A cross-cultural traditional Chinese medicine knowledge transmission method and system

Abstract

The application relates to the field of traditional Chinese medicine information technology and discloses a cross-cultural traditional Chinese medicine knowledge dissemination method and system, which aims to solve the problems of semantic distortion and cultural discount in cross-cultural dissemination of traditional Chinese medicine knowledge. The method and system are characterized in that a knowledge translation and multi-modal presentation framework driven by an audience cognitive model is constructed, the framework comprises a cultural cognitive model database, a cultural cognitive model generation module, a semantic deep mapping and narrative translation engine and a multi-modal presentation synthesis module. The engine generates explanatory narratives by deconstructing traditional Chinese medicine concepts, searching and generating explanatory narratives through cross-cognitive domain analogy. The module synthesizes and presents multi-modal content according to the cultural cognitive model. By adopting the technical scheme, the application can effectively overcome cultural barriers and realize accurate, efficient and barrier-free dissemination of traditional Chinese medicine knowledge worldwide.

Description

Technical Field

[0001] This invention belongs to the field of traditional Chinese medicine information technology, and specifically relates to a cross-cultural method and system for disseminating traditional Chinese medicine knowledge. Background Technology

[0002] With the deepening of global cultural exchange and the diversification of health concepts, Traditional Chinese Medicine (TCM), as a treasure of Chinese civilization, is receiving increasing attention and recognition from the international community for its unique theoretical system and practical value. Against this backdrop, how to effectively and accurately disseminate the profound, complex, and philosophically rich knowledge of TCM to audiences from different cultural backgrounds has become a key issue in promoting the internationalization of TCM, and has created an urgent need for related information technology support. To address this need, those skilled in the art have conducted many beneficial explorations, mainly focusing on two aspects: the digital organization and intelligent application of TCM knowledge.

[0003] In terms of knowledge digitization and structuring, existing technological solutions aim to address the fundamental problems of the vast amount of TCM classics, their fragmented content, and the difficulty of retrieval. For example, Chinese patent CN109215798B discloses a method for constructing a knowledge base for classical TCM texts. This method uses natural language processing technology to automatically extract key concepts (such as prescriptions, medicinal materials, syndromes, and acupoints) from massive amounts of classical TCM texts as knowledge entities, and identifies the complex semantic relationships between them, thereby constructing a vast, structured TCM knowledge graph. The contribution of this technological solution lies in its successful transformation of unstructured classical documents into a machine-readable and computable knowledge base, providing a solid technical foundation for the systematic research, data mining, and intelligent retrieval of TCM knowledge. It effectively solves the problem of organizing and managing TCM knowledge at its source, allowing the wisdom originally hidden in ancient documents to be presented systematically and in a standardized manner. In terms of intelligent applications, another type of technological solution focuses on applying the organized knowledge to specific scenarios. For example, Chinese Patent CN116682526B discloses a Traditional Chinese Medicine (TCM) knowledge recommendation system. This system can collect users' individual symptom information and intelligently match it with historical medical records and prescription data in a background knowledge base, thereby providing users with personalized health conditioning suggestions or knowledge content. The value of this solution lies in its transformation of a static knowledge base into a dynamic service, achieving precise matching between TCM knowledge and individualized needs, improving the relevance and practicality of knowledge delivery, and demonstrating good application results when serving user groups with specific cultural backgrounds.

[0004] However, as the application scenarios of TCM knowledge dissemination evolve from localization and specialization to globalization and universalization, the inherent limitations of the aforementioned technical solutions at the design principle level are becoming increasingly prominent. These technical solutions, whether in the construction of knowledge bases or the implementation of recommendation systems, are all built on a paradigm of "source-centric content." Specifically, their core objective is to ensure a faithful expression and internal logical consistency of the source knowledge (i.e., TCM classics and theoretical systems), with the entire information processing flow organized around "what the knowledge itself is." This paradigm is efficient and reasonable within a single cultural context, but its inherent contradictions become glaringly apparent when facing the new challenge of cross-cultural communication. The reason for this is that truly effective cross-cultural communication is not essentially a one-way "output" or "indoctrination" of knowledge, but a dynamic process of meaning "construction" and "adaptation" centered on "audience cognition." Due to its "source-centric" nature, the existing technical architecture naturally lacks a mechanism for deep modeling and dynamic adaptation to the audience's cultural background, cognitive models, and language habits. For example, core TCM concepts such as "qi," "yin and yang," and "internal heat" are not isolated terms, but rather complex collections of ideas rooted in traditional Chinese philosophy and life experience. Within the current technological framework, cross-cultural processing is often simplified to superficial multilingual translation. This approach fails to convey the cultural connotations and metaphorical systems behind the concepts, resulting in translations that may be literal but empty, incomprehensible, or even misleading in the target cultural context. This "semantic distortion" and "cultural discount" caused by the limitations of technological paradigms is the fundamental bottleneck restricting the international dissemination of TCM knowledge. Furthermore, current technologies also handle knowledge expression in a relatively singular way, mainly relying on text and static images. This also reflects the "source-centric" approach, prioritizing the complete recording of content rather than effective audience reception. This method ignores the differences in information reception preferences among different cultural groups and fails to fully utilize multimodal means such as multimedia and interactive experiences to build bridges for cultural communication, thus weakening the appeal and acceptance of the dissemination.

[0005] Therefore, while existing technologies have solved the problems of internal organization and initial application of TCM knowledge at a specific historical stage, their inherent "source-content-centric" design paradigm makes them unable to effectively address the deep-seated information transmission barriers caused by cultural, linguistic, and cognitive differences when dealing with cross-cultural communication. How to break through the limitations of existing technological paradigms and construct a new knowledge dissemination system that can shift the focus of dissemination from "faithful source content" to "adapting to audience cognition," achieving a deep understanding of the target audience's cultural background, and on this basis, dynamically, multi-dimensionally, and multi-modally intelligently adapting knowledge content, expression forms, and transmission strategies, has become a key technical challenge that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing technologies. Specifically, it aims to address the semantic distortion and cultural discounting issues caused by the "source-centric" approach to the cross-cultural dissemination of Traditional Chinese Medicine (TCM) knowledge. The core of existing solutions lies in ensuring faithful expression and internal logical consistency of source knowledge. However, when facing audiences with significant cultural, linguistic, and cognitive model differences, this paradigm cannot achieve deep modeling and dynamic adaptation of the audience's cognitive framework, thus severely restricting the effective international dissemination of TCM knowledge.

[0007] To achieve the aforementioned objectives, this invention provides a method and system for cross-cultural dissemination of Traditional Chinese Medicine (TCM) knowledge. Its core technical concept lies in constructing an adaptive knowledge translation and multimodal presentation framework driven by an audience cognitive model. This framework abandons the traditional direct language translation approach, instead creating and utilizing a structured "Cultural Cognitive Model (CCM)" to deeply quantify and model the target audience's cultural background, health concepts, metaphor systems, and information reception preferences. Based on this model, this invention further proposes a deep semantic mapping and narrative translation generation mechanism. This mechanism deconstructs the source TCM concepts into their basic semantic primitives and identifies highly analogous clusters in terms of function and relationship within the target culture's cognitive space. Then, based on these analogical relationships, it generates explanatory narratives that conform to the target culture's thinking habits and expression paradigms. Finally, according to the audience's information reception preferences revealed by the CCM model, the generated narrative content is dynamically synthesized into an optimal multimodal combination for presentation. This achieves a fundamental shift from "faithfully presenting the source content" to "adapting to the audience's cognition," resulting in high-fidelity and highly acceptable cross-cultural knowledge dissemination.

