Intelligent agent collaborative generation method based on dynamic sharing of office school interconnection teaching resources

Abstract

The invention discloses an agent collaborative generation method based on dynamic sharing of bureau-school interconnection teaching resources, which comprises the following steps: S1, constructing a bureau-school interconnection network system, and deploying a resource management agent which comprises a data acquisition module, a semantic analysis module, a decision module and a communication module; s2, collecting school teaching data, and sending the school teaching data to a central node of an education management department; s3, performing semantic annotation and classification on the received data, and constructing a dynamic teaching resource semantic database; s4, constructing an intelligent agent collaborative decision-making model, and dynamically adjusting a sharing strategy of teaching resources; s5, generating an optimal resource sharing scheme through the dynamic teaching resource semantic database and the intelligent agent collaborative decision-making model, and sending the optimal resource sharing scheme to the school child node; and S6, after sharing is completed, a resource sharing scheme is fed back to the center node, and the agent collaborative decision model is optimized. According to the invention, intellectualization, dynamics and precision of teaching resource sharing are realized, and a brand new technology is provided for balanced configuration of education resources.

Description

Technical Field

[0001] The present invention relates to the technical field of teaching resource sharing, and in particular to a method for collaboratively generating intelligent bodies based on dynamic sharing of teaching resources interconnected between bureaus and schools. Background Art

[0002] In the process of educational informatization, the efficient sharing of teaching resources is a core requirement for improving educational equity and quality. With the growing demand for data exchange between education administration departments and schools, the traditional static resource sharing model has exposed significant limitations. For example, the distribution of educational resources across regions is unbalanced, with high-quality resources concentrated in a small number of schools, while weaker schools struggle to access appropriate teaching content. Furthermore, the resource sharing process lacks a dynamic control mechanism, making it impossible to optimize transmission paths based on real-time demand and feedback, resulting in low resource utilization and high sharing costs.

[0003] Traditional teaching resource sharing solutions usually adopt a centralized architecture, where the education administration department uniformly stores resources and pushes them to schools in a one-way manner. This model relies on pre-set resource classification standards and achieves sharing through manual review and batch transmission. Its advantages lie in centralized management and standardized processes, making it suitable for large-scale distribution of standardized resources. However, the defects of this model are particularly prominent: resource updates lag, and it is impossible to collect usage data from the school side (such as browsing time, user reviews) in real time, resulting in a disconnect between resources and actual needs; there is a lack of intelligent decision-making mechanisms, and the resource allocation path is fixed, making it difficult to cope with changes in network topology and real-time traffic fluctuations, and transmission delays or bandwidth waste often occur; the security mechanism is weak, and no effective technical protection is provided for data privacy and transmission traceability, posing a risk of information leakage.

[0004] Existing technologies attempt to improve sharing efficiency through distributed storage. For example, some solutions incorporate blockchain technology for identity authentication and data traceability, and utilize edge computing to alleviate pressure on central nodes, improving data security and local processing efficiency to a certain extent. While edge computing can shorten the local preprocessing time for resource requests, it still cannot capture implicit connections between resources (such as the cross-application of interdisciplinary knowledge points). Furthermore, the coupling between global coordination and local decision-making is weak, making it difficult to balance resource utilization and sharing costs. Furthermore, the lack of integration of usage scenario data (such as real-time interaction records of smart terminal devices) makes it difficult to achieve accurate resource iteration. Summary of the Invention

[0005] Based on the above technical problems, this application discloses a method for collaboratively generating intelligent agents based on dynamic sharing of teaching resources between bureaus and schools, including:

[0006] S1. Build a bureau-school interconnected network system, connect the central node of the education administration department with the sub-nodes of each school through communication links, and deploy resource management agents on the central node and sub-nodes respectively. The resource management agent includes a data acquisition module, a semantic analysis module, a decision module, and a communication module;

[0007] S2. The resource management agent of each school sub-node collects the school teaching resource data, resource usage frequency and user evaluation dynamic information in real time through the data collection module, and sends the collected data to the central node of the education management department;

[0008] S3. The resource management agent of the central node uses the semantic analysis module to semantically annotate and classify all received teaching resource data, extract resource keywords and semantic features, and build a dynamic teaching resource semantic database;

[0009] S4. Build an agent collaborative decision-making model. By training the resource management agents of the central node and sub-nodes, using resource utilization, user satisfaction, and sharing cost as reward functions, each agent exchanges status information and decision plans through the communication module to dynamically adjust the sharing strategy of teaching resources.

[0010] S5. When a school sub-node generates resource demand, the resource management agent sends the demand information to the central node through the communication module. The central node optimizes the resource allocation path based on the dynamic teaching resource semantic database and the agent collaborative decision-making model, generates the optimal resource sharing plan, and sends it to the school sub-node;

[0011] S6. The resource management agent of each school sub-node completes the secure transmission and sharing of teaching resources through the communication module according to the received resource sharing plan. After the sharing is completed, the resource usage feedback information is uploaded to the central node to optimize the agent collaborative decision-making model.

[0012] Preferably, the bureau-school Internet network system adopts a layered distributed architecture, including a core layer, a convergence layer and an access layer; the core layer is deployed in the education management department, and is composed of a backbone network of high-performance switches and load balancing equipment; the convergence layer connects multiple schools in the region through optical fiber links, and deploys edge computing servers at each convergence node; the access layer is the internal network of each school, and dynamic regulation of network traffic is achieved through the SDN controller; in the network system, a distributed identity authentication mechanism based on blockchain is adopted between each node, combined with homomorphic encryption technology to achieve privacy protection and traceability of teaching resource data during shared transmission, and at the same time, edge computing servers are used to achieve local preprocessing and caching of resource requests.

