Teaching resource intelligent management system and method based on big data

Abstract

The invention discloses a teaching resource intelligent management system and method based on big data, and relates to the technical field of teaching resource intelligent management. Constructing a teaching resource arrangement model based on the mastery index of the micro knowledge points; solving the teaching resource arrangement model by adopting a two-stage heuristic decomposition algorithm to generate an optimized learning arrangement sequence; sequentially pushing the assembled micro learning units to a user terminal in a streaming manner according to the sequence of the optimized learning arrangement sequence, and obtaining process interaction data of a user; extracting process interaction data to analyze answer accuracy and response time stability; updating the total mastery index according to the two-dimensional normal probability model; and triggering re-optimization of the subsequent optimization learning arrangement sequence based on the updated total mastery index. Instant refreshing of the state is realized; the analysis efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent management technology for teaching resources, specifically a big data-based intelligent management system and method for teaching resources. Background Technology

[0002] Smart education is a new educational model that deeply integrates educational informatization or digitalization technologies with teaching and learning to facilitate the allocation of educational resources, teaching support, and the high-quality and balanced development of education. Its core lies in utilizing technologies such as artificial intelligence and big data to optimize the entire process of teaching, learning, assessment, and management, driving the transformation of educational services from standardized to personalized, and from experience-driven to data-driven.

[0003] To achieve adaptability in teaching, existing solutions typically incorporate technologies such as multi-dimensional behavioral perception and natural language processing for real-time assessment during student learning, and use complex matching algorithms to globally adjust pre-set learning plans. This adjustment model often suffers from sluggish response; assessment and adjustment occur between learning stages rather than during the learning process, failing to achieve true dynamic accompaniment. Furthermore, the granularity of resource adjustments is often limited to complete courses or large modules, lacking flexibility and making it difficult to instantly reorganize subsequent learning sequences and resources based on users' immediate feedback on specific knowledge points.

[0004] Therefore, this invention discloses an intelligent management system and method for teaching resources based on big data to solve the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent management system and method for teaching resources based on big data, so as to solve the problems raised in the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for intelligent management of teaching resources based on big data, the method comprising the following steps:

[0007] S1: Collect the event stream generated by the interaction between the target user and the interactive learning interface; analyze the mutation behavior in the event stream and extract the learning interval; analyze the mastery index of each micro-knowledge point based on the operation events within the learning interval;

[0008] S2: Based on the mastery index of micro-knowledge points, a teaching resource arrangement model is constructed; a two-stage heuristic decomposition algorithm is used to solve the teaching resource arrangement model to generate an optimized learning arrangement sequence;

[0009] S3: Assemble micro-learning units for each micro-knowledge point in the optimized learning orchestration sequence; sequentially push the assembled micro-learning units to the user terminal in the order of the optimized learning orchestration sequence, and obtain the user's process interaction data.

[0010] S4: Extract process interaction data to analyze answer accuracy and response time stability; update the overall mastery index based on the two-dimensional normal probability model; trigger re-optimization of subsequent learning orchestration sequences based on the updated overall mastery index.

[0011] According to the above scheme, S1 includes the following:

[0012] S101: With authorization from the target user, collect the event stream generated by the interaction between the target user and the interactive learning interface, E raw ={(t) i a i r i )|i∈[1,I]}, where t i To represent the timestamp of the i-th event, a i Let r be the operation type for the i-th event. i The identifier of the micro-knowledge point associated with the interactive learning interface content when the i-th event triggers an operation; I represents the total number of event streams; a cognitive complexity weight is preset for each micro-knowledge point; the operation types include, but are not limited to, clicking, dragging, and answering;

[0013] S102: The collected event stream is scanned using a sliding time window, and the boundary points in the event stream where the rate of change of operation frequency exceeds a preset transformation rate threshold are identified using a mutation point detection algorithm; the event stream is divided into learning intervals based on the boundary points; and the corresponding operation features are extracted for each learning interval. The operation features include the average operation frequency in the learning interval, the distribution vector of operation type, and a set of micro-knowledge points associated with all event streams in the learning interval.

[0014] S103: For each micro-knowledge point in the overall set of micro-knowledge points, extract all learning intervals containing the same micro-knowledge point, and analyze the ideal similarity of each extracted learning interval. The ideal similarity is equal to the similarity between the distribution vector of the operation type in the learning interval and the set ideal learning mode vector. The product of the ideal similarity, cognitive complexity weight, average operation frequency in the learning interval, and answer accuracy in the learning interval is recorded as the mastery index of the micro-knowledge point in the corresponding learning interval. The sum of the mastery indices of the same micro-knowledge point in all learning intervals is recorded as the total mastery index. The total mastery index is normalized, and the total mastery indices of all micro-knowledge points are arranged in a preset order to generate the user's dynamic knowledge state vector.

[0015] This invention employs a sliding time window and a mutation point detection algorithm to automatically divide learning intervals, avoiding the problem that fixed time windows may disrupt the continuous learning process, making interval division more in line with the actual cognitive rhythm; it transforms continuous event flows into discrete and meaningful learning intervals, reducing the complexity of subsequent feature extraction and calculation; by comprehensively considering operation frequency, ideal similarity, cognitive complexity, and answer accuracy, it calculates the mastery index, so that the mastery assessment of a single knowledge point no longer relies solely on test scores, but integrates process behavioral data, resulting in more robust results and stronger resistance to interference.

[0016] According to the above scheme, S2 includes the following:

[0017] S201: Let k denote the j-th micro-knowledge point in the total set of micro-knowledge points. j If we consider the micro-knowledge points k j If the corresponding learning unit is arranged at the m-th position of the learning sequence, then the decision variable x （j，m） =1; otherwise, the decision variable x （j，m） =0; Construct the teaching resource arrangement model F(j,m): ;

[0018] Among them, b j Represents micro-knowledge point k j The preset basic importance weights; t0 represents the current time; c j (t0) represents the micro-knowledge point k at the current time. j The overall mastery index; γ represents the position discount factor, which is a system preset constant; M represents the total number of positions in the learning sequence;

[0019] Constraints are imposed on the instructional resource arrangement model, including preorder constraints, cognitive load constraints, time budget constraints, and forgetting curve constraints; the preorder constraints are minor knowledge points k. a It is a micro-knowledge point k b The prerequisite knowledge points must be met when learning to arrange sequences. Where m' represents the position number in the learning sequence; the cognitive load constraint is the sum of the cognitive complexity weights of any consecutive positions in the learning sequence, which does not exceed the cognitive complexity threshold; the time budget constraint is the sum of the estimated learning times of all micro-knowledge points in the learning sequence, which does not exceed the time threshold; the forgetting curve constraint is that when the total mastery index is lower than the mastery threshold and the time interval from the last learning interval exceeds the time interval threshold, a review is arranged in the learning sequence.