[0008] To achieve the above objectives, the present invention provides a cross-cultural TCM knowledge dissemination system, the system comprising:

[0009] A TCM source knowledge graph module is provided. Internally, this module stores a structured TCM knowledge graph, which consists of knowledge entities, relationships between entities, and entity attributes. Knowledge entities are pre-defined as including syndrome entities, prescription entities, Chinese herbal medicine entities, acupoint entities, meridian entities, and theoretical concept entities. Relationships between entities are pre-defined as semantic connections such as "composition," "meridian tropism," "indications," and "location." Entity attributes are pre-defined as descriptive data such as medicinal properties, flavors, and efficacy. The entire knowledge graph is stored and managed using a Resource Description Framework (RDF) triple format, providing the system with standardized, machine-readable source TCM knowledge.

[0010] A cultural cognitive model database is provided to store multiple pre-constructed cultural cognitive models (CCMs) for different target cultural groups. Each CCM corresponds to a specific cultural identifier and has a unified, standardized data structure. The data structure includes: a cultural identifier field; a set of ontological anchor nodes, where each anchor node represents a core concept related to health, life, and body in the target culture; a set of weighted metaphorical links, where each link represents a commonly used metaphorical relationship between two ontological anchor nodes in the target culture, and is accompanied by a quantified weight value, which characterizes the frequency and importance of the metaphor in that culture; a pointer to a representative corpus of that culture; and an information modality preference vector, which quantifies the degree of preference of the cultural group for different information presentation modalities such as text, static images, animation, and interactive 3D models.

[0011] A cultural cognitive model generation module, connected to a cultural cognitive model database, is configured to automatically construct or update a corresponding cultural cognitive model for a specific target cultural group. The cultural cognitive model generation module specifically includes:

[0012] A corpus ingestion and processing unit is configured to acquire text and multimedia data in batches from pre-defined, diverse data sources representing the target culture, including but not limited to academic literature databases, public health forums, mainstream news media archives, and philosophical and sociological classics. After acquiring the data, the unit performs a series of natural language preprocessing operations, specifically including text segmentation, part-of-speech tagging, named entity recognition, syntactic dependency analysis, and stop word removal, thereby generating a standardized data stream to be analyzed.

[0013] A metaphor mining and abstraction unit integrates a statistical metaphor recognition algorithm based on source-target domain interaction mapping. The unit receives the preprocessed data stream and automatically identifies and extracts metaphorical expressions from the text by analyzing word co-occurrence patterns and semantic context, such as "the body is a machine" or "disease is war." For each identified metaphorical pattern, the unit calculates its frequency and distribution breadth across the entire corpus and assigns a quantified weight value accordingly, ultimately generating the weighted metaphor link set.

[0014] An ontology anchor point identification unit integrates a hierarchical Dirichlet Allocation (hLDA) topic modeling algorithm. This unit specifically processes a subset of the corpus related to health, medicine, and life sciences, using unsupervised learning to discover and extract several high-level topics with clear semantic cohesion from the text. These extracted topics, after manual verification and naming, are determined as the ontology anchor point nodes of the cultural cognition model.

[0015] A model graph construction unit integrates the ontological anchor nodes and weighted metaphorical links generated by the aforementioned units to construct a complete, directed acyclic graph structure of a cultural cognitive model with specific cultural identifiers. After construction, the unit serializes the model and stores it in the cultural cognitive model database for subsequent retrieval.

[0016] A semantic deep mapping and narrative translation engine, which is the core processing module of this invention, is configured to receive requests from users containing TCM concepts to be queried, and to communicate with the TCM source knowledge graph module and the cultural cognition model database. The engine's function is to perform a non-linear, adaptive translation from source TCM concepts to the target cultural cognition space. Specifically, the engine includes:

[0017] A TCM concept deconstruction unit, upon receiving a user request for a TCM concept to be queried, immediately accesses the TCM source knowledge graph module. By traversing the edges and nodes directly or indirectly connected to the concept node, the unit decomposes the single TCM concept into a set of basic semantic primitives. For example, for the concept "liver qi stagnation," the deconstructed set of semantic primitives is specifically {entity: liver, entity: qi, state: stagnation, associated symptoms: emotional depression, associated symptoms: hypochondriac pain}.

[0018] A cross-cognitive domain analogy search unit integrates a dual-encoder cross-linguistic sentence embedding model. The model's two encoders are deeply fine-tuned using a corpus of Traditional Chinese Medicine (TCM) theory and a representative corpus of the target culture, respectively, enabling the mapping of text fragments from different knowledge systems into a shared high-dimensional semantic vector space. Upon receiving a set of semantic primitives generated by a concept deconstruction unit, the unit converts each semantic primitive and each ontological anchor node in the target culture's cognitive model into a semantic vector within this shared space using their respective encoders. Subsequently, the unit executes a weighted k-nearest neighbor (k-NN) search algorithm in this vector space to find the target culture's ontological anchor node or node combination that is semantically closest to each TCM semantic primitive. The weights here are determined by the metaphorical link weights stored in the cultural cognitive model, prioritizing analogical relationships that conform to the mainstream metaphorical framework of the target culture. The unit ultimately outputs one or more highest-scoring analogy mapping pairs, where each pair contains a TCM semantic primitive and one or more target culture ontological anchors, along with a quantified analogy fitness score. The formula for calculating the analog fitness score is determined as follows:

[0019] S=α·cos(V tcm V ccm )+β·Wmetaphor

[0020] Where S is the final score, Vtcm is the semantic vector of the TCM semantic primitive, Vccm is the semantic vector of the target cultural anchor node, cos(·) is the cosine similarity between the two vectors, Wmetaphor is the metaphorical link weight connecting the relevant anchor nodes, and α and β are preset hyperparameters used to adjust the importance of semantic similarity and metaphorical consistency.

[0021] A constrained narrative generation unit employs a sequence-to-sequence (Seq2Seq) generative language model based on the Transformer architecture. Unlike general language models, the generation process of this unit is strictly constrained by the input structure. Its input is a structured prompt that precisely and sequentially concatenates the following information: the source TCM concept identifier to be explained, the highest-scoring analogy mapping pair determined by the analogy search unit, and the highest-weighted metaphorical framework related to that analogy mapping extracted from the target culture's cognitive model. Based on this constrained input, the unit generates a complete and fluent explanatory text narrative. The core logic of this narrative is no longer a direct translation of TCM terminology, but rather utilizes existing, frequently used core concepts and metaphors (i.e., analogies) in the target culture to explain the function, state, and interrelationships of the source TCM concepts, thereby ensuring the comprehensibility and acceptability of the generated content within the target cultural context.

