Educational resource intelligent recommendation method and system based on big data driving

Abstract

The invention relates to the technical field of educational resource recommendation, and discloses an educational resource intelligent recommendation method and system based on big data driving, and the system comprises a data fusion processing module, a portrait modeling module, a resource feature engine module, an intelligent recommendation core module and a closed-loop feedback module. By fusing knowledge state, cognitive ability and interest preference three-dimensional portraits, capturing user learning ability evolution in real time, dynamically updating knowledge mastery by adopting a knowledge tracking model, and perceiving interest migration in combination with an attention mechanism, the problem of learning cold start in a new field is solved, recommendation coverage range and accuracy are improved, and user experience is improved. The cognitive load sensitive ant colony optimization algorithm is designed, the learning path continuity is guaranteed through a heuristic function, the learning path structure reasonability is optimized, the cognitive burden of a user is reduced, a personalized knowledge attenuation model is constructed by fusing an Ebbinghaus forgetting curve, the knowledge long-term retention rate is increased, and the review resource release accuracy is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of educational resource recommendation, and in particular to a method and system for intelligently recommending educational resources based on big data. Background Art

[0002] With the rapid development of information technology and artificial intelligence, the demand for digital and intelligent educational resources is increasing. Traditional resource recommendation methods can no longer meet the needs of personalized learning. Currently, learners face problems such as a wide variety of educational resources, uneven quality, and difficulty in choosing. Especially in the fields of independent learning and online education, students often find it difficult to efficiently locate resources that suit their learning needs.

[0003] Although existing big data-driven educational resource recommendation methods and systems have achieved personalized resource recommendations to a certain extent, traditional recommendation technologies have the following limitations: 1. Static portrait defects: Existing methods construct a single interest tag portrait based on historical behavior, which cannot capture the dynamic changes in user knowledge status and cognitive ability. When users learn new areas, the recommendation accuracy decreases due to the cold start problem.

[0004] 2. Lack of resource association: Mainstream collaborative filtering algorithms only consider resource co-occurrence relationships and ignore the topological dependencies between knowledge points.

[0005] 3. Ignoring cognitive laws: The current system does not incorporate a memory decay model and cannot dynamically adjust the review strategy based on the Ebbinghaus forgetting curve. The user's knowledge retention rate decreases on average after 7 days.

[0006] 4. Rigid algorithm adaptation: Fixed recommendation strategies are difficult to adapt to the needs of users with different cognitive abilities.

[0007] Therefore, the present invention provides a new recommendation solution for educational resources based on big data-driven intelligent recommendation method and system integration of multimodal user portraits, cognitive load optimization and dynamic feedback mechanism to solve the core problems of cold start, cognitive bias and low long-term utility in the existing technology. Summary of the Invention

[0008] (1) Technical problems solved In response to the shortcomings of the existing technology, the present invention provides a method and system for intelligent recommendation of educational resources based on big data, which solves the problems raised in the above background technology.

[0009] (2) Technical solution To achieve the above objectives, the present invention provides the following technical solutions: a method and system for intelligent recommendation of educational resources based on big data, the method comprising the following steps: S1. Collect user learning behavior data and educational resource metadata to generate a user-resource interaction matrix; S2. Construct a multimodal user profile based on the user-resource interaction matrix to generate a user knowledge status profile, a cognitive ability profile, and an interest preference profile; S3. Extract semantic features and conduct knowledge graph association analysis on the educational resource library to generate resource feature vectors and topological relationship diagrams between resources; S4. Calculate the demand-resource matching degree based on the user knowledge status profile and the resource feature vector to generate an initial recommended resource set; S5. Perform dynamic path optimization on the initial recommended resource set based on an improved ant colony optimization algorithm, and generate an optimized recommendation sequence by combining the topological relationship diagram between resources and the user's cognitive ability profile; S6. Personalize the optimized recommendation sequence based on real-time user feedback data and a forgetting curve model to generate a dynamic educational resource recommendation list; S7. Update user profiles and resource feature vectors through reinforcement learning mechanisms to build an adaptive recommendation closed-loop system.