[0013] Preferably, the data acquisition module of the resource management agent in S1 is used to collect basic data, usage status data and user feedback data of on-campus teaching resources in real time; the semantic analysis module performs semantic analysis, annotation and classification on the collected data to build a resource semantic description system; the decision-making module generates resource sharing, allocation and optimization strategies based on the semantic analysis results and the agent collaborative decision-making model; the communication module is responsible for information interaction with other agents, receiving external instructions and transmitting local resource data and strategy execution result information, so as to realize two-way transmission of data and instructions between agents.

[0014] Preferably, the data acquisition module of the resource management intelligent body in S2 collects course schedules and question bank label data for structured teaching resource data; collects teaching videos and courseware documents for unstructured teaching resource data, and monitors users' browsing time, download times, and comment content on resources in real time. By combining Internet of Things sensors to collect usage scenario data of teaching resources on smart terminal devices on campus, multi-source heterogeneous data is formed, and a comprehensive data set containing basic resource attributes, usage status, and user interaction information is generated and uploaded to the central node of the education management department.

[0015] Preferably, the semantic analysis module of the resource management agent in S3 performs semantic annotation and classification on the teaching resource data in the following manner: first, construct an initial vocabulary W in the field of teaching resources = {w1, w2, ..., w n}, perform word segmentation on the received teaching resource data to obtain the vocabulary set T = {t1, t2, ..., t m}, by calculating the weight of the vocabulary in the resource data Perform semantic annotation, where f(t i ) is the vocabulary t i The frequency of occurrence in the current resource data, N is the total number of all teaching resource data, n(t i ) contains the word t i The number of resource data; according to the weight Score(t i ) sorts the words, combines the preset semantic classification threshold θ, and takes the semantic categories corresponding to the words with weights higher than θ as the core semantic labels of the teaching resources, completes the semantic annotation and classification of the teaching resource data, and constructs a dynamic teaching resource semantic database.

[0016] Preferably, the resource management agent of the central node of S3 constructs a dynamic teaching resource semantic database by extracting resource keywords and semantic features, specifically:

[0017] For the teaching resources that have completed semantic annotation and classification, the semantic label weight Score (t i )Calculate keyword relevance where αr is the adjustment coefficient, Cooccur(t i ,t j ) is the vocabulary t i With t j The number of times they appear together in the same resource; select the correlation Rel(t i ,t j ) is greater than the set threshold γ e The vocabulary pairs are used as core semantic feature combinations, and the teaching resource ID, semantic tag set, core semantic feature combination and corresponding weight value are combined according to DB re ={ID,{t i},{Rel(t i ,t j )}} structure to build a dynamic teaching resource semantic database.

[0018] Preferably, the agent collaborative decision-making model in S4 adopts a hierarchical reinforcement learning architecture, in which the upper central node agent acts as a global coordinator, responsible for receiving resource status information uploaded by the sub-node agents of each school and constructing a global resource distribution map; the lower sub-node agents act as local decision makers, generating local decision plans based on local resource demand and usage; the agent collaborative decision-making model adopts a dual-depth Q network algorithm, stores the historical state-action-reward sequence of each agent through an experience replay mechanism, and uses the attention mechanism to calculate the associated weights of different school resource demands and global resource distribution, and dynamically adjusts the weight coefficients of resource utilization, user satisfaction and sharing cost in the reward function to achieve a collaborative balance between global optimality and local optimality.

[0019] Preferably, when training the resource management agents of the central node and the child nodes in S4, the reward function R=R ut +R sa -R co ; Among them, R ut According to the ratio of actual usage of teaching resources to the total amount, R sa Based on the user's feedback on the resource, R co Determined based on resource transmission bandwidth occupancy and storage cost factors; each agent transmits its own state information S through the communication module in each decision cycle t t and decision plan A t Broadcast to the Internet and receive S from other agents t and A t After receiving the information, the actual reward value is calculated according to the reward function R, and the temporal difference algorithm is used Update the action value function Q and dynamically adjust the sharing strategy of teaching resources through continuous iterative training, where α is the learning rate and γ is the discount factor.

[0020] Preferably, the central node of S5 optimizes the resource allocation path through genetic algorithm, takes each school node as gene position, and resource allocation relationship as gene code to construct the initial population; calculates the path fitness Evaluate the individuals in the population, where L is the length of the resource allocation path, N su is the number of resources successfully transmitted in the path, N to For the total amount of resources planned to be transmitted in the path, two intersection points are randomly selected using the two-point crossover method to exchange gene fragments, and some gene positions are randomly mutated. By continuously repeating the selection, crossover and mutation operations until the fitness convergence conditions are met, the individual with the highest fitness is decoded into the optimal resource sharing plan, where the resource allocation path length L is determined according to the local school interconnection network topology.

[0021] Preferably, the S6 middle school sub-node structures the resource usage feedback information to form a feedback data packet containing resource ID, usage time, user rating, modification suggestions, and application scenario tags; after receiving it, the central node updates the semantic database record associated with the resource ID and calculates the semantic similarity. where s i is the original semantic label set, s new The new semantic label set extracted from the feedback information is used. When the similarity is lower than the threshold δ, the resource semantic label is supplemented or corrected. The resource utilization efficiency and user satisfaction related data in the feedback information are used as new state-reward samples and stored in the experience replay pool of the intelligent collaborative decision-making model. The model network parameters are updated to achieve dynamic optimization of the intelligent collaborative decision-making model.

[0022] Compared with the prior art, the technical solution of this application has the following technical effects:

[0023] The present invention realizes multi-dimensional dynamic perception of on-campus teaching resources through the data acquisition module of the resource management intelligent body. Whether it is a structured course schedule, question bank label, or unstructured teaching video, courseware document, it can collect and integrate dynamic information such as user browsing time, download times, comment content, etc. in real time, and at the same time combine with the Internet of Things sensor to capture the usage scenario data of smart terminal devices. On this basis, the semantic analysis module constructs a teaching resource domain vocabulary, word segmentation processing and weight calculation, sorts the vocabulary by weight and marks the core semantic tags to form a dynamic teaching resource semantic database. This mechanism breaks through the limitations of traditional static classification and can accurately capture the relationship between the implicit characteristics of resources and the actual needs of users. For example, it automatically classifies cross-domain resources according to the high-frequency co-occurrence of subject knowledge points, significantly improves the accuracy and real-time performance of resource annotation, and provides a high-precision data foundation for subsequent intelligent decision-making.

