Multi-language supported transnational dual-teacher lesson preparation intelligent auxiliary method and system

By constructing a multilingual, cross-border dual-teacher lesson preparation intelligent auxiliary system, the problems of heterogeneous resource formats, scattered storage, delayed updates, and insufficient semantic understanding in cross-border dual-teacher lesson preparation have been solved. It has realized intelligent matching and real-time collaboration of cross-language teaching resources, and improved the efficiency and quality of cross-border lesson preparation.

CN121767140APending Publication Date: 2026-03-31FUJIAN NETLONG PUTIAN EDUCATION TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202511599267.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-11-04
Publication Date
2026-03-31

AI Technical Summary

Technical Problem

The cross-border dual-teacher lesson preparation system suffers from heterogeneous teaching resource formats and scattered storage, lack of a unified semantic association framework, lagging resource update mechanism, inefficient unstructured data storage, system response delay, and insufficient semantic understanding of multilingual resources, resulting in low efficiency and quality of cross-border dual-teacher lesson preparation.

Method used

A multilingual, cross-border dual-teacher intelligent lesson preparation assistance system is constructed, including a unified access and preprocessing module for teaching resources, a multimodal and multilingual knowledge graph construction module, a cross-language semantic understanding and resource matching module, a real-time collaborative lesson preparation workspace module, and a system performance optimization and resource update module, to achieve intelligent matching and real-time collaboration of multilingual teaching resources.

Benefits of technology

The unified access and preprocessing module solves the problems of heterogeneous resource formats and scattered storage; the multilingual knowledge graph enables unified measurement of cross-language semantics; the cross-language semantic understanding module enables intelligent resource recommendation; the real-time collaborative editing module improves the efficiency of cross-border lesson preparation; the system performance optimization module ensures responsiveness; and the knowledge base is continuously updated to bridge the language and cultural gap between Chinese and foreign teachers.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121767140A_ABST
    Figure CN121767140A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of distributed storage, and particularly discloses a multi-language-supported intelligent auxiliary method and system for cross-country dual-teacher lesson preparation. The system comprises a teaching resource unified access and preprocessing module, a multi-mode multi-language knowledge graph construction module, a cross-language semantic understanding and resource matching module, a real-time collaborative lesson preparation work space module and a system performance optimization and resource updating module. Intelligent matching and recommendation of teaching resources are achieved through the multi-language knowledge graph and cross-language semantic coding, real-time collaborative editing and machine translation are combined, the problems of resource isomerism, semantic barrier and low collaboration efficiency in cross-country double-teacher lesson preparation are effectively solved, and lesson preparation efficiency and cross-culture teaching adaptability are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of distributed storage technology, specifically relating to a multilingual, cross-border dual-teacher intelligent lesson preparation assistance method and system. Background Technology

[0002] In the field of educational technology, especially in cross-border collaborative teaching scenarios, how to leverage information technology to improve the efficiency of teaching resource management and the intelligence level of collaborative lesson preparation has become a research hotspot. The integration of multilingual teaching resources and intelligent lesson preparation assistance are key aspects of this.

[0003] Among them, cross-border dual-teacher lesson preparation, as an emerging teaching model, aims to improve teaching quality by integrating teaching resources from different cultural backgrounds through collaboration between Chinese and foreign teachers. The core of this model lies in achieving efficient matching of cross-language and cross-cultural teaching resources and optimizing the lesson preparation process.

[0004] Existing technologies typically employ centralized resource repositories for teaching resource management, but these methods face limitations: teaching resources are heterogeneous in format and scattered in storage, lacking a unified semantic association framework, making it difficult for teachers to quickly locate suitable resources that match specific themes, difficulties, and cultural backgrounds; resource update mechanisms are lagging behind, unable to absorb new content contributed by teachers worldwide in real time, and unstructured data storage is inefficient, causing system response delays. Furthermore, traditional systems lack sufficient semantic understanding of multilingual resources, making it difficult to effectively bridge the language and cultural gap between Chinese and foreign teachers during lesson preparation. These issues severely restrict the efficiency and quality of cross-border dual-teacher lesson preparation, necessitating a solution that can intelligently match multilingual resources and optimize the lesson preparation process. Summary of the Invention

[0005] The technical problem this invention aims to solve is to overcome the shortcomings of existing cross-border dual-teacher lesson preparation systems, such as heterogeneous and scattered storage of teaching resources, lack of a unified semantic association framework, lagging resource update mechanisms, low efficiency of unstructured data storage, system response delays, and insufficient semantic understanding capabilities of multilingual resources. This invention provides a multilingual, cross-border dual-teacher lesson preparation intelligent assistance method and system. The invention aims to construct a lesson preparation assistance environment capable of intelligent matching, efficient storage, and real-time collaboration of multilingual teaching resources, fundamentally bridging the language and cultural gap between Chinese and foreign teachers during the lesson preparation process.

