A cross-platform education resource intelligent aggregation and accurate pushing method and system

Abstract

The present application relates to the technical field of education resource pushing, in particular to a cross-platform education resource intelligent aggregation and accurate pushing method and system, which is applied to an education resource service platform, and through collecting historical learning behavior records of learners on the platform and historical evaluation information of various resource providers before the learners request resources, a multi-dimensional data base is constructed; based on the above data, learning preferences and resource quality are analyzed, and multiple candidate resource aggregation schemes meeting the adaptation conditions are generated; then, the applicability and accuracy of each candidate scheme are evaluated, the knowledge relevance, learning matching degree and resource credibility are comprehensively considered, and the optimal recommended resource scheme is selected; finally, the scheme is accurately pushed to the learners, and personalized resource acquisition is realized. The method effectively breaks the cross-platform data island, improves the intelligent level of resource aggregation, enhances the accuracy and practicality of recommendation, and optimizes the learning experience.

Description

Technical Field

[0001] This invention relates to the field of educational resource delivery technology, and in particular to a method and system for intelligent aggregation and precise delivery of cross-platform educational resources. Background Technology

[0002] With the rapid development of educational informatization and the in-depth advancement of "Internet + Education," massive educational resources are distributed across multiple platforms and systems, resulting in a situation where data silos and resource fragmentation coexist. When faced with cross-platform, multi-source educational resources, learners often struggle to efficiently filter out high-quality content that matches their learning needs, cognitive levels, and interests, leading to low learning efficiency and insufficient resource utilization. Most existing educational resource recommendation systems are based on simple keyword matching or collaborative filtering algorithms, lacking in-depth analysis of learners' historical behavior and comprehensive evaluation of resource provider service quality, making it difficult to achieve personalized, precise, and intelligent resource recommendations.

[0003] Furthermore, most current platforms neglect the knowledge connections between educational resources during the resource aggregation process, failing to construct a unified knowledge graph system to support the deep integration of cross-domain content. Simultaneously, resource quality varies widely, and a credible, dynamic, and traceable evaluation mechanism is lacking, affecting the authority and reliability of recommendation results. Although some systems have introduced user rating or click feedback mechanisms, the evaluation data is susceptible to noise interference and lacks the ability to perform multi-dimensional data fusion and model-driven accuracy assessment.

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this invention is to provide a method and system for intelligent aggregation and precise delivery of cross-platform educational resources. This aims to solve the technical problems of existing educational resource delivery systems, such as cross-platform data silos, resource fragmentation, low recommendation accuracy, lack of multi-dimensional quality assessment and knowledge association, and difficulty in achieving personalized, intelligent, and precise aggregation and delivery of resources across the entire domain.

[0006] To achieve the above objectives, this invention provides a cross-platform intelligent aggregation and precise delivery method for educational resources, applied to an educational resource service platform. The method includes:

[0007] Before the learner obtains educational resources that match their learning needs, the historical evaluation information of each resource provider in the cross-platform educational resource library is obtained, and the learner's historical learning behavior records through the educational resource service platform are also obtained.

[0008] Based on the historical learning behavior records and the historical evaluation information, multiple candidate resource aggregation schemes that meet the adaptation conditions are determined;

[0009] The applicability accuracy of each candidate resource aggregation scheme is evaluated, and based on the results of the accuracy evaluation, a recommended resource scheme is determined from the candidate resource aggregation schemes.

[0010] The recommended resource plan is pushed to the learner's terminal so that relevant learners can access the educational resources in the recommended resource plan for learning through the learner's terminal.

[0011] Optionally, determining multiple candidate resource aggregation schemes that meet the adaptation conditions based on the historical learning behavior records and the historical evaluation information includes:

[0012] Based on the historical learning behavior records, determine the content preference coefficient of relevant learners when obtaining educational resources through the learner terminal;

[0013] Based on the historical evaluation information, the resource providers are arranged in descending order of evaluation, and a preliminary proportion of candidate resource providers are determined after the arrangement.

[0014] Based on the content preference coefficient and the candidate resource providers, multiple candidate resource aggregation schemes that meet the adaptation conditions are determined.

[0015] Optionally, determining multiple candidate resource aggregation schemes that meet the adaptation conditions based on the content preference coefficient and the candidate resource providers includes:

[0016] Based on the content preference coefficient, determine the multiple core resource types preferred by the relevant learners through the learner's terminal;

[0017] The core resource types and the candidate resource providers are combined using a knowledge graph to generate multiple initial resource schemes.

[0018] Each initial resource scheme is screened for suitability, and based on the results of the suitability screening, multiple candidate resource aggregation schemes that meet the suitability conditions are determined.

[0019] Optionally, the adaptation screening of each initial resource scheme includes:

[0020] The interest score corresponding to each core resource type is determined based on the content preference coefficient of each core resource type.

[0021] Based on the historical evaluation information of each candidate resource provider, the quality score corresponding to each candidate resource provider is determined;

[0022] Based on the preset weights, the interest score, and the quality score, the corresponding adaptation score for each initial resource scheme is determined.

[0023] Based on the adaptation score, the applicability of the relevant learner to select any one of the initial resource schemes is predicted and filtered.

[0024] Optionally, the accuracy assessment of the applicability of each candidate resource aggregation scheme includes:

[0025] The applicability accuracy of each candidate resource aggregation scheme is evaluated using a preset accuracy evaluation model to obtain the accuracy evaluation result. The preset accuracy evaluation model is obtained by training the model to be trained based on feature training samples, the learning effect labels of the feature training samples, and the knowledge matching degree labels of the feature training samples.

[0026] Optionally, the step of using a preset accuracy evaluation model to evaluate the applicability accuracy of each candidate resource aggregation scheme, and obtaining the accuracy evaluation result, includes:

[0027] Using a preset accuracy evaluation model, the knowledge structure feature information of the corresponding educational resources in each candidate resource aggregation scheme is extracted, and the historical service feature information of each resource provider in each candidate resource aggregation scheme is also extracted.

[0028] The knowledge structure feature information, the historical service feature information, and the corresponding adaptation scores of each candidate resource aggregation scheme are respectively processed by feature value transformation to obtain the feature value of each piece of information;

[0029] Based on the aforementioned feature values, the applicability accuracy of each candidate resource aggregation scheme is evaluated, and the accuracy evaluation result is calculated.

[0030] Optionally, before the learner obtains educational resources tailored to their learning needs, and before obtaining historical evaluation information of each resource provider in the cross-platform educational resource library, and before obtaining historical learning behavior records of the learner through the educational resource service platform, the method further includes:

[0031] Obtain resource-related data from each resource provider in the cross-platform educational resource library within a preset evaluation period, wherein the resource-related data is historical data within the preset evaluation period stored on a blockchain.

[0032] Based on preset evaluation indicators, the data content in the historical data is evaluated to obtain multiple evaluation results;

[0033] If all the evaluation results meet the preset evaluation thresholds, a comprehensive evaluation is performed on the multiple evaluation results to obtain historical evaluation information.

[0034] Furthermore, to achieve the above objectives, the present invention also provides a cross-platform intelligent aggregation and precise delivery system for educational resources, the system comprising:

[0035] The data acquisition module is used to acquire historical evaluation information of each resource provider in the cross-platform educational resource library before the learner obtains educational resources that match their learning needs, and to acquire historical learning behavior records of the learner through the educational resource service platform.

[0036] The initial screening module is used to determine multiple candidate resource aggregation schemes that meet the adaptation conditions based on the historical learning behavior records and the historical evaluation information.

[0037] The precision evaluation module is used to evaluate the applicability of each candidate resource aggregation scheme and determine the recommended resource scheme from the candidate resource aggregation schemes based on the results of the precision evaluation.

[0038] The intelligent push module is used to push the recommended resource scheme to the learner's terminal, so that the relevant learner can access the educational resources in the recommended resource scheme for learning through the learner's terminal.

[0039] Furthermore, to achieve the above objectives, the present invention also provides a cross-platform intelligent aggregation and precise push device for educational resources. The device includes: a memory, a processor, and a cross-platform intelligent aggregation and precise push program for educational resources stored in the memory and executable on the processor. The cross-platform intelligent aggregation and precise push program for educational resources is configured to implement the steps of the cross-platform intelligent aggregation and precise push method for educational resources as described above.

[0040] In addition, to achieve the above objectives, the present invention also provides a storage medium storing a cross-platform intelligent aggregation and precise push program for educational resources. When the cross-platform intelligent aggregation and precise push program for educational resources is executed by a processor, it implements the steps of the cross-platform intelligent aggregation and precise push method for educational resources as described above.

