Cultural element intelligent generation method and system based on large model, and medium
By building a regional database and multi-role intelligent cluster collaboration, the problem of insufficient regional knowledge in large models has been solved, and the precise extraction and diversified generation of cultural elements have been achieved, ensuring the scientific nature and compliance of the generated content.
Patent Information
- Application Number
- CN202510803111.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-16
- Publication Date
- 2025-09-12
AI Technical Summary
Existing large models lack regional knowledge during the pre-training stage, resulting in insufficient ability to generate culture in marginal areas, and difficulty in comprehensively considering multiple aspects of information, providing false information and understanding differences.
Build a regional database, collect historical and cultural data of the target area, establish a knowledge graph of cultural taboos, generate cultural exhibition planning plans through multi-role intelligent cluster collaboration, and combine text embedding models and large language models for multiple iterative optimizations to generate cultural product naming plans.
It improves the accuracy of cultural element extraction and the scientific nature of generated content, ensures the legality, compliance and diversified perspectives of cultural elements, and enhances the creative inspiration of cultural exhibition planning and the accuracy of cultural product naming.
Smart Images
Figure CN120632076A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of large model technology, and in particular to a method, system and medium for intelligently generating cultural elements based on large models. Background Art
[0002] Culture has both explicit and implicit characteristics. Explicit elements (such as architecture and clothing) are easy to extract, but implicit elements (such as values, customs, and myths and legends) are more difficult to extract directly and require in-depth analysis. Currently, the pre-training data for mainstream large-scale models mostly comes from open data on the internet, resulting in uneven coverage of regional knowledge. For example, data on traditional clothing patterns and oral epics of certain ethnic minorities in my country is lacking. The internet data that large-scale models rely on inherently exhibits a tendency toward "digital colonization," where cultural output from economically privileged regions dominates while cultures from marginalized regions are diluted or even obliterated. For example, the global recognition of India's Diwali festival far exceeds that of the Torch Festival in southwestern China, resulting in insufficient model ability to capture the latter's details. This lack of regional knowledge in the pre-training process of large-scale models can lead to hallucinations, where answers are often nonsensical and misleading. A single large-scale model cannot comprehensively consider multiple aspects of information, making it impossible to directly arrive at optimal content in a single step. Relying solely on the inherent knowledge of the agent cannot guarantee scientific evaluation. Due to differences in architecture and training data, different large-scale models may also have different understandings of the same target task. Summary of the Invention
[0003] The purpose of the present invention is to overcome the deficiencies of the prior art and to provide a method, system and medium for intelligently generating cultural elements based on a large model.
[0004] The object of the present invention is achieved through the following technical solutions: In a first aspect, the present invention provides: a method for intelligently generating cultural elements based on a large model, comprising the following steps: In the cultural element extraction stage, historical and cultural data of the target area are collected, pre-processed, a regional database is constructed, prompt words are generated, and a cultural taboo knowledge graph is established to optimize the extraction results and obtain cultural elements; During the cultural exhibition planning scheme generation phase, historical and cultural exhibition planning schemes are collected, a planning scheme database is established, and a cultural exhibition planning scheme is generated by combining prompt words and refined cultural elements. A cultural exhibition planning agent and a multi-role agent cluster are constructed to iterate the cultural exhibition planning scheme multiple times, thereby updating the planning scheme database. During the cultural product naming stage, users input cultural product design requirements, and the most relevant scheme is matched from the planning scheme database based on the cultural product design requirements, and finally a cultural product naming scheme is generated.
[0005] Preferably, the cultural element extraction stage further includes the following steps: Delete erroneous data, delete irrelevant data, and complete missing data from historical and cultural data, and then convert them into document data; Divide the document data into multiple data blocks based on sentences, contextual semantics, and punctuation marks. Use a pre-trained text embedding model to vectorize each data block, build an index based on the Faiss database, and store the vectorized data blocks offline to form a regional database. When a user inputs elements to refine requirements and target area-related information, a text embedding model is used to convert the user input text into an embedding vector. A first preset number of relevant data blocks are then retrieved from a regional database. The user input is then concatenated with each retrieved relevant data block. The BGE-Reranker model is used to interactively analyze the correlation between the user input and the relevant data blocks, thereby reordering the relevant data blocks and selecting a second preset number of relevant data blocks as the context for the prompt word to generate the prompt word. The prompt words are used to refine the cultural elements of the target area in conjunction with the context.