[0020] S202: A two-stage heuristic decomposition algorithm is used to solve the teaching resource arrangement model to generate the final learning arrangement sequence;

[0021] In the first stage, under the premise of satisfying the prior order relationship and time budget constraints, a greedy algorithm is used to select the knowledge points with the highest expected return per unit time to form an initial set of knowledge points. The expected return per unit time is equal to the difference between the preset basic importance weight multiplied by one and the total mastery index of the micro-knowledge points at the current moment, and then divided by the corresponding estimated learning time.

[0022] In the second stage, the initial set of knowledge points is sorted to obtain an initial sequence that satisfies all constraints; a neighborhood operation rule is preset, which includes: swapping the positions of two adjacent micro-knowledge points in the sequence that have no direct preorder constraints, or shifting a micro-knowledge point and all sub-blocks consisting of direct and indirect subsequent dependent knowledge points one position forward or backward in the sequence.

[0023] Starting with the initial sequence, an iterative optimization is performed using a metaheuristic search strategy. Each step of the iterative optimization includes: generating one or more candidate sequences based on the current sequence by applying the neighborhood operation rules; checking whether each candidate sequence satisfies the cognitive load constraint; calculating the total expected learning benefit value for candidate sequences that satisfy the cognitive load constraint according to the teaching resource arrangement model; selecting a sequence from the checked candidate sequences as the current sequence for the next iteration based on the adopted metaheuristic strategy; terminating the search when the preset iteration termination condition is met, and outputting the optimized learning arrangement sequence as the sequence that satisfies all constraints and has the highest total expected learning benefit value during the iteration process.

[0024] This invention addresses the challenges of high model complexity and difficulty in obtaining accurate solutions by proposing a two-stage algorithm combining greedy selection and metaheuristic search. The first stage (greedy selection) quickly selects a set of knowledge points with a high benefit-cost ratio, narrowing the search space and improving efficiency. The second stage (local search) iteratively optimizes the algorithm by defining reasonable neighborhood operations, effectively escaping local optima and approaching the global optimum, while satisfying complex constraints (especially nonlinear cognitive load constraints). This algorithm can generate high-quality, feasible solutions to complex instructional scheduling problems within an acceptable timeframe, meeting the real-time or near-real-time response requirements of online learning systems.

[0025] According to the above scheme, S3 includes the following:

[0026] S301: For each micro-knowledge point in the optimized learning orchestration sequence, retrieve the explanatory resources that match the micro-knowledge point; based on the user's latest dynamic knowledge state vector, call the exercise set generator to generate an exercise set; integrate the retrieved explanatory resources, exercise set, and pre-set feedback explanation templates, and package them into a micro-learning unit;

[0027] S302: The assembled micro-learning units are pushed to the user terminal in sequence according to the optimized learning orchestration sequence, and the user's procedural interaction data is obtained; the procedural interaction data includes, but is not limited to, the response time sequence of the answer, the sequence of answers submitted by the user, and additional operation events.

[0028] This invention does not push fixed course packages, but rather retrieves explanation resources in real time, generates adaptive practice questions, and packages them into units based on specific knowledge points in the sequence and the user's latest status. This decouples learning path planning from learning content presentation, greatly enhancing system flexibility; the collection of process interaction data not only focuses on the correctness of the final answer but also on the answering process, enabling the uncovering of deeper cognitive state information.

[0029] According to the above scheme, S4 includes the following:

[0030] S401: Extract the response time sequence of the responses and the sequence of answers submitted by the user to analyze the answer accuracy and response time stability; construct a two-dimensional normal probability model based on preset parameters to describe the performance distribution of users in two states: mastery and non-mastery; wherein, the two-dimensional normal probability model presets a mean vector and a covariance matrix for each state; input the answer accuracy and response time stability into the two-dimensional normal probability model, and calculate the first likelihood value of the procedural interaction data in the mastery state and the second likelihood value in the non-mastery state respectively; the product of the first likelihood value and the total mastery index is recorded as the first update index; the product of the second likelihood value and the difference between the first and the total mastery index is recorded as the second update index; the first update index is divided by the sum of the first update index and the first update index to record the updated total mastery index;

[0031] S402: Based on the updated total mastery index, the total mastery index of the related micro-knowledge points is disseminated and updated; the dissemination and update of the total mastery index of the related micro-knowledge points is equal to the product of the updated total mastery index and the preset update coefficient plus the historical total mastery index; if the absolute value of the difference between the dissemination and update of the total mastery index of the related micro-knowledge points and the historical total mastery index is greater than the difference ratio threshold, the optimized learning arrangement sequence is reordered according to the dissemination and update of the total mastery index.

[0032] This invention's two-dimensional normal model simultaneously considers accuracy and response time stability, enabling a more nuanced distinction between correct guesses and true knowledge. The Bayesian update is incremental, computationally inexpensive, and suitable for immediate updates after a user completes a small exercise, achieving near real-time state updates. Compared to predictions based solely on historical data, updates incorporating the latest performance evidence are more accurate and timely. It simulates the interconnectedness of knowledge structures, including how mastering a basic concept may enhance understanding of multiple related advanced concepts, making the system's perception of user states more aligned with the inherent connections within knowledge. By setting a "difference ratio threshold" as a trigger condition, the system only initiates computationally expensive reordering when the user's state changes significantly, avoiding unnecessary frequent adjustments and achieving a good balance between response sensitivity and system overhead. This ensures the learning path dynamically adapts to the user's latest progress or encountered difficulties, maintaining a consistently high level of personalization.