[0024] The proposed agent collaborative decision-making model employs a hierarchical reinforcement learning architecture. The upper-layer central node agent constructs a global resource distribution map, while the lower-layer sub-node agents generate decision plans based on local needs. Using a dual-deep Q-network algorithm and an attention mechanism, the model dynamically adjusts the weighting coefficients of resource utilization, user satisfaction, and sharing costs. Each agent exchanges state information and plans through a communication module during each decision cycle, and uses a temporal difference algorithm to update the action-value function. This achieves a coordinated balance between global and local optimality, avoiding the resource redundancy or shortage caused by traditional fixed strategies. This allows the teaching resource sharing strategy to be adaptively adjusted based on real-time dynamics, significantly improving the efficiency and rationality of resource allocation.

[0025] The hierarchical distributed architecture design of the bureau-school Internet network system of the present invention realizes dynamic regulation of network traffic and localized preprocessing of resource requests through high-performance switches and load balancing equipment at the core layer, edge computing servers at the aggregation layer, and SDN controllers at the access layer. The central node uses genetic algorithms to optimize resource allocation paths, uses school sub-nodes as gene positions, and resource allocation relationships as gene codes, and generates optimal sharing solutions through operations such as selection, crossover, and mutation. It automatically switches to backup paths when the network is congested, ensuring the stability and efficiency of resource transmission. Compared with traditional fixed-path transmission, it significantly improves the success rate and speed of resource transmission.

[0026] This invention uses resource usage feedback from school sub-nodes to build a complete closed-loop optimization mechanism. After structured processing, resource usage feedback data contains key information such as resource ID, usage duration, user ratings, and modification suggestions. Upon receiving the data, the central node updates the semantic database records in real time and calculates the semantic similarity between the original semantic label and the newly extracted label. When the similarity falls below a threshold, the resource semantic label is automatically supplemented or corrected. The feedback data is stored as a new state-reward sample in the experience replay pool of the intelligent agent collaborative decision-making model, and the model parameters are dynamically updated, forming a virtuous cycle of "collection-labeling-decision-making-feedback-optimization." This ensures that the teaching resource sharing system can continue to evolve and continuously improve user satisfaction and resource service quality.

[0027] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application so that it can be implemented in accordance with the contents of the specification, and to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following is a detailed description of the preferred embodiment of the present application in conjunction with the accompanying drawings.

[0028] Based on the detailed description of the specific embodiments of the present application in conjunction with the accompanying drawings below, those skilled in the art will become more aware of the above and other objects, advantages and features of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without inventive work. In all drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual scale.

[0030] Figure 1 This is a flow chart of the method for collaboratively generating intelligent agents based on dynamic sharing of teaching resources interconnected between bureaus and schools in the present invention;

[0031] Figure 2 This is a hierarchical distributed architecture diagram of the bureau-school Internet system;

[0032] Figure 3 Schematic diagram of the distributed deployment model of the resource management agent;

[0033] Figure 4 A graph comparing the resource matching accuracy of this application and the prior art over time;

[0034] Figure 5 A graph comparing the long-tail resource usage of this application and the prior art over time;

[0035] Figure 6 A comparison chart of the teaching resource sharing costs of this application and the existing technology;

[0036] Figure 7 This is a comparison chart of the decision response time of this application and the prior art. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. In the following description, specific details such as specific configurations and components are provided only to help fully understand the embodiments of the present application. Therefore, it should be clear to those skilled in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present application. In addition, for clarity and brevity, the description of known functions and structures has been omitted in the embodiments.

[0038] It should be understood that references throughout this specification to "one embodiment" or "this embodiment" mean that a particular feature, structure, or characteristic associated with the embodiment is included in at least one embodiment of the present application. Therefore, the appearance of "one embodiment" or "this embodiment" throughout this specification does not necessarily refer to the same embodiment. Furthermore, these particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0039] In addition, the present application may repeat reference numerals and / or letters in different examples. This repetition is for the purpose of simplicity and clarity and does not in itself indicate the relationship between the various embodiments and / or settings discussed.

[0040] The term "and / or" in this article is only a description of the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist at the same time. The term " / and" in this article describes another type of association object relationship, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " in this article generally indicates that the previous and subsequent associated objects are in an "or" relationship.

[0041] The term "at least one" in this article is merely a description of the association relationship between associated objects, indicating that three relationships may exist. For example, at least one of A and B can mean: A exists alone, A and B exist at the same time, and B exists alone.

[0042] It should also be noted that, in this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include," "comprises," or any other variations thereof are intended to cover non-exclusive inclusion.

[0043] Example 1

[0044] This embodiment mainly describes a method for collaboratively generating intelligent agents based on dynamic sharing of teaching resources between the bureau and the school. Figure 1 As shown, including:

[0045] S1. Build a bureau-school interconnected network system, connecting the central node of the education administration department with the sub-nodes of each school through communication links. Deploy resource management agents on the central node and sub-nodes respectively. The resource management agent includes a data collection module, a semantic analysis module, a decision module, and a communication module.

[0046] S2. The resource management agent of each school sub-node collects the school teaching resource data, resource usage frequency and user evaluation dynamic information in real time through the data collection module, and sends the collected data to the central node of the education management department;

[0047] S3. The resource management agent of the central node uses the semantic analysis module to semantically annotate and classify all received teaching resource data, extract resource keywords and semantic features, and build a dynamic teaching resource semantic database;

[0048] S4. Build an agent collaborative decision-making model. By training the resource management agents of the central node and sub-nodes, using resource utilization, user satisfaction, and sharing cost as reward functions, each agent exchanges status information and decision plans through the communication module to dynamically adjust the sharing strategy of teaching resources.