[0006] This invention provides a multilingual, cross-border, dual-teacher intelligent lesson preparation assistance system. The system includes a unified access and preprocessing module for teaching resources, a multimodal, multilingual knowledge graph construction module, a cross-language semantic understanding and resource matching module, a real-time collaborative lesson preparation workspace module, and a system performance optimization and resource update module.

[0007] The unified access and preprocessing module for teaching resources is responsible for collecting heterogeneous teaching resource data from distributed sources. This module has built-in standardized data interfaces, enabling it to access raw resources in various formats, including text, presentations, videos, audio, and interactive courseware.

[0008] Furthermore, this module includes a data cleaning and standardization submodule, which performs deduplication operations, removes completely duplicate resource entries, and parses semi-structured and unstructured data to extract text content, metadata information, and feature descriptions of multimedia files.

[0009] As one embodiment of the present invention, the module also integrates a format conversion engine, which can uniformly convert presentations from different sources into a preset standard slide format, and extract key frame sequences of video resources at specified time intervals to form standardized image sequence files, laying the foundation for subsequent in-depth processing.

[0010] The multimodal, multilingual knowledge graph construction module receives standardized data from the preprocessing module and is responsible for building and maintaining a dynamically evolving knowledge graph. At its core is a multilingual entity and relation extraction engine. This engine employs a deep neural network-based natural language processing model to automatically identify and extract teaching-related entities from text resources, including but not limited to subject concepts, teaching topics, knowledge points, skills, and cultural background elements. Simultaneously, the engine also extracts relevant information from video audio tracks and subtitles, and image descriptions.

[0011] Furthermore, this module includes a relation definition and linking submodule. This submodule defines semantic relationships between entities based on a preset ontology schema. These relationships include belonging, premise, relevance, instance, and cultural association. The knowledge graph is stored using a hybrid architecture of graph database and vector database. The graph database stores the topological structure of entities and relationships, while the vector database generates a high-dimensional semantic vector representation for each entity. This vector representation can transcend language boundaries and capture its deep semantic connotations.

[0012] The cross-lingual semantic understanding and resource matching module is the core of the system's intelligent recommendation mechanism. This module includes a cross-lingual semantic encoder, which is built based on a pre-trained multilingual Transformer model. This encoder can map query requests and target resources in different languages ​​to the same high-dimensional semantic vector space. In this space, semantically similar content, regardless of its surface language form, has vector representations that are close in distance.

[0013] Furthermore, this module includes a multi-dimensional matching strategy engine. This engine receives user input regarding lesson preparation requirements, including the target subject, teaching theme, student age group, language, and cultural compatibility. The matching strategy engine first vectorizes the requirements using a cross-lingual semantic encoder, then performs an approximate nearest neighbor search in the knowledge graph's vector database to initially filter out a set of semantically relevant candidate resources. Subsequently, the engine applies a series of configurable filtering rules and weight models to refine the ranking of the candidate set.

[0014] The filtering rules include resource format compatibility checks and difficulty level matching assessments. The weighted model then comprehensively calculates weighted scores for candidate resources and needs across multiple dimensions, including semantic relevance, pedagogical applicability, cultural fit, and user historical preferences, ultimately outputting a resource recommendation list sorted in descending order of overall matching degree.

[0015] The real-time collaborative lesson preparation workspace module provides Chinese and international teachers with an online collaborative editing environment that supports real-time multilingual interaction. This module achieves real-time collaborative editing based on operation conversion technology or conflict-free copy data type theory, ensuring eventual data consistency when multiple teachers edit the same lesson preparation document. The workspace also incorporates intelligent recommendation services provided by the aforementioned cross-language semantic understanding and resource matching module. While teachers are preparing lessons, the system can proactively recommend relevant multilingual teaching resources in real-time based on the context of the current editing content.

[0016] Furthermore, this module integrates a real-time machine translation service. This service is not a simple word-to-word translation, but rather combines a specialized terminology database for lesson preparation with contextual information to provide more accurate translations of teaching terms. These translations are presented as annotations or sidebar prompts to help teachers understand resource content in different language versions or colleagues' comments. In addition, the module offers version control, automatically saving different versions of the lesson preparation process and supporting comparison and retrospection of differences between versions.