[0041] This invention provides a method for intelligent aggregation and precise delivery of cross-platform educational resources. By integrating learners' historical learning behavior records with resource providers' historical evaluation information, the method constructs a personalized, multi-dimensional candidate resource aggregation scheme generation mechanism, achieving precise understanding of learning needs and intelligent matching of resource supply. Through model-driven applicability and accuracy evaluation of candidate schemes, the scientific rigor and reliability of the recommendation results are improved. Combining knowledge association and quality evaluation dimensions, the method effectively breaks down data silos and fragmentation issues in cross-platform educational resources, enhancing semantic connectivity between resources and the overall quality of recommended content. Ultimately, it achieves intelligent aggregation and precise delivery of full-domain educational resources for learners, significantly improving resource discovery efficiency, learning matching accuracy, and user experience, and promoting the development of educational services towards intelligence, personalization, and trustworthiness. Attached Figure Description

[0042] Figure 1 This is a flowchart illustrating an embodiment of the cross-platform intelligent aggregation and precise delivery method for educational resources according to the present invention;

[0043] Figure 2 This is a structural block diagram of an embodiment of the cross-platform intelligent aggregation and precise push system for educational resources of the present invention.

[0044] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0045] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0046] Reference Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the cross-platform intelligent aggregation and precise push method for educational resources according to the present invention, which presents an embodiment of the cross-platform intelligent aggregation and precise push method for educational resources according to the present invention.

[0047] In one embodiment, the cross-platform intelligent aggregation and precise delivery method for educational resources includes:

[0048] Step S100: Before the learner obtains educational resources that match their learning needs, obtain historical evaluation information of each resource provider in the cross-platform educational resource library, and obtain the learner's historical learning behavior records through the educational resource service platform.

[0049] The cross-platform educational resource library can be a collection of educational resources provided by multiple heterogeneous educational platforms, with inconsistent structures and formats. It can serve as the original data source for resource aggregation, supporting the integration and retrieval of educational resources across the entire platform. In this embodiment, the cross-platform educational resource library can access resource metadata and content from different educational platforms through API interfaces, web crawlers, or data exchange protocols. For example, the cross-platform educational resource library can include, but is not limited to, one or more of the following: video course libraries, electronic textbook libraries, and exercise question banks. Resource providers can be institutions or individuals contributing educational resources to the cross-platform educational resource library, such as schools, teachers, and educational enterprises. They can serve as the primary source of resource quality and service credibility, and their historical performance influences resource recommendation priority. Furthermore, resource providers can include, but are not limited to, university teaching teams, online education companies, and independent content creators.

[0050] Historical evaluation information can be a multi-dimensional feedback data set of resources previously provided by the resource provider, which can be used to assess the reliability, content quality, and service stability of the resource provider. In an exemplary embodiment, historical evaluation information can be obtained from structured or semi-structured data such as user ratings, expert reviews, completion rates, and reuse frequency. For example, historical evaluation information may include, but is not limited to, user satisfaction ratings, resource completion rates, and expert quality ratings. The learner end can be a terminal device or interactive interface through which learners access and use educational resources, and can be used as an entry point for behavioral data collection and an exit point for recommendation results presentation, realizing a closed loop of human-computer interaction. In one embodiment, the learner end may include, but is not limited to, mobile learning apps, web browsers, and smart teaching terminals. The educational resource service platform can be a centralized system platform that executes resource aggregation, evaluation, and push logic, and can be used to coordinate the entire process of cross-platform resource access, personalized modeling, and intelligent recommendation. Furthermore, the educational resource service platform can be deployed on a cloud or local server, integrating data processing, model inference, and service scheduling modules. Historical learning behavior records can be operation logs and interaction traces generated by learners on educational resource service platforms, which can be used to reflect learners' cognitive level, interest preferences, and learning path characteristics. In an exemplary embodiment, historical learning behavior records can be obtained by recording behavioral events such as clicks, viewing time, quiz accuracy, and resource collection through event tracking technology. For example, historical learning behavior records may include, but are not limited to, resource browsing sequences, practice quiz records, and learning progress logs.

[0051] Obtaining historical evaluation information from various resource providers in a cross-platform educational resource repository can be achieved by extracting evaluation data associated with each resource provider from the repository. Further, this operation can be implemented by pulling standardized scoring data from each platform in real time through a unified evaluation interface, or by periodically synchronizing evaluation logs from each platform and performing cleaning and normalization processes, thereby providing a basis for modeling resource quality and service credibility. Obtaining historical learning behavior records of learners through the educational resource service platform can be achieved by reading historical interaction data of a specified learner from the platform's behavior log system. In a specific embodiment, this operation can be achieved by querying behavior event streams in a distributed log database based on user ID, or by caching recent behaviors in real time through session context and uploading them in batches to the analysis engine, thereby providing a foundational data source for building personalized learner profiles.

[0052] Step S200: Based on historical learning behavior records and historical evaluation information, determine multiple candidate resource aggregation schemes that meet the adaptation conditions.

[0053] The adaptation criteria can be a set of matching rules used to screen candidate resource aggregation schemes, ensuring that the generated schemes meet the learner's current needs and resource quality requirements. For example, adaptation criteria may include, but are not limited to, knowledge coverage matching, difficulty level consistency, and resource timeliness thresholds. Candidate resource aggregation schemes can be multiple preliminary recommended resource sets formed by combining resources from across platforms based on the adaptation criteria. These can serve as intermediate products in the recommendation decision-making process for subsequent accuracy evaluation and screening. In an exemplary embodiment, candidate resource aggregation schemes are generated based on historical learning behavior records and historical evaluation information, which serve as input for the accuracy evaluation of their applicability.

[0054] Based on historical learning behavior records and evaluation information, multiple candidate resource aggregation schemes that meet the suitability criteria are identified. This can be achieved by integrating learner profiles and resource provider reputation, filtering and combining resources according to suitability criteria to form multiple schemes. Furthermore, this operation can be implemented by generating schemes based on knowledge graph path matching and weighted by resource provider ratings, or by using multi-objective optimization algorithms to balance dimensions such as interest preferences, difficulty matching, and resource authority, thereby generating a preliminary recommendation set that takes into account both personalized needs and resource quality.

[0055] Step S300: Perform an accuracy assessment on the applicability of each candidate resource aggregation scheme, and determine the recommended resource scheme from among the candidate resource aggregation schemes based on the accuracy assessment results.

[0056] The accuracy assessment of applicability can be a model-driven quantitative scoring process for candidate resource aggregation schemes. This process can be used to scientifically rank candidate schemes, suppress noise interference, and improve recommendation reliability. In one specific embodiment, the accuracy assessment of applicability can employ a machine learning model (such as a ranking model or reinforcement learning strategy) to fuse multi-dimensional features for scoring. The recommended resource scheme can be the optimal candidate resource aggregation scheme selected after the accuracy assessment. This scheme can be used as the final resource set pushed to learners, directly serving their learning activities. Furthermore, the recommended resource scheme, output from the accuracy assessment of applicability, is pushed to learners through the educational resource service platform.

[0057] The accuracy evaluation of the applicability of each candidate resource aggregation scheme can be achieved by inputting the candidate schemes into a pre-trained evaluation model and outputting a quantitative applicability score. For example, this operation can be performed by using a Pairwise Learning to Rank model to score each scheme pairwise, or by constructing an end-to-end neural network model to predict the future utility of the scheme using learner historical feedback as a supervision signal, thereby achieving scientific ranking and noise filtering of candidate schemes. Based on the accuracy evaluation results, the recommended resource scheme is determined from each candidate resource aggregation scheme, which can be done by selecting the candidate scheme with the highest evaluation score as the final recommended scheme. In an exemplary embodiment, this operation can be performed by directly selecting the Top-1 scoring scheme as the recommendation result, or by introducing diversity constraints into the Top-K schemes and then performing secondary screening to determine the final scheme implementation, thereby ensuring that the pushed content has high matching degree and high reliability.

[0058] Step S400: Push the recommended resource plan to the learner's terminal so that relevant learners can access the educational resources in the recommended resource plan for learning through the learner's terminal.

[0059] Educational resources can be digital content units that support learning activities, such as videos, texts, and quizzes, and can be used to form the basic content units for resource aggregation and delivery. For example, educational resources may include, but are not limited to, micro-lecture videos, interactive courseware, and adaptive practice questions. Pushing recommended resource solutions to learners can be achieved by sending a data package of recommended resource solutions to learners through the educational resource service platform. Furthermore, this operation can be implemented by pushing and updating the recommended list on the learning interface in real time via WebSocket, or by serializing the solutions into JSON format and responding to frontend requests via a RESTful API, thereby completing a closed loop from intelligent decision-making to user interaction.