[0010] Preferably, the S1 includes: S11. Obtain user online learning behavior data through a log collector, including video viewing time, exercise accuracy, knowledge point jump path, and interaction frequency, to generate user behavior time series data; S12. Extract resource structured metadata from the educational resource management platform, including resource type, knowledge tag, difficulty coefficient, related knowledge points, and historical usage statistics, to generate a resource feature matrix; S13. Perform tensor fusion on the user behavior time series data and the resource feature matrix to construct a three-dimensional user-resource interaction matrix: Among them, m is the number of users, n is the number of resources, and k is the behavior dimension.

[0011] Preferably, the S2 includes: S21. Process user behavior time series data based on the knowledge tracking model, predict the probability of users mastering knowledge points, and generate a user knowledge status profile: in, Indicates the mastery of the i-th knowledge point; S22. Analyze exercise response data through cognitive diagnostic models, calculate user memory strength, reasoning ability, and information processing speed, and generate cognitive ability profiles: in: is the memory strength, is the information processing speed, For logical reasoning ability; S23. Use the attention mechanism to extract resource browsing hotspot data and combine it with the collaborative filtering algorithm to generate interest preference portraits: in: Label for users The interest weight of , j is the total number of interest tags.

[0012] Preferably, the S3 includes: S31. Use the BERT model to semantically embed the text content of educational resources and generate resource semantic vectors: S32. Construct a resource association graph G = (V, E) based on the knowledge graph, where the vertex V represents the resource and the weight of the edge E is determined by the co-occurrence frequency of knowledge points between resources and the learning path dependency relationship; S33. Fusion of resource semantic vectors and graph topology features to generate resource feature vectors: is the vector concatenation operation, The importance score of the resource node in the graph.

[0013] Preferably, the S4 includes: S41. Calculate the cosine similarity between the user knowledge status profile and the resource feature vector: in, is the user knowledge state vector, Resource feature vector, the matching degree between user u and resource r; S42. Adjust the matching threshold based on the cognitive ability profile and generate an initial recommendation set: is the adaptive threshold coefficient, User information processing speed.

[0014] Preferably, the S5 includes: S51. Model the recommendation problem as a path optimization problem on the graph G and define the pheromone update rule: The pheromone concentration from resource node i to j, is the pheromone volatility coefficient, is the pheromone intensity constant, the cognitive load of path k; S52. Design heuristic function: in, For resources The difficulty coefficient, is the shortest path distance between resource nodes i, j in the graph; S53. Generate Pareto frontier solutions through multi-objective optimization and select the optimal recommendation sequence based on user cognitive ability profiles. .

[0015] Preferably, the S6 includes: S61. Calculate the knowledge decay weight based on the Ebbinghaus forgetting curve: in is the attenuation factor, is the time interval, is the memory strength parameter; S62. Dynamically adjust the recommendation list based on the real-time answer accuracy rate acc: Among them, the accuracy rate of acc users' real-time answers is Resources to review, Resources are already in hand.

[0016] Preferably, the S7 includes: S71. Build a deep reinforcement learning model to recommend rewards: Completion rate Knowledge gain is the reward function, and is the weight coefficient.

[0017] S72. Update user profiles and resource feature vectors through the dual-Q network to achieve online optimization of recommendation strategies.

[0018] Preferably, including: Data fusion processing module: including behavior collection unit, metadata extraction unit and tensor construction unit; Portrait modeling module: including knowledge status analysis unit, cognitive ability assessment unit and interest mining unit; Resource feature engine module: including semantic embedding unit, knowledge graph construction unit and feature fusion unit; Intelligent recommendation core module: including matching degree calculation unit, ant colony optimization unit and dynamic adjustment unit; Closed-loop feedback module: includes reinforcement learning agent unit and parameter updating unit.

[0019] Preferably, the behavior collection unit integrates the tracking SDK and API interface to capture user operation events in real time; The ant colony optimization unit is equipped with a dynamic parameter adjuster that automatically adjusts according to the system load: N is the size of the ant colony, and T is the maximum number of iterations; The reinforcement learning agent unit adopts a distributed architecture, including a mechanism for separating the experience replay pool and the target network.

[0020] (3) Beneficial effects Compared with the existing technology, the present invention provides a method and system for intelligent recommendation of educational resources based on big data, which has the following beneficial effects: 1. By integrating three-dimensional portraits of knowledge status, cognitive ability, and interest preferences, we can capture the evolution of users' learning abilities in real time, dynamically update knowledge mastery using a knowledge tracking model, and perceive interest migration using an attention mechanism, thereby solving the cold start problem of learning in new fields and improving the coverage and accuracy of recommendations.