[0049] S5. When a school sub-node generates resource demands, the resource management agent sends the demand information to the central node through the communication module. The central node optimizes the resource allocation path based on the dynamic teaching resource semantic database and the agent collaborative decision-making model, generates the optimal resource sharing plan, and sends it to the school sub-nodes;

[0050] S6. The resource management agent of each school sub-node completes the secure transmission and sharing of teaching resources through the communication module according to the received resource sharing plan. After the sharing is completed, the resource usage feedback information is uploaded to the central node to optimize the agent collaborative decision-making model.

[0051] Further, if Figure 2 As shown in the figure, the bureau-school interconnection network system adopts a layered distributed architecture, including a core layer, a convergence layer, and an access layer; the core layer is deployed in the education management department, and is composed of a backbone network of high-performance switches and load balancing equipment; the convergence layer connects multiple schools in the region through optical fiber links, and deploys edge computing servers at each convergence node; the access layer is the internal network of each school, and the SDN controller is used to realize dynamic regulation of network traffic; in the network system, a distributed identity authentication mechanism based on blockchain is adopted between each node, combined with homomorphic encryption technology to realize privacy protection and traceability of teaching resource data during shared transmission, and at the same time, edge computing servers are used to realize local preprocessing and caching of resource requests.

[0052] Furthermore, the data acquisition module of the resource management agent in S1 is used to collect basic data, usage status data and user feedback data of on-campus teaching resources in real time; the semantic analysis module performs semantic analysis, annotation and classification on the collected data to build a resource semantic description system; the decision-making module generates resource sharing, allocation and optimization strategies based on the semantic analysis results and the agent collaborative decision-making model; the communication module is responsible for information interaction with other agents, receiving external instructions and transmitting local resource data and strategy execution result information, so as to realize two-way transmission of data and instructions between agents.

[0053] Furthermore, the data collection module of the resource management agent in S2 collects course schedules and question bank label data for structured teaching resource data; collects teaching videos and courseware documents for unstructured teaching resource data, and monitors users' browsing time, download times, and comment content on resources in real time. By combining IoT sensors to collect usage scenario data of teaching resources on smart terminal devices on campus, multi-source heterogeneous data is formed, and a comprehensive data set containing basic resource attributes, usage status, and user interaction information is generated and uploaded to the central node of the education management department.

[0054] Furthermore, the semantic analysis module of the resource management agent in S3 uses the following method to semantically annotate and classify the teaching resource data: First, an initial vocabulary W = {w1, w2, ..., w n}, perform word segmentation on the received teaching resource data to obtain the vocabulary set T = {t1, t2, ..., t m}, by calculating the weight of the vocabulary in the resource data Perform semantic annotation, where f(t i ) is the vocabulary t i The frequency of occurrence in the current resource data, N is the total number of all teaching resource data, t(n i ) contains the word t i The number of resource data; according to the weight Score(t i ) sorts the words, combines the preset semantic classification threshold θ, and takes the semantic categories corresponding to the words with weights higher than θ as the core semantic labels of the teaching resources, completes the semantic annotation and classification of the teaching resource data, and constructs a dynamic teaching resource semantic database.

[0055] Furthermore, the resource management agent of the central node of S3 constructs a dynamic teaching resource semantic database by extracting resource keywords and semantic features. Specifically:

[0056] For the teaching resources that have completed semantic annotation and classification, the semantic label weight Score (t i )Calculate keyword relevance where α ris the adjustment coefficient, Cooccur(t i ,t j ) is the vocabulary t i With t j The number of times they appear together in the same resource; select the correlation Rel(t i ,t j ) is greater than the set threshold γ e The vocabulary pairs are used as core semantic feature combinations, and the teaching resource ID, semantic tag set, core semantic feature combination and corresponding weight value are combined according to DB re ={ID,{t i},{Rel(t i ,t j )}} structure to build a dynamic teaching resource semantic database.

[0057] Furthermore, the agent collaborative decision-making model in S4 adopts a hierarchical reinforcement learning architecture. The upper-level central node agent serves as the global coordinator, responsible for receiving resource status information uploaded by the sub-node agents of each school and constructing a global resource distribution map; the lower-level sub-node agents serve as local decision-makers, generating local decision-making plans based on local resource demand and usage; the agent collaborative decision-making model adopts a dual-depth Q-network algorithm, stores the historical state-action-reward sequence of each agent through the experience replay mechanism, and uses the attention mechanism to calculate the association weights between the resource demand of different schools and the global resource distribution, and dynamically adjusts the weight coefficients of resource utilization, user satisfaction and sharing cost in the reward function to achieve a collaborative balance between global optimality and local optimality.

[0058] Furthermore, when training the resource management agents of the central node and the child nodes in S4, the reward function R=R ut +R sa -R co ; Among them, R ut According to the ratio of actual usage of teaching resources to the total amount, R sa Based on the user's feedback on the resource, R co Determined based on resource transmission bandwidth occupancy and storage cost factors; each agent transmits its own state information S through the communication module in each decision cycle t t and decision plan A t Broadcast to the Internet and receive S from other agents t and A t After receiving the information, the actual reward value is calculated according to the reward function R, and the temporal difference algorithm is used Update the action value function Q and dynamically adjust the sharing strategy of teaching resources through continuous iterative training, where α is the learning rate and γ is the discount factor.

