A causal time series double-enhanced knowledge tracking method based on hypergraph clustering

Abstract

This invention discloses a causal temporal dual-enhanced knowledge tracing method based on hypergraph clustering, comprising: acquiring a dataset generated during online learning by learners and determining exercise interaction information; performing hypergraph clustering processing based on the exercise interaction information to construct a global knowledge dictionary; constructing a causal temporal dual-enhanced knowledge tracing model based on front-door adjustment logic and learning continuity based on the global knowledge dictionary; acquiring the current knowledge state and current exercise information of the object to be predicted, and generating a prediction result of the object's answer information based on the causal temporal dual-enhanced knowledge tracing model. This invention is highly efficient and can be widely applied in the field of computer technology.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, and particularly relates to a causal time sequence double-enhanced knowledge tracking method based on hypergraph clustering. BACKGROUND

[0002] Knowledge tracking is a modeling method for tracing the change of student knowledge state, and is the basis for realizing personalized learning path recommendation. The process data contains rich learning cognitive information, but due to the limitations of common sense logic and the learning behavior preferences, the data distribution has the characteristics of imbalance and continuity. Most of the existing mainstream knowledge tracking methods are developed in a single time sequence level, which is not sensitive to learning continuity and cannot adapt to the data extremely variable scene. SUMMARY

[0003] Therefore, the present application provides a high-efficiency causal time sequence double-enhanced knowledge tracking method based on hypergraph clustering.

[0004] In an aspect, the present application provides a causal time sequence double-enhanced knowledge tracking method based on hypergraph clustering, comprising:

[0005] acquiring a data set generated in the online learning process of a learner, and determining problem interaction information;

[0006] performing hypergraph clustering processing according to the problem interaction information, and constructing a global knowledge dictionary;

[0007] constructing a causal time sequence double-enhanced knowledge tracking model based on front door adjustment logic and learning continuity according to the global knowledge dictionary;

[0008] acquiring the current knowledge state and current problem information of a to-be-predicted object, and generating an answer information prediction result of the to-be-predicted object according to the causal time sequence double-enhanced knowledge tracking model.

[0009] Optionally, the data set generated in the online learning process of the learner is acquired, and the problem interaction information is determined,

[0010] comprising:

[0011] the student set, the problem set and the knowledge concept set generated in the online learning process of the learner are acquired;

[0012] a knowledge association matrix among the student set, the problem set and the knowledge concept set is constructed;

[0013] the answer response time information and the adjacent question interval time information of the learner are extracted according to the answer timestamp information of the learner;

[0014] Discretize the response time information and the adjacent question interval time information to obtain discretized time information;

[0015] Determine the exercise interaction information according to the knowledge association matrix and the discretized time information.

[0016] Optionally, the hypergraph clustering processing according to the exercise interaction information is performed to construct a global knowledge dictionary, including:

[0017] The relationship between a knowledge concept and a question is taken as a two-part graph, and a hyperedge weight of a vertex set contained in all hyperedges in the two-part graph is calculated according to preset support threshold and confidence threshold by using an association rule mining algorithm;

[0018] The hypergraph of the two-part graph is segmented according to a minimum weight edge to obtain a plurality of hypergraph subsets, and a plurality of data clusters are constructed;

[0019] According to the data clusters, each node is classified by an evaluation function between the node and the data cluster, and each node is assigned to a corresponding data cluster;

[0020] Each data cluster is taken as a clustering center for clustering analysis to obtain a final blocking result; wherein each block represents an interactive feature of a category;

[0021] All blocks are combined to form a global dictionary.

[0022] Optionally, the global knowledge dictionary is used to construct a causal and temporal double-enhanced knowledge tracking model based on a front door adjustment logic and learning continuity, including:

[0023] A causal enhancement module based on the front door adjustment logic, a temporal enhancement module based on the learning continuity, and a knowledge state prediction module are constructed according to the global knowledge dictionary;

[0024] A causal and temporal double-enhanced knowledge tracking model based on the front door adjustment logic and the learning continuity is constructed according to the causal enhancement module, the temporal enhancement module, and the knowledge state prediction module.