[0021] 2. Based on the knowledge graph, the topological dependencies between resources are explicitly modeled. By designing a cognitive load-sensitive ant colony optimization algorithm, the continuity of the learning path is ensured through the heuristic function, the rationality of the learning path structure is optimized, and the cognitive burden of users is reduced.

[0022] 3. By integrating the Ebbinghaus forgetting curve to build a personalized knowledge decay model, review resources are intelligently inserted and mastered content is removed based on real-time learning performance, improving the long-term retention rate of knowledge and enhancing the accuracy of review resource delivery.

[0023] 4. By establishing a dynamic feedback mechanism driven by the dual-Q network, we can achieve collaborative online optimization of user portraits and resource characteristics. By adaptively adjusting the optimization algorithm parameters to adapt to different cognitive characteristics, we can improve the system's adaptability to differentiated users and ensure long-term recommendation effectiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic diagram of the overall system architecture of the present invention.

[0025] Figure 2 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0026] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0027] See also Figure 1 The method and system for intelligent recommendation of educational resources based on big data drive include the following steps: S1. Collect user learning behavior data and educational resource metadata to generate a user-resource interaction matrix; S2. Construct a multimodal user profile based on the user-resource interaction matrix to generate a user knowledge status profile, a cognitive ability profile, and an interest preference profile; S3. Extract semantic features and conduct knowledge graph association analysis on the educational resource library to generate resource feature vectors and topological relationship diagrams between resources; S4. Calculate the demand-resource matching degree based on the user knowledge status profile and the resource feature vector to generate an initial recommended resource set; S5. Perform dynamic path optimization on the initial recommended resource set based on an improved ant colony optimization algorithm, and generate an optimized recommendation sequence by combining the topological relationship diagram between resources and the user's cognitive ability profile; S6. Personalize the optimized recommendation sequence based on real-time user feedback data and a forgetting curve model to generate a dynamic educational resource recommendation list; S7. Update user profiles and resource feature vectors through reinforcement learning mechanisms to build an adaptive recommendation closed-loop system.

[0028] S1 includes: S11. Obtain user online learning behavior data through a log collector, including video viewing time, exercise accuracy, knowledge point jump path, and interaction frequency, to generate user behavior time series data; S12. Extract resource structured metadata from the educational resource management platform, including resource type, knowledge tag, difficulty coefficient, related knowledge points, and historical usage statistics, to generate a resource feature matrix; S13. Perform tensor fusion on the user behavior time series data and the resource feature matrix to construct a three-dimensional user-resource interaction matrix: Where m is the number of users, n is the number of resources, and k is the behavior dimension.

[0029] S2 includes: S21. Process user behavior time series data based on the knowledge tracking model, predict the probability of users mastering knowledge points, and generate a user knowledge status profile: in Indicates the mastery of the i-th knowledge point; S22. Analyze exercise response data through cognitive diagnostic models, calculate user memory strength, reasoning ability, and information processing speed, and generate cognitive ability profiles: in: is the memory strength, is the information processing speed, For logical reasoning ability; S23. Use the attention mechanism to extract resource browsing hotspot data and combine it with the collaborative filtering algorithm to generate interest preference portraits: in: Label for users The interest weight of , j is the total number of interest tags.

[0030] S3 includes: S31. Use the BERT model to semantically embed the text content of educational resources and generate resource semantic vectors: S32. Construct a resource association graph G = (V, E) based on the knowledge graph, where the vertex V represents the resource and the weight of the edge E is determined by the co-occurrence frequency of knowledge points between resources and the learning path dependency relationship; S33. Fusion of resource semantic vectors and graph topology features to generate resource feature vectors:

[0031] is the vector concatenation operation, The importance score of the resource node in the graph.

[0032] S4 includes: S41. Calculate the cosine similarity between the user knowledge status profile and the resource feature vector: is the user knowledge state vector, Resource feature vector, the matching degree between user u and resource r; S42. Adjust the matching threshold based on the cognitive ability profile and generate an initial recommendation set: is the adaptive threshold coefficient, User information processing speed.