[0060] Taking personalized high school physics learning as an example, the cross-platform intelligent aggregation and precise push method for educational resources in this embodiment can be as follows: a high school student frequently watches mechanics-related micro-lessons on a mobile learning app (learner's end) but skips electromagnetism content. The system (educational resource service platform) records the student's viewing time, error patterns, and other historical learning behavior records. Simultaneously, the system obtains historical evaluation information from various resource providers (such as a key high school's teaching and research group or a well-known educational technology company) in the cross-platform educational resource library, including user ratings and completion rates. Based on this, the system generates multiple candidate resource aggregation schemes. For example, scheme A contains highly rated but difficult advanced placement (AP) content, while scheme B contains explanation videos of medium difficulty that are highly relevant to the student's incorrect answers. Subsequently, the applicability accuracy assessment model comprehensively considers the student's cognitive level and resource quality, determining that scheme B is superior. Finally, scheme B is pushed as the recommended resource scheme to the student's app homepage, achieving precise push.

[0061] This embodiment provides a method for intelligent aggregation and precise delivery of cross-platform educational resources. By acquiring historical evaluation information from various resource providers in a cross-platform educational resource database and historical learning behavior records from learners, multiple candidate resource aggregation schemes that meet the suitability criteria are determined based on the historical learning behavior records and historical evaluation information. The applicability and accuracy of each candidate resource aggregation scheme are evaluated, and a recommended resource scheme is determined from the candidate resource aggregation schemes based on the accuracy evaluation results. The recommended resource scheme is then pushed to the learner. This method can embed knowledge relevance and quality evaluation into the resource aggregation stage, breaking down data silos between platforms and enhancing the semantic connectivity of resources. At the same time, by replacing traditional keyword matching or single collaborative filtering with behavior and evaluation dual-drive, the recommendation shifts from passive retrieval to proactive intelligent delivery, improving resource discovery efficiency, demand matching degree, and user experience, and promoting the evolution of educational services towards intelligence, personalization, and trustworthiness.

[0062] In one embodiment, based on historical learning behavior records and historical evaluation information, multiple candidate resource aggregation schemes that meet the adaptation conditions are determined, including:

[0063] Based on historical learning behavior records, determine the content preference coefficient of relevant learners when obtaining educational resources through the learner terminal;

[0064] The content preference coefficient is a numerical indicator quantified based on learners' historical learning behavior records, representing their preference for specific educational resource types or themes. It can be used to fine-grainedly characterize learners' interests, cognitive levels, and learning path features, supporting personalized resource selection. In this embodiment, the content preference coefficient can be generated by analyzing resource browsing frequency, dwell time, repeated visits, and collection / skipping behaviors, using weighted statistics or embedding models to generate a preference vector and normalize it into coefficients. For example, the content preference coefficient may include, but is not limited to, one or more of the following: subject theme preference coefficient, difficulty level preference coefficient, and media format preference coefficient.

[0065] Based on historical learning behavior records, the content preference coefficients of relevant learners when accessing educational resources through the learner's end can be determined. This can be achieved by extracting features and quantitatively modeling historical learning behavior records, outputting coefficients that reflect learners' content preferences. Furthermore, this operation can be implemented by using TF-IDF or BM25 algorithms to weighted statistically analyze the resource tags accessed by learners to generate topic preference coefficients, or by using graph neural networks to aggregate learning paths on a knowledge graph, outputting preference intensity vectors for each knowledge point and normalizing them into coefficients. This allows for fine-grained modeling of individual learning preferences, providing demand-side basis for personalized aggregation.

[0066] Based on the historical evaluation information, the resource providers are arranged in descending order of evaluation, and a preliminary proportion of candidate resource providers are determined after the arrangement.

[0067] The ranking of resource providers from highest to lowest evaluation can be a sequence structure formed by sorting them in descending order based on historical evaluation information. This can be used to establish a dynamic reputation ranking of resource providers, providing a basis for quality screening. In an exemplary embodiment, the ranking of resource providers from highest to lowest evaluation can include, but is not limited to, ranking based on comprehensive scores, ranking based on completion rate, and ranking based on expert review weights. Arranging resource providers in descending order of evaluation based on historical evaluation information can be achieved by aggregating and scoring the historical evaluation information of resource providers and then ranking them in descending order of scores. Furthermore, this operation can be achieved by weighting and summing multi-dimensional evaluation indicators (such as user ratings, completion rate, and reuse frequency) before ranking, or by using Bayesian averaging or confidence intervals to correct for small sample evaluation biases before ranking, thereby constructing a dynamic and traceable resource provider quality ranking mechanism.

[0068] The pre-preset proportion of candidate resource providers can be a set of high-quality providers selected from a list of resource providers sorted from high to low evaluation, representing a fixed percentage (e.g., the top 20%). This set can be used to systematically filter out low-quality or unreliable resource providers, focusing on high-reputation supply sources. In one embodiment, the pre-preset proportion of candidate resource providers can be formed by truncating the sorted list after setting a percentage threshold. Determining the pre-preset proportion of candidate resource providers after sorting can be achieved by truncating the top providers from the sorted list according to a preset percentage (e.g., the top 30%). Furthermore, this operation can be implemented through fixed percentage truncation (e.g., always selecting the top 20%) or adaptive percentage adjustment (dynamically setting the truncation point based on the dispersion of evaluation distribution), thereby systematically excluding low-reputation resource sources and improving the overall quality baseline of candidate resources.

[0069] Based on the content preference coefficient and the candidate resource providers, multiple candidate resource aggregation schemes that meet the adaptation conditions are determined.

[0070] The candidate resource providers can be a subset of resource providers with high historical evaluations that have been selected through quality screening. These can serve as high-quality resource sources, limiting the resource source range for subsequent aggregation schemes and improving the credibility and stability of the recommended content. In this embodiment, the candidate resource providers can be generated by sorting historical evaluation information and selecting a predetermined proportion, serving as the supply-side input for the candidate resource aggregation scheme.

[0071] Based on content preference coefficients and candidate resource providers, multiple candidate resource aggregation schemes that meet the adaptation criteria are determined. This can be achieved by matching and combining the content preference coefficients with the resources owned by the candidate resource providers to generate multiple resource sets that meet the adaptation criteria. Furthermore, this operation can be implemented by weighted sampling of the resource libraries of candidate resource providers using the content preference coefficients as weights to generate diverse schemes, or by selecting resource combinations from candidate resource providers that cover preferred knowledge points and meet adaptation criteria such as difficulty and timeliness under the constraints of a knowledge graph. This allows for the cross-integration of demand-side preferences and supply-side quality, generating highly relevant and reliable candidate schemes.

[0072] For example, in the scenario of adaptive learning in junior high school mathematics, the cross-platform intelligent aggregation and precise push method for educational resources in this embodiment could be as follows: A junior high school student frequently watches videos about function graphs on a learning platform but repeatedly skips algebraic proof content. Based on this, the system calculates their "function topic preference coefficient" as 0.85 and their "proof type preference coefficient" as 0.2. Simultaneously, the system retrieves historical evaluation information from 100 mathematics resource providers in a cross-platform library, sorts them in descending order of comprehensive scores, and selects the top 25% (i.e., the top 25) as candidate resource providers. Subsequently, from these 25 high-reputation providers, the system prioritizes the selection of micro-lessons, interactive exercises, and other resources related to function graphs, combining them into three candidate resource aggregation schemes: Scheme 1 emphasizes dynamic visualization, Scheme 2 emphasizes the association of incorrect questions, and Scheme 3 integrates real-life examples. All schemes meet the adaptation conditions (such as knowledge point coverage and difficulty matching) for subsequent accuracy evaluation.

[0073] This embodiment provides a cross-platform intelligent aggregation and precise push method for educational resources. It determines content preference coefficients based on historical learning behavior records, sorts resource providers according to historical evaluation information and filters out a pre-set proportion of candidate resource providers, and determines multiple candidate resource aggregation schemes that meet the adaptation conditions based on the content preference coefficients and candidate resource providers. By quantitatively modeling learners' interests and cognitive characteristics, constructing a high-quality supply-side pool, and cross-integrating demand-side preferences and supply-side quality to generate candidate schemes, it achieves the dual constraints of fine-grained demand-side preferences and dynamic supply-side quality at the candidate scheme generation stage. This breaks through the limitations of traditional recommendations that rely on only a single dimension, forming an intelligent aggregation logic driven by both demand and supply. This provides a candidate set with high semantic consistency and high credibility for subsequent accuracy evaluation, effectively alleviating the problem of resource fragmentation and improving matching accuracy and user trust.

[0074] In one embodiment, based on the content preference coefficient and the candidate resource providers, multiple candidate resource aggregation schemes that meet the adaptation conditions are determined, including:

[0075] Based on the content preference coefficient, determine the core resource types preferred by relevant learners through the learner's terminal;

[0076] By combining core resource types and candidate resource providers using a knowledge graph, multiple initial resource solutions can be generated.