[0025] Optionally, the causal enhancement module based on the front door adjustment logic includes:

[0026] According to the characteristics of a learning object, an expression form of a knowledge tracking target after a causal intervention is determined;

[0027] Grid processing of the target: according to the expression form of the knowledge tracking target, a full connection network and an activation function layer are used for normalized weighted geometric average processing to perform grid processing on the knowledge tracking target;

[0028] Attentionalization of the local sampling process; based on the target expectation of local sampling, the local sampling process is subjected to attentionalization.

[0029] Attentionalization of the global sampling process: The global sampling process is subjected to attentionalization based on the content of the global sampling.

[0030] Fusion processing of dual sampling results: The results of local sampling and global sampling are fused to obtain a causal interaction embedding that removes confounding factors, thus completing the fusion of dual sampling results.

[0031] Optionally, the construction of the temporal enhancement module based on learning continuity includes:

[0032] Simulate the knowledge intake process: Multiply the knowledge association vectors related to the current problem in the knowledge association matrix with the knowledge state of the previous time step to obtain the associated knowledge state. Integrate causal interaction embedding and response time to perform feature diffusion in the time dimension, and construct an input gate for knowledge intake simulation.

[0033] Simulating the knowledge forgetting process: Integrating causal interactive embedding, relevant knowledge states and time intervals, and inputting the resulting vector into the forgetting gate to simulate the learner's forgetting process of mastered knowledge.

[0034] Optionally, the knowledge state prediction module includes:

[0035] The corresponding knowledge state is combined with the question, and the prediction result is generated after passing through a fully connected network and activating it with the sigmoid function.

[0036] Another aspect of this invention provides a causal temporal dual-enhancement knowledge tracing device based on hypergraph clustering, comprising:

[0037] The first module is used to acquire the dataset generated by learners during online learning and to determine the interactive information of exercises.

[0038] The second module is used to perform hypergraph clustering based on the exercise interaction information to construct a global knowledge dictionary;

[0039] The third module is used to construct a causal-temporal dual-enhanced knowledge tracing model based on the global knowledge dictionary and the front-door adjustment logic and learning continuity.

[0040] The fourth module is used to obtain the current knowledge status and current exercise information of the object to be predicted, and generate the prediction result of the answer information of the object to be predicted based on the causal time-series dual-enhanced knowledge tracing model.

[0041] Another aspect of the present invention provides an electronic device, including a processor and a memory;

[0042] The memory is used to store programs;

[0043] The processor executes the program to implement the method described above.

[0044] Another aspect of this invention provides a computer-readable storage medium storing a program that is executed by a processor to implement the methods described above.

[0045] This invention also discloses a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions, causing the computer device to perform the aforementioned method.

[0046] Embodiments of the present invention acquire datasets generated during online learning by learners and determine exercise interaction information; perform hypergraph clustering processing based on the exercise interaction information to construct a global knowledge dictionary; construct a causal-temporal dual-enhanced knowledge tracing model based on front-door adjustment logic and learning continuity based on the global knowledge dictionary; acquire the current knowledge state and current exercise information of the object to be predicted, and generate a prediction result of the object's answer information based on the causal-temporal dual-enhanced knowledge tracing model. The present invention is highly efficient. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a flowchart illustrating the overall steps of the present invention;

[0049] Figure 2 This is a schematic diagram illustrating the construction process of the knowledge tracking model provided in an embodiment of the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0051] To address the problems existing in the prior art, one aspect of this invention provides a causal temporal dual-enhancement knowledge tracing method based on hypergraph clustering, comprising:

[0052] Acquire datasets generated by learners during their online learning process to determine exercise interaction information;

[0053] Hypergraph clustering is performed based on the exercise interaction information to construct a global knowledge dictionary;

[0054] Based on the global knowledge dictionary, a causal-temporal dual-enhanced knowledge tracing model based on front-door adjustment logic and learning continuity is constructed.

[0055] Obtain the current knowledge state and current exercise information of the object to be predicted, and generate the prediction result of the answer information of the object to be predicted based on the causal-temporal dual-enhanced knowledge tracing model.

[0056] Optionally, the step of acquiring the dataset generated by learners during online learning and determining the exercise interaction information...