[0033] Another aspect of this application provides a big data-based intelligent management system for teaching resources. The system is applied to the above-mentioned big data-based intelligent management method for teaching resources. The system includes a knowledge status analysis module, a learning sorting module, an encapsulation interaction module, and an update and optimization module.

[0034] The knowledge state analysis module is used to collect the event stream generated by the interaction between the target user and the interactive learning interface; analyze the mutation behavior in the event stream and extract the learning interval; and analyze the mastery index of each micro-knowledge point based on the operation events within the learning interval.

[0035] The learning sorting module is used to construct a teaching resource arrangement model based on the mastery index of micro-knowledge points; a two-stage heuristic decomposition algorithm is used to solve the teaching resource arrangement model to generate an optimized learning arrangement sequence.

[0036] The encapsulation and interaction module is used to assemble micro-learning units for each micro-knowledge point in the optimized learning orchestration sequence; the assembled micro-learning units are then pushed sequentially to the user terminal in the order of the optimized learning orchestration sequence, and the user's process interaction data is obtained.

[0037] The update and optimization module is used to extract process interaction data to analyze the accuracy of answers and the stability of response time; update the total mastery index according to the two-dimensional normal probability model; and trigger the re-optimization of subsequent optimization learning orchestration sequences based on the updated total mastery index.

[0038] According to the above scheme, the knowledge state analysis module includes an operation feature analysis unit and a dynamic knowledge state vector analysis unit;

[0039] The operation feature analysis unit is used to collect the event stream generated by the interaction between the target user and the interactive learning interface with the authorization of the target user; identify the boundary points in the event stream where the rate of change of operation frequency exceeds a preset transformation rate threshold; divide the event stream into learning intervals based on the boundary points; and extract the corresponding operation features for each learning interval.

[0040] The dynamic knowledge state vector analysis unit is used to extract all learning intervals containing the same micro-knowledge point for each micro-knowledge point in the total set of micro-knowledge points, and analyze the ideal similarity of each extracted learning interval. The product of ideal similarity, cognitive complexity weight, average operation frequency within the learning interval, and answer accuracy within the learning interval is recorded as the mastery index of the micro-knowledge point in the corresponding learning interval. The sum of the mastery indices of the same micro-knowledge point in all learning intervals is recorded as the total mastery index. The total mastery index is normalized, and the total mastery indices of all micro-knowledge points are arranged in a preset order to generate the user's dynamic knowledge state vector.

[0041] According to the above scheme, the learning sorting module includes a teaching resource arrangement model construction unit and a model solving unit;

[0042] The teaching resource arrangement model construction unit is used to define the decision variables of micro-knowledge points, construct the teaching resource arrangement model based on the preset basic importance weight, total mastery index, decision variables and position discount factor, and impose constraints on the teaching resource arrangement model.

[0043] The model solving unit is used to solve the teaching resource arrangement model using a two-stage heuristic decomposition algorithm to generate the final learning arrangement sequence.

[0044] According to the above scheme, the encapsulation interaction module includes an encapsulation unit and an interaction data acquisition unit;

[0045] The encapsulation unit is used to retrieve explanatory resources that match each micro-knowledge point in the optimized learning orchestration sequence; to call the exercise set generator to generate an exercise set based on the user's latest dynamic knowledge state vector; and to integrate the retrieved explanatory resources, exercise set, and preset feedback explanation templates to form a micro-learning unit.

[0046] The method is used to push the assembled micro-learning units to the user terminal in the order of the optimized learning orchestration sequence, and to obtain the user's procedural interaction data; the procedural interaction data includes, but is not limited to, the response time sequence of the answer, the sequence of answers submitted by the user, and additional operation events.

[0047] According to the above scheme, the update and optimization module includes a total mastery index update unit and a sorting optimization unit;

[0048] The overall mastery index update unit is used to extract the response time sequence of the answers and the sequence of answers submitted by the user to analyze the answer accuracy and response time stability; construct a two-dimensional normal probability model based on preset parameters, input the answer accuracy and response time stability into the two-dimensional normal probability model, calculate the first likelihood value of the process interaction data in the mastery state and the second likelihood value in the non-mastery state respectively; and analyze and update the overall mastery index based on the first likelihood value and the second likelihood value.

[0049] The sorting optimization unit is used to propagate and update the total mastery index of related micro-knowledge points based on the updated total mastery index; if the absolute value of the difference between the propagated updated total mastery index of related micro-knowledge points and the historical total mastery index is greater than the difference ratio threshold, the optimized learning arrangement sequence is re-sorted according to the propagated updated total mastery index.

[0050] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention automatically divides learning intervals using a sliding time window and a mutation point detection algorithm, avoiding the problem that a fixed time window might disrupt the continuous learning process, making interval division more consistent with the actual cognitive rhythm; it transforms continuous event flows into discrete and meaningful learning intervals, reducing the complexity of subsequent feature extraction and computation; by comprehensively considering operation frequency, ideal similarity, cognitive complexity, and answer accuracy, it calculates the mastery index, ensuring that the mastery assessment of a single knowledge point no longer relies solely on test scores but integrates process-oriented behavioral data, resulting in more robust results and stronger anti-interference capabilities; this invention addresses the problems of high model complexity and difficulty in accurate solution. This paper proposes a two-stage algorithm combining greedy selection and metaheuristic search. The first stage quickly filters out a set of knowledge points with a high benefit-cost ratio, narrowing the search space and improving efficiency. The second stage, under complex constraints, iteratively optimizes through defined neighborhood operations, effectively escaping local optima and approaching the global optimum. This algorithm can generate high-quality, feasible solutions to complex instructional scheduling problems within an acceptable timeframe, meeting the real-time or near-real-time response requirements of online learning systems. This invention does not push fixed course packages but rather retrieves explanation resources, generates adaptive practice questions, and packages them into units in real time based on specific knowledge points in the sequence and the user's latest status. It decouples learning path planning from learning content presentation, greatly enhancing system flexibility. The collection of procedural interaction data not only focuses on the correctness of the final answer but also on the answering process, enabling the discovery of deeper cognitive state information. This invention's two-dimensional normal model simultaneously considers accuracy and response time stability, enabling a more nuanced distinction between correct guesses and true knowledge. The Bayesian update is incremental, computationally inexpensive, and suitable for immediate updates after a user completes a small exercise, achieving near real-time state updates. Compared to predictions based solely on historical data, updates incorporating the latest performance evidence are more accurate and timely. It simulates the interconnectedness of knowledge structures, making the system's understanding of user states more aligned with the inherent relationships within the knowledge base. By setting a "difference ratio threshold" as a trigger, the system only initiates a computationally expensive reordering process when the user's state changes significantly, avoiding unnecessary frequent adjustments and achieving a good balance between response sensitivity and system overhead. This ensures the learning path dynamically adapts to the user's latest progress or encountered difficulties, maintaining a consistently high level of personalization. Attached Figure Description