[0077] Each initial resource scheme is screened for suitability, and based on the results of the suitability screening, multiple candidate resource aggregation schemes that meet the suitability conditions are determined.

[0078] The core resource type can be a resource category with clear educational semantics and functional attributes, mapped from the content preference coefficient to reflect learner preferences. It can serve as a semantic intermediary connecting learner preferences and specific educational resources, guiding the direction of resource aggregation. In this embodiment, the core resource type can be clustered or classified based on the content preference coefficient to identify high-frequency and high-weight resource function types. For example, the core resource type can include, but is not limited to, one or more of video explanations, interactive exercises, and concept diagrams.

[0079] A knowledge graph can be a knowledge system that describes the semantic relationships between educational resources, knowledge points, resource types, and providers in the form of structured triples. It can be used as a semantic bridge for cross-platform resource integration, supporting the logical associations and contextual consistency modeling between resources. In an exemplary embodiment, the knowledge graph can integrate multi-source educational metadata through ontology modeling and construct a unified graph using entity alignment and relation extraction techniques. Furthermore, the knowledge graph can include, but is not limited to, subject knowledge graphs, resource type-knowledge point mapping graphs, and provider-resource capability graphs. Knowledge graph association and combination can be a process of structurally matching core resource types with the resources possessed by candidate resource providers based on semantic paths in the knowledge graph. It can be used to generate initial resource schemes with knowledge coherence and contextual consistency. In a specific embodiment, knowledge graph association and combination relies on the knowledge graph as its structural foundation, with core resource types and candidate resource providers as inputs and initial resource schemes as outputs.

[0080] The initial resource plan can be a preliminary set of resources generated through knowledge graph association and combination, which has not yet undergone adaptability screening. This set serves as the input for adaptability screening, carrying semantic associations and supply-side quality information. For example, the initial resource plan can include, but is not limited to, single-knowledge-point coverage plans, multi-knowledge-point progressive plans, and cross-modal fusion plans. Adaptability screening can be an evaluation process that filters and optimizes the initial resource plan based on preset adaptability conditions. It can be used to eliminate plans that do not meet learning objectives, cognitive levels, or timeliness requirements, thereby improving the quality of the candidate set. In this embodiment, adaptability screening can be based on a rule engine or a multi-dimensional scoring model to judge the compliance and matching degree of the plans. Furthermore, adaptability screening can include, but is not limited to, learning objective matching screening, cognitive difficulty adaptability screening, and resource timeliness screening.

[0081] Based on content preference coefficients, multiple core resource types preferred by learners through the learner's end are determined. This can be achieved by mapping content preference coefficients to a predefined resource type ontology and selecting types with scores above a threshold as core resource types. Furthermore, this operation can be implemented by selecting Top-K resource types as core types after softmax normalization of preference coefficients, or by grouping similar preference behaviors into several core resource type clusters using clustering algorithms. This enables the abstract transformation from behavioral data to educational semantic needs, improving the interpretability of resource matching. Core resource types and candidate resource providers are then associated and combined using a knowledge graph to generate multiple initial resource schemes. This can be achieved by querying educational resource nodes in the knowledge graph that simultaneously associate a specified core resource type with a candidate resource provider, forming multiple schemes according to different paths or combination strategies. Further, this operation can be achieved by combining multiple resource types under the same knowledge point through graph traversal based on the shortest semantic path, or by using a subgraph matching algorithm to extract resource subgraphs that satisfy the "provider-type-knowledge point" ternary constraint as scheme implementations. This enables semantic-level integration of resources across platforms, enhancing the knowledge coherence and logical consistency within the schemes.

[0082] Adaptability screening of initial resource solutions can be performed by verifying each solution individually based on adaptation criteria, retaining those that meet all or a weighted threshold. Furthermore, this operation can be achieved by setting hard rules (e.g., difficulty level must be ≤ learner's current level) for Boolean filtering, or by constructing a multi-dimensional adaptation scoring model to comprehensively score solutions and retain those above the threshold. This filters out semantically reasonable but practically inapplicable solutions, ensuring the candidate set highly matches the learner's current context. Based on the adaptation screening results, multiple candidate resource aggregation solutions that meet the adaptation criteria are determined. This can be achieved by directly using the initial resource solutions that passed the adaptation screening as candidate resource aggregation solutions. Further, this operation can be achieved by retaining all filtered solutions as candidates, or by deduplicating the filtered solutions based on diversity indicators before outputting them. This outputs a high-quality, semantically consistent candidate set for subsequent accuracy evaluation.

[0083] Taking personalized recommendations during the high school biology review stage as an example, the cross-platform intelligent aggregation and precise push method for educational resources in this embodiment can be as follows: A senior high school student's content preference coefficient shows a high preference for "concept diagram type" and "interactive exercise type" resources. Based on this, the system determines these two as core resource types. In the knowledge graph, the knowledge point of "photosynthesis" is associated with resources such as dynamic flowcharts from Education Company A (a candidate resource provider) and drag-and-drop fill-in-the-blank exercises from Research Team B. The system generates an initial resource plan through knowledge graph association and combination: Plan 1 includes diagrams + basic exercises, and Plan 2 includes diagrams + advanced reasoning questions. Subsequently, an adaptability screening is performed, and it is found that the student is currently in the first round of review. The system eliminates Plan 2, which contains advanced reasoning questions, and finally retains Plan 1 as one of the candidate resource aggregation plans to enter the subsequent evaluation process.

[0084] This embodiment provides a method for intelligent aggregation and precise delivery of cross-platform educational resources. It determines multiple core resource types preferred by learners based on content preference coefficients, and generates multiple initial resource schemes by associating these core resource types and candidate resource providers using a knowledge graph. Each initial resource scheme undergoes adaptability screening, and multiple candidate resource aggregation schemes that meet the adaptation criteria are determined based on the screening results. By mapping behavioral data to educational semantic needs, integrating semantic-level resources based on knowledge graphs, and optimizing candidate schemes based on multi-dimensional adaptation criteria, this method achieves the integration of semantic association, supply-side quality, and demand-side preferences. It overcomes the limitations of traditional recommendations that rely solely on statistical correlation, significantly enhancing the logical consistency, educational effectiveness, and interpretability of recommended content. This effectively alleviates the problems of cross-platform resource fragmentation and data silos, providing technical support for high-precision, high-reliability personalized education services.

[0085] In one embodiment, adaptability screening is performed on each initial resource scheme, including:

[0086] The interest score for each core resource type is determined based on the content preference coefficient corresponding to each core resource type.

[0087] The interest score can be a standardized numerical value representing the learner's interest in a resource type, derived from the content preference coefficient corresponding to the core resource type. It can be used to quantify the learner's subjective preference for different resource types, serving as a demand-side input for adapting the score calculation. In this embodiment, the interest score can be generated within the range of 0–1 or 0–100 by normalizing, scaling, or non-linearly mapping (such as Sigmoid) the content preference coefficient. Furthermore, the interest score can include, but is not limited to, one or more of the following: video interest score, exercise interest score, and text / image interest score.

[0088] Based on the magnitude of the content preference coefficients corresponding to each core resource type, the interest score corresponding to each core resource type is determined. This can be achieved by mapping the content preference coefficients to standardized interest scores while preserving the relative relationship of preference intensity. For example, this operation can compress the preference coefficients to the [0, 1] interval as interest scores through linear normalization, or enhance the discriminative power by nonlinearly amplifying the high preference region through a piecewise function. This allows for the transformation from raw behavioral preferences to a numerical representation that can participate in weighted calculations.

[0089] Based on the historical evaluation information of each candidate resource provider, the quality score corresponding to each candidate resource provider is determined;

[0090] The quality score can be a standardized rating reflecting the quality of resource supply, derived from the aggregation of historical evaluation information of candidate resource providers. It can be used to characterize the objective credibility and service stability of resource providers, serving as a supply-side input for the adaptation score calculation. In an exemplary embodiment, the quality score can be generated by normalizing a weighted average of multi-dimensional indicators (such as user ratings, completion rates, and expert ratings) from historical evaluation information. Furthermore, the quality score can include, but is not limited to, one or more of the following: content accuracy quality score, teaching effectiveness quality score, and service stability quality score.

[0091] Based on the historical evaluation information of each candidate resource provider, a quality score is determined for each candidate resource provider. This can be achieved by aggregating the historical evaluation information of candidate resource providers to generate a single quality score. In one specific embodiment, this operation can generate quality scores by performing principal component analysis on multidimensional evaluation indicators to reduce dimensionality, or by using a time-decayed weighted average to assign higher weight to recent evaluations to reflect dynamic reputation. This enables the construction of an objective and comparable measure of resource supply quality, supporting supply-side credibility modeling.

