Tourism culture content generation method and device based on large language model

Abstract

The present application relates to the field of digital technology of tourism and travel, and discloses a tourism and culture content generation method and device based on a large language model, wherein the device comprises a tourism and culture knowledge graph global construction and dynamic updating module, a scene demand semantic analysis and generation constraint standardization definition module, a culture narrative chain intelligent construction and core content original generation module, a multi-form content adaptation and immersive expression optimization module, a tourism and culture content culture authenticity verification and compliance verification module, and a content effect quantitative evaluation and generation model iterative optimization module; the method comprises the following steps: S1, a tourism and culture knowledge graph construction stage; S2, a demand analysis and constraint definition stage; S3, a narrative chain construction and core content generation stage; S4, a content form adaptation and immersive optimization stage; S5, a content verification and compliance verification stage; and S6, an effect evaluation and model iteration stage; the present application breaks through the problem that traditional tourism and culture content production relies on artificial creation and content homogenization is serious.

Description

Technical Field

[0001] This invention relates to the field of digital technology in cultural tourism, and in particular to a method and apparatus for generating tourism cultural content based on a large language model. Background Technology

[0002] With the continuous deepening of the integration of culture and tourism, the core competitiveness of the tourism industry has shifted from traditional natural landscape resources to the excavation and presentation of the cultural connotation of destinations. Tourists' demand for cultural experiences during the tourism process continues to rise, and high-quality, personalized, and scenario-based tourism cultural content has become the core element of tourism destination operation and cultural tourism product creation.

[0003] At the same time, generative artificial intelligence technology, represented by large language models, is developing rapidly, demonstrating mature application capabilities in text content creation, semantic understanding, and logic construction. It provides a feasible technical path for the intelligent and large-scale production of tourism and cultural content, and has become an important technical support for the digital transformation of culture and tourism.

[0004] Currently, the production and supply of tourism and cultural content are mainly divided into two modes. One is the traditional manual collection and editing mode, which relies on professional copywriters and cultural and tourism researchers to complete data collection, content writing and optimization. The other is a simple content generation mode based on a general generative model, which completes the simple generation of tourism and cultural content through basic text commands input by users.

[0005] Both models have obvious shortcomings in practical applications and cannot simultaneously meet the multiple core needs of the cultural tourism industry for content accuracy, scene adaptability, and production efficiency. There is an urgent need for a set of intelligent tourism cultural content generation technology solutions that are adapted to the characteristics of the cultural tourism industry. Summary of the Invention

[0006] The purpose of this invention is to provide a method and apparatus for generating tourism and cultural content based on a large language model, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] The tourism culture content generation device based on a large language model includes a module for the full-domain construction and dynamic updating of a tourism culture knowledge graph, a module for the semantic parsing of scenario-based needs and the standardized definition of generation constraints, a module for the intelligent construction of cultural narrative chains and the native generation of core content, a module for the adaptation of multi-form content and the optimization of immersive expression, a module for the verification of the authenticity and compliance of cultural tourism content, and a module for the quantitative evaluation of content effects and the iterative optimization of the generation model.

[0009] As a further improvement to this technical solution: the tourism culture knowledge graph full-domain construction and dynamic update module includes a cultural entity full-domain collection sub-device, a cultural relationship deep mining sub-device, and a knowledge graph dynamic update sub-device; the cultural entity full-domain collection sub-device connects to authoritative cultural tourism data channels to collect full-dimensional cultural entity data corresponding to tourist destinations, delineates seven core entity types: historical figures, intangible cultural heritage skills, folk tales, regional cultural context, historical events, cultural relics and historical sites, and works of art, and completes the standardized labeling of the basic attributes of each entity; the cultural relationship deep mining sub-device, based on a finely tuned large language model in the tourism culture field, mines four core relationships between cultural entities: causal relationships, temporal relationships, spatial relationships, and cultural heritage relationships, and completes the accuracy calibration of relationships through entity relationship quantification algorithms. The core calculation formula is: , in the formula: Representative Entity With entity The correlation score between them ranges from 0 to 1; Representative Entity Semantic embedding vectors and entities Cosine similarity between semantic embedding vectors; Representative Entity With entity The authoritative co-occurrence coefficient between them, with a value ranging from 0 to 1; Representative Entity With entity The normalized composite value of temporal and spatial distances between entities ranges from 0 to 1; the cultural association deep mining sub-device constructs a three-level association network of core associations, secondary associations, and extended associations for each entity based on association score; the knowledge graph dynamic update sub-device connects to authoritative cultural release channels to complete the dynamic update of knowledge graph entity attributes and associations, and sets authoritative source tags and credibility levels for each entity.

[0010] As a further improvement to this technical solution: the scenario-based requirement semantic parsing and generation constraint standardization definition module includes a requirement semantic deep parsing sub-device, a generation scenario adaptation sub-device, and a constraint condition standardization definition sub-device; the requirement semantic deep parsing sub-device, based on a large language model, extracts six core elements from the user's input natural language requirements: destination subject, core content theme, target audience, content carrier, narrative style, and content length, completing the structured transformation of requirement elements; the generation scenario adaptation sub-device has a built-in full-scenario adaptation rule library for cultural and tourism content, covering five core application scenarios: offline guided tours, short video dissemination, graphic science popularization, study tours, and immersive entertainment, setting four types of exclusive generation rules for each scenario: content structure, narrative rhythm, audience adaptation, and compliance red lines, completing the matching of user requirements with scenario rules, locking the target generation scenario, and retrieving the corresponding exclusive generation rules; the constraint condition standardization definition sub-device integrates the structured requirement elements and the target scenario's exclusive generation rules to generate standardized content generation constraint parameters.

[0011] As a further improvement to this technical solution: the intelligent construction and core content native generation module of the cultural narrative chain includes a narrative main line planning sub-device, a content node filling sub-device, and a core narrative generation sub-device; the narrative main line planning sub-device plans the main narrative line and core nodes of the content's development and transition based on standardized content generation constraint parameters and the entity association relationship of the tourism culture knowledge graph, locks in the core cultural highlights of the content, and completes the accurate matching of the main line with user needs through a narrative main line adaptation algorithm. The core calculation formula is: , in the formula: Represents the main narrative line The overall fit score ranges from 0 to 1; This represents the degree of fit between the narrative thread and the user-specified core theme, with a value ranging from 0 to 1. This represents the fit between the main narrative and the target generation scene, with a value ranging from 0 to 1. The textual logic coherence of the narrative thread is represented, with a value ranging from 0 to 1; , , These are the weighting coefficients of the three indicators, and their sum is 1. The narrative mainline planning sub-device determines the final narrative mainline based on the comprehensive adaptability score, clarifying the core positioning and content proportion of each node. The content node filling sub-device, based on the positioning and core requirements of the narrative mainline nodes, matches the cultural entities, allusions, details, and authoritative historical materials of the corresponding nodes from the tourism culture knowledge graph to complete the content material filling of each narrative node. The core narrative generation sub-device, based on a finely tuned large language model in the tourism culture field, uses standardized generation constraint parameters as boundaries, the determined narrative mainline as a framework, and the cultural materials filled in the nodes as the core to generate complete core cultural content text. It has a built-in cultural weight constraint mechanism to ensure that the proportion of cultural information in the generated content is not lower than the minimum threshold set by the target scene, and binds corresponding authoritative source tags to all core cultural expressions.