[0022] A multimodal presentation synthesis module, connected to the semantic deep mapping and narrative translation engine, receives the generated text narrative and communicates with the cultural cognitive model database. Its function is to transform the plain text narrative into an interactive final presentation interface rich in multiple media elements, based on the information reception preferences of the target audience. Specifically, the module includes:

[0023] A presentation modality optimization selection unit first retrieves and obtains the "information modality preference vector" of the current user's cultural group from a cultural cognitive model database. Then, the unit performs a matrix operation on this vector with a preset "content complexity evaluation vector" (which is output synchronously by the narrative generation unit when generating text, quantifying the level of abstraction of the narrative content). By solving an optimization problem aimed at maximizing the product of user preference and content suitability, the unit determines an optimal presentation modality combination scheme, for example, {text: 1.0, animation: 0.9, interactive 3D model: 0.75, static image: 0.5}.

[0024] A multimedia asset generation unit generates corresponding media assets in parallel or sequentially based on the modal combination scheme determined by the optimization selection unit. This unit is further divided into: a text formatter responsible for typesetting the original narrative text, adding titles and highlights; an image generator that integrates a diffusion model, which uses key sentences from the narrative text and visual style descriptors extracted from a cultural cognitive model as joint conditions to generate static illustrations highly matched to the narrative content and target cultural aesthetics; and an animation and 3D scene generator that accesses a parameterized 3D object library (pre-stored with abstract or concrete models such as cells, energy flows, and network nodes), and automatically assembles and renders an interactive animation or 3D scene based on the logical flow of the narrative text using procedural generation techniques, for visualizing the dynamic processes or complex structures described in the narrative.

[0025] An end-user interface compositor that integrates text, generated images, animations, and 3D scenes according to a preset interface layout template to generate a unified, complete, and user-interactive final presentation interface.

[0026] This invention also provides a cross-cultural method for disseminating TCM knowledge based on the above system, the method comprising the following steps:

[0027] Step S1: The system receives a knowledge query request initiated by the user. The request includes a TCM concept to be queried and the user's cultural background identifier.

[0028] Step S2: The semantic deep mapping and narrative translation engine loads the corresponding cultural cognitive model (CCM) from the cultural cognitive model database based on the cultural background identifier. If no perfectly matching model exists in the database, the cultural cognitive model generation module is activated, and the corresponding corpus is called in real time or near real time based on the user's cultural identifier (such as language, region, etc.) to generate a new temporary CCM or select the closest existing CCM.

[0029] Step S3: The TCM concept deconstruction unit in the engine is invoked, which accesses the TCM source knowledge graph module to parse the TCM concept to be queried into a set of basic semantic primitives.

[0030] Step S4: The cross-cognitive domain analogy search unit in the engine is activated. It uses the loaded CCM and the internal cross-lingual sentence embedding model to perform semantic analogy search for each TCM semantic primitive in the ontological anchor point of the target culture, calculates the analogy fitness score, and finally determines a set of optimal analogy mapping relationships.

[0031] Step S5: The constrained narrative generation unit in the engine receives the TCM concept, the optimal analogy mapping relationship, and the dominant metaphor framework extracted from CCM, and generates an explanatory text narrative that conforms to the cognitive habits of the target culture.

[0032] Step S6: The multimodal presentation synthesis module receives the text narrative, and its internal presentation modality optimization selection unit determines the optimal multimedia presentation combination based on the information modality preference vector in the CCM.

[0033] Step S7: The multimedia asset generation unit in the multimodal presentation synthesis module generates all necessary media assets, such as text, images, animations, or interactive 3D models, in parallel or serially according to the determined combination scheme.

[0034] Step S8: The final user interface compositor in the multimodal presentation compositing module integrates all generated media assets into a unified interactive interface and presents it to the user through the output device.

[0035] Furthermore, in the cultural cognition model generation module, the diverse data sources used by the corpus ingestion and processing unit are selected based on a comprehensive evaluation of the target culture in four dimensions: philosophy, medical history, popular culture, and daily language habits, to ensure the comprehensiveness and representativeness of the corpus.

[0036] Furthermore, in the semantic deep mapping and narrative translation engine, the training process of the dual-encoder cross-lingual sentence embedding model includes a specific adversarial training phase. In this phase, a discriminator is introduced, tasked with distinguishing whether a given semantic vector originates from the TCM encoder or the target culture encoder. The training objective of the encoders, in addition to minimizing the loss of the translation task, also includes maximizing the discriminator's confusion level. This aims to force the two encoders to learn a truly aligned, language-independent shared semantic space, thereby improving the accuracy of analogy search.

[0037] In a preferred embodiment of the present invention, the interactive logic of the multimodal presentation synthesis module when generating an interactive 3D scene is also constrained by the cultural cognition model. For example, if the CCM reveals the target cultural preference as "exploratory learning," the generated 3D scene will provide highly flexible perspective switching and model decomposition functions; if the CCM reveals the preference as "guided learning," the scene will unfold in a linear, step-by-step animation format, accompanied by mandatory narrative node synchronization.

[0038] In summary, this invention fundamentally changes the technological paradigm of cross-cultural knowledge dissemination by creating and applying a structured cultural cognition model and designing a complete technical process based on this model, from deep semantic analogy mapping to multimodal adaptive presentation. It is no longer a passive translation and transmission of content, but rather an active and intelligent "re-creation" and "reconstruction" of knowledge based on a profound understanding of the audience's cognitive world. This effectively overcomes cultural barriers and enables the precise, efficient, and barrier-free dissemination of complex professional knowledge, especially Traditional Chinese Medicine knowledge, globally, possessing extremely high practical value and broad application prospects. Attached Figure Description

[0039] Figure 1 This is a structural block diagram of the cross-cultural TCM knowledge dissemination system provided in this embodiment of the invention;

[0040] Figure 2 This is a flowchart illustrating the cross-cultural dissemination method of traditional Chinese medicine knowledge provided in an embodiment of the present invention;

[0041] Figure 3 This is a schematic diagram of the semantic deep mapping and narrative translation process provided in the embodiments of the present invention.

[0042] The attached figures are labeled as follows:

[0043] 10. Traditional Chinese Medicine Source Knowledge Graph Module; 20. Cultural Cognition Model Database; 30. Cultural Cognition Model Generation Module; 31. Corpus Acquisition and Processing Unit; 32. Metaphor Mining and Abstraction Unit; 33. Ontological Anchor Point Identification Unit; 34. Model Graph Construction Unit; 40. Semantic Deep Mapping and Narrative Translation Engine; 41. Traditional Chinese Medicine Concept Deconstruction Unit; 42. Cross-Cognitive Domain Analogy Search Unit; 43. Constrained Narrative Generation Unit; 50. Multimodal Presentation Synthesis Module; 51. Presentation Modality Optimization and Selection Unit; 52. Multimedia Asset Generation Unit; 53. Final User Interface Synthesizer. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0045] Please see Figure 1, a cross-cultural traditional Chinese medicine (TCM) knowledge dissemination method and system provided in this embodiment aims to fundamentally improve the accuracy and acceptance of TCM knowledge cross-cultural dissemination by constructing an adaptive knowledge translation framework centered on the audience cognitive model. At the hardware level, the system is deployed in one or more server clusters, which have high-performance computing units (CPU / GPU), large-capacity memory, and high-speed solid-state storage arrays, and communicate with external data sources and end-user devices through network interfaces. At the software architecture level, the system is designed as a microservices-based distributed system, where each functional module runs as an independent service and interacts through standard API protocols, thus ensuring the high availability, scalability, and maintainability of the system.

