Metadata structure, data processing method and storage medium

Abstract

The invention provides a metadata structure, a data processing method and a storage medium. The metadata structure is a structured data object on a computer readable medium, can be read and written by a computer program, and adopts a framework of three parts: a basic index part, which is used for uniquely identifying educational resources and supporting retrieval, and at least comprises a unique identification field and a resource classification / retrieval attribute field; the core attainment vector part is used for quantitatively representing an educational resource childbearing target for intelligent calculation and comprises an N-dimensional attainment vector, the dimension key is c01 to cN, and the value of each dimension is a real number in a preset numerical value interval; and the association application part is used for constructing an inter-resource relationship and learning evidence agreement which at least comprises a field for describing a resource dependency relationship and a rule field which is executable by a computer and is used for extracting evidence from learning interaction data. According to the scheme, structuralization and computability of the attestation information are realized.

Description

Technical Field

[0001] This disclosure belongs to the field of artificial intelligence education technology, and in particular relates to a metadata structure, data processing method and storage medium. Background Technology

[0002] With the deep integration of information technology into education, digital educational resources in the form of videos, interactive courseware, online assessments, and virtual experiments have experienced explosive growth. In order to effectively manage, retrieve, and apply these massive and heterogeneous resources, computer systems generally adopt metadata technology, that is, "data describing data".

[0003] Currently, the mainstream metadata technology solutions in this field mainly include the following categories:

[0004] The first category consists of description standards based on general attributes, such as IEEE LOM. This standard provides a multi-layered descriptive framework for learning objects, covering nine categories including general, technical, and educational. This approach addresses, to some extent, the basic discoverability and interoperability of educational resources. However, the granularity of its "educational" category description is very coarse. For example, the `learningResourceType` field can only distinguish between resource formats such as "practice" and "simulation," and the `difficulty` field is only a vague qualitative description. This descriptive approach limits computer systems to a shallow and superficial understanding of resources.

[0005] The second type: Indexing systems based on knowledge point tags. To improve the accuracy of retrieval and recommendation, major online education platforms and content providers generally adopt the method of tagging resources with "knowledge points." This approach links resources with specific knowledge points in the course syllabus, improving content relevance compared to the first type. However, it is essentially still a descriptive model centered on "knowledge transfer."

[0006] The third category consists of data specifications based on learning behavior tracking, such as xAPI and IMS Caliper. These technologies do not directly describe the educational resources themselves, but rather record the interactions between learners and the resources. While crucial for learning process analysis, they cannot answer a fundamental question: "What are the design goals of this resource itself? Besides delivering knowledge, what deeper abilities does it aim to cultivate in students?"

[0007] Modern educational theory has long since moved beyond the mere transmission of knowledge points, shifting towards the comprehensive cultivation of students' overall qualities. However, the aforementioned existing metadata technologies, when supporting next-generation intelligent education systems, have exposed the following serious technical bottlenecks:

[0008] The semantic gap in recommendations: Computer recommendation systems rely on metadata to understand content. When metadata only reaches the "knowledge point" level, the system can only make recommendations based on knowledge point associations, failing to understand and match the underlying literacy development goals of resources, resulting in recommendations that are out of touch with students' actual development needs.

[0009] The student profiles are often limited in scope: Due to the lack of metadata describing the content literacy objectives, computer systems struggle to effectively correlate and attribute students' learning behaviors to their literacy development. This results in computer-generated student profiles typically being limited to the level of knowledge mastery, failing to construct multidimensional, in-depth models that reflect students' comprehensive literacy development.

[0010] The analysis is costly and uncertain: Some systems attempt to use complex AI models to directly analyze raw resource content to guess its pedagogical intent. This approach is not only extremely computationally expensive, but the results are also highly uncertain and "black box" in nature, lacking the support and interpretability of educational theory.

[0011] In summary, the core deficiency of existing technologies lies in the lack of a structured description of the "competency development goals" carried by educational resources that transcend the knowledge points themselves. This inherent deficiency in underlying data has become a key technological bottleneck restricting the deep personalization, adaptability, and competency-based development of intelligent education systems. Summary of the Invention

[0012] The purpose of this disclosure is to address the problems in existing technologies where metadata descriptions are limited to a single dimension, failing to represent the deep educational value of resources, and leading to technical bottlenecks in intelligent recommendation and student profiling applications. To achieve the aforementioned objectives, this technical solution proposes a complete, closed-loop technical process that can be implemented by a computer system. The core of this solution is a three-part metadata structure, and a detailed explanation is provided of its method for processing educational resource data based on N-dimensional literacy representation in a computer system. By constructing this three-part metadata structure, abstract literacy goals are transformed into computer-computable N-dimensional vectors, and based on this, vectorized personalized recommendation and student profile updating methods are implemented.

[0013] To achieve the above objectives, the technical solution disclosed herein is summarized as follows:

[0014] On the one hand, this disclosure provides a metadata structure, which is a structured data object stored on a computer-readable medium and read and written by a computer program. The metadata structure adopts a three-part architecture, including: a basic index part, a core competency vector part, and an associated application part.

[0015] The basic index section is used to uniquely identify educational resources and support retrieval, and includes at least a unique identifier field and attribute fields for resource classification or retrieval; the core competency vector section is used to quantitatively represent the educational goals of educational resources for intelligent computing, and includes a competency vector with N dimensions, where the keys of the dimensions are c01 to cN, and the values ​​of each dimension are real numbers within a preset numerical range.

[0016] The associated application portion is used to construct relationships between resources and learning evidence conventions, including at least fields for describing resource dependencies and rule fields for defining rules that can be executed by a computer to extract evidence from learning interaction data.

[0017] The basic index section includes at least one attribute field for resource classification or retrieval: resource type, subject, grade range, knowledge point, and difficulty. The resource type is at least one of a predefined enumeration, including video, interactive module, assessment, reading material, and simulation / simulation categories. The subject uses national or international standard subject classification codes. The grade range is represented by a closed interval consisting of two integers [min, max]. The knowledge point field is bidirectionally associated with nodes in the external knowledge graph via a unique identifier or URI. The difficulty field takes a real number within a preset range, such as a real number in the range [0, 1], and is calibrated using item response theory.

[0018] The core competency vector component can be indexed in a vector database to support vector retrieval based on inner product or cosine similarity.

[0019] The fields used to describe resource dependencies in the associated application section include a set of preceding resources and a set of succeeding resources, which are used to construct a directed acyclic graph of the learning path. The rule fields used to define evidence in the associated application section include two parts: evidence type and rule description. The evidence type includes at least submission type, log event type, and test result type. The rule description is a parsable and executable rule language or script. The associated application section also includes a teaching intervention type field, which indicates the teaching strategy. The teaching strategy includes at least one of scaffolded, collaborative, and exploratory approaches. The associated application section also includes a permission and content security field, which is used for authorization control and content classification.

[0020] The structured data object can be serialized into JSON or an equivalent format, and consistency verification is performed through a predefined schema.

[0021] The unique identifier field in the basic index uses a time-sortable globally unique identifier format;

[0022] The basic index also includes a version number field, which follows the semantic versioning specification.

[0023] On the other hand, this disclosure provides a data processing method executed by a computer system, comprising the following steps:

[0024] Data acquisition steps: Acquire the digital educational resources to be processed, wherein the digital educational resources include at least one of video, text, and interactive courseware;

[0025] Vectorization representation generation steps: Input the digital educational resources into a pre-trained multimodal deep learning model, and generate an N-dimensional core competency vector through feature extraction and multi-label classification; each dimension of the N-dimensional core competency vector corresponds to a predefined competency evaluation index, and the value of each dimension represents the weight of the resource on the corresponding competency.