[0051] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0052] Figure 1 This is a flowchart illustrating an intelligent management method for teaching resources based on big data, as described in this invention.

[0053] Figure 2 This is a schematic diagram of the structure of an intelligent management system for teaching resources based on big data according to the present invention. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] Please see Figure 1 This invention provides a technical solution: an intelligent management method for teaching resources based on big data, which includes the following steps:

[0056] S1: Collect the event stream generated by the interaction between the target user and the interactive learning interface; analyze the mutation behavior in the event stream and extract the learning interval; analyze the mastery index of each micro-knowledge point based on the operation events within the learning interval;

[0057] S1 includes the following:

[0058] S101: With authorization from the target user, collect the event stream generated by the interaction between the target user and the interactive learning interface, E raw ={(t) i a i r i )|i∈[1,I]}, where t i To represent the timestamp of the i-th event, a i Let r be the operation type for the i-th event. i The identifier of the micro-knowledge point associated with the interactive learning interface content when the i-th event is triggered; I represents the total number of events in the event stream; a cognitive complexity weight is preset for each micro-knowledge point; the operation type includes, but is not limited to, clicking, dragging, and answering;

[0059] S102: The collected event stream is scanned using a sliding time window, and the boundary points in the event stream where the rate of change of operation frequency exceeds a preset transformation rate threshold are identified using a mutation point detection algorithm; the event stream is divided into learning intervals based on the boundary points; and the corresponding operation features are extracted for each learning interval. The operation features include the average operation frequency in the learning interval, the distribution vector of operation type, and a set of micro-knowledge points associated with all event streams in the learning interval.

[0060] Example 1: In this example, the predefined operation types are: {click, play, pause, answer, note, drag, ask for help, mark}; and are numbered according to the above order.

[0061] In the learning interval, the event type sequence is: [click, play, click, answer, pause, click, answer, note]; the statistics show that there were 3 clicks, 1 play, 1 pause, 2 answers, and 1 note; the total number of events is 8.

[0062] The distribution vector of operation types: [3 / 8, 1 / 8, 1 / 8, 2 / 8, 1 / 8, 0, 0, 0] = [0.375, 0.125, 0.125, 0.25, 0.125, 0, 0, 0];

[0063] S103: For each micro-knowledge point in the overall set of micro-knowledge points, extract all learning intervals containing the same micro-knowledge point, and analyze the ideal similarity of each extracted learning interval. The ideal similarity is equal to the similarity between the distribution vector of the operation type in the learning interval and the set ideal learning mode vector. The product of the ideal similarity, cognitive complexity weight, average operation frequency in the learning interval, and answer accuracy in the learning interval is recorded as the mastery index of the micro-knowledge point in the corresponding learning interval. The sum of the mastery indices of the same micro-knowledge point in all learning intervals is recorded as the total mastery index. The total mastery index is normalized, and the total mastery indices of all micro-knowledge points are arranged in a preset order to generate the user's dynamic knowledge state vector.

[0064] S2: Based on the mastery index of micro-knowledge points, a teaching resource arrangement model is constructed; a two-stage heuristic decomposition algorithm is used to solve the teaching resource arrangement model to generate an optimized learning arrangement sequence;

[0065] S2 includes the following:

[0066] S201: Let k denote the j-th micro-knowledge point in the total set of micro-knowledge points. j If we consider the micro-knowledge points k j If the corresponding learning unit is arranged at the m-th position of the learning sequence, then the decision variable x （j，m） =1; otherwise, the decision variable x （j，m） =0; Construct the teaching resource arrangement model F(j,m): ;

[0067] Among them, b j Represents micro-knowledge point k j The preset basic importance weights; t0 represents the current time; c j (t0) represents the micro-knowledge point k at the current time. j The overall mastery index; γ represents the position discount factor, which is a system preset constant; M represents the total number of positions in the learning sequence;

[0068] Constraints are imposed on the instructional resource arrangement model, including preorder constraints, cognitive load constraints, time budget constraints, and forgetting curve constraints; the preorder constraint is a minor knowledge point k. a It is a micro-knowledge point k b The prerequisite knowledge points must be met when learning to arrange sequences. Where m' represents the position number in the learning sequence; the cognitive load constraint is that the sum of the cognitive complexity weights of any consecutive positions in the learning sequence does not exceed the cognitive complexity threshold; the time budget constraint is that the sum of the estimated learning time of all micro-knowledge points in the learning sequence does not exceed the time threshold; the forgetting curve constraint is that when the total mastery index is lower than the mastery threshold and the time interval since the last learning interval exceeds the time interval threshold, a review is arranged in the learning sequence.

[0069] S202: A two-stage heuristic decomposition algorithm is used to solve the teaching resource arrangement model to generate the final learning arrangement sequence;

[0070] In the first stage, under the premise of satisfying the prior order relationship and time budget constraints, a greedy algorithm is used to select the knowledge points with the highest expected return per unit time to form the initial set of knowledge points. The expected return per unit time is equal to the difference between the preset basic importance weight multiplied by one and the total mastery index of the micro-knowledge points at the current moment, and then divided by the corresponding estimated learning time.