[0033] S5 includes: S51. Model the recommendation problem as a path optimization problem on G and define the pheromone update rule: The pheromone concentration from resource node i to j, is the pheromone volatility coefficient, is the pheromone intensity constant, the cognitive load of path k; S52. Design heuristic function: For resources The difficulty coefficient, is the shortest path distance between resource nodes i, j in the graph; S53. Generate Pareto frontier solutions through multi-objective optimization and select the optimal recommendation sequence based on user cognitive ability profiles. .

[0034] S6 includes: S61. Calculate the knowledge decay weight based on the Ebbinghaus forgetting curve: in is the attenuation factor, is the time interval, is the memory strength parameter; S62. Dynamically adjust the recommendation list based on the real-time answer accuracy rate acc: acc users' real-time answer accuracy rate, Resources to review, Resources are already in hand.

[0035] S7 includes: S71. Build a deep reinforcement learning model to recommend rewards: Completion rate Knowledge gain is the reward function, and is the weight coefficient.

[0036] S72. Update user profiles and resource feature vectors through the dual-Q network to achieve online optimization of recommendation strategies.

[0037] include: Data fusion processing module: including behavior collection unit, metadata extraction unit and tensor construction unit; Portrait modeling module: including knowledge status analysis unit, cognitive ability assessment unit and interest mining unit; Resource feature engine module: including semantic embedding unit, knowledge graph construction unit and feature fusion unit; Intelligent recommendation core module: including matching degree calculation unit, ant colony optimization unit and dynamic adjustment unit; Closed-loop feedback module: includes reinforcement learning agent unit and parameter updating unit.

[0038] The behavior collection unit integrates the tracking SDK and API interface to capture user operation events in real time; The ant colony optimization unit is equipped with a dynamic parameter adjuster that automatically adjusts according to the system load: N is the size of the ant colony, and T is the maximum number of iterations; The reinforcement learning agent unit adopts a distributed architecture, including a mechanism for separating the experience replay pool and the target network.

[0039] Example 1: Implementation process of intelligent recommendation method Step 1: Multimodal data acquisition and fusion 1. Real-time acquisition of user learning behavior data through a tracking collector integrated into the online education platform, including: Video viewing behavior characteristics: pause / fast-forward operation frequency, number of chapters watched repeatedly; Exercise interaction characteristics: response time, error type distribution, and knowledge point jump path; Resource browsing characteristics: duration of stay on the tab page, resource collection and download frequency; 2. Extract structured metadata from the educational resource management platform, including: Resource basic attributes: type (video / exercise / document), knowledge tag system, difficulty level; Resource association attributes: prerequisite knowledge dependency, historical usage statistics; 3. Data fusion processing: Perform time series alignment on user behavior data; Encode resource metadata as a feature vector; Construct a three-dimensional user-resource interaction tensor with dimensions of user ID, resource ID, and behavior characteristics; Step 2: Dynamic portrait construction 1. Knowledge status portrait construction: Enter the user's historical answer sequence; Calculate the probability vector of knowledge point mastery through the deep knowledge tracking model; Output format: in Indicates the mastery of the i-th knowledge point; 2. Cognitive Ability Profile Construction: The item response theory model was used to analyze the response patterns of the exercises; Calculate three core cognitive dimension parameters: Memory Strength Based on the change in correct rate of repeated answers; Information processing speed According to the response time distribution of the questions; Logical reasoning ability Performance assessment through complex problem solving; Output format:

[0040] 3. Construction of interest preference portraits: Apply attention mechanism to analyze resource browsing hot spots; Calculate interest weights using collaborative filtering algorithms; Output format:

[0041] Label for users The interest weight, Resource knowledge label; Step 3: Resource feature modeling 1. Semantic feature extraction: Use the pre-trained BERT model to process resource text content Generate 768-dimensional semantic embedding vector

[0042] 2. Knowledge graph construction: Define a resource node collection and associated edge sets