[0026] Metadata structuring encapsulation steps: Construct a three-part metadata structure object, encapsulate the N-dimensional core competency vector in the second layer of the metadata structure, encapsulate the basic attributes of the resource in the first layer, and encapsulate the associated application rules of the resource in the third layer to generate standardized resource metadata;

[0027] Data storage and indexing steps: Store the resource metadata in the database and establish a high-dimensional vector index for the N-dimensional core competency vector;

[0028] Intelligent application steps: In response to user requests, perform personalized recommendations or user profile updates based on the resource metadata.

[0029] The vector space defined by the N-dimensional core competency vector contains N orthogonal basis vectors, each corresponding to a preset dimension identifier, and the numerical range of each dimension output by the model is [0,1].

[0030] The three-part metadata structure specifically includes: a basic index part, a core competency vector part, and an associated application part;

[0031] The basic index includes: a globally unique resource identifier (resource_id), version number, resource type, subject classification, applicable grade level, and knowledge point tags; the core competency vector includes: the N-dimensional core competency vector (competency_profile), used to represent the educational goal weights of the resources; the related application includes: a list of precursor resources, a list of successor resources, expected assessment evidence rules (expected_evidence), and intervention strategy types.

[0032] In the vectorized representation generation step, the multimodal deep learning model adopts a dual-tower structure or a multi-stream structure, and the processing includes:

[0033] Multimodal feature extraction is performed on the input educational resources, including visual features, audio features, and text semantic features;

[0034] The extracted multimodal features are fused and mapped to the N-dimensional vector space through a fully connected layer;

[0035] The Sigmoid activation function is used to output the probability value of each dimension; combined with a pre-set expert rule base, the probability values ​​output by the model are corrected to obtain the final N-dimensional core competency vector.

[0036] The personalized recommendation operation in the intelligent application steps specifically includes:

[0037] Obtain the current literacy profile vector of the target user and the learning target strategy;

[0038] Based on the learning objective strategy, the current literacy profile vector is converted into a recommendation demand vector;

[0039] In the database, the vector similarity or inner product between the N-dimensional core competency vector of the candidate resource and the recommendation requirement vector is calculated;

[0040] The resources with the highest similarity or inner product are selected as the recommendation results.

[0041] Specifically, the user profile update operation in the intelligent application steps includes:

[0042] Receive event data of a user completing a learning session on a certain educational resource, the event data including a learning outcome score; read the N-dimensional core competency vector of the educational resource;

[0043] Based on the Bayesian update algorithm with forgetting effect, the forgetting weight is calculated by combining the time difference between the current timestamp and the previous update timestamp.

[0044] The increment of new evidence is calculated based on the learning outcome score and the literacy vector of the resource;

[0045] The user's literacy profile vector is updated using the forgetting weight and the new evidence increment.

[0046] Furthermore, this disclosure provides a personalized recommendation method based on the aforementioned metadata structure, comprising the following steps:

[0047] Obtain the N-dimensional literacy profile vector of the student;

[0048] Based on the predetermined recommendation strategy, the demand vector is calculated using the N-dimensional literacy profile vector;

[0049] In a vector database storing the core competency vectors of claim 1, a Top-K search based on the maximum inner product is performed to obtain candidate resources;

[0050] Based on resource dependency satisfaction and business constraints, the candidate resources are reordered and filtered.

[0051] Furthermore, this disclosure provides a method for dynamically updating student competency profiles based on the aforementioned metadata structure, comprising the following steps:

[0052] Receive learning events that include resource identifiers and learning outcomes;

[0053] Based on the N-dimensional literacy vector of the educational resources corresponding to the resource identifier, literacy evidence for each dimension is calculated;

[0054] Update the student's N-dimensional literacy profile based on the literacy evidence.

[0055] In another aspect, this disclosure provides a computer system or computer-readable storage medium having computer instructions stored thereon, which, when executed on a processor, are used to perform the method described in any one of the above descriptions.

[0056] By implementing the technical solution of this invention, the structured and computable nature of literacy information is achieved: This invention creates a novel data structure, transforming for the first time the abstract concept of "comprehensive literacy" in pedagogy into an N-dimensional numerical vector that can be directly read, stored, and computed by computers. This fundamentally changes the way computers perceive educational resources, enabling them to leap from understanding "knowledge points" to understanding "educational value," providing a data foundation for higher-level intelligent education applications. Simultaneously, it achieves automated and large-scale production of literacy data: This invention provides a human-computer collaborative computer generation method that can efficiently and consistently transform massive amounts of unstructured raw educational resources into standardized core literacy vectors. This solves the bottleneck problem of literacy data sources, making large-scale implementation of literacy-based intelligent education applications feasible. Furthermore, it constructs a literacy-based intelligent algorithm closed loop: This invention proposes a complete application method that transforms problems such as personalized recommendations and student profiling into vector operation problems in a high-dimensional literacy space. This not only improves the accuracy and interpretability of the application, but also builds a technical closed loop in the computer system of "content literacy vector -> recommendation -> learning feedback -> student literacy vector update", making "literacy" a fluid, computable and optimizable core data entity in the system, which greatly enhances the intelligence level and personalized service capabilities of the intelligent education system. Attached Figure Description

[0057] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0058] Figure 1 This is a structural diagram of a metadata structure according to this disclosure;

[0059] Figure 2 This is a schematic diagram of a data processing method disclosed herein;

[0060] Figure 3 This is a schematic diagram of the logical structure and data type structure of a metadata object in one embodiment of this disclosure;

[0061] Figure 4 This is a data processing flowchart of a core competency vector implemented in this public disclosure;

[0062] Figure 5 This is an architecture diagram of a personalized recommendation system based on vector retrieval, as disclosed in this publication.

[0063] Figure 6 This is a data flow diagram for the dynamic updating of student competence profiles, which is publicly implemented. Detailed Implementation

[0064] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0065] The technical solutions of this disclosure are illustrated below through specific embodiments. It should be understood that one or more steps mentioned in this disclosure do not preclude the existence of other methods and steps before or after the combined steps, or that other methods and steps may be inserted between these explicitly mentioned steps. It should also be understood that these examples are for illustrative purposes only and are not intended to limit the scope of this disclosure. Unless otherwise stated, the numbering of each method step is only for the purpose of identifying each method step, and not for limiting the order of each method or limiting the scope of implementation of this disclosure. Changes or adjustments to their relative relationships, without substantial changes to the technical content, can also be considered as within the scope of implementation of this disclosure.

[0066] The raw materials and instruments used in the examples are not subject to any specific restrictions on their source; they can be purchased from the market or prepared according to conventional methods known to those skilled in the art.

[0067] To address the shortcomings of traditional technologies, this disclosure proposes a metadata structure such as Figure 1As shown, it is a structured data object stored on a computer-readable medium and read and written by a computer program. The metadata structure adopts a three-part architecture, including: a basic index part L1, a core competency vector part L2, and an associated application part L3.

[0068] The basic index section is used to uniquely identify educational resources and support retrieval, and includes at least a unique identifier field and attribute fields for resource classification or retrieval; the core competency vector section is used to quantitatively represent the educational goals of educational resources for intelligent computing, and includes a competency vector with N dimensions, where the keys of the dimensions are c01 to cN, and the values ​​of each dimension are real numbers within a preset numerical range.

[0069] The associated application portion is used to construct relationships between resources and learning evidence conventions, including at least fields for describing resource dependencies and rule fields for defining rules that can be executed by a computer to extract evidence from learning interaction data.

[0070] The basic index section includes at least one attribute field for resource classification or retrieval: resource type, subject, grade range, knowledge point, and difficulty. The resource type is at least one of a predefined enumeration, which includes video, interactive module, assessment, reading material, and simulation / simulation categories. The subject adopts national or international standard subject classification codes. The grade range is represented by a closed interval consisting of two integers [min, max]. The knowledge point field is bidirectionally associated with nodes in the external knowledge graph through a unique identifier or URI. The difficulty field can take values ​​that are real numbers within a preset numerical range for each dimension, such as real numbers in the range [0, 1], and is calibrated using item response theory.