[0059] Furthermore, the central node of S5 optimizes the resource allocation path through genetic algorithm, takes each school node as gene position, and resource allocation relationship as gene coding to construct the initial population; by calculating the path fitness Evaluate the individuals in the population, where L is the length of the resource allocation path, N su is the number of resources successfully transmitted in the path, N to For the total amount of resources planned to be transmitted in the path, the two-point crossover method is used to randomly select two intersection points to exchange gene fragments and randomly mutate some gene positions; by repeatedly selecting, crossing over and mutating until the fitness convergence condition is met, the individual with the highest fitness is decoded into the optimal resource sharing plan, where the resource allocation path length L is determined according to the topological structure of the local school interconnection network.

[0060] Furthermore, the S6 middle school sub-node structures the resource usage feedback information to form a feedback data packet containing resource ID, usage time, user rating, modification suggestions, and application scenario tags; after receiving it, the central node updates the semantic database record associated with the resource ID and calculates the semantic similarity. where s i is the original semantic label set, s new The new semantic label set extracted from the feedback information is used. When the similarity is lower than the threshold δ, the resource semantic label is supplemented or corrected. The resource utilization efficiency and user satisfaction related data in the feedback information are used as new state-reward samples and stored in the experience replay pool of the intelligent collaborative decision-making model. The model network parameters are updated to achieve dynamic optimization of the intelligent collaborative decision-making model.

[0061] This embodiment describes in detail that this application constructs a two-layer architecture of "bureau-school interconnected network system + intelligent collaborative decision-making" to form a dynamic closed loop for the entire process. At the network layer, a layered distributed architecture is adopted. Through the collaboration of the core layer, the aggregation layer, and the access layer, dynamic traffic regulation and localized preprocessing are achieved, and blockchain and homomorphic encryption technology are used to ensure data security and traceability; at the intelligent body layer, multi-source heterogeneous data is captured in real time through the data acquisition module, and a dynamic semantic database is constructed by combining the improved TF-IDF algorithm with keyword association calculation to improve the accuracy of resource labeling; at the decision-making level, a layered reinforcement learning architecture is adopted. The reward function weight is dynamically adjusted through the dual-depth Q network algorithm and the attention mechanism to achieve collaborative optimization of global resource distribution and local demand, and at the same time form an iterative optimization path of "collection-labeling-decision-feedback", which shortens the resource semantic label update cycle to the real-time level, breaks through the "data island" and "decision-making separation" bottlenecks of the existing solution, and realizes the intelligent, dynamic and precise sharing of teaching resources, providing a new technical paradigm for the balanced allocation of educational resources.

[0062] Based on Example 1, this implementation describes in detail the resource management agent, such as Figure 3 As shown, specifically:

[0063] The resource management agent adopts a distributed deployment model, building heterogeneous instances at the central node of the education administration department and at the sub-nodes of the schools, forming a "global-local" collaborative architecture. The central node agent focuses on macro-resource regulation, integrating a high-performance semantic analysis module and a decision engine, and is responsible for semantic annotation, feature extraction, and global strategy generation for cross-regional resources. The school sub-node agent focuses on local resource perception and execution, capturing campus dynamics in real time through a lightweight data acquisition module. Each agent forms a bidirectional interactive link through the communication module, supporting sub-nodes to upload resource status and demand information to the central node, and allowing the central node to issue optimization strategies and sharing solutions to sub-nodes. The four functional modules (data acquisition, semantic analysis, decision generation, and communication interaction) do not operate independently, but rather achieve collaboration through the deep coupling of data flow and control flow. For example, user feedback data obtained by the data acquisition module is processed by the semantic analysis module to form structured semantic labels. The decision module generates sharing strategies based on the labels, and finally the communication module executes cross-node transmission, forming a closed-loop processing flow of "perception-analysis-decision-execution".

[0064] The data acquisition module breaks through the limitations of traditional single data type collection and builds a three-dimensional data capture system. For structured data, it adopts a mechanism that combines scheduled polling with event triggering. For example, the structured information of course schedules and question bank tags is updated regularly and synchronously through the API interface to ensure the accuracy and timeliness of basic data. For unstructured data such as teaching videos and courseware documents, optical character recognition (OCR) and audio and video feature extraction technology are introduced to analyze key knowledge points and media features in the content, such as extracting the speaker's blackboard text from the video and identifying the type of chart from the courseware. The data acquisition module embeds IoT sensors (such as the usage status sensor of the classroom terminal) to collect resource usage scenario data in real time, including device access type, user operation trajectory, and interaction duration, forming a three-dimensional data set that includes basic resource attributes, usage status, and user behavior. After standardization and cleaning, it is packaged into a comprehensive data set in a unified format.

[0065] The semantic analysis module solves the problem that traditional static classification cannot adapt to resource evolution by constructing a dynamic semantic description system. The semantic analysis module constructs an initial vocabulary based on the domain knowledge of teaching resources, and combines the improved TF-IDF algorithm (introducing the inverse document frequency logarithm calculation) to calculate the vocabulary weight, highlighting high-frequency and domain-distinguishing keywords, such as the core position of the terms "calculus" and "geometric model" in mathematics subject resources; on this basis, the keyword association algorithm (integrating label weights and co-occurrence counts) is used to mine implicit associations between words, such as discovering the high frequency co-occurrence of "trigonometric functions" and "analytic geometry" in the same resource, thereby establishing semantic links across knowledge points; the dynamic teaching resource semantic database is not statically stored, but is continuously updated based on real-time annotation results, forming a dynamic graph containing resource IDs, semantic label sets, and feature association networks. This mechanism upgrades resource descriptions from single-label classification to multi-dimensional semantic networks, supporting complex resource retrieval and intelligent matching. For example, when a user searches for "physics experiment videos", the system can recommend related "data acquisition methods" documents or "error analysis" courseware based on semantic associations;

[0066] The decision-making module uses a hierarchical reinforcement learning architecture. The upper-level central node agent identifies regional resource supply and demand trends through a global resource distribution map, while the lower-level sub-node agents generate candidate solutions based on local needs. A dual-deep Q-network algorithm (which introduces an attention mechanism to assign weights) balances resource utilization, user satisfaction, and sharing costs. For example, when a school applies for "programming teaching resources," the decision-making module first selects a set of resources with matching tags from the semantic database. It then optimizes the transmission path using a genetic algorithm—using the school node as a genetic location, path length, and transmission success rate as fitness indicators, and generating the optimal solution through crossover and mutation operations.