[0057] include:

[0058] The acquisition includes the student set, question set, and knowledge concept set generated by learners during their online learning process;

[0059] Construct a knowledge association matrix among the student set, the question set, and the knowledge concept set;

[0060] Based on the learner's answer timestamp information, extract the learner's answer response time information and the time interval between adjacent questions;

[0061] The response time information and the interval time information between adjacent questions are discretized to obtain discretized time information;

[0062] The exercise interaction information is determined based on the knowledge association matrix and the discretized time information.

[0063] Optionally, the step of performing hypergraph clustering based on the exercise interaction information to construct a global knowledge dictionary includes:

[0064] The relationship between knowledge concepts and questions is treated as a bipartite graph. An association rule mining algorithm is used to calculate the weight of each vertex set contained in the hyperedges of the bipartite graph according to a preset support threshold and confidence threshold.

[0065] The hypergraph of the bipartite graph is partitioned according to the edge with the minimum weight to obtain multiple hypergraph subsets and construct multiple data clusters;

[0066] Based on the data clusters, each node is classified using an evaluation function between the node and the data cluster, and each node is assigned to the corresponding data cluster.

[0067] Cluster analysis is performed using each data cluster as a cluster center to obtain the final block results; where each block represents the interaction features of a category;

[0068] Combine all blocks to form a global dictionary.

[0069] Optionally, the step of constructing a causal-temporal dual-enhanced knowledge tracing model based on front-door adjustment logic and learning continuity according to the global knowledge dictionary includes:

[0070] Based on the global knowledge dictionary, a causal enhancement module based on front-door adjustment logic, a temporal enhancement module based on learning continuity, and a knowledge state prediction module are constructed.

[0071] Based on the causal enhancement module, the temporal enhancement module, and the knowledge state prediction module, a causal-temporal dual-enhancement knowledge tracing model based on front-door adjustment logic and learning continuity is constructed.

[0072] Optionally, the construction of the causal enhancement module based on the front-door adjustment logic includes:

[0073] Based on the characteristics of the learners, determine the expression form of the knowledge tracking goals after using causal intervention;

[0074] Target gridding: Based on the expression form of the knowledge tracking target, the knowledge tracking target is gridded by normalized weighted geometric mean processing through a fully connected network and activation function layer;

[0075] Attentionalization of the local sampling process; based on the target expectation of local sampling, the local sampling process is subjected to attentionalization.

[0076] Attentionalization of the global sampling process: The global sampling process is subjected to attentionalization based on the content of the global sampling.

[0077] Fusion processing of dual sampling results: The results of local sampling and global sampling are fused to obtain a causal interaction embedding that removes confounding factors, thus completing the fusion of dual sampling results.

[0078] Optionally, the construction of the temporal enhancement module based on learning continuity includes:

[0079] Simulate the knowledge intake process: Multiply the knowledge association vectors related to the current problem in the knowledge association matrix with the knowledge state of the previous time step to obtain the associated knowledge state. Integrate causal interaction embedding and response time to perform feature diffusion in the time dimension, and construct an input gate for knowledge intake simulation.

[0080] Simulating the knowledge forgetting process: Integrating causal interactive embedding, relevant knowledge states and time intervals, and inputting the resulting vector into the forgetting gate to simulate the learner's forgetting process of mastered knowledge.

[0081] Optionally, the knowledge state prediction module includes:

[0082] The corresponding knowledge state is combined with the question, and the prediction result is generated after passing through a fully connected network and activating it with the sigmoid function.

[0083] Another aspect of this invention provides a causal temporal dual-enhancement knowledge tracing device based on hypergraph clustering, comprising:

[0084] The first module is used to acquire the dataset generated by learners during online learning and to determine the interactive information of exercises.

[0085] The second module is used to perform hypergraph clustering based on the exercise interaction information to construct a global knowledge dictionary;

[0086] The third module is used to construct a causal-temporal dual-enhanced knowledge tracing model based on the global knowledge dictionary and the front-door adjustment logic and learning continuity.