[0071] The core competency vector component can be indexed in a vector database to support vector retrieval based on inner product or cosine similarity.

[0072] The fields used to describe resource dependencies in the associated application section include a set of preceding resources and a set of succeeding resources, which are used to construct a directed acyclic graph of the learning path. The rule fields used to define evidence in the associated application section include two parts: evidence type and rule description. The evidence type includes at least submission type, log event type, and test result type. The rule description is a parsable and executable rule language or script. The associated application section also includes a teaching intervention type field, which indicates the teaching strategy. The teaching strategy includes at least one of scaffolded, collaborative, and exploratory approaches. The associated application section also includes a permission and content security field, which is used for authorization control and content classification.

[0073] The structured data objects can be serialized into JSON or an equivalent format, and consistency checks can be performed using a predefined schema.

[0074] The unique identifier field in the basic index uses a time-sortable globally unique identifier format;

[0075] The basic index also includes a version number field, which follows the semantic versioning specification.

[0076] To address the shortcomings of traditional technologies, this disclosure proposes a data processing method, such as... Figure 2 As shown, this method is executed by a computer system and includes the following steps:

[0077] 201: Data acquisition step: Acquire the digital educational resources to be processed, wherein the digital educational resources include at least one of video, text, and interactive courseware;

[0078] 202: Vectorization Representation Generation Step: The digital educational resources are input into a pre-trained multimodal deep learning model. Through feature extraction and multi-label classification, an N-dimensional core competency vector is generated. Each dimension of the N-dimensional core competency vector corresponds to a predefined competency evaluation index, and the value of each dimension represents the weight of the resource on the corresponding competency. The vector space defined by the N-dimensional core competency vector contains N orthogonal basis vectors, each corresponding to a preset dimension label. The numerical range of each dimension output by the model is [0,1]. The multimodal deep learning model in this step adopts a dual-tower structure or a multi-stream structure. The processing includes: extracting multimodal features from the input educational resources, including visual features, audio features, and text semantic features; fusing the extracted multimodal features and mapping them to the N-dimensional vector space through a fully connected layer; using the Sigmoid activation function to output the probability value of each dimension; and correcting the probability value output by the model using a preset expert rule base to obtain the final N-dimensional core competency vector.

[0079] 203: Metadata Structured Encapsulation Steps: Construct a three-part metadata structure object. The N-dimensional core competency vector is encapsulated in the second part of the metadata structure, the basic attributes of the resource are encapsulated in the first part, and the associated application rules of the resource are encapsulated in the third part, generating standardized resource metadata. Specifically, the three-part metadata structure includes: a basic index part, a core competency vector part, and an associated application part. The basic index part includes: a globally unique resource identifier (resource_id), version number, resource type, subject classification, applicable grade level, and knowledge point tags. The core competency vector part includes: the N-dimensional core competency vector (competency_profile), used to represent the educational goal weight of the resource. The associated application part includes: a list of precursor resources, a list of successor resources, expected evaluation evidence rules (expected_evidence), and intervention strategy types.

[0080] 204: Data storage and indexing steps: Store the resource metadata in the database, and establish a high-dimensional vector index for the N-dimensional core competency vector;

[0081] 205: Intelligent Application Steps: In response to user requests, perform personalized recommendations or user profile updates based on the resource metadata. This step specifically includes: obtaining the target user's current competency profile vector and learning objective strategy; converting the current competency profile vector into a recommendation demand vector according to the learning objective strategy; calculating the vector similarity or inner product between the N-dimensional core competency vector of candidate resources and the recommendation demand vector in the database; and selecting the resources with the highest similarity or inner product as the recommendation results.

[0082] The user profile update operation in the above-described intelligent application steps specifically includes:

[0083] The system receives event data showing that a user has completed learning from a certain educational resource, including a learning outcome score; it reads the N-dimensional core competency vector of the educational resource; it calculates a forgetting weight based on a Bayesian update algorithm with a forgetting effect, combined with the time difference between the current timestamp and the last update timestamp; it calculates a new evidence increment based on the learning outcome score and the competency vector of the resource; and it updates the user's competency profile vector using the forgetting weight and the new evidence increment.

[0084] Based on the above methods, this disclosure also proposes a personalized recommendation method based on the above-described metadata structure. The method includes: obtaining an N-dimensional literacy profile vector of a student; calculating a demand vector based on the N-dimensional literacy profile vector according to a predetermined recommendation strategy; and performing a Top-K search based on the maximum inner product in a vector database storing the core literacy vector of claim 1 to obtain candidate resources.

[0085] Based on resource dependency satisfaction and business constraints, the candidate resources are reordered and filtered. Based on the above method, this disclosure also proposes a method for dynamically updating student competency profiles based on the aforementioned metadata structure, specifically including the following steps: receiving learning events containing resource identifiers and learning outcomes; calculating competency evidence for each dimension based on the N-dimensional competency vector of the educational resource corresponding to the resource identifier; and updating the student's N-dimensional competency profile based on the competency evidence. It should be noted that updating the student's N-dimensional competency profile based on the competency evidence can be done dimension-by-dimensionally according to a Bayesian or forgetting factor weighting rule.

[0086] Based on the above methods, this disclosure also proposes a computer system or computer-readable storage medium having computer instructions stored thereon, which, when executed on a processor, are used to perform the methods described above.

[0087] To achieve the aforementioned objectives, this technical solution proposes a complete closed-loop technical process that can be implemented by a computer system. The core of this solution is a three-part, N-dimensional literacy-oriented metadata structure, and its data modeling, generation, and application methods in a computer system are elaborated in detail. Hereinafter, N=16, and a literacy-oriented metadata structure of 16 is used as an example to illustrate the specific implementation of this invention.

[0088] We'll begin with data modeling: The technical implementation of the three-part metadata structure is as follows:

[0089] The core of this technical solution's data modeling is the design and definition of a structured data object optimized for efficient processing and deep semantic understanding by computer systems. This data object aims to encapsulate all the necessary descriptive information of a digital educational resource into a logically clear, hierarchical, serializable, and easily extensible entity. To achieve this, this solution innovatively adopts a three-part architecture for data modeling, the specific definitions of which are shown in Table 1 below.

[0090] Table 1: Detailed technical specifications and implementation constraints for each field in the metadata structure

[0091] Level Field name (JSONKey) Data type (JSON Type) Technical Constraints and Implementation Details Example value L1 resource_id String It must be a globally unique identifier; the ULID format is recommended to ensure time-ordered nature. A unique index must be created in the database. "01H8X2J4P5N6M7R8S9T0V1W2X3 L1 version String It must comply with the SemVer 2.0.0 specification; it is used for content cache invalidation and version control in the system. "2.1.0" L1 resource_type String The values ​​must be from the predefined enumeration set: VIDEO, INTERACTIVE_MODULE, ASSESSMENT, READING, SIMULATION "INTERACTIVE_MODULE" L1 subject String It is recommended to use national or international standard subject classification codes to achieve cross-system interoperability. "G633.6" (Mathematics) L1 grade_level Array (ofIntegers) It must contain two integers [min, max], representing a closed interval; used for range queries. [8, 10] L1 knowledge points Array (ofStrings) The array elements are unique URIs or codes for knowledge points in the knowledge graph; a database inverted index must be created to speed up queries. ["KP:math-func-002", "KP:math-geo-005"] L1 difficulty Number (Float) The value must be a floating-point number in the range [0.0, 1.0]; it is recommended to calibrate using an item reaction theory (IRT) model to ensure objectivity. 0.72 L2 competency_profile Object (Map) Key innovations: Contains 16 key-value pairs, where the key must be a string from "c01" to "c16", and the value is a floating-point number in the range [0.0, 1.0]; an index needs to be created in a dedicated vector database. {"c01": 0.2, "c02": 0.5, ..., "c08": 0.9, ..., "c16": 0.3} L3 prerequisites Array (ofStrings) The array elements are the resource_ids of other resources; used to construct a directed acyclic graph of learning dependencies. ["01H8X2J4P5N6M7R8S9T0V1W2X0"] L3 postrequisites Array (ofStrings) The array elements are the resource_ids of other resources; used to define learning successor relationships. [] L3 expected_evidence Array (ofObjects) Used for automatic system evaluation; each object contains a type (enumeration: SUBMISSION, LOG_EVENT, QUIZ_RESULT) and a description (evaluation rule string / script). [{"type": "LOG_EVENT", "desc": "user.dragged_node >3"}] L3 intervention_type String The enumerated values ​​for instructional strategies guiding the AI ​​engine are: SCAFFOLDING (providing scaffolding), COLLABORATIVE (facilitating collaboration), and EXPLORATORY (encouraging exploration). "EXPLORATORY" L3 license_safety Object (Map) Includes license (e.g., "CC-BY-NC") and safety_level (e.g., "G") fields; used for access control and content filtering. {"license": "proprietary", "safety_level": "G"}