[0043] Calculate edge weights: Dependency strength is the association weight between resource i and resource j, is the dependence intensity coefficient; Fusion of semantic vectors and graph features: PageRank ; Step 4: Ant Colony Optimization Recommendation 1. Initialization phase: Set the ant colony size N=50 and the pheromone volatility coefficient ρ=0.2; Initialize the pheromone matrix: ; Path exploration phase: Define the state transition probability formula: in, is the probability of transferring from resource i to resource j, The pheromone concentration on path (i, j), Pheromone-inducing factor, is the heuristic function value, is the expectation heuristic factor; Dynamic optimization stage: Automatically adjust the ant colony size N∈[30,80] according to the system load; The pheromone matrix is ​​updated in real time after users complete resources; Step 5: Closed-loop feedback optimization Reinforcement learning mechanism implementation: 1. Define the state space: user portrait vector + resource feature vector 2. Define the action space: recommend a set of decisions 3. Design reward function: 、 is the weight coefficient, Improve the mastery of core knowledge points; 4. Build a dual-Q network architecture: The main network performs real-time recommendation decisions; The target network updates parameters regularly; Set up an experience replay pool to store historical state transition records; Example 2: System Implementation Scenario 1. Application scenario: a provincial online education platform; 2. System deployment architecture: Front-end interaction layer: Student terminal: embed behavior collection SDK to upload learning behavior in real time; Teacher terminal: provides a resource upload interface and generates structured metadata; Backend processing layer: Data fusion processing module: receives multi-source data and generates interaction tensors; Portrait modeling module: dynamically updates three types of user portraits; Resource feature engine: maintains knowledge graph and semantic feature library; Intelligent recommendation core: Execute ant colony optimization algorithm to generate recommendation sequence; Closed-loop feedback module: achieves system self-optimization through reinforcement learning; 3. Typical operation process: New user guidance phase: Perform cognitive ability assessment to initialize cognitive ability profile; Generate an initial interest preference profile through an interest questionnaire; Recommended learning resources with a cognitive load below 0.3; Stages of the learning process: Process behavioral data streams in real time; Update the user knowledge status portrait every 10 minutes; When the correct rate for three consecutive questions is less than 70%: Automatically insert related review resources; Dynamically reduce the difficulty coefficient of subsequent resources; System self-optimization stage: Daily offline batch training: Update semantic encoding model parameters; Retrain the knowledge tracking model; Real-time feedback processing: Asynchronously updating reinforcement learning networks; Dynamically adjust optimization algorithm parameters; Example 3: Vocational Education Skills Improvement Scenario Application Scenario The advanced professional skills training system for employed personnel focuses on solving challenges such as large differences in professional abilities, fragmented learning time, and rapid skill updates.

[0044] Implementation Process 1. Professional Profile Construction Multidimensional ability assessment: Integrate professional qualification certification data to build a basic capability map; Analyze work project experience to extract practical skill tags; Assess implicit competence items through situational testing; Dynamic demand capture: Real-time scanning of industry talent demand reports to generate skill popularity index; Monitor users’ professional social networks to identify development intentions; 2. Intelligent organization of resources Skill unit deconstruction: Break down the course into the smallest skill units; Mark workplace application dependencies between units; Contextual encapsulation: Package skill units according to workplace scenarios; Create cross-position composite resource packages; 3. Fragmented learning adaptation Time intelligence perception: Predict available study time based on user schedule data; Dynamically adjust content granularity, promote micro-courses during commuting hours, and promote practical training projects on weekends; Interrupt and resume mechanism: Record the context of each learning breakpoint; Automatically backtrack to related knowledge points when restarting; 4. Strengthening combat capabilities Virtual workplace sandbox: Build a simulation environment for typical business scenarios in the industry; Set up multi-role collaboration challenge tasks; Real-time feedback system: Corporate mentors remotely review practical assignments; Intelligent comparison of industry benchmark solutions; 5. Development path optimization Dual track system recommendation: Main path: advance systematically according to the target position competency model; Branch path: recommend quick solutions based on immediate work needs; Capacity gap warning: Compare corporate recruitment needs with individual capability maps; Recommend resources for learning emerging skills in advance; Implementation Effect

[0045] After deploying the employee training system for a certain technology company, covering 32 positions in five categories: 1. Core improvements: The speed of mastering key skills increased by 60%; The conversion rate of training content to work reached 85%; Employees’ proactive learning time increased threefold; 2. Typical scenarios: Transferred personnel: System detection of the need to switch from Java development positions to artificial intelligence engineers; Automatically generate the "AI Engineer Transformation Path": Pre-requisite Mathematics Foundation Strengthening Unit; The core recommendation is the Python machine learning practical package; With computer vision elective module; 3. Management promotion: Identify management capacity gaps; Push "Technical Manager Growth Package": Technology decision-making simulation sandbox; Cross-departmental collaboration case library; Strategic planning workshop.