[0087] The fourth module is used to obtain the current knowledge status and current exercise information of the object to be predicted, and generate the prediction result of the answer information of the object to be predicted based on the causal time-series dual-enhanced knowledge tracing model.

[0088] Another aspect of the present invention provides an electronic device, including a processor and a memory;

[0089] The memory is used to store programs;

[0090] The processor executes the program to implement the method described above.

[0091] Another aspect of this invention provides a computer-readable storage medium storing a program that is executed by a processor to implement the methods described above.

[0092] This invention also discloses a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions, causing the computer device to perform the aforementioned method.

[0093] The specific implementation process of the present invention will now be described in detail with reference to the accompanying drawings:

[0094] This invention addresses the problem of single temporal dimension and high data requirements in current knowledge tracing modeling and analysis. Combining research on hypergraph clustering and front-door adjustment in causal inference, it first constructs a global dictionary and then constructs dual temporal channels. It proposes a causal temporal dual-enhanced knowledge state tracking method based on hypergraph clustering to solve the problem of poor model analysis performance under complex and extreme learning data conditions.

[0095] Combination Figure 1 The following will explain each step in detail:

[0096] 1. Processing of learning interaction data:

[0097] Learners generate a lot of data during online learning, the most obvious of which is usually: student collections. Problem Set With a set of knowledge concepts that encompass all problems Where N s N q N c Let h represent the number of students, questions, and knowledge concepts, respectively. Assume the student's knowledge state at time t is h. t This indicates his mastery of various knowledge points. The knowledge association matrix (QC) represents the relationship between questions and concepts, and is encoded in binary format. When question q... i Includes knowledge concept c k At that time, QC jk =1, otherwise QC jk =0. When a student answers a question, there will be a response time rt, which is the total time taken to complete the current question. There will be an interval time between answering two adjacent questions, which is the interval between adjacent interactions it. The above two time information are extracted based on the answer timestamp information and discretized in seconds.

[0098] Therefore, when a student is learning on an online platform, their historical interaction records can be obtained {(q1, c1, rt1, a1), it1, (q2, c2, rt2, a2), it2, ..., (q t c t ,rt t a t ), it t}, where a t ∈{0,1} indicates that the student answered whether q was correct at time t. t Problem. This invention first processes the interactive data into interactive embeddings X = {x1, ..., x...}. t The specific method can be described as follows:

[0099]

[0100] here This represents the concat operation, W x It is the weight matrix, b x It is a bias term.

[0101] 2. Construction of a knowledge tracing model based on front-door adjustment logic and learning continuity

[0102] 2.1 Model Structure: The construction process of the knowledge tracing model provided in this embodiment of the invention is as follows: Figure 2 As shown, the specific construction process is as follows:

[0103] 2.2 Global dictionary construction based on hypergraph-K clustering:

[0104] Knowledge tracing tasks typically determine the relevance between information by analyzing the relationships between learning interactions, and control the dynamic changes in the corresponding knowledge state. However, due to learners' self-directedness and the inherent properties of the exercises, the interactive information is often unbalanced and insufficient, and the prediction results are easily misled by spurious relevance. To eliminate this impact, this invention uses causal relationships to remove confusing information instead of relational relationships. The key to removing confusion from learning information lies in the collection of global information. Therefore, constructing a global dictionary based on hypergraph clustering can promote data balance and remove confounding effects, and also improve the system's computational efficiency.

[0105] Hypergraphization of Bipartite Graphs: The relationship between knowledge concepts and questions can be viewed as a bipartite graph, where one question corresponds to multiple knowledge concepts, and one knowledge concept corresponds to multiple questions. As an extension of graphs, hypergraphs have hyperedges that differ from the edges of ordinary graphs, allowing connection of multiple vertices. This allows for the representation of concepts C = {c1, ..., c...}. m} can be viewed as a hyperedge E = {e1, ..., e} m The problem is Q = {q1, ..., q}. n} can be viewed as vertices V = {v1, ..., v} n Then we can obtain the hypergraph H = {V, E}, where V is a finite set and E satisfies and The set of vertices contained in e is denoted as F(e).