[0092] The metadata structure modeling adopts a three-part architecture, mainly focusing on technical layering and decoupling while maintaining functional cohesion. Each layer carries a specific type of computing task, allowing upper-layer applications to call upon tasks on demand, and allowing the underlying storage to be optimized independently.

[0093] The basic index section is designed for speed. It contains the most frequently used and least changed identifying information, optimized for high-speed retrieval, filtering, and classification operations in the database.

[0094] The core competency vector component: The design goal is intelligence. This is the core of the invention; it transforms abstract educational goals into a standardized numerical vector that directly serves as input for machine learning models and is the "fuel" for all AI algorithms.

[0095] Related Applications: The design goal is an ecosystem. It defines the location and operational attributes of resources within the entire educational content ecosystem, providing the necessary data interfaces for constructing complex learning paths, automating teaching strategies, and ensuring system security and compliance.

[0096] The specific implementation process of the structured data object is as follows:

[0097] In computer systems, the metadata structure is typically implemented as a JSON object due to its lightweight nature, readability, and widespread support by all modern programming languages ​​and databases. The following are the detailed technical specifications of this data object.

[0098] like Figure 3 The diagram shown is a structural illustration of a metadata object; this class diagram clearly illustrates the three main logical layers that make up a metadata object:

[0099] The L1 Foundation Index Layer is responsible for storing basic information and content characteristics of resources. This layer includes core fields such as resource ID, version, resource type enumeration, subject, grade level, knowledge points array, and difficulty value.

[0100] The L2 core competency vector layer (CoreCompetencyVectorLayer) depicts the competency profile cultivated by resources in vector form. Its core is a mapping structure `competency_profile`, where the key is a competency item (String) and the value is the weight or mastery level of that competency (Float).

[0101] The L3 Relational Operational Layer defines the relationships, application methods, and constraints of resources within the teaching process. This layer includes: learning path relationships (prerequisites and postrequisites), assessment requirements (expected evidence array), teaching intervention types (intervention type enumeration), and authorization and safety information (license safety object).

[0102] Relationship: MetadataObject aggregates these three layers through composition relationships, indicating that a complete metadata object must include these three parts, which are its inseparable components.

[0103] To ensure data consistency and validity, this solution provides the following JSON Schema definition. This Schema can be used for automatic validation of metadata during the data ingestion phase.

[0104] {

[0105] "$schema": "http: / / json-schema.org / draft-07 / schema#",

[0106] "title": "16D Competency-Oriented Metadata Schema",

[0107] "description": "Defines the three-layer metadata structure for smart education resources.",

[0108] "type": "object",

[0109] "properties": {

[0110] "resource_id": { "type": "string", "pattern": "^[0-9A-HJKMNP-TV-Z]{26}$"},

[0111] "version": { "type": "string", "pattern": "^(0|[1-9]\\d*)\\.(0|[1-9]\\d*)\\.(0|[1-9]\\d*)(?:-((?:0|[1-9]\\d*|\\d*[a-zA-Z-][0-9a-zA-Z-]*)(?:\\.(?:0|[1-9]\\d*|\\d*[a-zA-Z-][0-9a-zA-Z-]*))*))?(?:\\+([0-9a-zA-Z-]+(?:\\.[0-9a-zA-Z-]+)*))?$"},

[0112] "resource_type": { "enum": ["VIDEO", "INTERACTIVE_MODULE", "ASSESSMENT", "READING", "SIMULATION"]},

[0113] "subject": { "type": "string"},

[0114] "grade_level": { "type": "array", "items": [{ "type": "integer"}, { "type": "integer"}], "minItems": 2, "maxItems": 2},

[0115] "knowledge_points": { "type": "array", "items": { "type": "string"}},

[0116] "difficulty": { "type": "number", "minimum": 0.0, "maximum": 1.0},

[0117] "competency_profile": {

[0118] "type": "object",

[0119] "properties": {

[0120] "c01": { "type": "number", "minimum": 0.0, "maximum": 1.0},

[0121] "c02": { "type": "number", "minimum": 0.0, "maximum": 1.0},

[0122] "c03": { "type": "number", "minimum": 0.0, "maximum": 1.0},

[0123] "c04": { "type": "number", "minimum": 0.0, "maximum": 1.0},

[0124] "c05": { "type": "number", "minimum": 0.0, "maximum": 1.0},

[0125] "c06": { "type": "number", "minimum": 0.0, "maximum": 1.0},

[0126] "c07": { "type": "number", "minimum": 0.0, "maximum": 1.0},

[0127] "c08": { "type": "number", "minimum": 0.0, "maximum": 1.0},

[0128] "c09": { "type": "number", "minimum": 0.0, "maximum": 1.0},

[0129] "c10": { "type": "number", "minimum": 0.0, "maximum": 1.0},

[0130] "c11": { "type": "number", "minimum": 0.0, "maximum": 1.0},

[0131] "c12": { "type": "number", "minimum": 0.0, "maximum": 1.0},

[0132] "c13": { "type": "number", "minimum": 0.0, "maximum": 1.0},

[0133] "c14": { "type": "number", "minimum": 0.0, "maximum": 1.0},

[0134] "c15": { "type": "number", "minimum": 0.0, "maximum": 1.0},

[0135] "c16": { "type": "number", "minimum": 0.0, "maximum": 1.0}

[0136] },

[0137] "required": ["c01", "c02", "c03", "c04", "c05", "c06", "c07", "c08","c09", "c10", "c11", "c12", "c13", "c14", "c15", "c16"]

[0138] },

[0139] "prerequisites": { "type": "array", "items": { "type": "string"}},

[0140] "postrequisites": { "type": "array", "items": { "type": "string"}},

[0141] "expected_evidence": {

[0142] "type": "array",

[0143] "items": {

[0144] "type": "object",

[0145] "properties": {

[0146] "type": { "enum": ["SUBMISSION", "LOG_EVENT", "QUIZ_RESULT"]},

[0147] "description": { "type": "string"}

[0148] },

[0149] "required": ["type", "description"]

[0150] }

[0151] },

[0152] "intervention_type": { "enum": ["SCAFFOLDING", "COLLABORATIVE", "EXPLORATORY"]},

[0153] "license_safety": {

[0154] "type": "object",

[0155] "properties": {

[0156] "license": { "type": "string"},

[0157] "safety_level": { "type": "string"}

[0158] }

[0159] }

[0160] },

[0161] "required": ["resource_id", "version", "resource_type", "competency_profile"]

[0162] }

[0163] The data modeling scheme disclosed in this invention ensures high performance, high availability, and high intelligence of metadata in computer systems through its three-part architecture, precise field definitions, and strict technical constraints. It is not merely a static description file, but a dynamic and operable data entity tightly coupled with database indexes, AI algorithms, and system workflows, forming the data model foundation for the entire technical solution.