[0017] The system performance optimization and resource update module is dedicated to ensuring the efficient operation of the system and the continuous evolution of knowledge. This module includes a distributed caching and index management submodule, which caches hot resources, frequently queried semantic vectors, and parts of the knowledge graph, reducing data retrieval latency. Simultaneously, it maintains an efficient index structure, accelerating graph traversal and vector retrieval operations.

[0018] Furthermore, this module includes an incremental knowledge update engine. This engine continuously monitors newly added teaching resources and new content generated by teachers during collaborative lesson preparation. For new resources, the engine initiates a lightweight entity and relation extraction process and incrementally merges them into the existing knowledge graph, updating the corresponding semantic vectors. For content generated by teachers, the system analyzes its association with existing knowledge and optimizes the weight parameters of the recommendation model accordingly. As one embodiment of the invention, this module also implements a resource popularity and quality assessment mechanism, dynamically adjusting the priority of resources in the recommendation list based on the number of times resources are cited, teacher ratings, and collaborative usage effects, and identifying high-quality core resources.

[0019] In a preferred embodiment of the present invention, the training process of the cross-lingual semantic encoder employs a contrastive learning strategy. Specifically, the training objective of the encoder is to minimize the vector distance between sentence pairs from different languages ​​that describe the same teaching concept, while maximizing the vector distance between sentence pairs describing different teaching concepts, regardless of whether they are in the same language. This training process uses a large-scale multilingual teaching corpus to ensure that the encoder can effectively capture cross-lingual semantic consistency.

[0020] In a preferred embodiment of the present invention, the real-time machine translation service in the real-time collaborative lesson preparation workspace module integrates a domain-adaptive fine-tuning mechanism. This mechanism utilizes the bilingual lesson preparation materials and multilingual teaching resources accumulated by the system as parallel corpora, and periodically fine-tunes the basic machine translation model to make its output translation results more consistent with the language habits and professional terminology of the education field.

[0021] In a preferred embodiment of the present invention, the incremental knowledge update engine in the system performance optimization and resource update module employs a confidence-based conflict resolution strategy when merging new knowledge into the knowledge graph. When a newly extracted entity or relation has a potential conflict with information in the existing graph, the system comprehensively considers the authority of the information source, its timeliness, and the consistency of multiple independent sources to decide whether to update the existing knowledge, retain the original knowledge and mark the dispute, or introduce a new entity or relation branch.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention systematically solves the problems of heterogeneous teaching resource formats and scattered storage by constructing a unified access and preprocessing module. It transforms unstructured and semi-structured data into a standardized, deeply processable data format, laying a solid foundation for subsequent intelligent processing and improving the automation and standardization of data processing.

[0023] 2. This invention, by designing a multimodal, multilingual knowledge graph construction module, introduces for the first time a hybrid storage architecture that integrates graph structure and semantic vectors in the field of teaching resource management. This not only clearly expresses the complex semantic relationships between teaching concepts, but also achieves cross-language semantic uniformity measurement through vectorization, providing core support for accurate semantic matching.

[0024] 3. By integrating cross-language semantic understanding and resource matching modules, this invention utilizes a deep learning-based cross-language semantic encoder and multi-dimensional matching strategies to achieve truly language-independent intelligent resource recommendation. It can accurately match teaching topics, difficulty levels, and cultural backgrounds, effectively overcoming the bottleneck of traditional systems in multilingual semantic understanding.

[0025] 4. By deploying a real-time collaborative lesson preparation workspace module, this invention deeply integrates intelligent recommendation with real-time collaborative editing and context-aware machine translation, creating a seamless collaborative lesson preparation environment for Chinese and foreign teachers. This greatly improves the efficiency and experience of cross-border lesson preparation and directly bridges the language and cultural gap.

[0026] 5. By introducing a system performance optimization and resource update module, this invention ensures the system's response performance when facing massive amounts of data and high concurrency access. Furthermore, through incremental updates and dynamic evaluation mechanisms, the system's knowledge base can continuously evolve and self-optimize, maintaining the system's long-term vitality and usability. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the overall technical solution architecture of the multilingual support transnational dual-teacher lesson preparation intelligent auxiliary system proposed in this invention; Figure 2 This is a schematic diagram of the core principle framework of the cross-language semantic understanding and resource matching module in this invention; Figure 3 This is a logical flow diagram of the multimodal and multilingual knowledge graph construction module in this invention; Figure 4 This is a schematic diagram of the multi-level interaction relationship and data flow of the real-time collaborative lesson preparation workspace module in this invention; Figure 5 This is a flowchart of the main stages of the unified access and preprocessing module for teaching resources in this invention. Figure 6 This is a schematic diagram of the core principle framework of the system performance optimization and resource update module in this invention; Detailed Implementation