[0046] Example 4: Enterprise cross-departmental collaborative training system Application Scenario A collaborative capability enhancement platform for employees from multiple departments of large enterprises, focusing on resolving pain points such as cross-domain knowledge barriers, low collaboration efficiency, and insufficient innovation momentum.

[0047] Implementation Process 1. Three-dimensional capability modeling Job competency scan: Analyze the job descriptions of each department to extract core competency items; Mining implicit collaboration needs through historical project data; Establish a dynamic capability axis, with professional depth, collaboration breadth, and innovation sharpness; Knowledge flow graph construction: Track internal knowledge sharing paths to form a learning heat map; Identify key bottlenecks in cross-departmental knowledge transfer; 2. Intelligent team formation engine Complementary learning groups: Automatically match team members based on ability coordinate difference values; Ensure that each team includes three types of talents: technical, marketing, and management; Virtual Tutor Configuration: Assign senior staff to each group as "shadow coaches"; Monitor group collaboration data in real time and intervene stealthily; 3. Immersive sandbox simulation Reproduction of real business scenarios: Break down the company's annual key projects into training modules; Embed typical collaboration conflict points, resource competition and decision-making differences; Multi-role simultaneous deduction: Team members play the roles of project stakeholders The system dynamically generates emergency crisis event chains 4. Intelligent collaborative diagnosis Communication effectiveness analysis: Semantic analysis of meeting records to extract the proportion of effective decisions; Tracking the completeness of knowledge transfer, with information decay rate less than 15%; Innovation potential assessment: Record the number of solution iterations and the density of breakthrough ideas; Quantify the cross-domain knowledge integration index; 5. Adaptive Evolutionary System Dynamic course adjustments: Insert a communication skills micro-lesson when the group conflict resolution rate falls below the threshold; Automatically import industry disruption cases when the innovation index is insufficient; Ability Growth Navigation: Generate individual collaboration capability development reports; Recommend cross-departmental rotation practice paths; Implementation Effect After 6 months of implementation at a multinational manufacturing company: Core breakthrough: New product development cycles were shortened by 40%, and cross-departmental collaboration efficiency was improved; The number of innovative proposals increased by 220%, with a knowledge fusion effect; Meeting decision-making efficiency increased from 45% to 82%, and communication quality was optimized; Typical scenarios: Technology market collaboration: Mechanical engineers and marketing specialists team up to complete customer-customized solutions System real-time push technical parameter conversion script guide Virtual mentor triggers "value anchor" training module Crisis Response Drills: Simulate supply chain disruption events; Intelligent diagnosis exposes blind spots in inventory information sharing; Automatically generate supply chain visualization learning solutions.

[0048] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0049] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligently recommending educational resources based on big data, characterized by: The method comprises the following steps: S1. Collect user learning behavior data and educational resource metadata to generate a user-resource interaction matrix; S2. Construct a multimodal user profile based on the user-resource interaction matrix to generate a user knowledge status profile, a cognitive ability profile, and an interest preference profile; S3. Extract semantic features and conduct knowledge graph association analysis on the educational resource library to generate resource feature vectors and topological relationship diagrams between resources; S4. Calculate the demand-resource matching degree based on the user knowledge status profile and the resource feature vector to generate an initial recommended resource set; S5. Perform dynamic path optimization on the initial recommended resource set based on an improved ant colony optimization algorithm, and generate an optimized recommendation sequence by combining the topological relationship diagram between resources and the user's cognitive ability profile; S6. Personalize the optimized recommendation sequence based on real-time user feedback data and a forgetting curve model to generate a dynamic educational resource recommendation list; S7. Update user profiles and resource feature vectors through reinforcement learning mechanisms to build an adaptive recommendation closed-loop system.

2. The method for intelligently recommending educational resources based on big data according to claim 1, characterized in that: Said S1 comprises: S11. Obtain user online learning behavior data through a log collector, including video viewing time, exercise accuracy, knowledge point jump path, and interaction frequency, to generate user behavior time series data; S12. Extract resource structured metadata from the educational resource management platform, including resource type, knowledge tag, difficulty coefficient, related knowledge points, and historical usage statistics, to generate a resource feature matrix; S13. Perform tensor fusion on the user behavior time series data and the resource feature matrix to construct a three-dimensional user-resource interaction matrix: Among them, m is the number of users, n is the number of resources, and k is the behavior dimension.