[0012] As a further improvement to this technical solution: the multi-format content adaptation and immersive expression optimization module includes a content format conversion sub-device, an immersive expression optimization sub-device, and a multi-modal supporting content generation sub-device; the content format conversion sub-device, based on a large language model, converts the core content text into a corresponding format of content product according to the carrier requirements in the content generation constraint parameters; the immersive expression optimization sub-device performs full-dimensional immersive expression optimization of the content for the target scene and content format, including adjusting the narrative perspective, supplementing scene details, adjusting emotional rhythm, and embedding interactive content, and completes the immersion score of the optimized content through an immersion quantification evaluation algorithm, and completes the cyclic optimization of the content based on the score until the target scene threshold requirement is reached; the multi-modal supporting content generation sub-device generates adapted multi-modal supporting production instructions simultaneously based on the generated final content.

[0013] As a further improvement to this technical solution: the cultural authenticity verification and compliance verification module for cultural tourism content includes a cultural authenticity verification sub-device, a content compliance verification sub-device, and an automatic error correction sub-device; the cultural authenticity verification sub-device performs a full cross-comparison between the generated content and authoritative information from the tourism culture knowledge graph, identifies and marks cultural fact errors and unsubstantiated fictional content in the content, matches corresponding authoritative correction criteria, and completes a quantitative assessment of the authenticity of the generated content through a cultural fact accuracy algorithm. The core calculation formula is: , in the formula: The accuracy of cultural facts represented by the generated content, with a value ranging from 0 to 1; The number of cultural factual errors and fabricated content without authoritative sources identified in the representative content; The total number of core cultural fact statements contained in the representative content; the cultural authenticity verification sub-device sets a minimum threshold for the accuracy of cultural facts, and content that does not reach the threshold triggers an automatic correction process; the content compliance verification sub-device connects to the content compliance review rule library to complete the full-dimensional compliance verification of the generated content and marks the non-compliant content and compliance risk points; the automatic error correction sub-device, based on a large language model, combines the verification and validation results to automatically correct the marked content, and re-executes the verification and validation process after the correction is completed.

[0014] As a further improvement to this technical solution: the content effect quantitative evaluation and generation model iterative optimization module includes a feedback data collection sub-device, an effect index quantification sub-device, and a model fine-tuning optimization sub-device; the feedback data collection sub-device collects user feedback data and content dissemination data on the generated content to form a complete feedback data set; the effect index quantification sub-device constructs a content effect quantitative evaluation system, completes multi-dimensional quantitative scoring of content effect, selects high-scoring high-quality content samples, and extracts the core optimization directions from user feedback; the model fine-tuning optimization sub-device constructs a fine-tuning dataset based on high-quality content samples, and, combined with the optimization directions from user feedback, completes the continuous iterative optimization of the fine-tuned large language model in the tourism and culture field.

[0015] A method for generating tourism and cultural content based on a large language model includes the following steps:

[0016] S1. In the tourism culture knowledge graph construction stage, we collect full-dimensional cultural entity data of tourist destinations through authoritative cultural tourism data channels, complete the standardized labeling of entity attributes, explore the multi-level relationship between entities and complete the quantitative labeling of the relationship, construct a tourism culture-specific knowledge graph, and connect with authoritative release channels to complete the dynamic update of the knowledge graph and the entity credibility classification.

[0017] S2, the requirement analysis and constraint definition stage, extracts the core elements of user requirements through the large language model and completes the structure transformation, matches the built-in full-scenario adaptation rule library to lock the target generation scenario, and integrates requirement elements and scenario rules to generate standardized content generation constraint parameters;

[0018] S3, Narrative Chain Construction and Core Content Generation Stage: Based on content generation constraint parameters and knowledge graph entity association relationships, the narrative main line and core nodes are planned, the narrative main line adaptation quantification evaluation is completed and the final narrative framework is determined, corresponding cultural materials are matched and filled for each node, and a complete core cultural content text is generated based on a finely tuned large language model in the field of tourism culture.

[0019] S4. Content format adaptation and immersive optimization stage: Convert the core content text into the finished content corresponding to the target carrier, optimize the immersive expression of the content for the target scenario, complete the cyclical optimization of the content through immersive quantification evaluation, and generate supporting multimodal production instructions in sync.

[0020] S5. In the content verification and compliance verification stage, the generated content will be cross-referenced with authoritative information in the knowledge graph to complete the quantitative assessment of the accuracy of cultural facts. Simultaneously, full-dimensional compliance verification will be performed, and the marked errors and violations will be automatically corrected and re-verified.

[0021] S6. In the effect evaluation and model iteration stage, user feedback data and content dissemination data are collected to complete the multi-dimensional quantitative evaluation of content effect, select high-quality content samples to build a fine-tuning dataset, and complete the continuous iterative optimization of the large language model in the tourism and culture field.

[0022] Compared with the prior art, the beneficial effects of the present invention are:

[0023] 1. This invention breaks through the problems of traditional tourism cultural content production relying on manual creation, serious content homogenization, and insufficient expression of cultural core. It constructs a full-link intelligent system from the construction of cultural knowledge foundation to the native generation of content. It can complete the in-depth creation of content based on the core cultural context of tourist destinations, avoiding the problems of fragmented cultural information and broken cultural context logic in traditional content creation. At the same time, it can adapt to the exclusive needs of different application scenarios and content carriers, complete the immersive expression optimization of content, break the limitations of traditional cultural tourism content's one-way knowledge indoctrination, and greatly improve the adaptability of tourism cultural content to user needs and application scenarios, so that core tourism cultural resources such as regional culture, intangible cultural heritage skills, and historical context can be presented more completely and more in line with the needs of the audience.