[0028] This embodiment details the technical implementation of a multilingual, cross-border, dual-teacher intelligent lesson preparation support system. Please refer to the appendix. Figure 1The system consists of five core modules: a unified access and preprocessing module for teaching resources, a multimodal and multilingual knowledge graph construction module, a cross-language semantic understanding and resource matching module, a real-time collaborative lesson preparation workspace module, and a system performance optimization and resource update module. These modules communicate through a pre-defined data bus and message queue, forming a closed-loop intelligent lesson preparation support environment.

[0029] The unified access and preprocessing module for teaching resources is responsible for collecting heterogeneous teaching resources from multiple distributed source systems. Please refer to the appendix. Figure 5 This module contains three main processing stages.

[0030] The first phase is the resource access and protocol parsing phase. This module deploys various standardized data interface adapters, including an application programming interface adapter based on the Hypertext Transfer Protocol (HTTP) for accessing online educational resource libraries, a server connector based on the File Transfer Protocol (FTP) for acquiring shared resources within the institution, and an email parser based on the Simple Mail Transfer Protocol (SMLP) for receiving personal resources submitted by teachers.

[0031] These adapters can handle raw data in various formats, including text files, presentation files, video files, audio files, and interactive courseware packages. Upon receiving data, each adapter performs preliminary protocol parsing to extract metadata information of the resource, such as resource identifier, source system, original format, file size, and upload timestamp, and encapsulates this information into a unified resource description framework.

[0032] The second stage is the data cleaning and content standardization stage. This module internally runs a data cleaning and standardization submodule. This submodule first initiates the deduplication operation, the core of which is calculating the digital fingerprint of each resource file. The digital fingerprint is calculated using a cryptographic hash function, performing a hash operation on the file content to generate a 256-bit hash value.

[0033] The system maintains a global hash index of connected resources. When the hash value of a new resource exactly matches any existing value in the index, it is determined to be a completely duplicate resource. This resource will be marked and skipped from the subsequent processing.

[0034] For resources that are not completely duplicated, the submodule starts the content parsing engine.

[0035] For text-based resources, the parsing engine uses an open-source document parsing library to extract the plain text content and its structural information, such as chapter titles, paragraphs, and list items. For presentation files, the parsing engine decompresses the file package and reads the text content, image object descriptions, and animation sequence descriptions for each slide.

[0036] For video files, the parsing engine calls the video decoding library to extract keyframe images at a fixed interval of 2fps, and at the same time separates the audio track and any embedded subtitle track from the video container.

[0037] For audio files, the parsing engine performs speech recognition conversion, turning the speech content into text transcript.

[0038] All parsed text content will undergo a unified character encoding conversion to Unicode format, and basic natural language preprocessing will be performed, including sentence segmentation, word segmentation, and removal of meaningless stop words.

[0039] The third stage is the format conversion and feature encapsulation stage.

[0040] The integrated format conversion engine in this module performs standardized conversions for specific resource types.

[0041] For presentations, the engine parses all slide layouts, font styles, and image position information, and remaps them onto a preset standard slide template, generating a standardized presentation file with a uniform style in the Open Document Presentation Format (ODS) format. For keyframe sequences extracted from video resources, the engine adjusts each frame to a uniform 1280 pixel × 720 pixel resolution, converts it to the Joint Panel of Image Experts (JPA) format, and finally packages it into an ordered image sequence file.

[0042] After completing all processing, the module outputs a standardized resource object, which contains reference links to the original resource files, cleaned plain text content, extracted structured metadata, converted standard format files, and generated multimedia feature description files. The module then publishes this object to the system's message bus for downstream modules to consume.

[0043] The multimodal, multilingual knowledge graph construction module continuously listens to the message bus, receiving standardized resource objects from the preprocessing module. Please refer to the appendix. Figure 3 The core task of this module is to build and maintain a dynamically evolving multimodal and multilingual knowledge graph.

[0044] Its operation begins with a multilingual entity and relation extraction engine. This engine employs a pre-trained multilingual deep neural network model based on the Transformer architecture, specifically a variant of a large language model trained on multilingual corpora.