[0046] Specifically, the system includes a TCM source-end knowledge graph module 10, a cultural cognitive model database 20, a cultural cognitive model generation module 30, a semantic deep mapping and narrative translation engine 40, and a multimodal presentation synthesis module 50.

[0047] The TCM source-end knowledge graph module 10 is the knowledge cornerstone of the entire system, and a highly structured TCM knowledge graph is stored internally in a fixed manner. The knowledge graph uses the Resource Description Framework (RDF) as its underlying data model and is organized in the form of triples (Subject, Predicate, Object). To ensure data standardization and interoperability, the ontology layer of the graph is extended using mature vocabularies such as Dublin Core, SKOS (Simple Knowledge Organization System), and FOAF (Friend of a Friend), and special classes and attributes are defined for the particularity of the TCM field. Knowledge entities are strictly divided into multiple preset categories, including but not limited to syndrome entities (such as "liver qi stagnation"), formula entities (such as "Xiaoyao San"), traditional Chinese medicine (TCM) entities (such as "Bupleurum"), acupoint entities (such as "Taichong Point"), meridian entities (such as "Foot Jueyin Liver Meridian"), and theoretical concept entities (such as "yin and yang", "five elements"). The relationships between entities are represented by a series of precisely defined predicates. For example, the formula entity "Xiaoyao San" is connected to the TCM entity "Bupleurum" through the "hasComponent" relationship, the TCM entity "Bupleurum" is connected to the meridian entity "Foot Jueyin Liver Meridian" through the "entersMeridian" relationship, and the syndrome entity "liver qi stagnation" is connected to the formula entity "Xiaoyao San" through the "isTreatedBy" relationship. Entity attributes, such as the "nature", "taste", "meridian tropism", and "efficacy" of TCM, are stored as text or numerical data associated with the entity nodes. Take a specific RDF triple instance: <ex:Bupleurum> <tcm:hasproperty><ex:Cold nature>, <ex:Bupleurum chinense> <tcm:entersmeridian><ex:Foot Jueyin Liver Meridian>. The entire knowledge graph is stored in a high-performance graph database, such as Neo4j or Amazon Neptune, which supports efficient querying and reasoning through the SPARQL protocol, providing standardized and machine-readable source-end traditional Chinese medicine knowledge for subsequent modules.

[0048] The cultural cognitive model database 20 is the core data warehouse for the present invention to achieve audience cognitive adaptation. This database is constructed using a document-oriented database (such as MongoDB) to store multiple pre-constructed or dynamically generated cultural cognitive models (Cultural Cognitive Model, CCM) for different target cultural groups globally. Each CCM exists in the database as an independent JSON document and is identified by a unique cultural identifier (for example, "de-DE" represents German culture, "en-US" represents American culture) as the primary key. Each CCM document follows a unified and standardized data structure. This structure includes: a cultural identifier field for uniquely identifying the cultural group corresponding to the model; a set of ontological anchor nodes, which is stored in the form of an array, and each element in the array is a node object representing a core concept related to health, life, and body in the target culture. For example, for North American culture, it may include nodes such as "Immune System as a Defense Force", "Body as a finely-tuned Machine", "Metabolism as an Economic System", etc.; a set of weighted metaphorical link sets, which is also stored in the form of an array, and each element defines a common metaphorical relationship between two ontological anchor nodes and is attached with a floating-point weight value between 0 and 1. This weight value is obtained through statistical analysis of a large corpus and represents the usage frequency, universality, and influence of the metaphor in this culture; a pointer to the representative corpus of this culture, which is a URI pointing to the collection of original text and multimedia data stored in a distributed file system and used to generate this CCM; and an information modality preference vector, which is a multi-dimensional floating-point vector. The dimensions of the vector correspond one-to-one with the presentation modalities supported by the system (such as text, static_image, animation, interactive_3d). Each value in the vector quantifies the average preference degree of this cultural group for the corresponding information presentation modality. This data is sourced from large-scale user behavior analysis or sociological surveys.

[0049] The cultural cognitive model generation module 30 is key to realizing the system's adaptive capability. Its function is to automatically construct or iteratively update the corresponding cultural cognitive model for a specific target cultural group. This module is tightly coupled with the cultural cognitive model database 20 and consists of a series of collaborative processing units. Specifically, this module includes a corpus acquisition and processing unit 31, a metaphor mining and abstraction unit 32, an ontology anchor identification unit 33, and a model graph construction unit 34.

[0050] The corpus ingestion and processing unit 31 serves as the data entry point, configured with a complex ETL (Extract, Transform, Load) process. This unit can acquire massive amounts of text and multimedia data in batches from a pre-defined, highly diverse list of data sources, using APIs or web crawling techniques, based on identifiers of the target culture. The selection criteria for data sources are extremely stringent, aiming to ensure the comprehensiveness and representativeness of the corpus. It covers philosophical classics (such as the German works of Kant and Nietzsche), medical history documents (such as the historical archives of German medical journals), authoritative websites in the field of public health (such as publicly available publications from the Robert Koch Institute), archives of mainstream news media (such as the 30-year database of Der Spiegel), and social media and forums reflecting everyday language habits (such as online communities on German health and lifestyle). After acquiring the raw data, this unit initiates a distributed preprocessing pipeline based on the Spark framework. For text data, the pipeline executes the following steps sequentially: language detection, sentence segmentation, and word segmentation using the Stanford CoreNLP toolkit; high-precision part-of-speech tagging and named entity recognition are performed to identify names, places, organizations, and technical terms; dependency-based syntactic analysis is performed to construct the sentence's syntactic structure tree; and finally, high-frequency but non-informative words are removed based on a domain-specific stop word list. The output of the entire preprocessing pipeline is structured data with rich linguistic annotations, providing clean and standardized input for subsequent deep analysis units.

[0051] The metaphor mining and abstraction unit 32 is the core of constructing the metaphor network in CCM. It integrates a statistical metaphor identification algorithm based on source-target domain interaction mapping, specifically implemented as the Metaphor Identification Procedure (MIP). This unit receives the data stream output from the corpus ingestion and processing unit 31, and automatically identifies and extracts conceptual metaphors from the text by analyzing word co-occurrence patterns and semantic context. For example, when analyzing the sentence "His immune system fought off the virus's invasion," the algorithm identifies "immune system" as the target domain and "army / defense" as the source domain, thus abstracting the conceptual metaphor "IMMUNITY IS WARFARE." For each identified metaphor pattern, this unit traverses the entire corpus, using statistical methods such as point mutual information (PMI) and chi-square tests to calculate its frequency, distribution breadth, and association strength with specific health concepts, ultimately assigning it a quantified weight value. This process generates a weighted set of metaphor links in CCM.