[0164] The aforementioned computer data processing method for core competency vectors proposes and implements a method that automatically processes digital educational resource content and maps it to a predefined 16-dimensional competency space. The goal is to generate a numerical "competency DNA"—a core competency vector—for each educational resource. The definition of the 16-dimensional literacy space is as follows:

[0165] The first step of this method is to establish a standardized, multi-dimensional target vector space. This scheme defines a literacy space consisting of 16 orthogonal basis vectors. Each basis vector represents a core competency that is crucial to students' overall development.

[0166] Table 2: Definition and Explanation of Basis Vectors in the 16-Dimensional Core Competency Space

[0167] Dimension ID basis vector name Technical Explanation and Computer System Application Objectives Typical content features (used for model training) c01 Large Unit Introduction This dimension represents the ability of resources to construct knowledge structures and develop systematic thinking. The system utilizes this dimension to guide students in macro-level learning planning. Keywords: framework, review, overview, mind map, structure, system c02 Education on Mission and Responsibility This dimension represents the ability of resources to inspire social responsibility. The system utilizes this dimension for value guidance and team collaboration task recommendation. Keywords: dedication, responsibility, nation, hero, team, cooperation c03 Emotional Education This dimension represents the ability of resources to cultivate emotional experience, aesthetic appreciation, and empathy. The system utilizes this dimension to personalize the delivery of art and literature resources. Keywords: feelings, experiences, beauty, emotions, art, poetry c04 Intellectual Education This dimension represents the ability of resources to train logical reasoning, information discernment, and rational analysis. The system uses this dimension to recommend tasks such as debate and data analysis. Keywords: because...therefore..., analysis, evidence, logic, reasoning, data c05 Technical Education This dimension represents the ability of resources to improve tool use, engineering thinking, and hands-on skills. The system uses this dimension to recommend programming, experimental, and design tasks. Keywords: operation, implementation, construction, code, tool, design, steps c06 Science Education This dimension represents the ability of resources to cultivate scientific inquiry, hypothesis testing, and empirical thinking. The system utilizes this dimension to design inquiry-based learning pathways. Keywords: hypothesis, experiment, observation, variable, conclusion, verification c07 Culture and Education This dimension represents the ability of resources to enhance cultural understanding and intercultural communication. The system utilizes this dimension to broaden students' international perspectives and humanistic literacy. Keywords: tradition, customs, history, civilization, diversity, exchange c08 Philosophy Education This dimension represents the ability of resources to stimulate critical thinking and value judgment. The system utilizes this dimension to guide students in in-depth thinking and value exploration. Keywords: essence, meaning, why, value, right and wrong, paradox c09 Zhi Xing Education This dimension represents the ability of resources to facilitate the connection between theory and practice and to solve real-world problems. The system utilizes this dimension to connect classroom knowledge with social practice. Keywords: application, practice, case study, solution, reality, scenario c10 Project-based education This dimension represents the capabilities of resources in training objective planning, process management, and team collaboration. The system utilizes this dimension to support project-based learning. Keywords: project, plan, division of labor, milestone, delivery, summary c11 STEAM Integrated Education This dimension represents the ability to cultivate interdisciplinary integration and innovative design. The system utilizes this dimension to recommend complex tasks that require the integration of multidisciplinary knowledge. Interdisciplinary terms appear simultaneously, such as "programming + art" and "physics + design". c12 Innovative Education This dimension represents the ability of resources to stimulate divergent thinking and originality. The system utilizes this dimension to encourage students to propose unconventional solutions. Keywords: imagination, creation, new methods, breaking conventions, invention c13 Education on discerning truth, goodness, and beauty Representing resources enhances abilities in aesthetic judgment, moral discernment, and the pursuit of truth. The system utilizes this dimension to cultivate students' advanced critical thinking skills. Keywords: Distinguish, True or False, Good or Evil, Beauty or Ugliness, Judgment, Appreciation c14 Success Education This dimension represents the ability of resources to enhance resilience, self-efficacy, and growth mindset. The system utilizes this dimension for psychological resilience training. Keywords: Distinguish, True or False, Good or Evil, Beauty or Ugliness, Judgment, Appreciation c15 Happiness Education This dimension represents the ability of resources to promote positive emotions, self-awareness, and harmonious interpersonal relationships. The system utilizes this dimension to improve students' mental health. Keywords: happiness, gratitude, self, relationships, meaning, bliss c16 Academic Education This dimension represents the role of resources in cultivating academic standards, rigorous argumentation, and literature review skills. The system utilizes this dimension to prepare students for advanced learning. Keywords: citation, argumentation, literature review, research, paper

[0168] The technical process and algorithm implementation of computer generation

[0169] The data processing method of this invention is a multi-stage computer processing flow, the core of which is a specially designed and trained deep learning model. For example... Figure 2 Detailed explanation of the technical steps in the flowchart shown;

[0170] Phase 1: Input and Preprocessing

[0171] Step A1 -> A2: The computer system receives a raw educational resource R. The multimodal content parsing and splitting module (A2) first identifies the data modalities it contains.

[0172] Step A2 -> A3 / A4 / A5: For plain text or images containing text (such as PPT screenshots), the text processor (A3) extracts all readable text using OCR (Optical Character Recognition) technology; for audio in audio or video, the speech recognition module (A4) uses ASR (Automatic Speech Recognition) technology to convert it into a text stream; for video or images, the visual content processor (A5) extracts keyframes, performs scene recognition, or generates image description text.

[0173] Phase Goal: To transform all unstructured raw content into standardized text and visual data streams that can be further processed by computers. The meanings of the A1-A5 labels in this embodiment are shown in Table 3 below:

[0174] Table 3

[0175] label corresponding Figure 4 to block / module names illustrate A1 Original Educational Resources R The starting point of the process A2 Multimodal content parsing and splitting module Located below A1, it is responsible for distributing data streams. A3 Text processor The left branch is responsible for extracting the main text / OCR. A4 ASR (Automatic Speech Recognition) module The middle branch is responsible for audio-to-text conversion. A5 Visual content processor The right branch is responsible for extracting keyframes / image descriptions.

[0176] Phase Two: Deep Learning Model Inference

[0177] Steps A3 / A4 -> B1 & A5 -> B2: The standardized text and visual data streams are fed into two parallel, pre-trained deep learning encoders. The text feature extractor (B1) captures the semantic information of the content and generates high-dimensional text feature vectors. The visual feature extractor (B2) captures visual elements and scene information of the content, generating high-dimensional visual feature vectors.

[0178] Step B1 / B2 -> B3: The two feature vectors are input into the multimodal feature fusion layer (B3). This layer typically employs an attention mechanism, which dynamically calculates the contribution weights of textual and visual features to the final content understanding and generates a more comprehensive fused content representation vector that incorporates information from both modalities. .

[0179] Steps B3 -> B4: Fusion Vectors The data is fed into a multi-head mapping network (B4). This is a key parallel processing architecture containing 16 independent fully connected neural networks (or "heads"). Each "head" is specifically trained to identify content and its mapping to a particular literacy dimension (e.g., ...). or The strength of the association between .

[0180] Steps B4 -> B5: The output of each "head" is processed by the Sigmoid activation function to obtain a floating-point number in the interval [0, 1]. These 16 floating-point numbers together constitute the initial literacy vector. .

[0181] Phase Goal: To automatically and end-to-end generate a preliminary, quantitative assessment of literacy propensity from the content itself using a pure deep learning model. The meanings of the B1-B5 designations in this embodiment are shown in Table 4 below:

[0182] Table 4

[0183] label corresponding Figure 4 to the block / module names illustrate B1 Text Feature Extractor Boxes containing BERT / RoBERTa B2 Visual feature extractor Boxes containing ViT / ResNet B3 Multimodal feature fusion The box containing the Attention Mechanism receives the outputs of B1 and B2. B4 Multi-head mapping network The box containing "16 parallel fully connected layers" B5 Initial competence vector The output box contains "16 [0,1] values".