[0006] Preferably, the cultural taboo knowledge graph is established by the following steps: Collect legal and regulatory data, establish a legal and regulatory database, and collect cultural taboo data from various regions to form a cultural taboo knowledge map; During the extraction and optimization process, the retrieval enhancement generation technology RAG is used to allow the large language model to determine whether the extracted cultural elements are legal and compliant; then, based on the regional labels entered by the user, the cultural taboo knowledge graph is retrieved, and based on heuristic rules, it is determined whether the extracted cultural elements violate local customs; at the same time, real-time popular data from social media is collected to guide the large language model to determine whether the extracted cultural elements are in line with current hot topics.
[0007] Preferably, the cultural exhibition planning scheme generation stage further includes the following steps: Use the pre-trained text embedding model to vectorize the extracted cultural elements; Retrieving a third preset number of historical and cultural exhibition planning schemes from the planning scheme database based on the quantized cultural elements, concatenating the cultural elements with the retrieved historical and cultural exhibition planning schemes, and then using the BGE-Reranker model to determine relevance, and retaining a fourth preset number of relevant schemes; Based on the prompt words, a large language model is used to understand and optimize relevant plans from multiple dimensions of the target area to obtain a cultural exhibition planning plan.
[0008] Preferably, the cultural product naming stage further includes the following steps: Based on the prompt words and taking cultural products as the benchmark, a large language model is used to split the most relevant solutions. After splitting, a pre-trained text embedding model is used to vectorize and cluster them to generate a cluster of cultural product naming solutions. The most relevant scheme is decomposed into sub-themes. After each sub-theme is combined with each naming scheme in the cultural product naming scheme cluster, the correlation between them is determined using the BGE-Reranker model, and the fifth preset number of highly correlated naming schemes is selected. Based on the prompt words, a large language model is used to analyze the common design mechanisms between sub-themes and highly relevant naming schemes, and corresponding cultural product naming schemes are generated for the sub-themes.
[0009] Preferably, the process of constructing a cultural exhibition planning agent and a multi-role agent cluster to iterate the cultural exhibition planning scheme multiple times includes the following steps: Initialize the cultural exhibition planning agent. Based on the target area's cultural elements and planning scheme database, design role prompts to define its planning functions. Integrate the search enhancement generation module to call the planning scheme database to optimize scheme generation, and set up an introspection mechanism to verify the feasibility of the draft. Build a multi-role intelligent agent cluster, including three types of intelligent agents: cultural compliance experts, art design consultants, and market analysts. Each intelligent agent uses unique prompt words to define evaluation dimensions, and review the compliance, artistic quality, and market appeal of the plan respectively; An iterative optimization mechanism is established. After the cultural exhibition planning agent generates a draft, the multi-role agent cluster reviews it in parallel and provides feedback on structured defect reports. The cultural planning agent performs targeted optimization based on the structured defect reports and resubmits them, repeating the cycle until the preset number of iterations or scoring threshold is met.
[0010] Preferably, the cultural exhibition is a lantern exhibition, and the cultural products are lantern sets.
[0011] Preferably, the historical and cultural data include local chronicles, intangible cultural heritage archives and oral accounts of ethnic minorities.
[0012] The second aspect of the present invention provides: a system for intelligently generating cultural elements based on a large model, for implementing any of the above-mentioned methods for intelligently generating cultural elements based on a large model, comprising: The cultural element extraction module is used to collect historical and cultural data of the target area, pre-process the historical and cultural data, build a regional database, generate prompt words, and establish a cultural taboo knowledge graph to optimize the extraction results and obtain cultural elements; The cultural exhibition planning scheme generation module is used to collect historical and cultural exhibition planning schemes, establish a planning scheme database, combine prompt words and refined cultural elements to generate cultural exhibition planning schemes, and build a cultural exhibition planning agent and a multi-role agent cluster to iterate the cultural exhibition planning schemes multiple times, thereby updating the planning scheme database; The cultural product naming module is used for users to input cultural product design requirements, match the most relevant scheme from the planning scheme database based on the cultural product design requirements, and finally generate a cultural product naming scheme.
[0013] The third aspect of the present invention provides: a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are loaded and executed by a processor, any of the above-mentioned methods for intelligently generating cultural elements based on large models is implemented.
[0014] The beneficial effects of the present invention are: 1) By building a customized regional knowledge base, the efficiency of cultural element extraction is improved, overcoming the regional knowledge shortcomings of large language models in the pre-training stage, ensuring more accurate generation of content and extraction of cultural elements.
[0015] 2) Through the collaboration of multiple intelligent agents, and the integration of diverse perspectives and creative inspiration, cultural exhibition planning plans can be intelligently generated.