[0052] The ontology anchor identification unit 33 focuses on discovering the macro-conceptual framework of the target culture in the health field from the corpus. Internally, it implements a hierarchical Dirichlet Allocation (hLDA) topic modeling algorithm. Unlike standard LDA models, hLDA can discover a hierarchical topic structure from massive documents in an unsupervised manner, which coincides with the hierarchical nature of human conceptual systems. This unit specifically processes a subset of the corpus related to health, medicine, and life sciences (obtained through keyword filtering and classifier pre-screening). By setting reasonable hyperparameters (e.g., tree depth of 3), With parameters α = 10.0, γ = 1.0, the hLDA model iteratively clusters documents into different topics, forming a tree structure. The root node represents the most generalized concept, while the leaf nodes represent the most specific concepts. For example, when processing German health-related corpora, the model might identify at the top level... The grand themes such as "body mechanism" and "Lebensenergie" (the theory of life energy) are further subdivided into sub-themes such as "Stoffwechsel" (metabolism) and "Immunsystem" (immune system) at the next level. These high-level themes, discovered by algorithms and possessing high semantic cohesion, were formally identified as ontological anchor nodes of the cultural cognitive model after a small amount of manual verification and naming by domain experts.

[0053] The model graph construction unit 34 is the final step in the cultural cognitive model generation module. Its function is to integrate the weighted set of metaphorical links generated by the metaphor mining and abstraction unit 32 and the set of ontological anchor nodes generated by the ontological anchor identification unit 33, ultimately constructing a complete cultural cognitive model with a directed acyclic graph structure and specific cultural identifiers. After construction, this unit serializes this model graph into the aforementioned JSON document format and stores it in the cultural cognitive model database 20, or updates existing models with the same identifiers for subsequent queries and calls.

[0054] Please see Figure 3 The semantic deep mapping and narrative translation engine 40 is the processing hub for realizing the core technical concept of this invention. This engine is configured to receive requests from users containing the TCM concept to be queried, and to communicate in real time with the TCM source knowledge graph module 10 and the cultural cognition model database 20, performing a non-linear, adaptive translation from the source TCM concept to the target cultural cognition space. The internal logic of this engine is also divided into three closely cooperating units: a TCM concept deconstruction unit 41, a cross-cognitive domain analogy search unit 42, and a constrained narrative generation unit 43.

[0055] When the system receives a user request, such as a query for "liver qi stagnation" specifying the target culture as "de-DE", the engine is activated first. The TCM concept deconstruction unit 41 responds immediately, initiating a SPARQL query to the TCM source knowledge graph module 10, using the URI of the concept "liver qi stagnation" as the starting point. By executing a breadth-first or depth-first graph traversal algorithm, this unit recursively retrieves all relevant nodes and relationships directly or indirectly connected to the "liver qi stagnation" node within one to three hops. The traversal results are parsed and structured, thus decomposing this single, highly abstract TCM concept into a set of basic, more cross-culturally understandable semantic primitives. The semantic primitive set for "liver qi stagnation" might be: {source_concept:"liver qi stagnation",associated_entity:["liver","qi"],associated_state:"stagnation",function_impaired:"disperse",primary_symptom:["emotional repression","chest and rib pain"],secondary_symptom:"frequent sighing"}. This deconstruction process is crucial; it breaks down an indivisible, culturally specific concept into multiple more universal descriptions of functions, states, and phenomena.

[0056] Subsequently, this set of semantic primitives is passed to the cross-cognitive domain analogy search unit 42. This unit is key to achieving "deep mapping," and it integrates a specially trained dual-encoder cross-lingual sentence embedding model, specifically the Siamese-BERT architecture. The model has two BERT encoders: one is deeply fine-tuned using a corpus of millions of sentences of traditional Chinese medicine theory, including classics such as the *Huangdi Neijing* and *Shanghan Zabing Lun*; the other is fine-tuned using a representative German corpus pointed to by the target culture cognitive model (in this case, "de-DE" CCM). Furthermore, a specific adversarial training phase is introduced: a discriminator is added, tasked with distinguishing whether a given high-dimensional semantic vector originates from the traditional Chinese medicine encoder or the German encoder. In addition to minimizing the loss of the translation task, the training objective for both encoders includes maximizing the discriminator's confusion level. This adversarial training forces the two encoders to learn a truly aligned, language- and culture-independent shared semantic space.

[0057] Upon receiving the semantic primitive set of "liver qi stagnation," the unit maps each primitive in the set (e.g., "drainage," "emotional repression") and each ontological anchor node (e.g., "Stress-Regulationssystem," "emotionales Gleichgewicht," "Energiefluss") extracted from the loaded "de-DE" cultural cognitive model to the aforementioned shared high-dimensional semantic vector space (e.g., a 768-dimensional vector space) through its corresponding encoder. Next, for each TCM semantic primitive, the unit performs a weighted k-nearest neighbor (k-NN) search algorithm in this vector space to find the target cultural ontological anchor node or combination of nodes that is semantically closest to it. This search process is not a simple distance calculation but is guided by a specific analogy fitness score formula. This formula is precisely defined as:

[0058] S=α·cos(V tcm V ccm )+β·W metaphor ,

[0059] Here, S represents the final analogy fitness score, Vtcm is the semantic vector of the TCM semantic primitive, Vccm is the semantic vector of the target culture anchor node, and cos(·) represents the cosine similarity between the two vectors, used to measure pure semantic closeness. Wmetaphor is the metaphorical link weight extracted from the "de-DE" CCM that connects relevant anchor nodes; it represents whether the analogy conforms to the mainstream thinking framework of the target culture. α and β are preset hyperparameters, such as α = 0.7 and β = 0.3, used to dynamically adjust the importance of semantic similarity and metaphorical consistency in the final score. By calculating the S values ​​for all possible combinations, this unit ultimately outputs one or more analogy mapping pairs with the highest scores. For example, for the functional primitive "liver governs the free flow of Qi," a high-scoring mapping with "Stress-Regulationssystem" might be obtained, while "emotional repression" might form the best analogy with the disordered state of "emotionales Gleichgewicht."

[0060] Finally, these optimal analogy mappings are passed to the constrained narrative generation unit 43. This unit employs a Transformer-based sequence-to-sequence (Seq2Seq) generative language model, such as an instruction-tuned GPT model. Unlike general, open-ended text generation, the generation process in this unit is strictly constrained by a highly structured prompt. The prompt precisely and sequentially concatenates the following information in JSON format: the source TCM concept identifier to be explained ("liver qi stagnation"), the highest-scoring analogy pair determined by the analogy search unit ([{"tcm_primitive":"liver's function of dispersing and regulating qi","ccm_analogy":"Stress-Regulationssystem"},{"tcm_primitive":"qi stagnation","ccm_analogy":"Blockade imEnergiefluss"}]), and the highest-weighted metaphorical framework related to this analogy pair extracted from the "de-DE" CCM (e.g., "BODY IS A REGULATED SYSTEM"). Based on this constrained input, the unit generates a complete, logically fluent, and interpretive text narrative that conforms to German language habits and cultural context. The core logic of this narrative is no longer to rigidly translate terms such as "liver," "qi," and "stagnation," but rather to cleverly utilize core concepts and metaphors already used in German culture (i.e., analogies such as "stress regulation system" and "energy flow blockage") to systematically explain the functional disorders, state manifestations, and internal connections described by the original TCM concept of "liver qi stagnation," thereby ensuring that the generated content has extremely high comprehensibility and acceptance in the target cultural context.