[0071] In the second stage, the initial set of knowledge points is sorted to obtain an initial sequence that satisfies all constraints. Preset neighborhood operation rules include: swapping the positions of two adjacent micro-knowledge points in the sequence that have no direct preorder constraints, or shifting a micro-knowledge point and all its directly and indirectly dependent sub-blocks one position forward or backward in the sequence.

[0072] Starting with the initial sequence, an iterative optimization is performed using a metaheuristic search strategy. Each step of the iterative optimization includes: generating one or more candidate sequences based on the current sequence by applying neighborhood operation rules; checking whether each candidate sequence satisfies the cognitive load constraint; calculating the total expected learning benefit value for candidate sequences that satisfy the cognitive load constraint according to the teaching resource arrangement model; selecting a sequence from the checked candidate sequences as the current sequence for the next iteration based on the adopted metaheuristic strategy; terminating the search when the preset iteration termination condition is met, and outputting the optimized learning arrangement sequence as the sequence that satisfies all constraints and has the highest total expected learning benefit value during the iteration process.

[0073] Example 2: In this example, the initial knowledge point set contains micro-knowledge points k1, k2, and k3; the preset basic importance weight b1 of micro-knowledge point k1 is 10; the total mastery index c1(t0) of micro-knowledge point k1 at the current time is 0.3; the preset basic importance weight b2 of micro-knowledge point k2 is 8; the total mastery index c2(t0) of micro-knowledge point k2 at the current time is 0.5; the preset basic importance weight b3 of micro-knowledge point k3 is 6; the total mastery index c3(t0) of micro-knowledge point k3 at the current time is 0.7; the position discount factor γ=0.9; in this example, any order of micro-knowledge points k1, k2, and k3 satisfies the preorder constraint, cognitive load constraint, and time budget constraint;

[0074] For permutation scheme 1: [k1 k2 k3]; Decision variable value: x （1，1） =1, x （2，2） =1, x （3，3） =1; (the rest are 0); therefore, the expected learning benefit of a single micro-knowledge point k1 is: b1×(1-c1(t0))×x （1，1） ×γ 1-1 =7;

[0075] Similarly, the expected learning benefit per item for k2 is calculated to be 3.6; the expected learning benefit per item for k3 is calculated to be 1.458; the total expected learning benefit calculated by the teaching resource arrangement model is 12.058.

[0076] For permutation scheme 2: [k1 k3k2]; the expected learning benefit of micro-knowledge point k1 is 7; the expected learning benefit of micro-knowledge point k3 is 1.62; the expected learning benefit of micro-knowledge point k2 is 3.24; the total expected learning benefit calculated by the teaching resource arrangement model is 11.86.

[0077] For permutation scheme 3: [k2 k1 k3]; the expected learning benefit of micro-knowledge point k2 is 4; the expected learning benefit of micro-knowledge point k1 is 6.3; the expected learning benefit of micro-knowledge point k3 is 1.458; the total expected learning benefit calculated by the teaching resource arrangement model is 11.758.

[0078] For permutation scheme 4: [k2 k3 k1]; the expected learning benefit of micro-knowledge point k2 is 4; the expected learning benefit of micro-knowledge point k3 is 1.62; the expected learning benefit of micro-knowledge point k1 is 5.67; the total expected learning benefit calculated by the teaching resource arrangement model is 11.29.

[0079] For permutation scheme 5: [k3 k1 k2]; the expected learning benefit of micro-knowledge point k3 is 1.8; the expected learning benefit of micro-knowledge point k1 is 6.3; the expected learning benefit of micro-knowledge point k2 is 3.24; the total expected learning benefit calculated by the teaching resource arrangement model is 11.34.

[0080] For permutation scheme 5: [k3 k2 k1]; the expected learning benefit of micro-knowledge point k3 is 1.8; the expected learning benefit of micro-knowledge point k2 is 3.6; the expected learning benefit of micro-knowledge point k1 is 5.67; the total expected learning benefit calculated by the teaching resource arrangement model is 11.07.

[0081] Therefore, in this embodiment, the sequence with the highest total expected learning return value is [k1 k2 k3] for permutation scheme 1;

[0082] S3: Assemble micro-learning units for each micro-knowledge point in the optimized learning orchestration sequence; sequentially push the assembled micro-learning units to the user terminal in the order of the optimized learning orchestration sequence, and obtain the user's process interaction data.

[0083] S3 includes the following:

[0084] S301: For each micro-knowledge point in the optimized learning orchestration sequence, retrieve the explanatory resources that match the micro-knowledge point; based on the user's latest dynamic knowledge state vector, call the exercise set generator to generate an exercise set; integrate the retrieved explanatory resources, exercise set, and pre-set feedback explanation templates, and package them into a micro-learning unit;

[0085] S302: Push the assembled micro-learning units to the user terminal in the order of the optimized learning orchestration sequence, and obtain the user's procedural interaction data; the procedural interaction data includes, but is not limited to, the response time sequence of the answer, the sequence of answers submitted by the user, and additional operation events.

[0086] S4: Extract process interaction data to analyze answer accuracy and response time stability; update the overall mastery index based on the two-dimensional normal probability model; trigger re-optimization of subsequent learning orchestration sequences based on the updated overall mastery index.

[0087] S4 includes the following:

[0088] S401: Extract the response time series of responses and the sequence of answers submitted by users to analyze the accuracy of answers and the stability of response time; construct a two-dimensional normal probability model based on preset parameters to describe the performance distribution of users in the states of mastery and non-mastery; wherein, the two-dimensional normal probability model presets a mean vector and a covariance matrix for each state; input the accuracy of answers and the stability of response time into the two-dimensional normal probability model, and calculate the first likelihood value of the procedural interaction data in the mastery state and the second likelihood value in the non-mastery state respectively; the product of the first likelihood value and the total mastery index is recorded as the first update index; the product of the second likelihood value and the difference between the first and the total mastery index is recorded as the second update index; the first update index is divided by the sum of the first update index and the first update index to record the updated total mastery index;

[0089] A sequence of answer records is generated based on the response time sequence and the sequence of answers submitted by the user: R={(q n a n , t n , correct n )|n∈[1,N]};where q n Indicates the identifier for the nth question; a n This represents the user's answer to the nth question; t n This indicates the time taken to answer the nth question; correct n This represents the correct answer score for the nth question, where 1 is a correct answer and 0 is an incorrect answer; N represents the total number of questions; each question includes the estimated standard answering time and the question difficulty level.