[0184] Phase 3: Rule Correction and Confidence Calculation

[0185] Step A1 / B5 / C1 -> C2: Initialization Vector Along with the resource's basic metadata (such as subject and grade), it is fed into the rules engine (C2). The rules engine loads a predefined rule base (C1) from the teaching experts.

[0186] Steps C2 -> C3: The rule engine performs matching and transformation operations. For example, if a rule is "If the resource discipline is 'philosophy', then..." The weight of "(Philosophy Education)" is at least 0.7, but the model's predicted value is low, so the engine will automatically increase this value. This ensures that the results conform to basic teaching common sense and improves accuracy. The corrected vector is the final competency vector.

[0187] Steps C2 -> C4: Simultaneously, the system will calculate a confidence score for each dimension of the vector based on information such as the original probabilities output by the model and the strength of rule triggering, forming a confidence vector.

[0188] Phase Objective: To combine the prediction results of the AI ​​model with the domain knowledge of human experts, correct significant biases, and quantify the reliability of the results, providing decision support for the next phase of human review. The meanings of the C1-C4 designations in this embodiment are shown in Table 5 below:

[0189] Table 5

[0190] label correspond Figure 4 The box / module name illustrate C1 Teaching Expert Rule Base Located in the upper right corner, it serves as the input source for the rule engine. C2 Rule Engine The diamond-shaped decision / processing box receives inputs from B5 and C1. C3 Corrected vector The results output by the rule engine C4 Calculate confidence level The calculation module output to the right by the rule engine

[0191] Phase Four: Human-Machine Collaborative Audit

[0192] Step C3 / C4 -> D1: The final competency vector C(R) and its confidence vector are pushed to the pending queue of the human-machine collaborative review system (D1).

[0193] Step D1 <-> D2: The teaching expert (D2) sees the AI's processing results on the review interface. The system will highlight the dimensions that require the expert's focus based on the confidence score.

[0194] Steps D1 -> D3: Experts can confirm the AI ​​results with one click, and can also fine-tune some or all dimensions. The finally confirmed vector will be marked as "approved" and officially stored in the content database.

[0195] Phase Goal: To ensure the highest quality of the final data entering the database. Simultaneously, expert corrections can be used as new training data for future model iterations and optimizations, forming a closed loop of continuous improvement. The meanings of the D1-D3 labels in this embodiment are shown in Table 6 below:

[0196] Table 6

[0197] label correspond Figure 4 The box / module name illustrate D1 Teaching experts Manual intervention node D2 Human-machine collaborative auditing system Core Interactive Interface / System D3 Approved competency vectors / Added to database The final output points to the database.

[0198] Algorithm 1.1: Literacy Vector Generation Algorithm Based on Multimodal Fusion and Multi-head Mapping

[0199] This is an end-to-end computer algorithm that takes the original resources as input and outputs a structured object containing a 16-dimensional literacy vector and confidence scores.

[0200] import torch

[0201] import torch.nn as nn

[0202] from transformers import BertModel

[0203] from vit_pytorch import ViT

[0204] class CompetencyVectorGenerator(nn.Module):

[0205] def __init__(self, text_model_name='bert-base-uncased', image_size=256, patch_size=32, num_classes=16, dim=1024, depth=6, heads=16, mlp_dim=2048):

[0206] super().__init__()

[0207] # 1. Multimodal Feature Extraction Module

[0208] self.text_encoder = BertModel.from_pretrained(text_model_name)

[0209] self.visual_encoder = ViT(image_size=image_size, patch_size=patch_size, num_classes=1, dim=dim, depth=depth, heads=heads, mlp_dim=mlp_dim)

[0210] text_feature_dim = self.text_encoder.config.hidden_size

[0211] visual_feature_dim = dim

[0212] # 2. Feature Fusion Module (using simple concatenation and linear layers)

[0213] fused_dim = text_feature_dim + visual_feature_dim

[0214] self.fusion_layer = nn.Linear(fused_dim, mlp_dim)

[0215] # 3. Multi-head mapping module (16 independent classification heads)

[0216] self.competency_heads = nn.ModuleList([

[0217] nn.Sequential(

[0218] nn.Linear(mlp_dim, mlp_dim / / 2),

[0219] nn.ReLU(),

[0220] nn.Linear(mlp_dim / / 2, 1)

[0221] for _ in range(num_classes) ])

[0223] self.sigmoid = nn.Sigmoid()

[0224] def forward(self, text_input, image_input):

[0225] # Feature Extraction

[0226] # Use the [CLS] token output to represent the semantics of the entire text.

[0227] text_features = self.text_encoder(**text_input).last_hidden_state[:, 0, :]

[0228] # Visual encoder output

[0229] visual_features = self.visual_encoder(image_input) # Assuming the ViT model is modified to directly output feature vectors

[0230] # Fusion Features

[0231] fused_features = torch.cat((text_features, visual_features),dim=1)

[0232] fused_representation = torch.relu(self.fusion_layer(fused_features))

[0233] # Multiple Mapping

[0234] competency_logits = [head(fused_representation) for head inself.competency_heads]

[0235] # Stack and apply Sigmoid to get the final vector

[0236] competency_vector = torch.cat(competency_logits, dim=1)

[0237] return self.sigmoid(competency_vector)

[0238] # --- Model Training and Inference ---

[0239] # Training data: (Resource content, 16-dimensional vectors annotated by experts)

[0240] # Loss function: Binary Cross-Entropy (BCE) Loss, which calculates the loss independently for each of the 16 outputs and then sums them up.

[0241] # Reasoning Phase:

[0242] # 1. model = CompetencyVectorGenerator()

[0243] # 2. model.load_state_dict(torch.load('trained_model.pth'))

[0244] # 3. text, image = preprocess(raw_resource)

[0245] # 4. c_initial_vector = model(text, image)

[0246] # 5. c_final_vector, confidence = apply_rules(c_initial_vector, db_rules)

[0247] # 6. push_to_review_queue(c_final_vector, confidence)```

[0248] Loss function for model training:

[0249] For a training sample (resource) Expert-annotated vectors The model's output is a prediction vector. Loss function The sum of the binary cross-entropy loss commonly used in multi-label classification is employed:

[0250]

[0251] Mathematical expression of rule correction:

[0252] Rule correction can be modeled as a function :

[0253]

[0254] For example, a simple linear transformation rule:

[0255]

[0256] If resource R meets a certain condition.

[0257] Computer application methods based on this metadata

[0258] The metadata structure of this invention is not an isolated data definition; its core value lies in its operability, namely, serving as high-quality input for computer algorithms to drive the implementation of a series of intelligent educational applications. The following will elaborate on two of the most representative computer application methods: personalized recommendation based on metadata structure and dynamic updating of student competency profiles.

[0259] Application Method 1: Personalized recommendation methods based on the learning resource metadata structure, such as... Figure 5 The diagram shows the architecture of a personalized recommendation system based on vector retrieval.

[0260] This method aims to address the technical problems of existing recommendation systems that result in "one-size-fits-all" recommendations or shallow matching based solely on knowledge points. This solution transforms the recommendation problem into a similarity calculation problem in a high-dimensional vector space, achieving more accurate and interpretable personalized content distribution. The personalized recommendation method described above is based on vector retrieval, and computer-implemented personalized recommendation systems include:

[0261] 1) Offline processing stage

[0262] Starting point: 16-dimensional metadata of all digital courseware;

[0263] Core processing: Construct vector indexes using efficient vector indexing algorithms such as HNSW or Faiss;

[0264] Output: A well-established high-dimensional vector database, providing basic support for online queries;

[0265] 2) Online real-time recommendation stage

[0266] Triggered by: User login or sending a recommendation request;

[0267] User profile: Obtain the 16-dimensional literacy profile vector P(S) of the user;

[0268] Strategy calculation: Calculate the specific "demand vector" D(S) based on the recommendation strategy;

[0269] Vector retrieval: Queries the vector database using demand vectors and returns the Top-K most similar resource IDs;

[0270] Intelligent optimization: Through reordering and filtering modules, factors such as difficulty and user history are comprehensively considered;

[0271] Final output: The final recommendation list generated and presented to the user;

[0272] It should be noted that the offline / online separation ensures system performance, with offline index building and online rapid response; vectorized retrieval utilizes similarity calculations in a high-dimensional vector space to achieve accurate matching; personalized strategies dynamically adjust demand vectors based on user profiles and recommendation strategies; and multi-factor optimization intelligently re-ranks recommendations based on initial retrieval to improve recommendation quality.