[0061] The multimodal presentation synthesis module 50, as the final output of the system, is responsible for transforming the plain text narrative generated by the aforementioned engine into a final presentation interface rich in various media elements and possessing interactive capabilities. This module is connected to the output of the semantic deep mapping and narrative translation engine 40 and communicates with the cultural cognitive model database 20. Internally, it is specifically divided into a presentation modality optimization selection unit 51, a multimedia asset generation unit 52, and a final user interface synthesizer 53.

[0062] Upon receiving the text narrative, the presentation modality optimization selection unit 51 first retrieves and obtains the "information modality preference vector" of the current user's cultural group ("de-DE") from the cultural cognitive model database 20. Simultaneously, the constrained narrative generation unit 43 outputs a "content complexity evaluation vector" when generating the text. This vector quantifies the abstractness of the narrative content, the complexity of the dynamic process, and the relevance of the spatial structure. This unit performs a dot product or more complex matrix operation on these two vectors to solve an optimization problem aimed at maximizing the product of "user preference" and "content suitability." The output is an optimal presentation modality combination scheme and its weights. For example, for the concept of "liver qi stagnation," which involves dysfunction and emotional association, the output scheme might be {text:1.0, animation:0.9, interactive_3d:0.4, static_image:0.75}, indicating that a text-based approach supplemented with explanatory animations and static charts is the optimal strategy.

[0063] Based on this optimization scheme, the multimedia asset generation unit 52 launches multiple internal sub-generators in parallel or serially to create corresponding media assets. Internally, it includes: a text formatter responsible for professionally formatting the original narrative text, such as dividing it into paragraphs, adding titles, and highlighting keywords; an image generator that integrates an advanced Latent Diffusion Model, which uses key sentences in the narrative text (such as "It's like the body's pressure regulation system malfunctioning, causing energy to flow sluggishly") and visual style descriptors extracted from "de-DE" CCM (such as "sachlich," meaning objective, realistic style) as joint conditions to generate static illustrations or infographics highly matching the narrative content and target cultural aesthetics; and an animation and 3D scene generator that accesses a parameterized 3D object library (which pre-stores abstract or concrete models such as human organs, energy flow particle systems, and network nodes), and automatically assembles and renders an interactive animation or 3D scene according to the logical flow of the narrative text through procedural generation technology. For example, it can generate an animation that visually shows a gauge needle, symbolizing a "pressure regulation system," shifting from the "normal" zone to the "overload" zone, causing congestion in the pipes, symbolizing "energy flow."

[0064] The final user interface compositor 53 is the last link in the presentation process. Using a modern front-end framework (such as React or Vue.js), it intelligently integrates the text content, static images, animation clips, and interactive 3D scene modules generated by each unit into a unified, complete, cross-platform final presentation interface based on a pre-defined, responsive interface layout template. This interface is ultimately delivered to the user's browser or client application via a web service for direct access, reading, and interaction. As a preferred implementation, the interaction logic within this interface is also constrained by the CCM. For example, if the "de-DE" CCM reveals a preference for "exploratory learning" in the target culture, the generated 3D scene will offer highly flexible perspective switching, model decomposition, and parameter adjustment functions; if the CCM of another culture reveals a preference for "guided learning," the scene will unfold in a linear, step-by-step animation format, accompanied by mandatory narrative node synchronization, ensuring that the user understands the scene according to a pre-defined logical path.

[0065] Please see Figure 2 The present invention also provides a cross-cultural method for disseminating traditional Chinese medicine knowledge based on the above system, the method specifically including the following steps:

[0066] Step S1: The system's front-end interface receives the knowledge query request initiated by the user. The request is submitted through a structured form or API call, which must contain at least two fields: one is the TCM concept to be queried (e.g., "spleen and stomach deficiency and cold"), and the other is the user's cultural background identifier (e.g., "ja-JP", representing Japanese culture).

[0067] Step S2: The request is routed to the semantic deep mapping and narrative translation engine 40. Based on the received cultural background identifier "ja-JP", this engine queries the cultural cognitive model database 20 and loads the corresponding cultural cognitive model (CCM). If a pre-built "ja-JP" model exists in the database with an exact match, it is loaded directly. If not, the cultural cognitive model generation module 30 is dynamically activated. Based on the "ja-JP" identifier, this module calls a pre-defined Japanese cultural corpus (such as the Aozora Bunko database, CiNii academic database, etc.) to execute the aforementioned CCM generation process in real-time or near real-time, generating a new, temporary CCM for this query. Alternatively, the system can select an existing CCM that is culturally closest (such as "zh-CN") as a substitute.

[0068] Step S3: The TCM concept deconstruction unit 41 in the engine is invoked, which accesses the TCM source knowledge graph module 10 to parse the TCM concept to be queried, "spleen and stomach deficiency and cold", into a set of basic semantic primitives, such as {associated_entity:["spleen","stomach"],associated_state:["deficiency","cold"],function_impaired:"transformation",primary_symptom:["loss of appetite","diarrhea"],associated_phenomenon:"prefers warmth and pressure"}.

[0069] Step S4: The cross-cognitive domain analogy search unit 42 in the engine is activated. Utilizing the "ja-JP" CCM loaded in Step S2 and its internal cross-linguistic sentence embedding model, adversarially trained on a Chinese-Japanese bilingual corpus, it performs a weighted semantic analogy search for each semantic primitive of "spleen and stomach deficiency and cold" within the ontological anchors of "ja-JP" culture (which may include "digestion and absorption capacity," "maintaining body temperature," "enel-gear production," etc.). By calculating the analogy fitness score, a set of optimal analogy mappings is ultimately determined.

[0070] Step S5: The constrained narrative generation unit 43 in the engine receives the source concept "spleen and stomach deficiency and cold", the optimal analogical mapping relationship determined in step S4, and the dominant metaphorical framework extracted from the "ja-JP" CCM (which may be related to core concepts in Japanese culture such as "nature", "harmony", and "ki"). Based on this highly structured input, the unit generates an explanatory text narrative that conforms to Japanese cultural cognitive habits and language expression paradigms.

[0071] Step S6: The generated text narrative is passed to the multimodal presentation synthesis module 50. Its internal presentation modality optimization selection unit 51 determines the optimal multimedia presentation combination based on the information modality preference vector (which may prefer static, elaborate images and simple text) stored in the "ja-JP" CCM, combined with the complexity of the content.

[0072] Step S7: The multimedia asset generation unit 52 in the multimodal presentation synthesis module 50 generates all necessary media assets in parallel according to the determined combination scheme. For example, the image generator may generate a schematic diagram with a "mono no aware" or "wabi-sabi" aesthetic style that depicts a weakened digestive function.