[0090] Example 3: In this example, the answer accuracy rate is equal to the ratio of the total number of correct answers to the total number of questions;

[0091] Example 4: In this example, the answer accuracy rate is equal to the sum of the products of the correct answer index and the difficulty weight, divided by the sum of the difficulty weights;

[0092] Example 5: Analyze the relative answering time ratio for each question; the relative answering time ratio is equal to the ratio of the time used to answer the question to the estimated standard answering time; calculate the mean and standard deviation of the relative answering time ratio, and record the ratio of the mean and standard deviation as the coefficient of variation (CV);

[0093] The self-similarity score S is analyzed based on the relative answering time ratio. pattern ; ;

[0094] Among them, R n R represents the ratio of the relative answering time for the nth question. n+1 This represents the relative answering time ratio for the (n+1)th question, and τ represents the preset scale parameter;

[0095] Response time stability was analyzed based on the normalized coefficient of variation and self-similarity score.

[0096] Response time stability = 1 - α × (1 - CV) - β × S pattern Where α and β represent preset proportional parameters;

[0097] S402: Based on the updated total mastery index, the total mastery index of the related micro-knowledge points is disseminated and updated; the dissemination and update of the total mastery index of the related micro-knowledge points is equal to the product of the updated total mastery index and the preset update coefficient plus the historical total mastery index; if the absolute value of the difference between the dissemination and update of the total mastery index of the related micro-knowledge points and the historical total mastery index is greater than the difference ratio threshold, the optimized learning arrangement sequence is reordered according to the dissemination and update of the total mastery index.

[0098] Please see Figure 2 The present invention provides a technical solution: an intelligent management system for teaching resources based on big data, which includes a knowledge status analysis module, a learning ranking module, an encapsulation interaction module, and an update and optimization module;

[0099] The knowledge state analysis module is used to collect event streams generated by the interaction between the target user and the interactive learning interface; analyze the mutation behavior in the event stream and extract the learning interval; and analyze the mastery index of each micro-knowledge point based on the operation events within the learning interval.

[0100] The learning ranking module is used to construct a teaching resource arrangement model based on the mastery index of micro-knowledge points; a two-stage heuristic decomposition algorithm is used to solve the teaching resource arrangement model to generate an optimized learning arrangement sequence;

[0101] The encapsulated interaction module is used to assemble micro-learning units for each micro-knowledge point in the optimized learning orchestration sequence; the assembled micro-learning units are then pushed sequentially to the user terminal in the order of the optimized learning orchestration sequence, and the user's process interaction data is obtained.

[0102] The update and optimization module is used to extract process interaction data to analyze answer accuracy and response time stability; update the overall mastery index according to the two-dimensional normal probability model; and trigger the re-optimization of subsequent optimization learning orchestration sequences based on the updated overall mastery index.

[0103] The knowledge state analysis module includes an operational feature analysis unit and a dynamic knowledge state vector analysis unit;

[0104] The operation feature analysis unit is used, with the authorization of the target user, to collect the event stream generated by the interaction between the target user and the interactive learning interface; identify the boundary points in the event stream where the rate of change of operation frequency exceeds a preset transformation rate threshold; divide the event stream into learning intervals based on the boundary points; and extract the corresponding operation features for each learning interval.

[0105] The dynamic knowledge state vector analysis unit is used to extract all learning intervals containing the same micro-knowledge point for each micro-knowledge point in the overall set of micro-knowledge points, and analyze the ideal similarity of each extracted learning interval. The product of ideal similarity, cognitive complexity weight, average operation frequency within the learning interval, and answer accuracy within the learning interval is recorded as the mastery index of the micro-knowledge point in the corresponding learning interval. The sum of the mastery indices of the same micro-knowledge point in all learning intervals is recorded as the total mastery index. The total mastery index is normalized, and the total mastery indices of all micro-knowledge points are arranged in a preset order to generate the user's dynamic knowledge state vector.

[0106] The learning sequencing module includes a teaching resource arrangement model construction unit and a model solving unit;

[0107] The teaching resource arrangement model construction unit is used to define the decision variables of micro-knowledge points. Based on the preset basic importance weights, total mastery index, decision variables and position discount factors, the teaching resource arrangement model is constructed, and constraints are imposed on the teaching resource arrangement model.

[0108] The model solving unit is used to solve the instructional resource arrangement model using a two-stage heuristic decomposition algorithm to generate the final learning arrangement sequence.

[0109] The encapsulated interaction module includes an encapsulation unit and an interactive data acquisition unit;

[0110] The encapsulation unit is used to retrieve the explanatory resources that match each micro-knowledge point in the optimized learning orchestration sequence; to call the exercise set generator to generate an exercise set based on the user's latest dynamic knowledge state vector; and to integrate the retrieved explanatory resources, exercise set, and pre-set feedback explanation templates to form a micro-learning unit.

[0111] This is used to push the assembled micro-learning units to the user terminal in the order of the optimized learning orchestration sequence, and to obtain the user's procedural interaction data; the procedural interaction data includes, but is not limited to, the response time sequence of answers, the sequence of answers submitted by the user, and additional operation events.

[0112] The update and optimization module includes an overall mastery index update unit and a sorting optimization unit;

[0113] The overall mastery index update unit is used to extract the response time series of responses and the sequence of answers submitted by users to analyze the accuracy of answers and the stability of response time; a two-dimensional normal probability model based on preset parameters is constructed, and the accuracy of answers and the stability of response time are input into the two-dimensional normal probability model to calculate the first likelihood value of procedural interaction data in the mastery state and the second likelihood value in the non-mastery state; the overall mastery index is updated based on the first likelihood value and the second likelihood value.