[0273] Algorithm 2.1: Recommendation Algorithm Based on Maximum Inner Product Search (MIPS)

[0274] This algorithm is the core of the aforementioned system architecture and is executed by the computer in real time.

[0275] 1) Input: Student literacy profile vector ;

[0276] Pre-built vectors containing all resource literacy The vector database VDB; the recommendation strategy (e.g., "address weaknesses", "leverage strengths"); the number of recommendations. .

[0277] 2) Calculate the demand vector This is a policy transition step performed by a computer, as follows:

[0278] if Strategy == "Addressing weaknesses":

[0279] , in It is a vector consisting entirely of 1s.

[0280] else if Strategy == "Play to your strengths":

[0281]

[0282] else if Strategy == "Interest-oriented":

[0283] (This can be obtained by weighted averaging of the competency profile from historical interaction data.)

[0284] 3) Perform vector retrieval. Here, we utilize the efficient retrieval engine of the vector database to perform a Top-K maximum inner product search:

[0285]

[0286] This step aims to find the vector of demand. Largest inner product Each resource vector is used to solve for:

[0287]

[0288] 4) Reordering and Filtering: This step reorders and filters the original Top-K list. We will refine the process to improve the user experience.

[0289] Algorithm 2.2: Reordering Sub-algorithm; details are as follows:

[0290] def rerank_and_filter(raw_list, student_profile, history_list,params):

[0291] final_list = []

[0292] for resource_id, sim_score in raw_list:

[0293] metadata = get_metadata(resource_id)

[0294] # a) Filter already learned content

[0295] if resource_id in history_list:

[0296] continue

[0297] # b) Difficulty matching penalty

[0298] difficulty_diff = abs(student_profile.avg_level - metadata['difficulty'])

[0299] difficulty_penalty = params['lambda_d'] * difficulty_diff

[0300] # c) Diversity promotion (compared to similarity penalty for recent learning history)

[0301] diversity_penalty = 0

[0302] recent_vectors = get_recent_vectors(history_list, 5)

[0303] current_vector = metadata['competency_profile_vector']

[0304] if recent_vectors:

[0305] max_hist_sim = max([cosine_similarity(current_vector, v)for v in recent_vectors])

[0306] diversity_penalty = params['lambda_s'] * max_hist_sim

[0307] # d) Calculate the final score

[0308] final_score = sim_score - difficulty_penalty - diversity_penalty

[0309] if final_score > params['score_threshold']:

[0310] final_list.append((resource_id, final_score))

[0311] # Sort by final score

[0312] final_list.sort(key=lambda x: x[1], reverse=True)

[0313] return final_list

[0314] Output: Returns the final list of recommendations after reordering and filtering.

[0315] Application Method Two: A method for dynamically updating student competency profiles using metadata structures, such as... Figure 6 The diagram shows the data flow for the dynamic update of student competency profiles; this method aims to solve the technical problems of static and lagging student profile updates. This solution provides a real-time computer update model based on learning activity feedback and incorporating the forgetting effect. Algorithm 3.1, a Bayesian update algorithm with the forgetting effect, is used; this algorithm is automatically triggered by the computer upon receiving a learning event.

[0316] 1) Input: Students complete learning activities by watching videos and / or submitting quiz data; the event handler extracts key information, such as student ID, resource ID, score, and timestamp; specifically including the following:

[0317] current portrait (including levels of each dimension) and last update timestamp )

[0318] Resource metadata (Including competency vectors) )

[0319] Activity Results

[0320] Current timestamp

[0321] Hyperparameter: Learning rate Basic update weight Forgetting constants in each dimension

[0322] 2) Initialization: This involves starting the portrait update service and creating a new portrait object. The initial value is equal to .

[0323] 3) Iterative update: By obtaining the student's current profile and resource metadata, the profile update algorithm module completes the profile update, resulting in the updated profile; the specific steps are as follows:

[0324] for to do:

[0325] / / Extract variables from input

[0326]

[0327]

[0328]

[0329] / / Calculate the time difference

[0330]

[0331] / / Step a: Calculate the dynamically updated weight lambda_i (core)

[0332] / / This weighting simulates the forgetting curve of human memory, with a higher weight given to skills that have not been practiced recently.

[0333]

[0334] / / Step b: Calculate the new evidence E_i provided in this activity

[0335] / / The strength of evidence is directly proportional to the activity score and the weight of the resources themselves in contributing to that competency.

[0336]

[0337] / / Step c: Perform Bayesian update

[0338] / / The new literacy level is a weighted average of the old level and new evidence.

[0339]

[0340] / / Step d: Update the portrait object

[0341]

[0342]

[0343] end for

[0344] 4) Output: Returns the updated complete portrait object. , store / persistentize the image.

[0345] The two application methods described above fully demonstrate the technical value of the metadata structure of this invention: it is a computable, operable core technical component deeply embedded in the AI ​​algorithm process. By vectorizing educational resources and student profiles in the same 16-dimensional literacy space, this solution provides a solid technical foundation for realizing a series of advanced, precise, and personalized intelligent education functions.

[0346] It should be noted that although this embodiment uses 16 dimensions as an example, those skilled in the art can set N to 8, 20, 24, or other positive integers according to actual teaching needs, and the dimension definition is not limited to the contents described in Table 2 above. Furthermore, the metadata structure is not limited to JSON format; it can also be implemented using XML, YAML, Protocol Buffers, or database binary storage formats, etc., which will not be repeated here.

[0347] The core innovation of this invention lies in creating a data structure that transforms the abstract concept of "comprehensive literacy" in education into a standardized, multi-dimensional numerical vector (i.e., "core literacy vector") that can be directly computed by a computer, and provides a complete set of computer implementation methods for data generation and application around this structure.

[0348] This core innovation can be further broken down into three closely linked and indispensable sub-innovations: At the "data modeling" level, it achieves for the first time the structuring and computability of competencies; the technical solution of this invention creates a completely new data "container." This container (i.e., the N-dimensional metadata structure) no longer only loads surface information such as "knowledge points" as existing technologies, but for the first time can structurally "load" and "quantify" deep-seated, previously elusive comprehensive competencies such as "critical thinking," "innovation ability," and "teamwork." This is a breakthrough from 0 to 1, providing an unprecedented data paradigm for computers to "understand" the essential goals of education.

[0349] At the "data generation" level: For the first time, the automated and large-scale production of literacy data has been achieved. This invention's technical solution is a "machine" that "processes" educational content into literacy data. It efficiently and consistently transforms massive amounts of unstructured raw educational resources, such as videos and texts, into standardized core literacy vectors. This solves the fundamental problem of where literacy data comes from, making large-scale application possible. At the "data application" level: For the first time, a closed-loop intelligent education algorithm based on literacy vectors has been realized. By transforming personalized recommendations, student profiling, and other issues into vector operation problems in a high-dimensional literacy space, it not only fundamentally improves the accuracy and interpretability of these applications but, more importantly, constructs a technical closed loop: Content (literacy vector) -> Recommendation -> Student Learning -> Feedback -> Student Profiling (literacy vector) Update. In this closed loop, "literacy" is no longer an empty slogan but a core data entity that flows, is calculated, and is optimized within the system.

[0350] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the methods described above; the electronic device includes the processor, which is configured to perform the methods described above.

[0351] The example electronic devices described in this disclosure are intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic devices may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their patterns are merely examples and are not intended to limit the implementations of this disclosure described and / or claimed herein.