[0092] Based on the preset weights, the interest score, and the quality score, the corresponding adaptation score for each initial resource scheme is determined.

[0093] The preset weights can be configurable parameters used to adjust the relative importance of interest scores and quality scores in the adaptation score calculation. They can be used to achieve a strategic balance between personalized preferences and resource quality, supporting adjustments to the system's recommendation bias in different scenarios. For example, preset weights can include, but are not limited to, one or more of fixed equilibrium weights, adaptive weights for the learning stage, and user-controllable preference weights. The adaptation score can be an overall matching score of the initial resource plan, calculated by combining interest scores and quality scores and weighting them according to the preset weights. It can be used as a unified quantitative indicator to predict learners' preference for plans and support screening decisions. In this embodiment, the adaptation score is generated jointly by the interest score, quality score, and preset weights, and is the direct basis for applicability prediction.

[0094] Based on preset weights, interest scores, and quality scores, the corresponding adaptation score for each initial resource scheme is determined. This can be achieved by weighting the interest scores for resource types and the quality scores for providers within the scheme according to resource composition, and then merging these weights into a single adaptation score. Furthermore, this operation can be performed by calculating the adaptation score for each resource item within the scheme separately (interest weight multiplied by interest score plus quality weight multiplied by quality score), and then averaging the results. Alternatively, the overall interest score and overall quality score of the scheme can be aggregated first, and then weighted according to a global preset weight. This allows for the establishment of a unified multi-objective optimization scoring system that balances subjective preferences and objective quality.

[0095] Based on the adaptation score, the applicability of the relevant learner to select any one of the initial resource schemes is predicted and filtered.

[0096] The suitability prediction process can be a model-based estimation of the likelihood of a learner adopting a particular initial resource option based on the fit score. This can be used to transform static scoring into dynamic behavior prediction, improving the foresight and practicality of the screening results. In a specific embodiment, suitability prediction can employ logistic regression, click-through rate prediction models, or reinforcement learning strategies, using the fit score as the primary feature to predict the selection probability.

[0097] Based on the fit score, the applicability of any solution chosen by a learner from the initial resource options is predicted and filtered. This can be done by using the fit score as the core feature to predict the probability of learners adopting each solution, and retaining solutions that are above a threshold or ranked highly. For example, this operation can be achieved by setting a fit score threshold to retain only solutions above that threshold, or by using a ranking learning model to jointly predict the selection probability and filter the Top-K solutions using the fit score and other contextual features. This transforms static scoring into behavioral prediction, improving the practicality and predictive power of the filtering results.

[0098] Taking university students' introductory programming learning as an example, the cross-platform educational resource intelligent aggregation and precise push method in this embodiment can be as follows: A student's content preference coefficient for "interactive practice type" resources is 0.92, corresponding to an interest score of 92; for "video explanation type" resources, it is 0.65, with a score of 65. Candidate resource provider A (historical evaluation comprehensive score of 4.8 / 5) has a corresponding quality score of 96, and B has 85. The system configures preset weights for the current learning stage: interest accounts for 60%, and quality accounts for 40%. Initial resource scheme 1 is interactive practice provided by A, with an adaptation score equal to the interest weight multiplied by the interest score plus the quality weight multiplied by the quality score, resulting in 93.6; scheme 2 is video explanation provided by B, with an adaptation score of 73.0. Based on the adaptation scores, the system predicts that the student is more likely to choose scheme 1, so scheme 1 is retained in the adaptation screening, scheme 2 is eliminated, and high-quality candidate schemes are output.

[0099] This embodiment provides a cross-platform intelligent aggregation and precise push method for educational resources. It determines interest scores based on the content preference coefficients corresponding to each core resource type, quality scores based on historical evaluation information of each candidate resource provider, and suitability scores based on preset weights, interest scores, and quality scores. Based on the suitability scores, it predicts and filters the applicability of any option from the initial resource plan for relevant learners. By converting content preference coefficients into interest scores to numerically represent the intensity of interest on the demand side, converting historical evaluation information into quality scores to construct an objective measure of the quality reputation on the supply side, introducing preset weights to weight and fuse the two to form a unified suitability score, and using this score to drive a behavioral prediction model for plan selection, it achieves collaborative quantitative modeling of demand-side interest intensity and supply-side quality reputation. This constructs a unified and interpretable multi-objective evaluation system, overcoming the problems of separation between preference and quality and single evaluation dimensions in traditional recommendations. It significantly improves the scientific rigor, robustness, and educational suitability of candidate plans, providing a high-confidence decision-making basis for precise push, and enhancing the system's intelligence level and user trust.

[0100] In one embodiment, the applicability accuracy assessment of each candidate resource aggregation scheme includes:

[0101] The applicability accuracy of each candidate resource aggregation scheme is evaluated using a pre-defined accuracy evaluation model, and the results of the accuracy evaluation are obtained. The pre-defined accuracy evaluation model is obtained by training the model to be trained based on the feature training samples, the learning effect labels of the feature training samples, and the knowledge matching degree labels of the feature training samples.

[0102] The preset accuracy evaluation model can be a machine learning model trained based on multi-dimensional supervised signals, used to quantitatively evaluate the applicability of candidate resource aggregation schemes. It can be used to score the comprehensive performance of candidate resource aggregation schemes in two dimensions: cognitive effectiveness and knowledge structure adaptability. In this embodiment, the preset accuracy evaluation model can be obtained through supervised training of the model to be trained using feature training samples and their corresponding learning effect labels and knowledge matching degree labels. Further, the preset accuracy evaluation model can be one or more of the following: a ranking learning model, a multi-task neural network model, and a graph neural network evaluation model. The feature training samples can be input data instances used to train the accuracy evaluation model, containing multi-dimensional feature representations of candidate resource aggregation schemes. They can be used as the basic data units for model training, carrying joint features of resource combination and learner context. For example, feature training samples can include, but are not limited to, behavior-resource joint feature vectors, knowledge graph embedding features, and multimodal resource meta-features.

[0103] Learning outcome tags can be supervisory signals reflecting the actual learning effectiveness of learners after using a resource or solution. They can be used to guide the model to learn the causal relationship between resources and learning outcomes, thereby improving the educational effectiveness of recommendations. In an exemplary embodiment, learning outcome tags may include, but are not limited to, knowledge point mastery rate, task completion rate, and post-test score improvement. Knowledge matching tags can be supervisory signals measuring the degree of fit between candidate resource aggregation solutions and target learning needs at the knowledge structure level. They can be used to ensure that recommended content has correct prerequisite dependencies, concept levels, or cross-domain associations in the knowledge graph. In a specific embodiment, knowledge matching tags may include, but are not limited to, prerequisite knowledge coverage, target knowledge point path consistency, and cross-disciplinary concept association strength.

[0104] The model to be trained can be an untrained machine learning model architecture used to fit the mapping relationship between feature training samples and multidimensional labels. It can serve as the initial form of a pre-defined accuracy evaluation model, which is then transformed into a usable evaluator through the training process. Furthermore, the model to be trained can employ deep feedforward networks, Transformer encoders, graph convolutional networks, etc. The accuracy evaluation result can be a quantitative score or ranking value output by the pre-defined accuracy evaluation model for each candidate resource aggregation scheme. This can be used as a basis for selecting and recommending resource schemes, reflecting the comprehensive performance of the scheme in terms of educational effectiveness and knowledge adaptability.

[0105] By using a pre-defined accuracy evaluation model, the applicability of each candidate resource aggregation scheme is assessed. This can be achieved by inputting the feature representation of the candidate resource aggregation scheme into the pre-defined accuracy evaluation model to obtain its applicability score. Furthermore, this operation can be implemented by embedding the knowledge graph of each resource in the scheme and concatenating it with the learner's current knowledge state before inputting it into the model, or by constructing a graph structure representation of the scheme as a whole (such as resource nodes + learner state nodes) and performing end-to-end evaluation through a graph neural network. This allows for a scientific quantitative evaluation of candidate schemes in terms of both cognitive effectiveness and knowledge structure adaptability.

[0106] Based on feature training samples, their learning effect labels, and their knowledge matching labels, a pre-defined accuracy evaluation model is trained on the model to be trained. This can be achieved by using feature training samples as input and learning effect labels and knowledge matching labels as joint supervision signals to optimize the parameters of the model to be trained. For example, this operation can be performed using a multi-task learning framework, sharing the underlying feature extraction module, and outputting separate prediction heads for learning effect and knowledge matching, or by designing a weighted loss function to dynamically balance the training weights of the two types of labels. This avoids a single objective dominating the model optimization direction, enabling the model to simultaneously predict learning effectiveness and knowledge fit, thus enhancing the educational significance and interpretability of the evaluation.