[0024] 2. This invention establishes a full-process cultural authenticity verification and compliance check mechanism. Through cross-comparison with authoritative knowledge graphs, it fundamentally avoids common problems in AI-generated cultural and tourism content, such as historical errors and unfounded fabrications, ensuring the authority and rigor of tourism and cultural content. At the same time, it constructs a complete closed loop of content effect feedback and model iteration, which can continuously optimize content generation capabilities based on content usage effects and user feedback. This significantly reduces the creation threshold and labor costs of tourism and cultural content, improves the production efficiency of cultural and tourism content, and provides a standardized and reusable intelligent solution for the cultural dissemination and brand building of tourist destinations, helping to promote the precise and efficient dissemination of tourism and cultural resources.

[0025] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it according to the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Specific embodiments of the present invention are given in detail below with reference to the accompanying drawings. Attached Figure Description

[0026] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0027] Figure 1 This is a schematic diagram of the method and apparatus for generating tourism and cultural content based on a large language model. Detailed Implementation

[0028] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are for illustrative purposes only and are not intended to limit the scope of the invention. The invention is described more specifically in the following paragraphs by way of example with reference to the accompanying drawings. It should be noted that the drawings are in a very simplified form and use non-precise proportions, and are only used to facilitate and clarify the illustration of the embodiments of the present invention.

[0029] Please see Figure 1 In this embodiment of the invention, the tourism culture content generation device based on a large language model includes a tourism culture knowledge graph full-domain construction and dynamic update module, a scenario-based demand semantic parsing and generation constraint standardization definition module, a cultural narrative chain intelligent construction and core content native generation module, a multi-form content adaptation and immersive expression optimization module, a cultural tourism content authenticity verification and compliance verification module, and a content effect quantitative evaluation and generation model iterative optimization module.

[0030] Specifically, the six modules form a closed-loop architecture that covers the entire chain of knowledge foundation construction, requirement analysis, content generation, content optimization, quality verification, and model iteration. This architecture covers the entire process of tourism and cultural content from requirement input to finished product output and continuous optimization. Each module provides input and support for the modules before and after it, forming a complete technical solution.

[0031] The tourism culture knowledge graph construction and dynamic update module includes a cultural entity full-domain collection sub-device, a cultural relationship in-depth mining sub-device, and a knowledge graph dynamic update sub-device. The cultural entity full-domain collection sub-device connects to authoritative cultural tourism data channels to collect multi-dimensional cultural entity data corresponding to tourist destinations, defining seven core entity types: historical figures, intangible cultural heritage skills, folk tales, regional cultural context, historical events, cultural relics and historical sites, and works of art, and completing standardized labeling of the basic attributes of each entity. The cultural relationship in-depth mining sub-device, based on a finely tuned large language model in the tourism culture field, mines four core relationships among cultural entities: causal relationships, temporal relationships, spatial relationships, and cultural heritage relationships. It completes the accuracy calibration of relationships through an entity relationship metric algorithm; the core calculation formula is: , in the formula: Representative Entity With entity The correlation score between them ranges from 0 to 1; Representative Entity Semantic embedding vectors and entities The cosine similarity between semantic embedding vectors; Representative Entity With entity The authoritative co-occurrence coefficient between them, with a value ranging from 0 to 1; Representative Entity With entity The normalized composite value of temporal and spatial distances between entities ranges from 0 to 1; the cultural association deep mining sub-device constructs a three-level association network of core associations, secondary associations, and extended associations for each entity based on association score; the knowledge graph dynamic update sub-device connects to authoritative cultural release channels to complete the dynamic update of entity attributes and associations in the knowledge graph, and sets authoritative source tags and credibility levels for each entity.

[0032] Specifically, the Cultural Entity Comprehensive Data Collection Sub-device: Purpose: Responsible for the comprehensive collection and standardized labeling of core tourism cultural entities, providing a basic data source for knowledge graph construction; Functional Expansion: This sub-device can connect to authoritative cultural and tourism data channels such as the official database of the Ministry of Culture and Tourism, local intangible cultural heritage protection centers, authoritative historical research institutions, cultural relics and archaeology release platforms, and formal local chronicle databases. It collects comprehensive cultural entity data corresponding to tourist destinations, clearly defining seven core entity types: historical figures, intangible cultural heritage skills, folk tales, regional cultural context, historical events, cultural relics and historical sites, and works of art. At the same time, it completes the standardized labeling of basic attributes such as the dynasty, geographical affiliation, cultural value, and authoritative source for each entity, eliminating informal cultural information without authoritative sources, and ensuring the authority and standardization of basic entity data.

[0033] Cultural Relationship Deep Dive Sub-device: Purpose: To uncover multi-level relationships between cultural entities, accurately quantify the degree of association, and construct a relationship network between entities, providing core support for subsequent narrative chain construction; Functional Expansion: Based on a finely tuned large language model in the tourism and culture field, this sub-device uncovers four core relationships among seven categories of cultural entities: causal relationships, temporal relationships, spatial relationships, and cultural heritage relationships. It uses an entity association quantification algorithm to accurately calibrate the degree of association. The core calculation formula is: , in the formula: Representative Entity With entity The correlation score between the two entities ranges from 0 to 1. The higher the score, the closer the cultural connection between the two entities, and the higher the priority for them to be presented together in content generation. Representative Entity semantic embedding vector With entity semantic embedding vector The cosine similarity between the entities is used to extract entity semantic embedding vectors through a finely tuned large language model encoder in the tourism and culture domain, which is used to measure the degree of basic semantic association between the two entities. Representative Entity With entity The authoritative co-occurrence coefficient between the two entities ranges from 0 to 1. It is calculated based on the frequency of co-occurrence of the two entities in authoritative historical materials, official cultural and tourism data, and formal research literature. The higher the co-occurrence frequency, the closer the coefficient is to 1, which is used to ensure the authority of the association. Representative Entity With entity The normalized composite value of temporal and spatial distance between the entities ranges from 0 to 1. The temporal distance is the normalized result of the time difference between the historical periods of the two entities, and the spatial distance is the normalized result of the spatial distance between the two entities. The closer the distance, the smaller the value, and the higher the correlation score. Supplementary function: Based on the correlation score, this sub-device constructs a three-level correlation network for each entity: core correlation, secondary correlation, and extended correlation. A correlation score ≥ 0.7 is a core correlation, 0.3 ≤ correlation score < 0.7 is a secondary correlation, and correlation score < 0.3 is an extended correlation, providing a clear hierarchical basis for matching node materials in the subsequent narrative. Formula function: This formula is used to quantify the closeness of the correlation between two cultural entities. It calculates a standardized correlation score through multi-dimensional indicators, providing a precise quantitative basis for the hierarchical division of entity correlation and the linkage of entities in the narrative, ensuring the coherence of the textual context and logic.