[0114] The sorting optimization unit is used to propagate and update the total mastery index of related micro-knowledge points based on the updated total mastery index; if the absolute value of the difference between the propagated updated total mastery index of related micro-knowledge points and the historical total mastery index is greater than the difference ratio threshold, the optimized learning arrangement sequence is re-sorted according to the propagated updated total mastery index.

[0115] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0116] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A big data-based intelligent management method for teaching resources, characterized in that, The method comprises the following steps: S1: collecting an event stream generated by the target user interacting with the interactive learning interface; analyzing the mutation behavior in the event stream and extracting a learning interval; According to the operation event in the learning interval, the mastery degree index of each micro-knowledge point is analyzed; S2: based on the mastery degree index of the micro-knowledge point, a teaching resource arrangement model is constructed; a two-stage heuristic decomposition algorithm is used to solve the teaching resource arrangement model to generate an optimized learning sequence; S3: Assembling a micro learning unit for each micro knowledge point in the optimized learning sequence; the assembled micro learning unit is sequentially pushed to the user terminal according to the order of the optimized learning sequence, and the process interaction data of the user is obtained; S4: Extracting process interaction data to analyze answer accuracy and response time stability; updating the total mastery degree index according to the two-dimensional normal probability model; based on the updated total mastery degree index, triggering the re-optimization of the subsequent optimized learning sequence. 2.The big data-based teaching resource intelligent management method according to claim 1, characterized in that: In S1, the following content is included: S101: authorized by the target user, collect the event stream generated by the target user interacting with the interactive learning interface, E raw = { (t i , a i , r i ) | i ∈ [1, I]}, wherein t i is the timestamp indicating the occurrence of the ith event, a i is the operation type of the ith event, r i is the micro knowledge point identifier associated with the content of the interactive learning interface when the ith event triggers the operation; I represents the total number of event streams; a cognitive complexity weight is preset for each micro knowledge point; the operation type includes but is not limited to clicking, dragging and answering; S102: Scan the collected event stream using a sliding time window, and use a mutation point detection algorithm to identify boundary points in the event stream whose operation frequency change rate exceeds a preset threshold; based on the boundary points, the event stream is divided into learning intervals; for each learning interval, extract the corresponding operation characteristics, including the average operation frequency in the learning interval, the distribution vector of the operation type, and the set of micro-knowledge points associated with all event streams in the learning interval; S103: For each micro-knowledge point in the total set of micro-knowledge points, extract all learning intervals containing the same micro-knowledge point, analyze the ideal similarity of each extracted learning interval, which is equal to the similarity between the distribution vector of the operation type of the learning interval and the ideal learning mode vector; the product of the ideal similarity, the cognitive complexity weight, the average operation frequency in the learning interval, and the answer accuracy in the learning interval is recorded as the mastery degree index of the micro-knowledge point in the corresponding learning interval; the sum of the mastery degree indexes of the same micro-knowledge point in all learning intervals is recorded as the total mastery degree index; normalize the total mastery degree index, and arrange all micro-knowledge points according to the total mastery degree index in a preset order to generate a dynamic knowledge state vector of the user. 3.The big data-based teaching resource intelligent management method according to claim 2, characterized in that: In S2, the following content is included: S201: record the jth micro-knowledge point in the total set of micro-knowledge points as k j ; If the micro knowledge point k j The corresponding learning unit is arranged to learn at the mth position of the learning sequence, and the decision variable x （j，m） =1; otherwise the decision variable x （j，m） =0; build a teaching resource arrangement model; Constraints are imposed on the teaching resource arrangement model, including precedence constraints, cognitive load constraints, time budget constraints, and forgetting curve constraints. The precedence constraints are that if micro-knowledge point k a is a prerequisite for micro-knowledge point k b , then the following condition needs to be met in the learning sequence: ; where m' represents the sequence number of the position in the learning sequence. The cognitive load constraints are that the cognitive complexity weight sum of any continuous micro-knowledge points in the learning sequence does not exceed the cognitive complexity threshold. The time budget constraints indicate that the total estimated learning time of all micro-knowledge points in the learning sequence does not exceed the time threshold. The forgetting curve constraints are that when the total mastery degree index is below the mastery degree threshold and the time interval from the last learning interval exceeds the time interval threshold, a review is arranged in the learning sequence. S202: Use a two-stage heuristic decomposition algorithm to solve the teaching resource arrangement model to generate the final learning sequence; The first stage uses a greedy algorithm to select the knowledge point with the highest expected return per unit time under the premise of satisfying the precedence relationship and time budget constraint to form an initial knowledge point set; The unit time expected return is equal to the difference between the preset basic importance weight and the total mastery degree index of the micro-knowledge point at the current time, divided by the corresponding estimated learning time; In the second stage, the initial knowledge point set is sorted to obtain an initial sequence that satisfies all constraint conditions; The preset neighborhood operation rule comprises: exchanging the positions of two micro knowledge points adjacent to each other and having no direct precedence constraint in the sequence, or moving a micro knowledge point and a sub-block formed by all directly and indirectly subsequent dependent knowledge points as a whole by one position in the sequence forward or backward; Starting from the initial sequence, meta-heuristic search strategy is adopted for iterative optimization, each step of the iterative optimization comprising: based on the current sequence, one or several candidate sequences are generated by applying the neighborhood operation rule; for each candidate sequence, it is checked whether it satisfies the cognitive load constraint; for the candidate sequence satisfying the cognitive load constraint, the total expected learning benefit value is calculated according to the teaching resource arrangement model; according to the adopted meta-heuristic strategy, one sequence is selected from the checked candidate sequences as the current sequence of the next round of iteration; when the preset iteration termination condition is reached, the search is terminated, and the sequence satisfying all constraints and having the highest total expected learning benefit value in the iteration process is output as the optimized learning arrangement sequence. 4.The big data-based teaching resource intelligent management method according to claim 3, characterized in that: In S3, the following contents are included: S301: for each micro knowledge point in the optimized learning arrangement sequence, a lecture resource matching the micro knowledge point is retrieved; an exercise set generator is called to generate an exercise set according to the latest dynamic knowledge state vector of the user; the retrieved lecture resource, the exercise set and the preset feedback explanation template are integrated and packaged to form a micro learning unit; S302: the assembled micro learning unit is sequentially pushed to the user terminal according to the order of the optimized learning arrangement sequence, and process interaction data of the user is obtained; the process interaction data includes but is not limited to the response time sequence of the answer, the answer sequence submitted by the user and additional operation events. 5.The big data-based teaching resource intelligent management method according to claim 4, characterized in that: In S4, the following contents are included: S401: the answer correctness rate and the response time stability are analyzed by extracting the response time sequence of the answer and the answer sequence submitted by the user; a two-dimensional normal probability model based on preset parameters is constructed to describe the performance distribution of the user in the mastered and unmastered states; the two-dimensional normal probability model presets a mean vector and a covariance matrix for each state; the answer correctness rate and the response time stability are input into the two-dimensional normal probability model to calculate the first likelihood value of the process interaction data in the mastered state and the second likelihood value in the unmastered state; the product of the first likelihood value and the total mastery index is recorded as the first update index; the product of the second likelihood value and the difference between one and the total mastery index is recorded as the second update index; the first update index is divided by the sum of the first update index and the first update index to obtain the updated total mastery index; S402: based on the update total mastery index, the total mastery index of the associated micro knowledge point is updated; the updated total mastery index of the associated micro knowledge point is equal to the product of the update total mastery index and the preset update coefficient plus the historical total mastery index; if the absolute value of the difference between the updated total mastery index of the associated micro knowledge point and the historical total mastery index divided by the historical total mastery index is greater than the difference value threshold, the optimized learning sequence is reordered according to the updated total mastery index.