[0107] Taking the advanced learning of the concept of functions in junior high school mathematics as an example, the cross-platform intelligent aggregation and precise push method of educational resources in this embodiment can identify that a student has completed the learning of linear functions and needs to advance to quadratic functions. Candidate resource aggregation scheme A contains short videos with high click rates but lacking explanations of image transformations, while scheme B contains interactive courseware with low exposure but complete coverage of vertex-style, image translation, and comparison with linear functions. The feature training samples consist of resource usage records of similar learners in the past, and their learning effect label is the accuracy rate of subsequent tests. The knowledge matching degree label is calculated from the path completeness between the "quadratic function" node and the prerequisite "linear function" in the knowledge graph. After multi-task training, the preset accuracy evaluation model gives a higher score to scheme B because it is better than scheme A in both knowledge matching degree (coverage of key sub-concepts) and learning effect (high mastery rate of historical users). Based on this, the system pushes scheme B, realizing the leap from popular to effective recommendation.

[0108] This embodiment provides a cross-platform intelligent aggregation and precise push method for educational resources. It evaluates the applicability of each candidate resource aggregation scheme using a pre-set accuracy assessment model, yielding the accuracy assessment results. The pre-set accuracy assessment model is trained on a model based on feature training samples, learning effect labels of the feature training samples, and knowledge matching degree labels of the feature training samples. By introducing a pre-set accuracy assessment model jointly supervised by learning effect labels and knowledge matching degree labels, the candidate resource aggregation scheme undergoes a dual evaluation of cognitive effectiveness and knowledge structure adaptability. Furthermore, by using feature training samples and their multi-dimensional labels to train the model, the evaluation process transcends traditional shallow feedback. During the evaluation stage, the model can identify high-value but low-exposure long-tail resources and effectively filter content with only superficial popularity. This supports the scientific rigor and reliability of recommendation mechanisms, promoting resource aggregation from data integration to semantic-level educational effectiveness optimization, achieving a technological leap from accurate recommendations to effective learning.

[0109] In one embodiment, a preset accuracy evaluation model is used to evaluate the applicability accuracy of each candidate resource aggregation scheme, and the accuracy evaluation results are obtained, including:

[0110] Using a pre-defined accuracy evaluation model, the knowledge structure feature information of the corresponding educational resources in each candidate resource aggregation scheme is extracted, and the historical service feature information of each resource provider in each candidate resource aggregation scheme is also extracted.

[0111] The knowledge structure feature information, historical service feature information, and the corresponding adaptation scores of each candidate resource aggregation scheme are respectively processed by feature value transformation to obtain the feature value of each piece of information;

[0112] Based on the eigenvalues, the applicability accuracy of each candidate resource aggregation scheme is evaluated, and the accuracy evaluation results are calculated.

[0113] The knowledge structure feature information can be structured information describing the semantic position, conceptual level, and relationships with other knowledge points of each educational resource in the knowledge graph within the candidate resource aggregation scheme. This information is used to measure the consistency and completeness of the resource content and learning objectives at the knowledge system level. In this embodiment, the knowledge structure feature information can be extracted based on a unified educational knowledge graph, including embedding vectors, prerequisite / successor paths, and cross-domain connection strength of the knowledge points covered by the resources. For example, the knowledge structure feature information may include, but is not limited to, one or more of the following: knowledge point embedding vectors, prerequisite / successor path lengths, and interdisciplinary concept bridging degree. The historical service feature information can be a multi-dimensional set of indicators reflecting the past service quality of each resource provider in the candidate resource aggregation scheme. This information can be used to evaluate the credibility, stability, and service continuity of the resource providers. In an exemplary embodiment, the historical service feature information can be generated by aggregating historical evaluation information, including dimensions such as response timeliness, content update frequency, and user complaint rate. Further, the historical service feature information may include, but is not limited to, one or more of the following: service response timeliness, content version update cycle, and user negative feedback rate.

[0114] The fit score can be a quantitative score of the degree of matching between the candidate resource aggregation scheme and the learner's needs obtained in the initial screening stage. It can be used as an initial matching basis and participate in feature fusion for subsequent accuracy evaluation. In a specific embodiment, the fit score can be calculated based on historical learning behavior records and historical evaluation information in step S200. The feature value can be a numerical representation after standardization or normalization, used to uniformly express knowledge structure feature information, historical service feature information, and the fit score. It can be used to achieve comparability of heterogeneous features and model input capability, supporting multi-source feature fusion evaluation. For example, the feature value can include, but is not limited to, one or more of the following: normalized fit score, knowledge graph embedding projection value, service reputation standardization index, etc.

[0115] By using a pre-defined accuracy evaluation model, the knowledge structure feature information of the corresponding educational resources in each candidate resource aggregation scheme is extracted. This can be achieved by calling a knowledge graph interface to parse the knowledge points and topological relationships associated with each educational resource in the scheme, generating structured features. Further, this operation can be achieved by extracting the TransE or RotatE graph embedding vectors of the knowledge points covered by each resource and aggregating them into a scheme-level representation, or by retrieving its location and association path in the curriculum standards within the educational ontology based on knowledge tags in the resource metadata. This allows resource content to be mapped to a unified semantic space, supporting deep matching at the knowledge level. The pre-defined accuracy evaluation model also extracts the historical service feature information of each resource provider in each candidate resource aggregation scheme. This can be achieved by extracting statistical indicators related to service quality from the resource provider's historical evaluation information, forming feature vectors. In a specific embodiment, this operation can be achieved by calculating the resource provider's average update interval and timeliness compliance rate over the past three months, or by aggregating its historical resource user complaint rate, customer service response time, and other operational indicators. This introduces a supply-side credibility dimension, enhancing the authority and stability of the recommendation results.

[0116] Transforming knowledge structure feature information into feature values ​​can involve standardizing, reducing dimensionality, or embedding projections of the original knowledge structure features to convert them into numerical feature values. For example, this operation can be achieved by using Min-Max normalization to map discrete indicators such as path length to the [0, 1] interval, or by using a graph attention network to aggregate multi-hop knowledge relationships into a fixed-dimensional vector, thus allowing semantic structure information to be processed uniformly by the model. Transforming historical service feature information into feature values ​​can involve normalizing or weighted synthesis of service indicators to generate feature values ​​with a uniform scale. In an exemplary embodiment, this operation can be achieved by using Z-score standardization to process continuous variables such as service response time, or by using expert rules to weighted summation of multiple indicators to generate a comprehensive service reputation score, thereby eliminating differences in the dimensions of different service indicators and supporting multi-dimensional fusion.

[0117] The matching scores of each candidate resource aggregation scheme are transformed using feature values. This can involve scaling or nonlinear transformation of the initial matching scores to bring them within a fusionable range with other features. Further, this operation can be achieved by applying a sigmoid function to compress the matching scores to the (0, 1) interval, or by retaining the original scores but adding a smoothing bias to avoid extreme values ​​dominating the result. This ensures that the initial matching signal participates reasonably in the weight allocation of the evaluation model. Based on the feature values, the applicability accuracy of each candidate resource aggregation scheme is evaluated. This can be achieved by concatenating or weighting the three types of feature values ​​and inputting them into a preset accuracy evaluation model, outputting a comprehensive applicability score. In a specific embodiment, this operation can be achieved by using a fully connected neural network to perform nonlinear fusion scoring of the concatenated features, or by using an attention mechanism to dynamically weight the importance of different feature dimensions. This allows for joint decision-making regarding knowledge adaptability, service credibility, and initial matching degree. The calculated accuracy evaluation results can be used to summarize the model output and generate the final evaluation score or ranking number of each candidate resource aggregation scheme, thus providing a quantitative basis for determining the recommended resource scheme.

[0118] Taking the recommendation of introductory university programming courses as an example, the cross-platform intelligent aggregation and precise push method for educational resources in this embodiment can be used for a computer science freshman who needs to learn Python basics. The system generates two candidate resource aggregation schemes: Scheme A includes MOOC videos from well-known universities (with high adaptation scores), but the course has not been updated in the past two years; Scheme B comes from an emerging educational platform, covering the latest syntax features and providing interactive exercises, but the platform has a shorter history. The preset accuracy evaluation model extracts the following features: Scheme A's knowledge structure features show that it covers core syntax but lacks new knowledge points such as asynchronous programming, and its historical service features show that updates have stagnated; Scheme B has a complete knowledge structure and is highly aligned with the current teaching syllabus, and its service features show high-frequency updates and a low complaint rate. After feature value transformation, the three types of features are input into the model. Scheme B, due to its superior knowledge structure and high service activity, obtains a higher accuracy evaluation result and is ultimately pushed, reflecting the dual advantages of being suitable for learning and trustworthy.