[0034] Knowledge Graph Dynamic Update Sub-device: Purpose: Responsible for the dynamic updating of the knowledge graph and the management of entity credibility, ensuring the timeliness and authority of the knowledge base; Functional Expansion: This sub-device can connect in real time to the latest data released by authoritative cultural research institutions, archaeological release platforms, and official cultural and tourism channels to complete the dynamic update of entity attributes and relationships in the knowledge graph; at the same time, it sets authoritative source tags and credibility levels for each entity, classifying them into four levels according to the authority of the source: national authoritative source, local official source, formal academic source, and other compliant source, providing a basis for the authenticity verification of subsequent content generation.

[0035] The scenario-based requirement semantic parsing and generation constraint standardization definition module includes a requirement semantic deep parsing sub-device, a generation scenario adaptation sub-device, and a constraint standardization definition sub-device. The requirement semantic deep parsing sub-device, based on a large language model, extracts six core elements from the user's input natural language requirements: destination subject, core content theme, target audience, content carrier, narrative style, and content length, completing the structured transformation of these requirement elements. The generation scenario adaptation sub-device has a built-in rule library for adapting to all scenarios of cultural and tourism content, covering five core application scenarios: offline guided tours, short video dissemination, graphic science popularization, study tours, and immersive entertainment. It sets four exclusive generation rules for each scenario: content structure, narrative rhythm, audience adaptation, and compliance red lines, matching user requirements with scenario rules, locking in the target generation scenario, and retrieving the corresponding exclusive generation rules. The constraint standardization definition sub-device integrates structured requirement elements with the target scenario's exclusive generation rules to generate standardized content generation constraint parameters.

[0036] Specifically, the semantic deep analysis sub-device for user needs: Purpose: Responsible for extracting and structuring the core elements of user needs, converting unstructured natural language needs into structured elements that can be processed by the system; Functional expansion: Based on the named entity recognition and intent understanding capabilities of the large language model, this sub-device extracts six core elements from the user's input natural language needs: destination subject, core content theme, target audience, content carrier, narrative style, and content length, completing the structuring transformation of the needs elements, while marking missing core needs elements and triggering user confirmation to supplement them;

[0037] Scene Adaptation Sub-device: Purpose: To match content generation scene rules with user needs, providing scene-based constraints for subsequent content generation; Functionality: This sub-device has a built-in rule library for adapting cultural and tourism content across all scenes, covering five core application scenarios: offline guided tours, short video dissemination, graphic science popularization, study tours, and immersive entertainment. It sets four exclusive generation rules for each scenario: content structure, narrative rhythm, audience adaptation, and compliance red lines. It can accurately match user needs with scene rules, lock the target generation scene, and retrieve the corresponding exclusive generation rules, ensuring that the final generated content fully adapts to the dissemination and usage requirements of the target application scenario.

[0038] Standardized Constraint Definition Sub-device: Purpose: To integrate user needs elements and scenario rules to generate standardized content generation constraint parameters, defining clear boundaries and core requirements for subsequent content generation; Functional Expansion: This sub-device integrates the parsed structured needs elements with the target scenario's exclusive generation rules to generate standardized content generation constraint parameters, clarifying content length ranges, narrative perspectives, cultural priorities, content structure frameworks, and compliance requirements, and outputting a standardized set of constraint parameters as the core input for subsequent content generation.

[0039] The intelligent construction and core content native generation module of the cultural narrative chain includes a narrative mainline planning sub-device, a content node filling sub-device, and a core narrative generation sub-device. The narrative mainline planning sub-device, based on standardized content generation constraint parameters and the entity association relationships of the tourism culture knowledge graph, plans the main narrative line and core nodes of the content's structure, identifies the core cultural highlights, and achieves precise matching between the main line and user needs through a narrative mainline adaptation algorithm. The core calculation formula is: , in the formula: Represents the main narrative line The overall fit score ranges from 0 to 1; This represents the degree of fit between the narrative thread and the user-specified core theme, with a value ranging from 0 to 1. This represents the fit between the main narrative and the target generation scene, with a value ranging from 0 to 1. The textual logic coherence of the narrative thread is represented, with a value ranging from 0 to 1; , , These are the weighting coefficients of the three indicators, and their sum is 1. The narrative mainline planning sub-device determines the final narrative mainline based on the comprehensive adaptability score, clarifying the core positioning and content proportion of each node. The content node filling sub-device, based on the positioning and core requirements of the narrative mainline nodes, matches the cultural entities, allusions, details, and authoritative historical materials of the corresponding nodes from the tourism culture knowledge graph to complete the content material filling of each narrative node. The core narrative generation sub-device, based on the finely tuned large language model in the tourism culture field, uses standardized generation constraint parameters as boundaries, the determined narrative mainline as a framework, and the cultural materials filled in the nodes as the core to generate complete core cultural content text. It has a built-in cultural weight constraint mechanism to ensure that the proportion of cultural information in the generated content is not lower than the minimum threshold set by the target scene, and binds corresponding authoritative source tags to all core cultural expressions.

[0040] Specifically, the narrative thread planning sub-device: Purpose: To plan the narrative logic and node structure of the content, build the core framework for content generation, and ensure the logical coherence of the content's context, adapting to user needs and scenario requirements; Functional Expansion: Based on standardized content generation constraint parameters and the entity association relationships of the tourism culture knowledge graph, this sub-device plans the narrative thread and four core nodes (introduction, development, transition, and conclusion), identifies the core cultural highlights of the content, and achieves precise matching between the narrative thread and user needs through a narrative thread adaptability algorithm. The core calculation formula is: , in the formula: Represents the main narrative line The overall suitability score ranges from 0 to 1. The higher the score, the better the main storyline suits the user's needs. The main storyline with the highest score is selected as the core narrative storyline. Represents the degree of fit between the narrative thread and the user-specified core theme, with a value ranging from 0 to 1. It is calculated based on the sum of the correlation between the core cultural entities covered by the main thread and the user-specified theme. Represents the compatibility between the main narrative and the target generation scene, with a value ranging from 0 to 1. It is calculated based on the degree of matching between the narrative rhythm, node settings and scene rules of the main narrative. For example, short video scenes are adapted to fast-paced main narratives with strong climaxes, while offline guided tour scenes are adapted to main narratives that are gradual and spatially linked. This represents the logical coherence of the narrative thread, with a value ranging from 0 to 1. It is calculated based on the average entity correlation between the main thread nodes to ensure the logical coherence of the narrative thread without logical breaks. , , These are the weighting coefficients for the three indicators, with a sum of 1. The default values ​​are 0.4, 0.4, and 0.2, respectively, and can be dynamically adjusted according to user needs. Supplementary function: This sub-device determines the final narrative thread based on the comprehensive adaptability score, clarifying the core positioning, content proportion, and core entity coverage of the four major nodes (introduction, development, transition, and conclusion), thus building a complete narrative framework for subsequent content generation. Formula function: This formula is used to quantitatively evaluate the comprehensive adaptability of the planned narrative thread to user needs and target scenarios. A standardized adaptability score is calculated through multi-dimensional indicators, enabling quantitative screening of multiple narrative threads and ultimately determining the optimal narrative framework.