[0106] Global dictionary construction based on hypergraph clustering:

[0107] Step 1: Calculate the weights of the hyperedges. Using an association rule mining algorithm, calculate the weights of the vertices F(e) contained in all hyperedges e according to the set support threshold minSup and confidence threshold minConfidence.

[0108] Step 2: Hypergraph Partitioning. Cut the hypergraph using the edge with the least weight, repeating this principle k times until k hypergraph subsets are obtained. The internal nodes of these k hypergraph branches have strong similarity, thus forming data clusters.

[0109] Step 3: Node Classification. The hypergraph is partitioned into k data clusters. A node's cluster affiliation is determined using an evaluation function between the node and its cluster. Here, an evaluation function is used. Where Node represents a node, and the data clusters obtained after hypergraph partitioning are represented by DC. i express.

[0110] Step 4: Using the k data clusters obtained after hypergraph segmentation as cluster centers for the K-means algorithm, perform cluster analysis to obtain the final block results. Each block represents the interaction features of a category. Combine all blocks to form a global dictionary X. GD This method uses K-means clustering to generate an approximate and deterministic global dictionary for all samples. This not only ensures that the acquisition of global information is within a controllable range, but also guarantees that the subsequent estimates in global and local sampling are based on the same distribution.

[0111] 2.2 Causal reinforcement module based on front-door adjustment logic:

[0112] To overcome the problem of false relevance affecting results due to the presence of the confounding factor of knowledge composition, this invention improves the self-attention mechanism by utilizing the front-door adjustment logic in causal inference, thereby uncovering deeper true causal relationships and obtaining interactive embeddings with enhanced causality.

[0113] Front-door adjustment mechanism: Based on the characteristics of learning, the knowledge tracking goal after causal intervention can be represented as P(Y|do(X)). Introducing front-door adjustment logic can formally decompose the goal into:

[0114]

[0115] Where m represents the cognitive ability that the attention mechanism focuses on, x′ represents the possible feature inputs, local sampling means that the sampled m only comes from the current interactive feature inputs, and global sampling means sampling from global information, that is, sampling all possible x′.

[0116] The networked construction of the front-door adjustment mechanism:

[0117] Step 1. Target Networking: To introduce the front-door adjustment mechanism into the attention mechanism, this invention, based on equation (2), parameterizes P(Y|M, X) through a fully connected network g(·). The last layer of the network is a Softmax activation function layer, and the idea of ​​Normalized Weighted Geometric Mean (NWGM) is used to reduce computational cost, specifically as follows:

[0118]

[0119]

[0120]

[0121] Step 2. Attentionalization of the Local Sampling Process: For the target expectation of local sampling, it can be regarded as a basic attention computation process, which can be described as:

[0122]

[0123] All of Q L K L V L All of these originate from the current input use case, i.e., the original set of interaction features, W Q W K W v Let A be the weight matrix. L These are attention weights, representing the network estimate for probability P(M|X).

[0124] Step 3. Attentionalization of the global sampling process: For the global sampling part, the attention logic of "query: Q, key: K, value: V" can also be used to estimate... This desired outcome can be described as:

[0125]

[0126] Among them, X GD Q is a global dictionary formed by compressing all the interaction data of all students. G From the current input sample, K G V G W is obtained from the global dictionary. Q , W k , W V , where A is the weight matrix. G These are attention weights, representing the network estimate for probability P(X′|X).

[0127] Step 4. Fusion of Dual Sampling Results: Based on the theoretical basis of front-door adjustment, the results of local sampling and global sampling are fused to obtain a causal interaction embedding free from obfuscation factors.

[0128]

[0129] in This represents a fusion operation, which involves first performing a splicing operation and then passing it through a single-layer fully connected network.

[0130] 2.3 Temporal Enhancement Module Based on Learning Continuity:

[0131] Considering that learners' cognitive abilities change continuously during the learning process, in order to overcome the shortcomings of quantitative and temporal changes in knowledge, this invention opens up a dual time-domain channel based on process-oriented time data to simulate the knowledge intake and forgetting process in the cognitive process, thereby obtaining a more reliable and interpretable data source for subsequent knowledge state prediction.