6. A big data-based teaching resource intelligent management system, which is applied to the big data-based teaching resource intelligent management method of any one of claims 1-5, characterized in that, The system comprises a knowledge state analysis module, a learning sequencing module, an encapsulation interaction module and an update optimization module. The knowledge state analysis module is used to collect the event stream generated by the interaction between the target user and the interactive learning interface; analyze the mutation behavior in the event stream and extract the learning interval; According to the operation event in the learning interval, the mastery index of each micro knowledge point is analyzed; The learning sequencing module is used to construct a teaching resource scheduling model based on the mastery index of the micro knowledge point; a two-stage heuristic decomposition algorithm is used to solve the teaching resource scheduling model to generate an optimized learning sequence. The encapsulation interaction module is used to assemble a micro learning unit for each micro knowledge point in the optimized learning sequence; the assembled micro learning unit is sequentially pushed to the user terminal in the order of the optimized learning sequence, and the process interaction data of the user is obtained. The update optimization module is used to analyze the answer accuracy and response time stability of the process interaction data; the total mastery index is updated according to the two-dimensional normal probability model; based on the updated total mastery index, the re-optimization of the subsequent optimized learning sequence is triggered. 7.The big data-based teaching resource intelligent management system according to claim 6, characterized in that: The knowledge state analysis module comprises an operation feature analysis unit and a dynamic knowledge state vector analysis unit. The operation feature analysis unit is used to collect the event stream generated by the interaction between the target user and the interactive learning interface after authorization by the target user; identify the boundary points in the event stream whose operation frequency change rate exceeds the preset change rate threshold; The event stream is split into learning intervals based on the boundary points; The corresponding operation features of each learning interval are extracted; The dynamic knowledge state vector analysis unit is used to extract all learning intervals containing the same micro knowledge point for each micro knowledge point in the total set of micro knowledge points, analyze the ideal similarity of each extracted learning interval; the product of the ideal similarity, the cognitive complexity weight, the average operation frequency in the learning interval and the answer accuracy in the learning interval is recorded as the mastery index of the micro knowledge point in the corresponding learning interval; the sum of the mastery indexes of the same micro knowledge point in all learning intervals is recorded as the total mastery index; The total mastery indexes of all micro knowledge points are arranged in a preset order to generate the dynamic knowledge state vector of the user after normalization processing of the total mastery index. 8.The big data-based teaching resource intelligent management system according to claim 6, characterized in that: The learning sequencing module comprises a teaching resource scheduling model construction unit and a model solving unit; The teaching resource arrangement model construction unit is configured to define decision variables of the micro knowledge points, construct a teaching resource arrangement model based on preset basic importance weights, a total mastery index, the decision variables, and a position discount factor, and impose constraint conditions on the teaching resource arrangement model; The model solving unit is configured to solve the teaching resource arrangement model by using a two-stage heuristic decomposition algorithm to generate a final learning arrangement sequence. 9.The big data-based teaching resource intelligent management system according to claim 6, characterized in that: The packaging interaction module includes a packaging unit and an interaction data collection unit. The packaging unit is configured to, for each micro knowledge point in the optimized learning arrangement sequence, retrieve an explanation resource matching the micro knowledge point, generate an exercise set by calling an exercise set generator according to a latest dynamic knowledge state vector of the user, and integrate the retrieved explanation resource, the exercise set, and a preset feedback explanation template to form a micro learning unit. The micro learning unit is pushed to the user terminal in the order of the optimized learning arrangement sequence, and process interaction data of the user is obtained, where the process interaction data includes, but is not limited to, a response time sequence of an answer, an answer sequence submitted by the user, and an additional operation event. 10.The big data-based teaching resource intelligent management system according to claim 6, characterized in that: The update optimization module includes a total mastery index update unit and an arrangement optimization unit. The total mastery index update unit is configured to extract the response time sequence of the answer and the answer sequence submitted by the user to analyze an answer accuracy and a response time stability. A two-dimensional normal probability model based on preset parameters is constructed, the answer accuracy and the response time stability are input into the two-dimensional normal probability model, a first likelihood value of the process interaction data in a mastery state and a second likelihood value of the process interaction data in a non-mastery state are respectively calculated, and the total mastery index is updated based on the first likelihood value and the second likelihood value. The arrangement optimization unit is configured to propagate and update the total mastery index of the associated micro knowledge points based on the updated total mastery index, and if an absolute value of a difference between a propagated and updated total mastery index of the associated micro knowledge points and a historical total mastery index divided by the historical total mastery index is greater than a difference ratio threshold value, the optimized learning arrangement sequence is reordered based on the propagated and updated total mastery index. ​

Images