[0041] Content Node Filling Sub-device: Purpose: To match corresponding cultural materials to each node of the main narrative, serving as a fundamental support for the generation of the core narrative; Functional Expansion: Based on the node positioning and core requirements of the main narrative, this sub-device matches cultural entities, anecdotal details, authoritative historical materials, and scene-based materials from the tourism culture knowledge graph, prioritizing core relevance and authoritative sources, to complete the content material filling for each narrative node, while ensuring the interconnectivity of materials between nodes and avoiding content duplication;

[0042] Core Narrative Generation Sub-device: Purpose: Based on the narrative framework and node materials, it generates complete core cultural tourism content text, serving as the core execution unit for content generation in the entire solution; Functional Expansion: This sub-device, based on a finely tuned large language model in the tourism and culture field, uses standardized generation constraint parameters as boundaries, a defined narrative thread as the framework, and cultural materials filled in at nodes as the core, to generate core cultural content text that is logically coherent, fully expresses its cultural core, and meets the requirements of the scenario; it also incorporates a cultural weight constraint mechanism, with the constraint formula as follows: ,in The proportion of cultural information in the generated content. The length of the text containing the core cultural information. This represents the total length of the content text. The minimum threshold for the proportion of cultural information is set for the target scenario, and different thresholds are set for different scenarios, such as study tour scenarios. =0.7, Short video dissemination scenarios =0.4; In addition, during the generation process, this sub-device binds authoritative source tags from the knowledge graph to all core cultural expressions, providing a basis for subsequent authenticity verification.

[0043] The multi-format content adaptation and immersive expression optimization module includes a content format conversion sub-device, an immersive expression optimization sub-device, and a multi-modal supporting content generation sub-device. The content format conversion sub-device, based on a large language model, converts the core content text into a corresponding format of finished content according to the carrier requirements in the content generation constraint parameters. The immersive expression optimization sub-device optimizes the content's immersive expression across all dimensions for the target scene and content format. Optimization directions include adjusting the narrative perspective, supplementing scene-specific details, adjusting emotional rhythm, and embedding interactive content. It uses an immersion metric evaluation algorithm to score the immersion of the optimized content and iteratively optimizes the content based on the score until it reaches the target scene threshold requirements. The multi-modal supporting content generation sub-device generates corresponding multi-modal supporting production instructions simultaneously based on the generated final content.

[0044] Specifically, the content format conversion sub-device: Purpose: To convert core content text into a content format adapted to the target carrier, enabling the content to be implemented in multiple scenarios; Functional expansion: Based on the strong instruction compliance capability of the large language model, this sub-device converts core content text into corresponding content products according to the carrier requirements in the content generation constraint parameters, including scenic spot guide texts, short video narration scripts, WeChat official account article texts, folk story scripts, study tour teaching materials, immersive script murder plots, etc., strictly adhering to the format specifications and content structure requirements of the corresponding carrier;

[0045] Immersive Expression Optimization Sub-device: Purpose: To optimize the immersiveness and user engagement of content, enabling users to deeply receive the content; Functionality: This sub-device optimizes the immersive expression of content across all dimensions, targeting specific scenarios and content formats. Optimization directions include adjusting first-person and second-person narrative perspectives, supplementing the contextual details of destination spaces and sensory experiences, adjusting the emotional rhythm to suit the scenario, and integrating Q&A, check-in, and interactive guidance content. Simultaneously, through an immersion quantification and evaluation algorithm, the optimized content is scored for immersion. Based on the score, the content is iteratively optimized until the score reaches the threshold requirements of the target scenario, ensuring that the immersiveness of the content meets the scenario's needs.

[0046] Multimodal Content Generation Sub-device: Purpose: To generate supporting multimodal content production instructions to support the multi-format implementation of content; Functional Expansion: Based on the generated final content, this sub-device synchronously generates adapted multimodal production instructions, including short video storyboards, scene descriptions, audio and music suggestions, interactive element design, AR navigation trigger point settings, etc., providing standardized support instructions for video production, offline implementation, and multimodal presentation of content, realizing full-chain support from text content to multimodal finished products.

[0047] The cultural authenticity verification and compliance check module for cultural tourism content includes a cultural authenticity verification sub-device, a content compliance check sub-device, and an automatic error correction sub-device. The cultural authenticity verification sub-device performs a full cross-comparison of the generated content with authoritative information from the tourism culture knowledge graph, identifies and marks cultural factual errors and unsubstantiated fabricated content, matches corresponding authoritative correction criteria, and completes a quantitative assessment of the authenticity of the generated content through a cultural factual accuracy algorithm. The core calculation formula is: , in the formula: The accuracy of cultural facts represented by the generated content, with a value ranging from 0 to 1; The number of cultural factual errors and fabricated content without authoritative sources identified in the representative content; The total number of core cultural facts expressed in the representative content; the cultural authenticity verification sub-device sets a minimum threshold for the accuracy of cultural facts, and content that does not reach the threshold triggers an automatic correction process; the content compliance verification sub-device connects to the content compliance review rule library to complete the full-dimensional compliance verification of the generated content and marks the non-compliant content and compliance risk points; the error content automatic correction sub-device is based on a large language model and combines the verification and validation results to complete the automatic correction of the marked content, and after the correction is completed, the verification and validation process is re-executed;

[0048] Specifically, the cultural authenticity verification sub-device: Purpose: To verify the cultural authenticity and historical accuracy of the generated content, avoiding cultural factual errors and unsubstantiated fabrications; Functional Expansion: This sub-device performs a full cross-comparison of the generated content with authoritative entity information and historical sources in the tourism culture knowledge graph, identifying cultural factual errors, historical time discrepancies, misunderstandings of cultural lineage, and unsubstantiated fabrications. It accurately marks erroneous content and matches corresponding authoritative correction evidence; A cultural factual accuracy algorithm is used to quantitatively evaluate the authenticity of the generated content. The core calculation formula is: , in the formula: The accuracy of cultural facts represented by the generated content ranges from 0 to 1, with values ​​closer to 1 indicating higher authenticity. The number of cultural factual errors and fabricated content without authoritative sources identified in the representative content; The total number of core cultural fact statements contained in the content represents the total number of such statements. Supplementary function: This sub-device sets a minimum threshold for cultural fact accuracy, with a default threshold of 0.95. This means the content's cultural fact accuracy must reach 95% or higher; content that does not reach this threshold will trigger an automatic correction process. Formula function: This formula is used to quantitatively evaluate the cultural fact accuracy of the generated content. By comparing the number of erroneous contents with the total number of cultural statements, a standardized accuracy score is obtained, providing a quantitative basis for content quality control and ensuring the cultural authenticity of the generated content.