[0119] This embodiment provides a cross-platform intelligent aggregation and precise push method for educational resources. It extracts the knowledge structure feature information of corresponding educational resources from each candidate resource aggregation scheme using a pre-set accuracy evaluation model, and extracts the historical service feature information corresponding to each resource provider from each candidate resource aggregation scheme. The knowledge structure feature information, historical service feature information, and the corresponding adaptation scores of each candidate resource aggregation scheme are then processed to obtain feature values ​​for each piece of information. Based on these feature values, the applicability accuracy of each candidate resource aggregation scheme is evaluated, and the accuracy evaluation result is calculated. This is achieved by mapping the knowledge structure feature information to a unified semantic space. By supporting deep matching at the knowledge level, aggregating service quality indicators from historical evaluation information to enhance recommendation authority, standardizing heterogeneous features to achieve multi-source fusion, and jointly inputting three types of feature values ​​into the evaluation model to complete comprehensive decision-making, this approach can incorporate knowledge semantic location, service credibility, and demand matching degree into the same quantitative framework. This allows the evaluation to go beyond superficial popularity and focus on educational effectiveness and supply reliability. It effectively integrates the knowledge relevance and service quality of cross-platform resources, breaks down data silos, and enhances the authority, personalization, and overall teaching value of recommended content, achieving the technical effect of efficient, reliable, and intelligent precise delivery of educational resources.

[0120] In one embodiment, before the learner obtains educational resources tailored to their learning needs, and before obtaining historical evaluation information from various resource providers in the cross-platform educational resource library, and before obtaining the learner's historical learning behavior records through the educational resource service platform, the method further includes:

[0121] Obtain resource-related data from each resource provider in the cross-platform educational resource library within a preset evaluation period. The resource-related data is historical data within the preset evaluation period stored on blockchain.

[0122] Based on the preset evaluation indicators, various data contents in the historical data are evaluated to obtain multiple evaluation results;

[0123] If multiple evaluation results meet the preset evaluation thresholds, a comprehensive evaluation of the multiple evaluation results is performed to obtain historical evaluation information.

[0124] The preset evaluation period can be used to define the time window for collecting and evaluating historical data from resource providers, ensuring that the evaluation is based on the latest, continuous, and representative behavioral data, and supporting a dynamic update mechanism. In this embodiment, the preset evaluation period can include, but is not limited to, one or more of the following: a semester cycle, a monthly cycle, or a project-based learning cycle. Resource-related data can be the original operation and feedback data related to the educational resources published by the resource provider within the preset evaluation period, which can be used as the basic input for generating historical evaluation information, reflecting the actual usage effect and service quality of the resources. Furthermore, resource-related data can be obtained by extracting log data such as user interactions, completion rates, and reuse frequencies associated with a specific resource provider from a cross-platform educational resource library. For example, resource-related data can include, but is not limited to, one or more of the following: resource access frequency, learning completion rate, and user feedback records.

[0125] Blockchain can be a distributed, immutable ledger technology architecture that provides trusted storage and traceability mechanisms for historical data within a preset evaluation period, preventing data forgery or tampering. In an exemplary embodiment, blockchain can use smart contracts to write the hash values ​​of resource-related data into consortium blockchains or private blockchain nodes to achieve data notarization. For example, blockchains can employ one or more of consortium blockchains, private blockchains, and hybrid blockchains. The historical data within the preset evaluation period can be a set of original resource-related data generated by the resource provider within a specified time window and notarized by the blockchain. This data can serve as an objective basis for multi-dimensional evaluation, ensuring the authenticity and timeliness of the data source. Preset evaluation indicators can be multi-dimensional quality metrics used for structured evaluation of historical data content, transforming subjective quality judgments into calculable and comparable quantitative dimensions. In a specific embodiment, preset evaluation indicators can include, but are not limited to, one or more of content accuracy indicators, teaching effectiveness indicators, and user satisfaction indicators.

[0126] The historical data can consist of specific data items that can be independently evaluated within a preset evaluation period. These items can be used as input variables for different preset evaluation indicators, supporting fine-grained evaluation. For example, the historical data may include, but is not limited to, one or more of the following: video completion rate, exercise accuracy distribution, and resource update frequency. Multiple evaluation results can be quantitative scores or grades calculated separately for each data item based on preset evaluation indicators. These can be used to construct a multi-dimensional characterization of the resource provider's service quality, serving as a prerequisite for threshold filtering and comprehensive evaluation. The preset evaluation threshold can be a minimum passing standard set for each evaluation result, serving as a quality filtering threshold to ensure that only qualified resource providers enter the comprehensive evaluation stage. In an exemplary embodiment, the preset evaluation threshold may include, but is not limited to, one or more of the following: content accuracy threshold, teaching effectiveness threshold, and user satisfaction threshold.

[0127] A holistic comprehensive evaluation can be a high-level assessment process that weights and integrates multiple evaluation results or logically combines them after all individual evaluation results meet their corresponding thresholds. This can be used to generate unified and reliable historical evaluation information for subsequent resource aggregation and recommendation. Furthermore, the holistic comprehensive evaluation can employ weighted average methods, fuzzy comprehensive evaluation methods, or multi-attribute decision models (such as TOPSIS) for fusion. Obtaining resource-related data from each resource provider within a preset evaluation period in the cross-platform educational resource database can be achieved by extracting original usage and feedback data associated with each resource provider from the cross-platform educational resource database according to time windows. Further, this operation can be implemented by querying resource call records of each resource provider in a distributed log database within a timestamp range, or by calling standardized data interfaces provided by each platform to periodically retrieve structured resource usage statistics, thus providing a timely and comprehensive data foundation for reliable evaluation.

[0128] Based on blockchain storage, historical data within a preset evaluation period can be stored by writing resource-related data or its hash digests into the blockchain network to achieve tamper-proof evidence preservation. Furthermore, this operation can be achieved by encrypting the original data and storing it off-chain (such as IPFS), only uploading the hash value to the blockchain, or by automatically triggering the data uploading process to the blockchain periodically in batches via smart contracts. This ensures the authenticity, integrity, and auditability of historical data, enhancing the credibility of the evaluation mechanism. Based on preset evaluation indicators, various data items in the historical data are evaluated to obtain multiple evaluation results. This can be achieved by mapping each data item to a corresponding preset evaluation indicator and performing quantitative calculations to generate individual scores. Further, this operation can be achieved by using a rule engine to match data content and indicator weights to output standardized scores, or by training a lightweight regression model to use data content as features to predict the scores of each indicator. This allows for a multi-dimensional and structured characterization of the resource provider's service quality. When multiple evaluation results meet preset evaluation thresholds, a comprehensive evaluation of the multiple evaluation results is performed to obtain historical evaluation information. This can be achieved by first verifying whether all individual evaluations meet the standards, then performing fusion calculations on the compliant items to output the final historical evaluation information. Furthermore, this operation can be achieved by using a weighted summation method to assign weights to the compliance items according to their importance and then synthesizing the overall evaluation, or by constructing a multi-level decision tree to dynamically select a comprehensive evaluation strategy based on the threshold satisfaction status. This can ensure that historical evaluation information has comprehensiveness, reliability, and high-quality thresholds, and prevent low-quality resources from being mixed into the recommendation process.

[0129] Taking the resource access assessment of university MOOC platforms as an example, the cross-platform intelligent aggregation and precise push method for educational resources in this embodiment can involve a provincial-level educational resource service platform accessing MOOC courses from multiple universities. The system sets a preset evaluation period of one semester. At the end of each semester, it automatically obtains resource-related data from each university platform, such as course completion rate, test pass rate, and student comment sentiment, and writes the hash values ​​of this data into the educational consortium blockchain. Subsequently, the system scores each data point based on preset evaluation indicators (such as content accuracy ≥ 0.85, teaching effectiveness ≥ 0.8, and satisfaction ≥ 4.0 / 5.0). If a university course meets all three indicators, a comprehensive evaluation is triggered, and a weighted average method is used to generate its historical evaluation information (such as a comprehensive score of 4.3). This information is then used to generate subsequent candidate resource aggregation schemes, ensuring that only high-credibility courses are included in the recommendation pool.