[0045] The model receives preprocessed text content as input and performs named entity recognition and relation extraction tasks. Entity recognition covers a variety of teaching-related types, including subject concept entities, such as the Pythagorean theorem in mathematics and Newton's second law in physics; teaching theme entities, such as environmental protection and ancient civilizations; knowledge point entities, such as solving quadratic equations and the process of photosynthesis; skill-related entities, such as data analysis ability and critical thinking; and cultural background element entities, such as the Chinese traditional festival Spring Festival and Western historical events like the Renaissance.

[0046] Relation extraction is based on a predefined instructional ontology schema, which is stored in the form of a web ontology language file.

[0047] The core semantic relationships defined include: belonging relationship, which indicates that a knowledge point belongs to a certain discipline or topic; premise relationship, which indicates that mastering one knowledge point is a prerequisite for learning another knowledge point; relevance relationship, which indicates that two knowledge points are related in teaching; instance relationship, which indicates that a specific case is an instance of a certain abstract concept; and cultural association relationship, which connects teaching concepts with their related cultural background elements.

[0048] In addition to text resources, the engine also processes multimedia resources. For video, the engine analyzes the transcribed text of its audio track and embedded subtitle text, and uses image recognition models to analyze keyframe sequences to identify visual entities such as teaching tools, experimental devices, and geographical landscapes. For audio, the engine analyzes its transcribed text.

[0049] Each extracted entity and relation is accompanied by a confidence score, output by a deep neural network model, indicating the degree of confidence in the prediction. The extraction results are then fed into the relation definition and linking submodule.

[0050] This submodule first performs entity linking, which involves aligning the newly extracted entities with existing entities in the knowledge graph.

[0051] The alignment process is based on the semantic vector representation of entities and string similarity. For new entities that cannot be linked to existing entities, the submodule creates new entity nodes in the knowledge graph.

[0052] Subsequently, the submodule creates or updates semantic relationship edges between entities based on the ontology schema.

[0053] The physical storage of the knowledge graph employs a hybrid architecture.

[0054] The graph database portion uses a native graph database to store the topological structure of entity nodes and relation edges.

[0055] Each node has a unique identifier, node type, and set of attributes.

[0056] Each edge has a relation type, a starting node identifier, a target node identifier, and a set of attributes.

[0057] The vector database section uses a database specifically optimized for high-dimensional vector retrieval to store the semantic vector representation of each entity node.

[0058] The semantic vector is a 1768-dimensional floating-point vector generated by the encoder layer in the aforementioned multilingual deep neural network model.

[0059] The vector generation process ensures that semantically similar teaching concepts, regardless of whether their surface language is Chinese, English, or other languages, have a small Euclidean distance or cosine similarity in the vector space.

[0060] Graph databases and vector databases are linked by unique identifiers for entity nodes to ensure data consistency.

[0061] The knowledge graph construction module also provides a graph query interface that supports complex semantic queries based on graph query language and approximate semantic retrieval based on vectors.

[0062] The cross-language semantic understanding and resource matching module is the core of the system's intelligent resource recommendation mechanism. Please refer to the appendix. Figure 2 This module contains two core components. The first core component is a cross-language semantic encoder.

[0063] The encoder itself is a pre-trained multilingual model based on the Transformer architecture, and its parameters have been trained on a large-scale multilingual text corpus before system deployment. In the specific application of this system, the training process of the encoder is further optimized using a contrastive learning strategy.

[0064] The training objective is to learn a semantic vector space in which the vector representations of sentence pairs from different languages ​​that describe the same teaching concept are as close as possible, while the vector representations of sentence pairs that describe different teaching concepts are as far apart as possible.

[0065] Let the pairs of positive sample sentences from different languages ​​but with the same semantics be denoted as . Its encoded vector is Negative sample sentence pairs with different semantics are Its encoded vector is The training process aims to minimize the following contrastive loss function: Where sim represents the cosine similarity function, is a temperature hyperparameter, usually set to 0.1, and N is the total number of samples in a training batch.

[0066] By minimizing this loss, the encoder learns to map semantically identical cross-lingual content to similar locations in the vector space.

[0067] The second core component is the multi-dimensional matching strategy engine. This engine is responsible for handling users' lesson preparation needs.

[0068] User requirements are typically entered through graphical user interface forms, and include multiple dimensions such as target subject, teaching theme, student age group, expected resource language, and cultural compatibility requirements.

[0069] The engine first converts text-based fields, such as teaching topic descriptions, into a high-dimensional semantic vector called the demand vector through a cross-language semantic encoder.

[0070] Meanwhile, numerical and categorical fields, such as student age group and cultural suitability level, are encoded into numerical feature vectors. Subsequently, the matching strategy engine initiates a two-stage retrieval and ranking process.