[0049] Content Compliance Verification Sub-device: Purpose: To verify the full-dimensional compliance of generated content and avoid ideological and content compliance risks; Functional Expansion: This sub-device connects to the content compliance review rule library of the Ministry of Culture and Tourism and the relevant requirements of the regulations on the governance of online content ecology, and completes compliance verification in four dimensions: historical perspective compliance, ethnic culture compliance, cultural relic protection compliance, and public order and good customs compliance. It accurately marks illegal content and compliance risk points, and clarifies the types of violations and rectification requirements.

[0050] Automatic Error Correction Sub-device: Purpose: To automatically correct marked errors and violations, ensuring the authenticity and compliance of the final output content; Functionality: Based on a large language model, this sub-device combines the results of cultural authenticity verification and compliance checks to automatically correct marked errors and violations. The correction process strictly follows the authoritative source information and compliance rules in the knowledge graph. After correction, the verification and validation process is re-executed until the accuracy and compliance of the content reach the threshold requirements, generating a final draft that meets the requirements.

[0051] The content effectiveness quantitative evaluation and generation model iterative optimization module includes a feedback data collection sub-device, an effectiveness index quantification sub-device, and a model fine-tuning optimization sub-device. The feedback data collection sub-device collects user feedback data and content dissemination data to form a complete feedback data set. The effectiveness index quantification sub-device constructs a content effectiveness quantitative evaluation system, completes multi-dimensional quantitative scoring of content effectiveness, selects high-scoring high-quality content samples, and extracts the core optimization directions from user feedback. The model fine-tuning optimization sub-device constructs a fine-tuning dataset based on high-quality content samples, and combines the optimization directions from user feedback to complete the continuous iterative optimization of the fine-tuned large language model in the tourism and culture field.

[0052] Specifically, the feedback data collection sub-device: Purpose: To collect user feedback data and content dissemination data across all dimensions of the generated content, providing a data source for model iteration; Functional expansion: This sub-device can collect users' manual ratings, modification suggestions, and usage scenario feedback on the generated content. At the same time, it connects to the content publishing platform to collect dissemination data such as the number of views, completion rate, likes, and collections, forming a complete set of feedback data.

[0053] Quantitative Evaluation Sub-device for Effectiveness Metrics: Purpose: To conduct multi-dimensional quantitative evaluation of the effectiveness of content usage, screen high-quality content samples, and extract optimization directions; Functional Expansion: This sub-device constructs a quantitative evaluation system for content effectiveness, and completes multi-dimensional quantitative scoring of content effectiveness from four core dimensions: cultural accuracy, scenario adaptability, audience acceptance, and dissemination effect. It screens high-scoring high-quality content samples and extracts core optimization directions based on user feedback.

[0054] Model Fine-tuning and Optimization Sub-device: Purpose: To complete the continuous iterative optimization of the large language model, forming a complete closed loop of generation, feedback, and optimization; Functional Expansion: Based on the selected high-scoring and high-quality content samples, this sub-device constructs a fine-tuning dataset for tourism and cultural content generation. Combining the optimization direction based on user feedback, it performs low-rank adaptation (LoRA) fine-tuning on the large language model for tourism and culture, continuously optimizing the model's narrative ability, scene adaptation ability, and the quality of cultural content generation.

[0055] A method for generating tourism and cultural content based on a large language model includes the following steps:

[0056] S1. In the tourism culture knowledge graph construction stage, we collect full-dimensional cultural entity data of tourist destinations through authoritative cultural tourism data channels, complete the standardized labeling of entity attributes, explore the multi-level relationship between entities and complete the quantitative labeling of the relationship, construct a tourism culture-specific knowledge graph, and connect with authoritative release channels to complete the dynamic update of the knowledge graph and the entity credibility classification.

[0057] S2, the requirement analysis and constraint definition stage, extracts the core elements of user requirements through the large language model and completes the structure transformation, matches the built-in full-scenario adaptation rule library to lock the target generation scenario, and integrates requirement elements and scenario rules to generate standardized content generation constraint parameters;

[0058] S3, Narrative Chain Construction and Core Content Generation Stage: Based on content generation constraint parameters and knowledge graph entity association relationships, the narrative main line and core nodes are planned, the narrative main line adaptation quantification evaluation is completed and the final narrative framework is determined, corresponding cultural materials are matched and filled for each node, and a complete core cultural content text is generated based on a finely tuned large language model in the field of tourism culture.

[0059] S4. Content format adaptation and immersive optimization stage: Convert the core content text into the finished content corresponding to the target carrier, optimize the immersive expression of the content for the target scenario, complete the cyclical optimization of the content through immersive quantification evaluation, and generate supporting multimodal production instructions in sync.

[0060] S5. In the content verification and compliance verification stage, the generated content will be cross-referenced with authoritative information in the knowledge graph to complete the quantitative assessment of the accuracy of cultural facts. Simultaneously, full-dimensional compliance verification will be performed, and the marked errors and violations will be automatically corrected and re-verified.

[0061] S6. In the effect evaluation and model iteration stage, user feedback data and content dissemination data are collected to complete the multi-dimensional quantitative evaluation of content effect, select high-quality content samples to build a fine-tuning dataset, and complete the continuous iterative optimization of the large language model in the tourism and culture field.

[0062] Specifically, in the S1 stage of constructing the tourism culture knowledge graph, we collect full-dimensional cultural entity data of tourist destinations through authoritative cultural tourism data channels, complete the standardized labeling of entity attributes, explore the multi-level relationships between entities and complete the quantitative labeling of relationships, construct a tourism culture-specific knowledge graph, and connect with authoritative release channels to complete the dynamic updating of the knowledge graph and the leveling of entity credibility.