[0016] 3) Relying on external knowledge guidance in the naming of cultural products, through precise matching and semantic reasoning, it is possible to avoid inaccurate and unreasonable naming caused by the inherent knowledge limitations of the model, and significantly improve the overall usability and attractiveness. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Figure 1 is a flow chart of the method of the lamp of the present invention; Figure 2 4 is a flow chart of the overall method in the lighting display embodiment of the present invention. DETAILED DESCRIPTION
[0018] The following will clearly and completely describe the technical solutions of the present invention in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work shall fall within the scope of protection of the present invention.
[0019] See Figure 1 The first aspect of the present invention provides: a method for intelligently generating cultural elements based on a large model, comprising the following steps: In the cultural element extraction stage, historical and cultural data of the target area are collected, pre-processed, a regional database is constructed, prompt words are generated, and a cultural taboo knowledge graph is established to optimize the extraction results and obtain cultural elements; During the cultural exhibition planning scheme generation phase, historical and cultural exhibition planning schemes are collected, a planning scheme database is established, and a cultural exhibition planning scheme is generated by combining prompt words and refined cultural elements. A cultural exhibition planning agent and a multi-role agent cluster are constructed to iterate the cultural exhibition planning scheme multiple times, thereby updating the planning scheme database. During the cultural product naming stage, users input cultural product design requirements, and the most relevant scheme is matched from the planning scheme database based on the cultural product design requirements, and finally a cultural product naming scheme is generated.
[0020] In this embodiment, the cultural exhibition is a lantern show, and the cultural products are lantern sets. Retrieval-enhanced generation technology is used to construct a customized database using regional unstructured document content. During the database construction phase, data granularity and index structure are optimized. During the database retrieval phase, a content ranking scheme generation method and a sliding window strategy are used to re-rank the relevance of the searched content, improving the accuracy and reliability of the contextual responses during the agent's question-and-answer process. A regional folklore expert agent is designed to accept basic design requirements input and extract traditional cultural knowledge related to the region, such as history, folklore, and beliefs, based on contextual searches of the customized database.
[0021] Through the domain cultural expert intelligent agent, cultural elements are further refined from regional traditional cultural knowledge, and the generated content is iterated multiple times based on compliance, usability, uniqueness, and market recognition within the introspection framework to ensure that the cultural elements do not violate taboos in content, are expressible in form, and are novel in conception; finally, the lantern design expert intelligent agent conducts a correlation analysis between current hot data and cultural elements to discover potential hot elements and ensure the organic combination of traditional cultural elements and modern culture.
[0022] In some embodiments, the cultural element extraction stage further includes the following steps: Delete erroneous data, delete irrelevant data, and complete missing data from historical and cultural data, and then convert them into document data; Divide the document data into multiple data blocks based on sentences, contextual semantics, and punctuation marks. Use a pre-trained text embedding model to vectorize each data block, build an index based on the Faiss database, and store the vectorized data blocks offline to form a regional database. When a user inputs elements to refine requirements and target area-related information, a text embedding model is used to convert the user input text into an embedding vector. A first preset number of relevant data blocks are then retrieved from a regional database. The user input is then concatenated with each retrieved relevant data block. The BGE-Reranker model is used to interactively analyze the correlation between the user input and the relevant data blocks, thereby reordering the relevant data blocks and selecting a second preset number of relevant data blocks as the context for the prompt word to generate the prompt word. The prompt words are used to refine the cultural elements of the target area in conjunction with the context.
[0023] In this embodiment, historical and cultural data, including unstructured data such as local chronicles, intangible cultural heritage archives, and oral histories of ethnic minorities, is used to construct a regionally customized database. The collected historical and cultural data for the target region undergoes preprocessing, including deleting erroneous and irrelevant data and supplementing missing data. This data is then converted into a unified document format. The document data is then segmented into sentences, using punctuation marks such as line breaks, periods, question marks, and exclamation points as the basis for contextual semantics. Long documents are then divided into multiple small data chunks (chunks), with the sentences within these chunks logically connected.
[0024] Then, each chunk is vectorized using the pre-trained text embedding model BGE, and an index is built based on Faiss. The vectorized chunks are stored offline to form a regionally customized database. When the user enters design requirements and relevant information about the target area, the BGE model is first used to convert the user's input text into an embedding vector. Then, the top-K relevant chunks are retrieved from the regionally customized database (where K is user-defined). Next, the user input is concatenated with each retrieved chunk, and the BGE-Reranker model is used to interactively analyze the correlation between the user input and the chunk, accurately calculating the correlation between the two. This allows the K chunks to be reordered, and the top-N chunks are selected from them as the context of the prompt word.