[0071] The first stage is coarse screening, where the engine performs an approximate nearest neighbor search in the vector database of the knowledge graph.

[0072] The search algorithm employs a hierarchical navigable small world graph algorithm, using the demand vector as the query point to quickly find the K nearest entity vectors in the vector space, where K is typically set to 100.

[0073] The resources corresponding to these entities are retrieved through graph database association, forming a preliminary set of candidate resources.

[0074] The second stage is fine ranking, where the engine applies a configurable, weighted, multi-dimensional scoring model to calculate the ranking for each resource in the candidate set.

[0075] The scoring model comprehensively considers the following dimensions: semantic relevance, which calculates the cosine similarity between the semantic vector of the resource itself and the demand vector; this dimension typically has a weight of 0.4. Instructional applicability, which assesses the matching degree between the resource's preset difficulty tags and the target student age group, as well as the suitability of the resource type and lesson preparation activities; this dimension typically has a weight of 0.3. Cultural fit, which evaluates whether the resource content contains elements that are consistent with or contradict the cultural background requirements specified in the demand; this dimension typically has a weight of 0.2. User historical preference, which constructs a user profile based on the user's past clicks, favorites, and ratings of similar resources, and calculates the matching degree between the resource and the user profile; this dimension typically has a weight of 0.1.

[0076] The scores for each dimension are normalized and then weighted and summed to obtain the overall matching score for the resource.

[0077] Finally, the engine generates a resource recommendation list based on the overall matching score from high to low, and returns it to the caller through an application programming interface, such as a real-time collaborative lesson preparation workspace.

[0078] The real-time collaborative lesson preparation workspace module provides a shared, collaborative online lesson preparation environment for Chinese and foreign teachers located in different regions and using different languages. Please refer to the attached document. Figure 4 This module is built on a web browser and cloud server architecture.

[0079] Its core is a real-time collaborative editing engine, which is based on the theory of operation transformation.

[0080] When a teacher edits a shared lesson preparation document on a client side—such as inserting, deleting, or formatting text—the client immediately generates an operation object. This object contains the operation type, operation location, operation content, operator identifier, and a logical timestamp. This operation object is then sent in real-time to the collaborative editing server via a network socket connection. Upon receiving the operation, the server first broadcasts it to all other clients editing the same document.

[0081] Before applying remote operations, each client will first use an operation conversion algorithm to convert the received remote operations based on its local operation history, so as to ensure that the document state after applying the operations remains consistent across all clients.

[0082] For example, when user A inserts character A at position five and user B inserts character B at position five, the operation conversion algorithm will ensure that all clients ultimately see documents with both A and B at position five, rather than one overwriting the other.

[0083] The workspace module deeply integrates services from cross-language semantic understanding and resource matching modules.

[0084] The integration method involves deploying a context listener on the client side.

[0085] This listener continuously monitors the paragraph where the user's current editing cursor is located, the selected text content, and the overall structure of the document.

[0086] When the listener detects that the user has not made any input operation within a certain period of time, such as 2 seconds, it will send the current editing context information, including the semantic vector of the surrounding text, to the matching strategy engine.

[0087] Based on this context, the engine calculates in real time and returns a set of relevant multilingual teaching resource recommendations.

[0088] These recommendations are dynamically displayed in the sidebar or pop-up window of the work interface in the form of cards. Teachers can preview them directly or drag and drop the resources into their lesson preparation documents.

[0089] In addition, the workspace module integrates a real-time machine translation service. This service is not a simple word-to-word translation, but rather combines a specialized terminology database for lesson preparation with an understanding of the context.

[0090] The terminology database was built by analyzing a large amount of bilingual lesson preparation materials and multilingual teaching resources accumulated within the system, and includes standard translations of subject-specific terms. The translation service is triggered when a teacher hovers their mouse over text in another language, or receives comments from colleagues who speak different languages.

[0091] It first performs word segmentation and syntactic analysis on the source text, then combines the context of the current lesson preparation topic to perform matching and replacement in a terminology database, finally generating a translation result that better conforms to the language habits of the education field. The translation result is presented in the form of a semi-transparent floating layer or a sidebar tooltip.

[0092] The workspace module also provides comprehensive version management functionality. The server automatically records every collaborative operation sequence of a document and periodically creates a document version snapshot, such as every 5 minutes or each time a user manually saves. Each version comes with metadata such as version number, creation timestamp, and major contributors.

[0093] Teachers can view a list of different versions through the version history interface and select any two versions for comparison. The differences will be highlighted, making it easy for teachers to trace the evolution of their lesson preparation ideas.