[0132] Step 1. Simulation of Knowledge Intake Process: The time it takes for students to acquire new knowledge typically fluctuates. Very short-term knowledge intake usually manifests as instantaneous memory, with only a temporary storage effect and little impact on the brain's knowledge state. Longer-term knowledge intake, however, leaves a knowledge imprint that is reflected in subsequent practice interactions. To further extract the correlation between answering time, an input gate is constructed for knowledge intake simulation. Its construction process is as follows: Each interaction usually corresponds to a different question, and the corresponding knowledge state update is also different. To ensure that the interaction at time t matches the existing knowledge state, the knowledge association vector QC related to the current question in the knowledge association matrix is ​​first... qt Knowledge state h from the previous moment t-1 Multiplication yields the associated knowledge state at time t-1, and the causal interaction embedding and response time are fused to perform feature diffusion in the time dimension. This process can be modeled as follows:

[0133]

[0134] Here, tanh represents a non-linear activation function. Let b be the weight matrix. li For the bias term, rt t This represents the response time when answering a question.

[0135] Since the acquired knowledge cannot fully influence the learner's knowledge state, an input gate was designed to control the ability to absorb knowledge.

[0136]

[0137] Here, σ represents the sigmoid non-linear activation function. This is the weight matrix. This is a bias term.

[0138] Step 2. Simulation of Knowledge Forgetting Process: When students are not engaged in interactive practice, their knowledge is constantly being lost, and the rate of loss varies depending on the interval between interactions. To extract the memory associations during interaction intervals, a forgetting gate is constructed to simulate the knowledge forgetting process. Its construction process is as follows: Causal interaction embeddings, relevant knowledge states, and interval times are integrated, and the resulting vector is input into the forgetting gate to simulate the learner's forgetting process of acquired knowledge.

[0139]

[0140] Here, σ represents the sigmoid non-linear activation function. Let b be the weight matrix. lf As a bias term, it t This represents the response time when answering a question.

[0141] η t with li t Multiplication yields the learner's actual knowledge input, and the forgetting factor is added to the learner's previous knowledge state to obtain the learner's new knowledge state h after the learning interaction at time t. t :

[0142] h t =lf t *h t -1+η t *li t #(12)

[0143] 2.4 Knowledge State Prediction Module:

[0144] In real learning scenarios, students face problem q t+1 When answering a question, the system uses relevant knowledge states based on the knowledge concepts associated with the question. Therefore, the prediction module first combines the relevant knowledge states with the question, then passes them through a fully connected network and activates them with a sigmoid function before outputting the result.

[0145]

[0146] Here W out It is the weight matrix, b out It's a bias term. The output y t+1 This indicates a prediction of the learner's performance on the next question q. t+1 The score ranges from (0, 1).

[0147] This invention chooses to use cross-entropy log loss to train the model, that is, to minimize the objective function of the predicted value versus the true value for each interaction:

[0148]

[0149] Where y t Let a be the predicted answer at time t. t This represents the actual answer to the question at time t.

[0150] Finally, the model training for knowledge state tracking: Each module involved in this invention is differentiable, therefore the entire knowledge tracking network model is also differentiable, and conventional deep learning optimization methods can be used to train the model proposed in this invention globally.

[0151] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this invention are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is altered and sub-operations described as part of a larger operation are executed independently.

[0152] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the described functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.