[0063] S2, the requirement analysis and constraint definition stage, extracts the core elements of user requirements through the large language model and completes the structure transformation, matches the built-in full-scenario adaptation rule library to lock the target generation scenario, and integrates requirement elements and scenario rules to generate standardized content generation constraint parameters;

[0064] S3, Narrative Chain Construction and Core Content Generation Stage: Based on content generation constraint parameters and knowledge graph entity association relationships, the narrative main line and core nodes are planned, the narrative main line adaptation quantification evaluation is completed and the final narrative framework is determined, corresponding cultural materials are matched and filled for each node, and a complete core cultural content text is generated based on a finely tuned large language model in the field of tourism culture.

[0065] S4. Content format adaptation and immersive optimization stage: Convert the core content text into the finished content corresponding to the target carrier, optimize the immersive expression of the content for the target scenario, complete the cyclical optimization of the content through immersive quantification evaluation, and generate supporting multimodal production instructions in sync.

[0066] S5. In the content verification and compliance verification stage, the generated content will be cross-referenced with authoritative information in the knowledge graph to complete the quantitative assessment of the accuracy of cultural facts. Simultaneously, full-dimensional compliance verification will be performed, and the marked errors and violations will be automatically corrected and re-verified.

[0067] S6. In the effect evaluation and model iteration stage, user feedback data and content dissemination data are collected to complete the multi-dimensional quantitative evaluation of content effect, select high-quality content samples to build a fine-tuning dataset, and complete the continuous iterative optimization of the large language model in the tourism and culture field.

[0068] The method of use and working principle of this invention are as follows:

[0069] Usage: The initial deployment and initial configuration of the system must be completed first, including the basic construction of the tourism culture knowledge graph and the pre-setting of a rule base for full-scenario adaptation. Users can submit relevant requests for tourism culture content generation through the corresponding system entry. After receiving the request, the system first extracts and structures the core elements of the request, matches the exclusive generation rules for the corresponding application scenario, and generates standardized content generation constraint parameters. Simultaneously, it links with the already constructed tourism culture knowledge graph to complete the matching of cultural entities and the preparation of content materials. Subsequently, based on a finely tuned large language model in the tourism culture field, it sequentially completes the narrative planning, core content generation, content form adaptation, and immersive expression optimization. The generated content undergoes full cultural authenticity verification and full-dimensional compliance verification. Content that does not meet the requirements is automatically corrected before the final draft is output to the user. The system simultaneously collects user feedback data and usage effect data on the generated content, and selects high-quality content samples to continuously iterate and optimize the content generation model.

[0070] Working Principle: Based on a knowledge graph specific to the tourism and culture sector, and using a large language model fine-tuned for tourism and culture scenarios as the core execution vehicle, the system constructs a closed-loop operational logic for the scenario-based, precise, and compliant generation of tourism and culture content. Through comprehensive data collection of tourism and culture entities and multi-level relational mining, it builds a standardized cultural knowledge system with complete contextual logic. Through deep semantic analysis of user needs and scenario-based rule adaptation, it achieves precise matching between generated content and target application scenarios and actual user needs. Through the narrative logic construction and native content generation capabilities of the large language model, it completes the intelligent creation of cultural and tourism content. Through cross-verification of cultural authenticity and a multi-dimensional compliance verification mechanism, it ensures the cultural accuracy and compliance of the generated content. Finally, through user feedback collection and model iteration optimization mechanisms, it continuously improves content generation capabilities, completing intelligent support for the entire process from demand input to content implementation.

[0071] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any way. Those skilled in the art can readily implement the present invention based on the description and drawings above. However, any modifications, alterations, and variations made by those skilled in the art without departing from the scope of the present invention using the disclosed technical content are equivalent embodiments of the present invention. Furthermore, any modifications, alterations, and variations made to the above embodiments based on the essential technology of the present invention are still within the protection scope of the present invention.

Claims

1. A tourism and cultural content generation device based on a large language model, characterized in that, The device includes a module for the full-domain construction and dynamic updating of a tourism culture knowledge graph, a module for the semantic parsing of scenario-based needs and the standardized definition of generation constraints, a module for the intelligent construction of cultural narrative chains and the native generation of core content, a module for the adaptation of multi-form content and the optimization of immersive expression, a module for the verification of the cultural authenticity and compliance of cultural tourism content, and a module for the quantitative evaluation of content effects and the iterative optimization of the generation model.

2. The tourism and cultural content generation device based on a large language model according to claim 1, characterized in that, The tourism culture knowledge graph full-domain construction and dynamic update module includes a cultural entity full-domain collection sub-device, a cultural relationship deep mining sub-device, and a knowledge graph dynamic update sub-device. The cultural entity full-domain collection sub-device connects to authoritative cultural tourism data channels to collect multi-dimensional cultural entity data corresponding to tourist destinations, defining seven core entity types: historical figures, intangible cultural heritage skills, folk tales, regional cultural context, historical events, cultural relics and historical sites, and works of art, and completing standardized labeling of the basic attributes of each entity. The cultural relationship deep mining sub-device, based on a finely tuned large language model in the tourism culture field, mines four core relationships among cultural entities: causal relationships, temporal relationships, spatial relationships, and cultural heritage relationships. It completes the accuracy calibration of relationships through an entity relationship quantification algorithm; the core calculation formula is: , in the formula: .

3. The tourism and cultural content generation device based on a large language model according to claim 1, characterized in that, The scenario-based requirement semantic parsing and generation constraint standardization definition module includes a requirement semantic deep parsing sub-device, a generation scenario adaptation sub-device, and a constraint standardization definition sub-device. The requirement semantic deep parsing sub-device, based on a large language model, extracts six core elements from the user's input natural language requirements: destination subject, core content theme, target audience, content carrier, narrative style, and content length, completing the structured transformation of these requirement elements. The generation scenario adaptation sub-device has a built-in full-scenario adaptation rule library for cultural and tourism content, covering five core application scenarios: offline guided tours, short video dissemination, graphic science popularization, study tours, and immersive entertainment. It sets four exclusive generation rules for each scenario: content structure, narrative rhythm, audience adaptation, and compliance red lines, matching user requirements with scenario rules, locking in the target generation scenario, and retrieving the corresponding exclusive generation rules. The constraint standardization definition sub-device integrates structured requirement elements with the target scenario's exclusive generation rules to generate standardized content generation constraint parameters.