[0073] Step S8: The final user interface synthesizer 53 integrates all generated media assets into a unified interactive interface with a design style that conforms to the aesthetics of Japanese users, and presents it to the user through the output device.

[0074] The technical effects of the present invention will be further illustrated below through a specific embodiment.

[0075] Example 1

[0076] This example aims to explain the traditional Chinese medicine concept of "the upward flaring of heart fire" to a user with a cultural background identifier of "en-GB" (UK).

[0077] 1. System initialization and request reception: The system receives the request {concept: "the upward flaring of heart fire", ccm_id: "en-GB"}.

[0078] 2. CCM loading: The system loads the pre-built "en-GB" model from the cultural cognitive model database 20. This model contains ontological anchors such as "Central Nervous System (CNS) Regulation", "Emotional Processing Unit", "System Overheating / Burnout", etc., and weighted metaphorical links such as "MIND IS A MACHINE" (weight: 0.82), "EMOTION IS TEMPERATURE" (weight: 0.75).

[0079] 3. Concept deconstruction: The traditional Chinese medicine concept deconstruction unit 41 deconstructs "the upward flaring of heart fire" into a set of semantic primitives: {entity: "heart", attribute: "fire", phenomenon: "upward flaring", primary_symptom: ["insomnia", "mouth and tongue ulcers"], associated_emotion: "irritability"}. Among them, "heart" is associated with the function of "Governs the mind / spirit" in the knowledge graph.

[0080] 4. Analogy search: The cross-cognitive domain analogy search unit 42 performs operations.

[0081] "The heart governs the mind / spirit" obtains a high cosine similarity with the "CNS Regulation" and "Emotional Processing Unit" anchors in the "en-GB" CCM through semantic vector calculation.

[0082] "Fire" and "upward flaring" are highly correlated with the "System Overheating / Burnout" anchor.

[0083] "Irritability" and "insomnia" are very close in the semantic space to the consequences of "CNS Regulation" disorder.

[0084] The analog fitness score formula S = 0.7 * cos(V_tcm, V_ccm) + 0.3 * W_metaphor is applied. Since the metaphor weights of "MIND IS A MACHINE" and "EMOTION IS TEMPERATURE" are high, the path that maps "heart fire flaring up" to "CNS / Emotional Processing Unit Overheating" obtains the highest score. The finally determined optimal analog mapping is: [{"tcm_primitive": "heart (spirit)", "ccm_analogy": "Central Nervous / Emotional Processing Unit"}, {"tcm_primitive": "fire / flaring up", "ccm_analogy": "Overheating / Burnout"}].

[0085] 5. Narrative generation: The Constrained Narrative Generation Unit 43 receives the above information and generates the following core narrative (a Chinese paraphrase is provided here for clarity): "You can understand the 'heart' in traditional Chinese medicine as the central processor or emotional regulation center of your body. The condition described by 'heart fire flaring up' is similar to this core system being 'overheated' or 'burned out' due to long-term stress or excessive thinking. This 'overheated' state will interfere with its normal function, causing the brain to be unable to calm down, resulting in insomnia and restlessness, just like an overheated computer running unstably. At the same time, this internal 'heat' will spread upward, manifested as ulcers in the mouth and tongue on the body."

[0086] 6. Multimodal synthesis: According to the preference vector (preference information chart and short animation) of the "en-GB" CCM, the system generates an infographic juxtaposing the analogy images of the brain and CPU, highlighting the "overheated" state with red highlights and thermometer symbols. At the same time, a 30-second animation is generated, showing the process of a CPU overheating and causing the entire system to malfunction. The final interface integrates and presents the narrative text, infographic, and animation.

[0087] In summary, the present invention creates and applies a structured cultural cognitive model, and designs a complete technical process from semantic deep analog mapping to multimodal adaptive presentation based on this model, fundamentally changing the technical paradigm of cross-cultural knowledge dissemination. It is no longer the passive translation and transmission of content, but an active, intelligent knowledge "re-creation" and "re-construction" based on a profound understanding of the audience's cognitive world, thereby effectively overcoming cultural barriers and achieving the precise, efficient, and barrier-free dissemination of complex professional knowledge, especially traditional Chinese medicine knowledge, globally, with extremely high practical value and broad application prospects.

[0088] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.< / tcm:entersmeridian> < / tcm:hasproperty>

Claims

1. A cross-cultural system for disseminating traditional Chinese medicine knowledge, characterized in that, The system includes: The TCM source knowledge graph module (10) is configured to store a structured TCM knowledge graph consisting of knowledge entities, relationships between entities and entity attributes, in order to provide standardized source TCM knowledge; A cultural cognitive model database (20) is configured to store at least one cultural cognitive model associated with a specific cultural identifier. The cultural cognitive model quantitatively models the cultural background, health concepts, metaphor system, and information reception preferences of the target audience. The cultural cognitive model has a standardized data structure, including: a cultural identifier field, used to uniquely identify the cultural group corresponding to the model; a set of ontological anchor nodes, where each anchor node represents a core concept related to health, life, or body in the target culture; a set of weighted metaphor links, where each link represents a common metaphorical relationship between two ontological anchor nodes in the target culture, and is accompanied by a weight value that quantifies the frequency and importance of the metaphor; and an information modality preference vector, which is a multi-dimensional vector that quantifies the degree of preference of the cultural group for different information presentation modalities such as text, static images, animation, and interactive 3D models. The semantic deep mapping and narrative translation engine (40) communicates with the TCM source knowledge graph module (10) and the cultural cognition model database (20), and is configured to receive requests containing TCM concepts to be queried, and translate the TCM concepts to be queried into explanatory text narratives that conform to the cognitive habits of the target culture according to the cultural cognition model; specifically, it includes: a TCM concept deconstruction unit (41), configured to access the TCM source knowledge graph module (10), and decompose the TCM concepts to be queried into a set of basic semantic primitives composed of function, state and phenomenon descriptions through graph traversal. The set of words; the cross-cognitive domain analogy search unit (42) is configured to use an internally integrated dual encoder cross-linguistic sentence embedding model to map the semantic primitives and ontological anchor nodes in the cultural cognitive model to a shared high-dimensional semantic vector space, and then execute a weighted search algorithm in the vector space to determine the optimal analogy mapping relationship; the weighted search algorithm executed by the cross-cognitive domain analogy search unit (42) is based on a preset analogy fitness score formula to score each potential analogy mapping relationship, and the analogy fitness score formula is determined as follows: ； Where S is the final analogy fitness score, Vtcm is the semantic vector of the TCM semantic primitive, Vccm is the semantic vector of the target culture ontology anchor node, cos(·) is the cosine similarity between the two vectors, Wmetaphor is the metaphor link weight extracted from the cultural cognitive model and related to the ontology anchor node, and α and β are preset hyperparameters used to adjust the importance of semantic similarity and metaphor consistency. The training process of the dual-encoder cross-lingual sentence embedding model includes an adversarial training phase, which introduces a discriminator to distinguish the source of the semantic vectors and forces the two encoders to learn an aligned, language-independent shared semantic space by maximizing the confusion of the discriminator. The constrained narrative generation unit (43) is configured to receive the query TCM concept, the optimal analogy mapping relationship, and the dominant metaphor framework extracted from the cultural cognitive model, and generate the explanatory text narrative based on this structured information. The core logic of the narrative is to use the analogy mapping relationship to explain the query TCM concept. The constrained narrative generation unit (43) adopts a sequence-to-sequence generative language model based on the Transformer architecture. The generation process is strictly constrained by a highly structured prompt, which precisely and sequentially concatenates the following information in JSON format: the source TCM concept identifier to be explained, the highest-scoring analogy mapping pair determined by the analogy search unit, and the highest-weighted metaphor framework related to the analogy mapping extracted from the cultural cognitive model. Based on this constrained input, the unit generates the explanatory text narrative. A multimodal presentation synthesis module (50) is connected to the semantic deep mapping and narrative translation engine (40) and communicates with the cultural cognitive model database (20). It is configured to receive the explanatory text narrative and synthesize the text narrative into a user presentation interface containing multiple media elements based on the information receiving preferences recorded in the cultural cognitive model. The cultural cognition model generation module (30) includes an ontology anchor identification unit (33), which is configured to extract high-level, semantically cohesive topics from a subset of health-related data after preprocessing, and identify them as ontology anchor nodes. The ontology anchor identification unit (33) integrates a hierarchical Dirichlet Allocation (hLDA) topic modeling algorithm, which discovers and extracts a hierarchical topic structure from the text through unsupervised learning, and identifies the high-level topics in the structure as ontology anchor nodes.