[0025] Prompt words give the Large Language Model (LLM) the role of a domain cultural expert, extracting cultural elements from regional traditional cultural knowledge. To ensure the accuracy and usability of the extracted cultural elements, a self-reflection framework is designed to constrain the content generated by the LLM.
[0026] In some embodiments, the cultural taboo knowledge graph is established by the following steps: Collect legal and regulatory data, establish a legal and regulatory database, and collect cultural taboo data from various regions to form a cultural taboo knowledge map; During the extraction and optimization process, the retrieval enhancement generation technology RAG is used to allow the large language model to determine whether the extracted cultural elements are legal and compliant; then, based on the regional labels entered by the user, the cultural taboo knowledge graph is retrieved, and based on heuristic rules, it is determined whether the extracted cultural elements violate local customs; at the same time, real-time popular data from social media is collected to guide the large language model to determine whether the extracted cultural elements are in line with current hot topics.
[0027] In this embodiment, legal and regulatory data is collected to establish a legal and regulatory database; data on cultural taboos in various regions is collected to form a cultural taboo knowledge graph. First, using search-augmented generation (RAG) technology, the LLM determines whether the extracted cultural elements are legal and compliant. Next, based on the regional tags in the user's input, the cultural taboo knowledge graph is searched and, using heuristic rules, the generated cultural elements are judged to determine whether they violate local customs. Simultaneously, real-time trending data from social media platforms such as WeChat official accounts, Xiaohongshu, Douyin, Kuaishou, and Weibo is collected. Based on prompts, the LLM is guided to determine whether the extracted cultural elements align with current trends. Based on the analysis of the compliance, usability, uniqueness, and market acceptance of the extracted cultural elements, the LLM, a domain cultural expert, is guided to extract cultural elements through multiple iterations (the number of iterations is user-defined) to ensure that the cultural elements are content-appropriate, expressive, and novel, effectively integrating traditional cultural elements with modern culture.
[0028] In some embodiments, the cultural exhibition planning scheme generation stage further includes the following steps: Use the pre-trained text embedding model to vectorize the extracted cultural elements; Retrieving a third preset number of historical and cultural exhibition planning schemes from the planning scheme database based on the quantized cultural elements, concatenating the cultural elements with the retrieved historical and cultural exhibition planning schemes, and then using the BGE-Reranker model to determine relevance, and retaining a fourth preset number of relevant schemes; Based on the prompt words, a large language model is used to understand and optimize relevant plans from multiple dimensions of the target area to obtain a cultural exhibition planning plan.
[0029] In this example, outstanding historical lantern exhibition theme design proposals are collected and uniformly converted into document form. The document-based design proposals are divided into multiple chunks based on sentence-by-sentence and punctuation marks such as line breaks, periods, question marks, and exclamation marks, according to contextual semantics. Each chunk is vectorized using the pre-trained text embedding model BGE, and an index is constructed based on Faiss. The vectorized chunks are then stored offline to form a database of outstanding historical lantern exhibition theme design proposals.
[0030] The creative design process for the lantern show's core theme is divided into two key steps: generation and decision-making. In the generation phase, the extracted lantern show cultural elements are vectorized using the BGE model. The top-K outstanding lantern show theme design proposals are then retrieved from a database of historically outstanding lantern show theme design proposals, where K is user-defined. Next, the lantern show cultural elements and the retrieved K design proposals are concatenated and fed into the BGE-Reranker model to determine the correlation between the cultural elements and the design proposals. The top-N design proposals, where N is user-defined, are retained. Based on prompts, the LLM interprets the retrieved N design proposals from the perspectives of the target region's cultural elements, location information, and the integration of modern aesthetics and traditional culture. Furthermore, the LLM combines the generated cultural elements to generate a lantern show core theme that is original, artistic, and infused with local characteristics.
[0031] During the decision-making phase, a Light Show Planning Agent and a review team composed of a cluster of multi-role Agents were constructed to iterate through a "generate -> evaluate -> regenerate" process (the number of iterations is user-defined) to improve the Light Show's overall planning scheme. The specific process is as follows: First, the Light Show Planning Agent is initialized and assigned the "Light Show Chief Planner" role using a structured prompt word template. This template explicitly requires that, based on the input list of cultural elements and the ID of a historical proposal library, a draft be output, including a theme design (within 20 words), a zoning plan, and a topological diagram of the light group layout. Constraints are set, such as an innovation self-score of ≥ 8 / 10. A RAG module is integrated to search the planning proposal database in real time for additional context. Next, we built a multi-role intelligent agent cluster with three professional roles to conduct the review: cultural compliance experts used APIs to connect to the cultural taboo knowledge graph to detect sensitive elements; art design consultants compared the generated solution with the historical top 5 solutions and output a JSON-formatted artistic quantitative report; and market analysts used Weibo's hot search API to analyze and generate a hot topic relevance report. Finally, a closed-loop iterative mechanism is established, in which a multi-role intelligent agent cluster generates a combined evaluation report, and the light show planning intelligent agent performs targeted optimization based on the reported defects. This process is executed repeatedly until the artistic scores are ≥8 points and the market rating reaches "medium" or above, or the upper limit of 5 iterations is reached, thereby achieving a systematic improvement in the quality of the plan.