4. The tourism and cultural content generation device based on a large language model according to claim 1, characterized in that, The intelligent construction and core content native generation module of the cultural narrative chain includes a narrative mainline planning sub-device, a content node filling sub-device, and a core narrative generation sub-device. The narrative mainline planning sub-device, based on standardized content generation constraint parameters and the entity association relationship of the tourism culture knowledge graph, plans the main narrative line and core nodes of the content's structure, identifies the core cultural highlights, and achieves precise matching between the main line and user needs through a narrative mainline adaptation algorithm. The core calculation formula is: , in the formula: Represents the main narrative line The overall fit score ranges from 0 to 1; This represents the degree of fit between the narrative thread and the user-specified core theme, with a value ranging from 0 to 1. This represents the fit between the main narrative and the target generation scene, with a value ranging from 0 to 1. The textual logic coherence of the narrative thread is represented, with a value ranging from 0 to 1; , , These are the weighting coefficients of the three indicators, and their sum is 1. The narrative mainline planning sub-device determines the final narrative mainline based on the comprehensive adaptability score, clarifying the core positioning and content proportion of each node. The content node filling sub-device, based on the positioning and core requirements of the narrative mainline nodes, matches the cultural entities, allusions, details, and authoritative historical materials of the corresponding nodes from the tourism culture knowledge graph to complete the content material filling of each narrative node. The core narrative generation sub-device, based on a finely tuned large language model in the tourism culture field, uses standardized generation constraint parameters as boundaries, the determined narrative mainline as a framework, and the cultural materials filled in the nodes as the core to generate complete core cultural content text. It has a built-in cultural weight constraint mechanism to ensure that the proportion of cultural information in the generated content is not lower than the minimum threshold set by the target scene, and binds corresponding authoritative source tags to all core cultural expressions.

5. The tourism and cultural content generation device based on a large language model according to claim 1, characterized in that, The multi-format content adaptation and immersive expression optimization module includes a content format conversion sub-device, an immersive expression optimization sub-device, and a multi-modal supporting content generation sub-device. The content format conversion sub-device, based on a large language model, converts the core content text into a corresponding format of finished content according to the carrier requirements in the content generation constraint parameters. The immersive expression optimization sub-device optimizes the content's immersive expression across all dimensions for the target scene and content format. The optimization directions include adjusting the narrative perspective, supplementing scene-specific details, adjusting emotional rhythm, and embedding interactive content. An immersion metric evaluation algorithm is used to score the immersion of the optimized content, and the content is iteratively optimized based on the score until the target scene threshold requirements are met. The multi-modal supporting content generation sub-device generates corresponding multi-modal supporting production instructions simultaneously based on the generated final content.

6. The tourism and cultural content generation device based on a large language model according to claim 1, characterized in that, The cultural authenticity verification and compliance check module for cultural tourism content includes a cultural authenticity verification sub-device, a content compliance check sub-device, and an automatic error correction sub-device. The cultural authenticity verification sub-device performs a full cross-comparison of the generated content with authoritative information from the tourism culture knowledge graph, identifies and marks cultural factual errors and unsubstantiated fabricated content, matches corresponding authoritative correction criteria, and completes a quantitative assessment of the authenticity of the generated content through a cultural factual accuracy algorithm. The core calculation formula is: , in the formula: The accuracy of cultural facts represented by the generated content, with a value ranging from 0 to 1; The number of cultural factual errors and fabricated content without authoritative sources identified in the representative content; The total number of core cultural fact statements contained in the representative content; the cultural authenticity verification sub-device sets a minimum threshold for the accuracy of cultural facts, and content that does not reach the threshold triggers an automatic correction process; The content compliance verification sub-device connects to the content compliance review rule base to complete the full-dimensional compliance verification of the generated content and mark the non-compliant content and compliance risk points; the automatic error correction sub-device is based on a large language model and combines the verification and validation results to automatically correct the marked content, and after the correction is completed, the verification and validation process is re-executed.

7. The tourism and cultural content generation device based on a large language model according to claim 1, characterized in that, The content effect quantitative evaluation and generation model iteration optimization module includes a feedback data acquisition sub-device, an effect indicator quantification sub-device, and a model fine-tuning optimization sub-device. The feedback data acquisition sub-device collects user feedback data and content dissemination data to form a complete feedback data set; the effect index quantification sub-device constructs a content effect quantification evaluation system, completes multi-dimensional quantitative scoring of content effect, selects high-scoring high-quality content samples, and extracts the core optimization directions from user feedback; the model fine-tuning optimization sub-device constructs a fine-tuning dataset based on high-quality content samples, and combines the optimization directions from user feedback to complete the continuous iterative optimization of the fine-tuned large language model in the tourism and culture field.

8. A method for generating tourism and cultural content based on a large language model, applied to the tourism and cultural content generation apparatus based on a large language model as described in any one of claims 1-7, characterized in that, Includes the following steps: S1. In the tourism culture knowledge graph construction stage, we collect full-dimensional cultural entity data of tourist destinations through authoritative cultural tourism data channels, complete the standardized labeling of entity attributes, explore the multi-level relationship between entities and complete the quantitative labeling of the relationship, construct a tourism culture-specific knowledge graph, and connect with authoritative release channels to complete the dynamic update of the knowledge graph and the entity credibility classification. S2, the requirement analysis and constraint definition stage, extracts the core elements of user requirements through the large language model and completes the structure transformation, matches the built-in full-scenario adaptation rule library to lock the target generation scenario, and integrates requirement elements and scenario rules to generate standardized content generation constraint parameters; S3, Narrative Chain Construction and Core Content Generation Stage: Based on content generation constraint parameters and knowledge graph entity association relationships, the narrative main line and core nodes are planned, the narrative main line adaptation quantification evaluation is completed and the final narrative framework is determined, corresponding cultural materials are matched and filled for each node, and a complete core cultural content text is generated based on a finely tuned large language model in the field of tourism culture. S4. Content format adaptation and immersive optimization stage: Convert the core content text into the finished content corresponding to the target carrier, optimize the immersive expression of the content for the target scenario, complete the cyclical optimization of the content through immersive quantification evaluation, and generate supporting multimodal production instructions in sync. S5. In the content verification and compliance verification stage, the generated content will be cross-referenced with authoritative information in the knowledge graph to complete the quantitative assessment of the accuracy of cultural facts. Simultaneously, full-dimensional compliance verification will be performed, and the marked errors and violations will be automatically corrected and re-verified. S6. In the effect evaluation and model iteration stage, user feedback data and content dissemination data are collected to complete the multi-dimensional quantitative evaluation of content effect, select high-quality content samples to build a fine-tuning dataset, and complete the continuous iterative optimization of the large language model in the tourism and culture field.

Images