[0094] The system performance optimization and resource update module is crucial for ensuring the system's efficient, stable operation and continuous evolution. Please refer to the appendix. Figure 6 This module contains three core sub-modules. The first is the distributed caching and index management sub-module.

[0095] This submodule builds a multi-level caching system on system memory and high-speed solid-state drives.

[0096] The first-level cache stores the most frequently used resource object entities and semantic vectors, which are managed using the least recently used algorithm.

[0097] The second-level cache stores frequently queried knowledge graph subgraphs, such as the core knowledge points and their relationships under a popular subject.

[0098] The caching subsystem monitors all data access requests and dynamically adjusts the cached content, retaining frequently accessed data in the cache.

[0099] Meanwhile, this submodule is responsible for maintaining and optimizing various indexes.

[0100] For graph databases, it ensures that the attribute indexes on nodes and relationships are up-to-date.

[0101] For vector databases, it periodically, for example, at 2 a.m. every day, rebuilds or optimizes the hierarchical navigable small world graph index to maintain efficient near nearest neighbor search performance.

[0102] The second core submodule is the incremental knowledge update engine. This engine continuously monitors two main sources of knowledge.

[0103] The first is the standardized resource objects that newly enter the system through the unified access and preprocessing module for teaching resources. For these new resources, the engine initiates a lightweight entity and relation extraction process.

[0104] This process uses the same model as the main knowledge graph building module, but may employ a higher confidence threshold to speed up processing.

[0105] The extracted new knowledge elements, including entities and relationships, are fed into a confidence-based conflict resolution strategy processor.

[0106] When newly extracted information is inconsistent with existing information in the knowledge graph, for example, a new resource indicates that concept A and concept B have a premise relationship, while the graph records a correlation relationship, the processor will comprehensively evaluate the authority of the information source, the timeliness of the new resource, and the degree of consistency of the relationship statements from multiple independent sources.

[0107] Based on the evaluation results, the processor decides whether to update the existing relationships in the graph, retain the original knowledge and create a new relationship edge with disputed annotations for the new information, or, in extreme cases, introduce a new entity node branch.

[0108] Secondly, there is new content generated by teachers in the real-time collaborative lesson preparation workspace module, including newly created documents, added annotations, and inserted resource links.

[0109] The engine analyzes the relationship between this content and the existing knowledge graph, such as identifying new knowledge points or connections between new concepts mentioned in the document.

[0110] This information is used to optimize the weight parameters of the recommendation model. For example, if multiple teachers frequently associate a certain type of resource with a specific topic, the system will appropriately increase the weight of that type of resource under that topic.

[0111] The third core sub-module is the resource popularity and quality assessment mechanism. This mechanism dynamically calculates a comprehensive score for each resource based on multiple indicators.

[0112] The metrics include the number of times a resource is cited, i.e., how often it is cited by other lesson preparation documents; explicit teacher ratings, i.e., ratings from one to five stars; and implicit collaborative usage effectiveness data, such as the average dwell time of the resource in the lesson preparation session and the modification rate after it is inserted into the document. These metrics are normalized and weighted to obtain a quality score between 0 and 1.

[0113] The system performance optimization and resource update module will periodically, such as weekly, sort resources according to this quality score, give higher priority to high-quality core resources in the recommendation list, and add a high-quality label to them in the user interface.

[0114] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0115] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A cross-country double-teacher lesson preparation intelligent assistance system with multi-language support, characterized in that, Comprise: Teaching resource unified access and preprocessing module, for collecting heterogeneous teaching resource data from scattered source end, the module has multiple standardized data interfaces, can access text, presentation, video, audio and interactive courseware in multiple formats of original resources, the module contains a data cleaning and standardization submodule, the submodule performs de-duplication operation, eliminates completely repeated resource entries, and analyzes semi-structured and unstructured data, extracts text content, metadata information and multimedia file feature description; Multimodal multilingual knowledge graph construction module, for receiving standardized data from preprocessing module, and responsible for constructing and maintaining a dynamically evolving knowledge graph, the core of the module is a multilingual entity and relation extraction engine, which uses deep neural network based natural language processing model to automatically identify and extract teaching related entities from text resources, including discipline concepts, teaching topics, knowledge points, skill items and cultural background elements; Cross-language semantic understanding and resource matching module, for realizing intelligent resource recommendation, the module contains a cross-language semantic encoder, which is based on pre-trained multilingual Transformer model and can map different language query requests and target resources to the same high-dimensional semantic vector space; Real-time collaborative lesson planning workspace module, which provides an online, multi-language real-time interactive collaborative editing environment for Chinese and foreign teachers, which realizes real-time collaborative editing based on operation conversion technology or conflict-independent copy data type theory, and the workspace is embedded with intelligent recommendation services provided by the cross-language semantic understanding and resource matching module; System performance optimization and resource update module, for ensuring efficient operation of the system and continuous evolution of knowledge, the module contains a distributed cache and index management submodule, which caches hot resources, high-frequency query semantic vectors and part of the knowledge graph subgraph.