[0067] The communication module adopts blockchain-based distributed identity authentication and homomorphic encryption technology to ensure the security and traceability of resource transmission, and uses edge computing servers to implement localized preprocessing of requests to reduce the load on central nodes. The deep collaboration between decision-making and communication is reflected in the following: the encrypted transmission instructions generated by the decision-making module are executed in real time by the communication module, and the network status during the transmission process (such as bandwidth occupancy and delay) is input into the decision-making model as a feedback signal, forming a closed loop of policy iteration, so that the resource sharing process remains efficient and robust in a dynamic environment.

[0068] This embodiment describes in detail how the resource management intelligent body realizes dynamic collection of teaching resources, precise semantic annotation and intelligent decision-making through multi-module collaboration, builds a dynamic semantic database and optimizes sharing strategies, combines secure transmission with a closed-loop feedback mechanism, improves resource utilization and user satisfaction, reduces sharing costs, and realizes efficient collaborative sharing of resources between bureaus and schools.

[0069] Based on Example 1, this implementation describes in detail the specific implementation effects of this application, specifically:

[0070] Simulate a teaching resource sharing scenario for a regional education system to verify the performance of a collaborative agent generation method based on bureau-school interconnection (hereinafter referred to as the "technique of this application") in terms of core indicators such as resource utilization, sharing cost, and user satisfaction, and compare it with a traditional centralized sharing solution (hereinafter referred to as the "existing technology"). Set a central node including one education administration department and 32 school sub-nodes, covering a total of 86,000 structured (question bank, curriculum) and unstructured (teaching videos, courseware) resources. The simulation period is 30 days, and key data on resource requests, transmissions, and feedback are recorded daily.

[0071] The experimental environment and parameter settings are shown in Table 1;

[0072] Table 1 Experimental environment and parameters

[0073]

[0074] The technology of this application: The data collection module collects resource usage data of each school every day, including browsing time (average 20.45±5.21 minutes / time), number of downloads (average 1216.5±213.7 times per day), number of comments (average 451.8±89.3 comments per day), and combines IoT sensors to capture classroom terminal usage scenario data (such as device access type distribution: PC accounts for 69.8%, tablet accounts for 23.5%, and mobile phones account for 8.7%).

[0075] The semantic analysis module calculates vocabulary weights using an improved TF-IDF algorithm. For example, the weight of the word "calculus" in mathematical resources is 0.875, which is higher than the threshold θ = 0.5 and is marked as a core tag. The keyword correlation calculation shows that the correlation between "trigonometric function" and "analytic geometry" is Rel = 0.739, which is higher than the threshold γ e =0.6, forming a semantic link;

[0076] Dynamic semantic database scale: 1,200-1,500 new semantic tags are added and 800-1,000 relationships are updated daily.

[0077] Existing technology: Manually collect resource metadata on a regular basis and annotate static tags (such as "Mathematics - High School - Compulsory 1") without collecting user behavior data; static tag library size: approximately 500 fixed tags, with less than 100 updates per month.

[0078] Comparison of resource requests and allocation strategies, the technology of this application: When school A initiates a request for "Python programming teaching video", the central node intelligent agent matches the label set {programming, Python, teaching video} through the semantic database, retrieves 2627 matching resources, and uses the genetic algorithm to optimize the path, selecting the shortest path (number of hops = 3) and the lowest cost (bandwidth occupancy 23.5Mbps) transmission plan, with a generation time of 0.87 seconds; during the transmission process, the SDN controller dynamically adjusts the traffic, with an average transmission delay of 120.4ms and a success rate of 99.16%.

[0079] Existing technology: After manually reviewing the request, tag-matching resources (about 500 copies) are retrieved from the central library, and the transmission path is manually selected. The solution generation time is about 37 minutes; the fixed path transmission delay fluctuates greatly (200-500ms), and the success rate is 76.2%.

[0080] In this application technology, school A provided feedback on a courseware with a "user rating of 3.2 / 5 and 12 modification suggestions." The central node calculated the semantic similarity Sim = 0.65 (lower than the threshold δ = 0.7), triggered a label correction, added a "error-prone point analysis" label, and stored the feedback data in the experience replay pool. The model parameter update cycle was 1 hour. After 30 days of iteration, the decision accuracy of the intelligent agent increased from the initial 78.5% to 89.2%. The existing technology collects feedback quarterly, manually corrects labels, and has no model optimization mechanism. The decision accuracy is maintained at 65%-70% for a long time.

[0081] As shown in Table 2, the technology of this application achieves an average daily resource access of 15432.6±1204.5 times through dynamic semantic annotation, intelligent agent collaborative decision-making and multi-source data collection, which is a significant improvement over the 8926.3±987.2 times of the existing technology; Figure 4 As shown in Figure 2, the effective resource utilization rate reaches 89.2%, far exceeding the 65.4% of the existing technology, indicating that it can more accurately match user needs and reduce invalid resource usage. In addition, Figure 5 As shown, the utilization rate of long-tail resources (low-frequency resources) in the present application technology reaches 37.8%, while the utilization rate of existing technologies is only 12.5%. This shows that the dynamic semantic association and global resource scheduling effectively activate unpopular resources and enhance the overall value of the resource pool.