[0130] This embodiment provides a method for intelligent aggregation and precise delivery of cross-platform educational resources. It acquires resource-related data from various resource providers within a preset evaluation period, where the data is historical data stored on a blockchain within the preset evaluation period. Based on preset evaluation indicators, it evaluates various data items in the historical data, obtaining multiple evaluation results. When all multiple evaluation results meet preset evaluation thresholds, a comprehensive evaluation is performed to obtain historical evaluation information. By acquiring resource-related data from resource providers within the preset evaluation period and storing historical data immutably on a blockchain to ensure the authenticity and traceability of the data source, and by performing a structured evaluation of the historical data content based on multi-dimensional preset evaluation indicators to generate multiple evaluation results, a preset evaluation threshold is set as a quality filtering mechanism. The comprehensive evaluation is only performed when all key dimensions meet the standards. This method produces highly reliable, dynamically updated historical evaluation information, replacing the traditional single-scoring method susceptible to noise interference. It provides high-quality, multi-dimensional, and reliable input for candidate resource aggregation schemes, effectively supporting quality screening and knowledge association integration in precise delivery, and promoting the evolution of educational resources from fragmented accumulation to reliable intelligent aggregation.

[0131] In addition, refer to Figure 2 To achieve the above objectives, the present invention also provides a cross-platform intelligent aggregation and precise delivery system for educational resources, the system comprising:

[0132] The data acquisition module 10 is used to acquire historical evaluation information of each resource provider in the cross-platform educational resource library before the learner obtains educational resources that are suitable for their learning needs, and to acquire historical learning behavior records of the learner through the educational resource service platform.

[0133] The initial screening module 20 is used to determine multiple candidate resource aggregation schemes that meet the adaptation conditions based on the historical learning behavior records and the historical evaluation information.

[0134] The precision evaluation module 30 is used to perform a precision evaluation on the applicability of each candidate resource aggregation scheme, and to determine the recommended resource scheme from the candidate resource aggregation schemes based on the results of the precision evaluation.

[0135] The intelligent push module 40 is used to push the recommended resource scheme to the learner's terminal, so that the relevant learner can obtain the educational resources in the recommended resource scheme for learning through the learner's terminal.

[0136] Other embodiments or specific implementations of the cross-platform educational resource intelligent aggregation and precise push system described in this invention can be referred to the above-mentioned method embodiments, and will not be repeated here.

[0137] Furthermore, to achieve the above objectives, the present invention also provides a cross-platform intelligent aggregation and precise push device for educational resources. The device includes: a memory, a processor, and a cross-platform intelligent aggregation and precise push program for educational resources stored in the memory and executable on the processor. The cross-platform intelligent aggregation and precise push program for educational resources is configured to implement the steps of the cross-platform intelligent aggregation and precise push method for educational resources as described above.

[0138] In addition, to achieve the above objectives, the present invention also provides a storage medium storing a cross-platform intelligent aggregation and precise push program for educational resources. When the cross-platform intelligent aggregation and precise push program for educational resources is executed by a processor, it implements the steps of the cross-platform intelligent aggregation and precise push method for educational resources as described above.

[0139] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for intelligent aggregation and precise delivery of cross-platform educational resources, characterized in that, Applied to an educational resource service platform, the method includes: Before the learner obtains educational resources that match their learning needs, the historical evaluation information of each resource provider in the cross-platform educational resource library is obtained, and the learner's historical learning behavior records through the educational resource service platform are also obtained. Based on the historical learning behavior records and the historical evaluation information, multiple candidate resource aggregation schemes that meet the adaptation conditions are determined; The applicability accuracy of each candidate resource aggregation scheme is evaluated, and based on the results of the accuracy evaluation, a recommended resource scheme is determined from the candidate resource aggregation schemes. The recommended resource plan is pushed to the learner's terminal so that relevant learners can access the educational resources in the recommended resource plan for learning through the learner's terminal.

2. The cross-platform intelligent aggregation and precise delivery method for educational resources as described in claim 1, characterized in that, The step of determining multiple candidate resource aggregation schemes that meet the adaptation conditions based on the historical learning behavior records and the historical evaluation information includes: Based on the historical learning behavior records, determine the content preference coefficient of relevant learners when obtaining educational resources through the learner terminal; Based on the historical evaluation information, the resource providers are arranged in descending order of evaluation, and a preliminary proportion of candidate resource providers are determined after the arrangement. Based on the content preference coefficient and the candidate resource providers, multiple candidate resource aggregation schemes that meet the adaptation conditions are determined.

3. The method for intelligent aggregation and precise delivery of cross-platform educational resources as described in claim 2, characterized in that, The step of determining multiple candidate resource aggregation schemes that meet the adaptation conditions based on the content preference coefficient and the candidate resource providers includes: Based on the content preference coefficient, determine the multiple core resource types preferred by the relevant learners through the learner's terminal; The core resource types and the candidate resource providers are combined using a knowledge graph to generate multiple initial resource schemes. Each initial resource scheme is screened for suitability, and based on the results of the suitability screening, multiple candidate resource aggregation schemes that meet the suitability conditions are determined.

4. The method for intelligent aggregation and precise delivery of cross-platform educational resources as described in claim 3, characterized in that, The adaptation screening of each initial resource scheme includes: The interest score corresponding to each core resource type is determined based on the content preference coefficient of each core resource type. Based on the historical evaluation information of each candidate resource provider, the quality score corresponding to each candidate resource provider is determined; Based on the preset weights, the interest score, and the quality score, the corresponding adaptation score for each initial resource scheme is determined. Based on the adaptation score, the applicability of the relevant learner to select any one of the initial resource schemes is predicted and filtered.

5. The method for intelligent aggregation and precise delivery of cross-platform educational resources as described in claim 1, characterized in that, The accuracy assessment of the applicability of each candidate resource aggregation scheme includes: The applicability accuracy of each candidate resource aggregation scheme is evaluated using a preset accuracy evaluation model to obtain the accuracy evaluation result. The preset accuracy evaluation model is obtained by training the model to be trained based on feature training samples, the learning effect labels of the feature training samples, and the knowledge matching degree labels of the feature training samples.

6. The cross-platform intelligent aggregation and precise delivery method for educational resources as described in claim 5, characterized in that, The step of using a preset accuracy evaluation model to evaluate the applicability of each candidate resource aggregation scheme and obtaining the accuracy evaluation results includes: Using a preset accuracy evaluation model, the knowledge structure feature information of the corresponding educational resources in each candidate resource aggregation scheme is extracted, and the historical service feature information of each resource provider in each candidate resource aggregation scheme is also extracted. The knowledge structure feature information, the historical service feature information, and the corresponding adaptation scores of each candidate resource aggregation scheme are respectively processed by feature value transformation to obtain the feature value of each piece of information; Based on the aforementioned feature values, the applicability accuracy of each candidate resource aggregation scheme is evaluated, and the accuracy evaluation result is calculated.

7. The method for intelligent aggregation and precise delivery of cross-platform educational resources as described in claim 1, characterized in that, Before the learner obtains educational resources tailored to their learning needs on the learner's end, and before obtaining historical evaluation information of each resource provider in the cross-platform educational resource library, and before obtaining historical learning behavior records of the learner through the educational resource service platform, the method further includes: Obtain resource-related data from each resource provider in the cross-platform educational resource library within a preset evaluation period, wherein the resource-related data is historical data within the preset evaluation period stored on a blockchain. Based on preset evaluation indicators, the data content in the historical data is evaluated to obtain multiple evaluation results; If all the evaluation results meet the preset evaluation thresholds, a comprehensive evaluation is performed on the multiple evaluation results to obtain historical evaluation information.

8. A cross-platform intelligent aggregation and precise delivery system for educational resources, characterized in that, The system includes: The data acquisition module is used to acquire historical evaluation information of each resource provider in the cross-platform educational resource library before the learner obtains educational resources that match their learning needs, and to acquire historical learning behavior records of the learner through the educational resource service platform. The initial screening module is used to determine multiple candidate resource aggregation schemes that meet the adaptation conditions based on the historical learning behavior records and the historical evaluation information. The precision evaluation module is used to evaluate the applicability of each candidate resource aggregation scheme and determine the recommended resource scheme from the candidate resource aggregation schemes based on the results of the precision evaluation. The intelligent push module is used to push the recommended resource scheme to the learner's terminal, so that the relevant learner can access the educational resources in the recommended resource scheme for learning through the learner's terminal.

9. A cross-platform intelligent aggregation and precise delivery device for educational resources, characterized in that, The device includes: a memory, a processor, and a cross-platform intelligent aggregation and precise push program for educational resources stored in the memory and executable on the processor, wherein the cross-platform intelligent aggregation and precise push program for educational resources is configured to implement the steps of the cross-platform intelligent aggregation and precise push method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium stores a cross-platform educational resource intelligent aggregation and precise push program. When the processor executes the cross-platform educational resource intelligent aggregation and precise push program, it implements the steps of the cross-platform educational resource intelligent aggregation and precise push method as described in any one of claims 1 to 7.

Images