2. The cross-country dual teacher lesson planning intelligent assistant system with multi-language support according to claim 1, characterized in that, The teaching resource unified access and preprocessing module also integrates a format conversion engine, which can convert different source presentations into a preset standard slide format, and extract key frame sequences of video resources at specified time intervals to form standardized image sequence files.

3. The cross-country dual teacher lesson planning intelligent assistant system of claim 1, wherein, The multilingual entity and relation extraction engine also extracts relevant information from the audio track and subtitles of videos and the description text of images; The multimodal multilingual knowledge graph construction module has a relationship definition and linking submodule, which defines semantic relationships between entities according to a preset ontology schema, including belongs, premise, related, instance, and cultural association; The storage of knowledge graph adopts a hybrid architecture of graph database and vector database, where the graph database is used to store the topological structure of entities and relationships, and the vector database generates high-dimensional semantic vector representation for each entity.

4. The cross-country double teacher lesson planning intelligent assistant system of multi-language support according to claim 1, characterized in that, The cross-language semantic understanding and resource matching module has a multi-dimensional matching strategy engine, which receives user's lesson planning requirement input, including target subject, teaching topic, student age, language type, and cultural adaptability requirement; The matching strategy engine first vectorizes the demand using a cross-language semantic encoder, then performs an approximate nearest neighbor search in the vector database of the knowledge graph, preliminarily filters out a set of semantically related candidate resources, and then applies a series of configurable filtering rules and weight models to the candidate set for fine sorting. The filtering rules include resource format compatibility check and difficulty level matching degree evaluation, and the weight model comprehensively calculates the weighted score of the candidate resources and the demand in the dimensions of semantic relevance, teaching applicability, cultural fit, and user historical preferences. Finally, a resource recommendation list ranked in descending order of comprehensive matching degree is output.

5. The cross-country dual teacher lesson planning intelligent assistant system of claim 1, wherein, The real-time collaborative lesson planning workspace module integrates a real-time machine translation service that provides teaching language translation in combination with a professional term library and contextual context in the field of lesson planning. The module also provides version management functions, automatically saves different versions of the lesson planning process, and supports difference comparison and rollback between versions.

6. The cross-country double teacher lesson planning intelligent assistant system of multi-language support according to claim 1, characterized in that, The system performance optimization and resource update module is provided with an incremental knowledge update engine that continuously monitors newly accessed teaching resources and new content generated by teachers during collaborative lesson planning. The module also implements a resource popularity and quality evaluation mechanism based on the number of citations, teacher ratings, and collaborative use effects to dynamically adjust the priority of resources in the recommendation list.

7. The cross-country double teacher lesson planning intelligent assistant system of multi-language support according to claim 1, characterized in that, The training process of the cross-language semantic encoder uses a contrastive learning strategy. The training goal of the encoder is to minimize the vector distance of sentence pairs from different languages but describing the same teaching concept, while maximizing the vector distance of sentence pairs describing different teaching concepts. The training process uses a large-scale multilingual teaching corpus.

8. The cross-country double teacher lesson planning intelligent assistant system of multi-language support according to claim 1, characterized in that, When merging new knowledge into the knowledge graph, the incremental knowledge update engine uses a confidence-based conflict resolution strategy. When there is a potential conflict between the newly extracted entities or relationships and the existing information in the graph, the system will consider the authority, time freshness, and consistency of multiple independent sources.

9. The cross-country double teacher lesson planning intelligent assistant system of multi-language support according to claim 1, characterized in that, The real-time machine translation service integrates a domain-adaptive fine-tuning mechanism that uses the system's accumulated bilingual lesson planning materials and multilingual teaching resources as parallel corpora to periodically fine-tune the basic machine translation model.

10. A method for supporting a cross-country double-teacher lesson preparation intelligent assistance in multiple languages, characterized by, The multilingual-supported cross-border dual-teacher lesson planning intelligent assistant system of any one of claims 1-9 is used to implement cross-border dual-teacher lesson planning intelligent assistance.