[0082] Table 2 Resource utilization

[0083] index This application technology Existing technology Average daily resource visits (times) 15432.6±1204.5 8926.3±987.2 Effective resource utilization 89.2％ 65.4％ Long-tail resource utilization 37.8% (proportion of low-frequency resources) 12.5％

[0084] The average daily shared cost distribution for the two technical solutions is as follows: Figure 6As shown, the technology in this application uses edge computing server cache resources, SDN dynamic traffic control and blockchain lightweight authentication, with an average daily cost of 1206.56 yuan, of which bandwidth occupation cost is 780.01 yuan and storage cost is 415.75 yuan (edge ​​cache reduces central storage pressure by 30%). In contrast, the existing technology adopts a centralized architecture, lacks edge caching and intelligent traffic management, and has an average daily cost of 2316.67 yuan, with bandwidth and storage costs each accounting for 50%. The lack of edge caching leads to excessive central load.

[0085] As shown in Table 3, from the user experience perspective, the resource matching accuracy of this application technology reached 89.2%, significantly higher than the 65.4% of the existing technology, thanks to the precise recommendation capabilities of the dynamic semantic database and reinforcement learning model. The transmission delay satisfaction score was 7.8 / 10, better than the 5.2 / 10 of the existing technology, reflecting the improvement in transmission efficiency achieved by the layered network architecture and intelligent path optimization. The functional completeness score was 8.5 / 10 (including functions such as intelligent recommendation and real-time feedback), far exceeding the 6.0 / 10 of the existing technology (which only supports basic search), indicating that it provides more comprehensive user services through intelligent agent collaboration.

[0086] Table 3 User experience dimensions

[0087] Dimensions This application technology (N=1000) Existing technology (N=1000) Resource matching accuracy 89.2％ 65.4％ Transmission delay satisfaction 7.8 / 10 (average rating) 5.2 / 10 Functionality 8.5 / 10 (including smart recommendations) 6.0 / 10 (basic search)

[0088] like Figure 7 As shown, the decision response time is compared in the form of a box plot. The technology of this application uses real-time interaction between a hierarchical reinforcement learning model and a communication module, with an average response time of only 1.21 seconds, and 95% of the requests are completed within 2 seconds, with fast response speed and high stability. The existing technology relies on manual review and fixed strategies, with an average response time of up to 30 minutes, and there are significant delay fluctuations (minimum 15 minutes, maximum 1 hour). The data shows that the technology of this application completely solves the inefficiency problem of traditional solutions through automated intelligent decision-making and meets the demand for real-time resource sharing.

[0089] This implementation describes in detail how the technology of this application solves the problem of coordination between global and local decision-making through a hierarchical reinforcement learning model. The attention mechanism improves the efficiency of reward function weight adjustment by more than 50%. The dynamic semantic database supports real-time label updates, which is 90% more efficient than static annotation, adapting to the needs of rapid iteration of educational resources.

[0090] The above are only preferred embodiments of the present invention, which do not limit the scope of protection of the present invention. For those skilled in the art, the present invention can be modified and varied in various ways. Any changes, modifications, replacements, integrations and parameter changes to these embodiments through conventional substitutions or that can achieve the same functions without departing from the principles and spirit of the present invention fall within the scope of protection of the present invention.

Claims

1. A collaborative agent generation method based on dynamic sharing of teaching resources between bureaus and schools, characterized by: include: S1. Build a bureau-school interconnected network system, connect the central node of the education administration department with the sub-nodes of each school through communication links, and deploy resource management agents on the central node and sub-nodes respectively. The resource management agent includes a data acquisition module, a semantic analysis module, a decision module, and a communication module; S2. The resource management agent of each school sub-node collects the school teaching resource data, resource usage frequency and user evaluation dynamic information in real time through the data collection module, and sends the collected data to the central node of the education management department; S3. The resource management agent of the central node uses the semantic analysis module to semantically annotate and classify all received teaching resource data, extract resource keywords and semantic features, and build a dynamic teaching resource semantic database; S4. Build an agent collaborative decision-making model. By training the resource management agents of the central node and sub-nodes, using resource utilization, user satisfaction, and sharing cost as reward functions, each agent exchanges status information and decision plans through the communication module to dynamically adjust the sharing strategy of teaching resources. S5. When a school sub-node generates resource demand, the resource management agent sends the demand information to the central node through the communication module. The central node optimizes the resource allocation path based on the dynamic teaching resource semantic database and the agent collaborative decision-making model, generates the optimal resource sharing plan, and sends it to the school sub-node; S6. The resource management agent of each school sub-node completes the secure transmission and sharing of teaching resources through the communication module according to the received resource sharing plan. After the sharing is completed, the resource usage feedback information is uploaded to the central node to optimize the agent collaborative decision-making model.

2. The method for collaboratively generating intelligent agents based on dynamic sharing of teaching resources between bureaus and schools according to claim 1 is characterized in that: The bureau-school interconnection network system adopts a layered distributed architecture, including a core layer, a convergence layer and an access layer; the core layer is deployed in the education management department, and is composed of a backbone network of high-performance switches and load balancing equipment; the convergence layer connects multiple schools in the region through optical fiber links, and deploys edge computing servers at each convergence node; the access layer is the internal network of each school, and dynamic regulation of network traffic is achieved through an SDN controller; in the network system, a distributed identity authentication mechanism based on blockchain is adopted between each node, combined with homomorphic encryption technology to achieve privacy protection and traceability of teaching resource data during shared transmission, and at the same time, edge computing servers are used to achieve local preprocessing and caching of resource requests.

3. The method for collaboratively generating intelligent agents based on dynamic sharing of teaching resources between bureaus and schools according to claim 1 is characterized in that: The data collection module of the resource management agent in S1 is used to collect basic data, usage status data and user feedback data of on-campus teaching resources in real time; the semantic analysis module performs semantic analysis, annotation and classification on the collected data to build a resource semantic description system; the decision-making module generates resource sharing, allocation and optimization strategies based on the semantic analysis results and the agent collaborative decision-making model; The communication module is responsible for information exchange with other intelligent agents, receiving external instructions and transmitting local resource data and strategy execution result information, realizing two-way transmission of data and instructions between intelligent agents.