[0153] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0154] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0155] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0156] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0157] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0158] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0159] The above is a detailed description of the preferred embodiments of the present invention, but the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A causal temporal double-enhanced knowledge tracing method based on hypergraph clustering, characterized in that, The method comprises the following steps: acquiring a data set generated in the online learning process of a learner, and determining problem interaction information; performing hypergraph clustering processing according to the problem interaction information, and constructing a global knowledge dictionary; constructing a causal and temporal double-enhanced knowledge tracking model based on a front door adjustment logic and learning continuity according to the global knowledge dictionary; acquiring a current knowledge state and current problem information of a to-be-predicted object, and generating an answer information prediction result of the to-be-predicted object according to the causal and temporal double-enhanced knowledge tracking model; the step of acquiring the data set generated in the online learning process of the learner and determining the problem interaction information comprises: acquiring a student set, a problem set and a knowledge concept set generated in the online learning process of the learner; constructing a knowledge association matrix among the student set, the problem set and the knowledge concept set; extracting response time information and adjacent question interval time information of the learner according to a response timestamp information of the learner; performing discretization processing on the response time information and the adjacent question interval time information to obtain discretized time information; determining the problem interaction information according to the knowledge association matrix and the discretized time information; the step of constructing the causal and temporal double-enhanced knowledge tracking model based on the front door adjustment logic and the learning continuity according to the global knowledge dictionary comprises: constructing a causal enhancement module based on the front door adjustment logic, a temporal enhancement module based on the learning continuity and a knowledge state prediction module according to the global knowledge dictionary; constructing the causal and temporal double-enhanced knowledge tracking model based on the front door adjustment logic and the learning continuity according to the causal enhancement module, the temporal enhancement module and the knowledge state prediction module; the step of constructing the causal enhancement module based on the front door adjustment logic comprises: determining an expression form of a knowledge tracking target after a causal intervention according to characteristics of a learning object; grid processing of the target: performing normalized weighted geometric average processing on the knowledge tracking target through a full connection network and an activation function layer according to the expression form of the knowledge tracking target, and performing grid processing on the knowledge tracking target; attention processing of a local sampling process: performing attention processing on the local sampling process according to a target expectation of local sampling; attention processing of a global sampling process: performing attention processing on the global sampling process according to contents of global sampling; fusion processing of double sampling results: fusing results of local sampling and global sampling to obtain a causal interaction embedding in which confusion factors are removed, and completing fusion of the double sampling results; the step of constructing the temporal enhancement module based on the learning continuity comprises: simulating a knowledge intake process: multiplying a knowledge association vector related to a current question in the knowledge association matrix and a knowledge state at a previous time to obtain an associated knowledge state, fusing the causal interaction embedding and the response time to perform feature diffusion in a time dimension, and constructing an input door for knowledge intake simulation; simulating a knowledge forgetting process: fusing the causal interaction embedding, the associated knowledge state and the interval time, and inputting a vector obtained to a forgetting door to simulate a forgetting process of the learner on mastered knowledge.

2. The method of claim 1, wherein, the step of performing hypergraph clustering processing according to the problem interaction information and constructing the global knowledge dictionary comprises: The knowledge concept and the question are taken as a bipartite graph, a correlation rule mining algorithm is used, and the vertex set contained in all hyper-edges of the bipartite graph is subjected to hyper-edge weight calculation according to preset support threshold and confidence threshold; The hypergraph of the bipartite graph is segmented according to the minimum weight edge, a plurality of hypergraph subsets are obtained, and a plurality of data clusters are constructed; According to the data clusters, each node is classified through an evaluation function between the node and the data cluster, and each node is assigned to a corresponding data cluster; Each data cluster is taken as a clustering center for clustering analysis, and a final block result is obtained; wherein each block represents an interactive feature of a category; All blocks are combined to form a global dictionary.

3. The method of claim 1, wherein, A knowledge state prediction module is constructed, including: The corresponding knowledge state and the question are combined, a prediction result is generated after passing through a full connection network and being activated by a sigmoid function.

4. An apparatus for implementing the hypergraph clustering based causal temporal dual augmented knowledge tracing method according to any one of claims 1-3, characterized in that, Including: A first module is configured to obtain a data set generated in an online learning process of a learner and determine exercise interaction information; A second module is configured to perform hypergraph clustering processing according to the exercise interaction information and construct a global knowledge dictionary; A third module is configured to construct a causal and time-series double-enhanced knowledge tracking model based on a front-door adjustment logic and learning continuity according to the global knowledge dictionary; A fourth module is configured to obtain a current knowledge state and current exercise information of a to-be-predicted object, and generate a prediction result of answer information of the to-be-predicted object according to the causal and time-series double-enhanced knowledge tracking model.

5. An electronic device, comprising: A processor and a memory are included; The memory is configured to store a program; The processor executes the program to implement the method in any one of claims 1 to 3.

6. A computer readable storage medium characterized by The storage medium stores a program, and the program is executed by the processor to implement the method in any one of claims 1 to 3.

Images