3. The method for intelligently recommending educational resources based on big data according to claim 1, characterized in that: The S2 includes: S21. Process user behavior time series data based on the knowledge tracking model, predict the probability of users mastering knowledge points, and generate a user knowledge status profile: in, Indicates the mastery of the i-th knowledge point; S22. Analyze exercise response data through cognitive diagnostic models, calculate user memory strength, reasoning ability, and information processing speed, and generate cognitive ability profiles: in: is the memory strength, is the information processing speed, For logical reasoning ability; S23. Use the attention mechanism to extract resource browsing hotspot data and combine it with the collaborative filtering algorithm to generate interest preference portraits: in: Label for users The interest weight of , j is the total number of interest tags.

4. The method for intelligently recommending educational resources based on big data according to claim 1, characterized in that: The S3 includes: S31. Use the BERT model to semantically embed the text content of educational resources and generate resource semantic vectors: S32. Construct a resource association graph G = (V, E) based on the knowledge graph, where the vertex V represents the resource and the weight of the edge E is determined by the co-occurrence frequency of knowledge points between resources and the learning path dependency relationship; S33. Fusion of resource semantic vectors and graph topology features to generate resource feature vectors: in, is the vector concatenation operation, The importance score of the resource node in the graph.

5. The method for intelligently recommending educational resources based on big data according to claim 1, characterized in that: The S4 includes: S41. Calculate the cosine similarity between the user knowledge status profile and the resource feature vector: is the user knowledge state vector, Resource feature vector, the matching degree between user u and resource r; S42. Adjust the matching threshold based on the cognitive ability profile and generate an initial recommendation set: in, is the adaptive threshold coefficient, User information processing speed.

6. The method for intelligently recommending educational resources based on big data according to claim 1, characterized in that: The S5 includes: S51. Model the recommendation problem as a path optimization problem on the graph G and define the pheromone update rule: The pheromone concentration from resource node i to j, is the pheromone volatility coefficient, is the pheromone intensity constant, the cognitive load of path k; S52. Design heuristic function: in, For resources The difficulty coefficient, is the shortest path distance between resource nodes i, j in the graph; S53. Generate Pareto frontier solutions through multi-objective optimization and select the optimal recommendation sequence based on user cognitive ability profiles. .

7. The method for intelligently recommending educational resources based on big data according to claim 1, characterized in that: The S6 includes: S61. Calculate the knowledge decay weight based on the Ebbinghaus forgetting curve: in is the attenuation factor, is the time interval, is the memory strength parameter; S62. Dynamically adjust the recommendation list based on the real-time answer accuracy rate acc: acc users' real-time answer accuracy rate, Resources to review, Resources are already in hand.

8. The method for intelligently recommending educational resources based on big data according to claim 1, characterized in that: The S7 includes: S71. Build a deep reinforcement learning model to recommend rewards: Completion rate Knowledge gain is the reward function, and is the weight coefficient; S72. Update user profiles and resource feature vectors through the dual-Q network to achieve online optimization of recommendation strategies.

9. An intelligent recommendation system for educational resources driven by big data, used to implement the method according to any one of claims 1 to 8, characterized in that: include: Data fusion processing module: including behavior collection unit, metadata extraction unit and tensor construction unit; Portrait modeling module: including knowledge status analysis unit, cognitive ability assessment unit and interest mining unit; Resource feature engine module: including semantic embedding unit, knowledge graph construction unit and feature fusion unit; Intelligent recommendation core module: including matching degree calculation unit, ant colony optimization unit and dynamic adjustment unit; Closed-loop feedback module: includes reinforcement learning agent unit and parameter updating unit.

10. The big data-driven intelligent recommendation system for educational resources according to claim 9, characterized in that: The behavior collection unit integrates the tracking SDK and API interface to capture user operation events in real time; The ant colony optimization unit is equipped with a dynamic parameter adjuster that automatically adjusts according to the system load: N is the size of the ant colony, and T is the maximum number of iterations; The reinforcement learning agent unit adopts a distributed architecture, including a mechanism for separating the experience replay pool and the target network.

Images