4. The method for collaboratively generating intelligent agents based on dynamic sharing of teaching resources between bureaus and schools according to claim 1 is characterized in that: The data acquisition module of the resource management intelligent body in S2 collects course schedules and question bank label data for structured teaching resource data; collects teaching videos and courseware documents for unstructured teaching resource data, and monitors users' browsing time, download times, and comment content on resources in real time. By combining IoT sensors to collect usage scenario data of teaching resources on smart terminal devices on campus, multi-source heterogeneous data is formed, and a comprehensive data set containing basic resource attributes, usage status, and user interaction information is generated and uploaded to the central node of the education management department.

5. The method for collaboratively generating intelligent agents based on dynamic sharing of teaching resources between bureaus and schools according to claim 1 is characterized in that: The semantic analysis module of the resource management agent in S3 performs semantic annotation and classification on the teaching resource data in the following manner: First, an initial vocabulary W in the field of teaching resources is constructed, which is: {w1, w2, ..., w n }, perform word segmentation on the received teaching resource data to obtain the vocabulary set T = {t1, t2, ..., t m }, by calculating the weight of the vocabulary in the resource data Perform semantic annotation, where f(t i ) is the vocabulary t i The frequency of occurrence in the current resource data, N is the total number of all teaching resource data, n(t i ) contains the word t i The number of resource data; according to the weight Score(t i ) sorts the words, combines the preset semantic classification threshold θ, and takes the semantic categories corresponding to the words with weights higher than θ as the core semantic labels of the teaching resources, completes the semantic annotation and classification of the teaching resource data, and constructs a dynamic teaching resource semantic database.

6. The method for collaboratively generating intelligent agents based on dynamic sharing of teaching resources between bureaus and schools according to claim 1 or 5, characterized in that: The resource management agent of the central node of S3 constructs a dynamic teaching resource semantic database by extracting resource keywords and semantic features, specifically: For the teaching resources that have completed semantic annotation and classification, the semantic label weight Score (t i )Calculate keyword relevance where α r is the adjustment coefficient, Cooccur(t i ,t j ) is the vocabulary t i With t j the number of times they co-occur in the same resource; Select the correlation Rel(t i ,t j ) is greater than the set threshold γ e The vocabulary pairs are used as core semantic feature combinations, and the teaching resource ID, semantic tag set, core semantic feature combination and corresponding weight value are combined according to DB re ={ID,{t i },{Rel(t i ,t j )}} structure to build a dynamic teaching resource semantic database.

7. The method for collaboratively generating intelligent agents based on dynamic sharing of teaching resources between bureaus and schools according to claim 1 is characterized in that: The agent collaborative decision-making model in S4 adopts a hierarchical reinforcement learning architecture. The upper-level central node agent serves as the global coordinator, responsible for receiving resource status information uploaded by the sub-node agents of each school and constructing a global resource distribution map; the lower-level sub-node agents serve as local decision makers, generating local decision plans based on local resource demand and usage; the agent collaborative decision-making model adopts a dual-depth Q-network algorithm, stores the historical state-action-reward sequence of each agent through an experience replay mechanism, uses the attention mechanism to calculate the association weights between the resource demand of different schools and the global resource distribution, and dynamically adjusts the weight coefficients of resource utilization, user satisfaction and sharing cost in the reward function to achieve a collaborative balance between global optimality and local optimality.

8. The method for collaboratively generating intelligent agents based on dynamic sharing of teaching resources between bureaus and schools according to claim 1 is characterized in that: When training the resource management agents of the central node and the child nodes in S4, the reward function R=R ut +R sa -R co ; Among them, R ut According to the ratio of actual usage of teaching resources to the total amount, R sa Based on the user's feedback on the resource, R co Determined based on resource transmission bandwidth occupancy and storage cost factors; each agent transmits its own state information S through the communication module in each decision cycle t t and decision plan A t Broadcast to the Internet and receive S from other agents t and A t After receiving the information, the actual reward value is calculated according to the reward function R, and the temporal difference algorithm is used Update the action value function Q and dynamically adjust the sharing strategy of teaching resources through continuous iterative training, where α is the learning rate and γ is the discount factor.

9. The method for collaboratively generating intelligent agents based on dynamic sharing of teaching resources between bureaus and schools according to claim 1 is characterized in that: The central node of S5 optimizes the resource allocation path through genetic algorithm, takes each school node as gene position, and resource allocation relationship as gene coding to construct the initial population; by calculating the path fitness Evaluate the individuals in the population, where L is the length of the resource allocation path, N su is the number of resources successfully transmitted in the path, N to For the total amount of resources planned to be transmitted in the path, two intersection points are randomly selected using the two-point crossover method to exchange gene fragments, and some gene positions are randomly mutated. By continuously repeating the selection, crossover and mutation operations until the fitness convergence conditions are met, the individual with the highest fitness is decoded into the optimal resource sharing plan, where the resource allocation path length L is determined according to the local school interconnection network topology.

10. The method for collaboratively generating intelligent agents based on dynamic sharing of teaching resources between bureaus and schools according to claim 1 is characterized in that: The S6 middle school sub-node structures the resource usage feedback information to form a feedback data packet containing resource ID, usage time, user rating, modification suggestions, and application scenario tags; after receiving it, the central node updates the semantic database record associated with the resource ID and calculates the semantic similarity. where s i is the original semantic label set, s new The new semantic label set extracted from the feedback information is used. When the similarity is lower than the threshold δ, the resource semantic label is supplemented or corrected. The resource utilization efficiency and user satisfaction related data in the feedback information are used as new state-reward samples and stored in the experience replay pool of the intelligent collaborative decision-making model. The model network parameters are updated to achieve dynamic optimization of the intelligent collaborative decision-making model.

Images