containing four types of information: practice, knowledge, answer, and cognitive level. The padding data processing module is used to set hyperparameters based on the length of the original quadruple of the learner's answer feature, and process the original quadruple of the answer feature into a data format of equal length; The multi-dimensional feature embedding and fusion module is used to encode the feature information ID, correct or incorrect answer, and assessment level from different dimensions of the knowledge point contained in the learner's answer feature quadruple, and then embed and splice them to obtain the final hierarchical embedding representation of each knowledge point; The deep cognitive tracking module is used to encapsulate the answer embedding representation to obtain the required model input. It constructs a cognitive tracking unit based on four gate structures: Slip Gate, Guess Gate, Level Gate, and Output Gate. This constructs a deep cognitive tracking model that integrates gate structures. It models and tracks the learner's cognitive state of different knowledge points during the learning process, as well as the parameter information of the knowledge points themselves. The cognitive level output module is used to dynamically output the learner's cognitive level at each knowledge point at different moments based on the learner's cognitive state information during the tracking process; The future performance prediction module is used to predict the learner's future performance on the knowledge point based on the learner's cognitive state information and knowledge point parameter information during the tracking process; The personalized resource recommendation module is used to recommend personalized learning resources to learners based on the learner's cognitive level information, knowledge point parameter information and test question knowledge cognition tensor TKC information during the tracking process.

6. A computer device, characterized in that: The computer device includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes the steps of the learner cognitive tracking method for learning process evaluation as described in any one of claims 1 to 4.

7. A computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to execute the steps of the learner cognitive tracking method for learning process evaluation according to any one of claims 1 to 4.

8. An information data processing terminal, characterized in that: The information data processing terminal is used to implement the learner cognitive tracking system for learning process evaluation as described in claim 5.

Images