2. The system according to claim 1, characterized in that, The cultural cognitive model generation module (30) is connected to the cultural cognitive model database (20) and is configured to automatically construct or update the corresponding cultural cognitive model for a specific target cultural group; the cultural cognitive model generation module (30) further includes: Corpus ingestion and processing unit (31) is configured to acquire and preprocess text and multimedia data from diverse data sources representing the target culture; Metaphor mining and abstraction unit (32) is configured to identify conceptual metaphors from the preprocessed data and calculate their weights to generate the weighted metaphor link set; The model graph construction unit (34) is configured to integrate the ontological anchor node and the weighted metaphor link to construct a complete cultural cognition model and store it in the cultural cognition model database (20).

3. The system according to claim 2, characterized in that, The metaphor mining and abstraction unit (32) integrates a statistical metaphor recognition algorithm based on source-target domain interaction mapping, which identifies and extracts metaphorical expressions by analyzing the co-occurrence patterns and semantic context of words.

4. The system according to claim 1, characterized in that, The multimodal presentation synthesis module (50) specifically includes: Presentation modality optimization selection unit (51) is configured to retrieve the current user's "information modality preference vector" from the cultural cognitive model database (20) and, in conjunction with the content complexity assessment of the explanatory text narrative, determine an optimal presentation modality combination scheme by solving an optimization problem; Multimedia asset generation unit (52) is configured to generate all necessary media assets according to the optimal presentation modality combination scheme; The final user interface synthesizer (53) is configured to integrate the text narrative with the generated media assets according to a preset interface layout template to generate the final user presentation interface.

5. The system according to claim 4, characterized in that, The multimedia asset generation unit (52) is further divided into: A text formatter configured to format, add headings, and highlight the explanatory text narrative; An image generator, which integrates a diffusion model with key sentences in the narrative text and visual style descriptors extracted from the cultural cognitive model as joint conditions, is used to generate static illustrations that match the narrative content and the target cultural aesthetics. An animation and 3D scene generator is configured to access a parameterized 3D object library and, based on the logical flow of the narrative text, automatically assemble and render an interactive animation or 3D scene for visualizing dynamic processes or complex structures using procedural generation technology. When the animation and 3D scene generator generates an interactive 3D scene, its interaction logic is also constrained by the cultural cognition model. Specifically, when the cultural cognition model reveals the target cultural preference of "exploratory learning", the generated 3D scene is configured to provide highly flexible perspective switching and model decomposition functions. When the cultural cognition model reveals the target cultural preference of "guided learning", the generated 3D scene is configured to unfold in a linear, step-by-step animation form, accompanied by mandatory narrative node synchronization.

6. A method for cross-cultural dissemination of traditional Chinese medicine knowledge based on the above system according to any one of claims 1-5, characterized in that, The method includes the following steps: Step S1: The system receives a knowledge query request initiated by the user. The request includes a TCM concept to be queried and the user's cultural background identifier. Step S2: The semantic deep mapping and narrative translation engine (40) loads the corresponding cultural cognitive model from the cultural cognitive model database (20) based on the cultural background identifier. If there is no completely matching model in the database, the cultural cognitive model generation module is activated and calls the corresponding corpus in real time or near real time based on the user's cultural identifier to generate a new temporary cultural cognitive model or select the most similar existing cultural cognitive model. Step S3: The TCM concept deconstruction unit (41) in the engine is invoked, which accesses the TCM source knowledge graph module (10) to parse the TCM concept to be queried into a set of basic semantic primitives; Step S4: The cross-cognitive domain analogy search unit (42) in the engine is activated. It uses the loaded cultural cognitive model and the internal cross-linguistic sentence embedding model to perform semantic analogy search for each TCM semantic primitive in the ontological anchor point of the target culture, calculates the analogy fitness score, and finally determines a set of optimal analogy mapping relationships. Step S5: The constrained narrative generation unit (43) in the engine receives the TCM concept, the optimal analogy mapping relationship and the dominant metaphor framework extracted from the cultural cognitive model, and generates an explanatory text narrative that conforms to the cognitive habits of the target culture. Step S6: The multimodal presentation synthesis module (50) receives the text narrative, and its internal presentation modality optimization selection unit (51) determines the optimal multimedia presentation combination based on the information modality preference vector in the cultural cognitive model. Step S7: The multimedia asset generation unit (52) in the multimodal presentation synthesis module (50) generates all necessary media assets such as text, images, animations or interactive 3D models in parallel or serially according to the determined combination scheme; Step S8: The final user interface compositor (53) in the multimodal presentation compositing module (50) integrates all generated media assets into a unified interactive interface and presents it to the user through the output device.

7. The method according to claim 6, characterized in that, In the cultural cognition model generation module, the corpus ingestion and processing unit (31) uses a variety of data sources. The selection criteria are based on a comprehensive evaluation of the target culture in four dimensions: philosophy, medical history, popular culture and daily language habits, so as to ensure the comprehensiveness and representativeness of the corpus. In the semantic deep mapping and narrative translation engine (40), the training process of the dual encoder cross-language sentence embedding model includes a specific adversarial training phase; At this stage, a discriminator is introduced, whose task is to distinguish whether a given semantic vector originates from the TCM encoder or the target culture encoder; the training objective of the encoder, in addition to minimizing the loss of the translation task, also includes maximizing the confusion of the discriminator, which aims to force the two encoders to learn a truly aligned, language-independent shared semantic space, thereby improving the accuracy of analogy search. When generating an interactive 3D scene, the interactive logic of the multimodal presentation synthesis module (50) is also constrained by the cultural cognition model.

Images