[0352] The electronic device includes a computing unit that can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) or a computer program loaded from a storage unit into a random access memory (RAM). The RAM may also store various programs and data required for the operation of the electronic device. The computing unit, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0353] Multiple components in an electronic device are connected to an I / O interface, including: input units such as keyboards and mice; output units such as various types of displays and speakers; storage units such as disks and optical discs; and communication units such as network interface cards (NICs), modems, and wireless transceivers. The communication unit allows the electronic device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0354] The computing unit can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of computing units include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit performs the various methods and processes described above, such as traffic control methods. For example, in some embodiments, the traffic control method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on an electronic device via ROM and / or a communication unit. When the computer program is loaded into RAM and executed by the computing unit, one or more steps of the traffic control method described above may be performed. Alternatively, in other embodiments, the computing unit may be configured to perform the traffic control method by any other suitable means (e.g., by means of firmware).

[0355] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0356] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or a traffic control device, such that when executed by the processor or controller, the program code causes the patterns / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0357] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0358] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0359] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0360] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0361] The foregoing description of specific exemplary embodiments of this disclosure is for illustrative and explanatory purposes. These descriptions are not intended to limit this disclosure to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of this disclosure and their practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of this disclosure, as well as various different choices and variations. The scope of this disclosure is intended to be defined by the claims and their equivalents.

Claims

1. A metadata structure, which is a structured data object stored on a computer-readable medium and read / written by a computer program, said metadata structure adopting a three-part architecture, including: The basic index section, the core competency vector section, and the related application section; The basic index section is used to uniquely identify educational resources and support retrieval, and includes at least a unique identifier field and attribute fields for resource classification or retrieval. The core competency vector part is used to quantitatively represent the educational goals of educational resources for intelligent computing. It includes a competency vector with N dimensions, where the keys of the dimensions are c01 to cN, and the values ​​of each dimension are real numbers within a preset range. The associated application portion is used to construct relationships between resources and learning evidence conventions, including at least fields for describing resource dependencies and rule fields for defining rules that can be executed by a computer to extract evidence from learning interaction data.

2. The metadata structure as described in claim 1, characterized in that, The attribute fields used for resource classification or retrieval in the basic index section include at least one: resource type, subject, grade range, knowledge point, and difficulty. The resource type is at least one of a predefined enumeration, which includes video, interactive module, assessment, reading material, and simulation / simulation categories. The subject adopts national or international standard subject classification codes. The grade range is represented by a closed interval consisting of two integers [min, max]. The knowledge point field is bidirectionally associated with nodes in the external knowledge graph through a unique identifier or URI. The difficulty field takes a real number within a preset range and is calibrated using item response theory. The core competency vector component can be indexed in a vector database to support vector retrieval based on inner product or cosine similarity. The fields used to describe resource dependencies in the associated application section include a set of preceding resources and a set of succeeding resources, which are used to construct a directed acyclic graph of the learning path. The rule fields used to define evidence in the associated application section include two parts: evidence type and rule description. The evidence type includes at least submission type, log event type, and test result type. The rule description is a parsable and executable rule language or script. The associated application section also includes a teaching intervention type field, which indicates the teaching strategy. The teaching strategy includes at least one of scaffolded, collaborative, and exploratory approaches. The associated application section also includes a permission and content security field, which is used for authorization control and content classification.

3. The metadata structure as described in claim 1, characterized in that, The structured data objects can be serialized into JSON or an equivalent format, and consistency checks can be performed using a predefined schema. The unique identifier field in the basic index uses a time-sortable globally unique identifier format; The basic index also includes a version number field, which follows the semantic versioning specification.

4. A data processing method, characterized in that, This method is executed by a computer system and includes the following steps: Data acquisition steps: Acquire the digital educational resources to be processed, wherein the digital educational resources include at least one of video, text, and interactive courseware; Vectorization representation generation steps: Input the digital educational resources into a pre-trained multimodal deep learning model, and generate an N-dimensional core competency vector through feature extraction and multi-label classification; each dimension of the N-dimensional core competency vector corresponds to a predefined competency evaluation index, and the value of each dimension represents the weight of the resource on the corresponding competency. Metadata structuring encapsulation steps: Construct a three-part metadata structure object, encapsulate the N-dimensional core competency vector in the second layer of the metadata structure, encapsulate the basic attributes of the resource in the first layer, and encapsulate the associated application rules of the resource in the third layer to generate standardized resource metadata; Data storage and indexing steps: Store the resource metadata in the database and establish a high-dimensional vector index for the N-dimensional core competency vector; Intelligent application steps: In response to user requests, perform personalized recommendations or user profile updates based on the resource metadata.

5. The method according to claim 4, characterized in that, The vector space defined by the N-dimensional core competency vector contains N orthogonal basis vectors, each corresponding to a preset dimension identifier, and the numerical range of each dimension output by the model is a preset range.

6. The method according to claim 4, characterized in that, The three-part metadata structure specifically includes: a basic index part, a core competency vector part, and a related application part; The basic index includes: a globally unique resource identifier (resource_id), version number, resource type, subject classification, applicable grade level, and knowledge point tags; the core competency vector includes: the N-dimensional core competency vector (competency_profile), used to represent the educational goal weights of the resources; the related application includes: a list of precursor resources, a list of successor resources, expected assessment evidence rules (expected_evidence), and intervention strategy types.

7. The method according to claim 4, characterized in that, In the vectorized representation generation step, the multimodal deep learning model adopts a dual-tower structure or a multi-stream structure, and the processing includes: Multimodal feature extraction is performed on the input educational resources, including visual features, audio features, and text semantic features; The extracted multimodal features are fused and mapped to the N-dimensional vector space through a fully connected layer; The Sigmoid activation function is used to output the probability value of each dimension; combined with a pre-set expert rule base, the probability values ​​output by the model are corrected to obtain the final N-dimensional core competency vector.

8. The method according to claim 4, characterized in that, The personalized recommendation operation in the intelligent application steps specifically includes: Obtain the current literacy profile vector of the target user and the learning target strategy; Based on the learning objective strategy, the current literacy profile vector is converted into a recommendation demand vector; In the database, the vector similarity or inner product between the N-dimensional core competency vector of the candidate resource and the recommendation requirement vector is calculated; The resources with the highest similarity or inner product are selected as the recommendation results.

9. The method according to claim 4, characterized in that, The user profile update operation in the intelligent application steps specifically includes: Receive event data of a user completing a learning session on a certain educational resource, wherein the event data includes a learning outcome score; Read the N-dimensional core competency vector of this educational resource; Based on the Bayesian update algorithm with forgetting effect, the forgetting weight is calculated by combining the time difference between the current timestamp and the previous update timestamp. The increment of new evidence is calculated based on the learning outcome score and the literacy vector of the resource; The user's literacy profile vector is updated using the forgetting weight and the new evidence increment.

10. A personalized recommendation method based on the metadata structure of claim 1, characterized in that, Includes the following steps: Obtain the N-dimensional literacy profile vector of the student; Based on the predetermined recommendation strategy, the demand vector is calculated using the N-dimensional literacy profile vector; In a vector database storing the core competency vectors of claim 1, a Top-K search based on the maximum inner product is performed to obtain candidate resources; Based on resource dependency satisfaction and business constraints, the candidate resources are reordered and filtered.

11. A method for dynamically updating student competency profiles based on the metadata structure described in claim 1, characterized in that, Includes the following steps: Receive learning events that include resource identifiers and learning outcomes; Based on the N-dimensional literacy vector of the educational resources corresponding to the resource identifier, literacy evidence for each dimension is calculated; Update the student's N-dimensional literacy profile based on the literacy evidence.

12. A computer system or computer-readable storage medium having computer instructions stored thereon, characterized in that, When the computer instructions are executed on a processor, they are used to perform the method of any one of claims 1 to 11.

Images