[0032] The evaluation process employed a profiling approach to assess the core themes and planning proposals for the Lantern Festival. Specifically, prompts were designed to assess the originality, appeal, and distinctiveness of the core themes, based on factors such as similarity to previous cultural events, alignment with current hot topics, and inclusion of local content. For the overall planning proposal, prompts were used to instruct the LLM to consider factors such as the overall plan's coherence and alignment with the theme, ultimately leading to an assessment of the rationality and completeness of the planning proposal.
[0033] In some embodiments, the cultural product naming stage further includes the following steps: Based on the prompt words and taking cultural products as the benchmark, a large language model is used to split the most relevant solutions. After splitting, a pre-trained text embedding model is used to vectorize and cluster them to generate a cluster of cultural product naming solutions. The most relevant scheme is decomposed into sub-themes. After each sub-theme is combined with each naming scheme in the cultural product naming scheme cluster, the correlation between them is determined using the BGE-Reranker model, and the fifth preset number of highly correlated naming schemes is selected. Based on the prompt words, a large language model is used to analyze the common design mechanisms between sub-themes and highly relevant naming schemes, and corresponding cultural product naming schemes are generated for the sub-themes.
[0034] In this example, the user inputs a lighting group design requirement (measured in lighting group areas), which is vectorized using the BGE model. The top-K historical lighting group design proposals with a high degree of similarity to the current lighting group design requirement are retrieved from the database of historically outstanding lighting show theme design proposals constructed in the previous phase. The lighting group design requirement and these K design proposals are then concatenated and fed into the BGE-Reranker model to determine their correlation and select the top-N design proposals with the highest correlation. Based on the prompt word, the LLM is guided to split the N retrieved design proposals based on the lighting group area. All split lighting group area design proposals are vectorized and clustered using the BGE model to generate a cluster of historical lighting group area naming proposals. The light show planning scheme given by the light show theme creativity and planning scheme generation module is decomposed into sub-themes. After splicing each sub-theme and each naming scheme in each naming scheme cluster, the BGE-Reranker model is used to judge the correlation between them, and M highly correlated naming schemes are selected. Hint words are designed to let the LLM analyze the commonality of the design mechanism between the sub-themes and the highly correlated naming schemes, and generate corresponding light group naming schemes for the sub-themes.
[0035] A quality assessment framework for lamp group naming schemes based on model cluster analysis is constructed. Through indicators such as cultural element fit, design requirement completion, and current hot topic relevance, design prompts guide the large model cluster to iteratively evaluate and optimize the accuracy, rationality, and attractiveness of the lamp group naming scheme based on the original design requirements and relevant historical and cultural data to ensure the quality of the lamp group naming scheme.
[0036] In some embodiments, the process of constructing a cultural exhibition planning agent and a multi-role agent cluster to iterate the cultural exhibition planning plan multiple times includes the following steps: Initialize the cultural exhibition planning agent. Based on the target area's cultural elements and planning scheme database, design role prompts to define its planning functions. Integrate the search enhancement generation module to call the planning scheme database to optimize scheme generation, and set up an introspection mechanism to verify the feasibility of the draft. Build a multi-role intelligent agent cluster, including three types of intelligent agents: cultural compliance experts, art design consultants, and market analysts. Each intelligent agent uses unique prompt words to define evaluation dimensions, and review the compliance, artistic quality, and market appeal of the plan respectively; An iterative optimization mechanism is established. After the cultural exhibition planning agent generates a draft, the multi-role agent cluster reviews it in parallel and provides feedback on structured defect reports. The cultural planning agent performs targeted optimization based on the structured defect reports and resubmits them, repeating the cycle until the preset number of iterations or scoring threshold is met.
[0037] In some embodiments, the cultural exhibition is a lantern exhibition, and the cultural products are lantern sets.
[0038] In some embodiments, the historical and cultural data include local chronicles, intangible cultural heritage archives, and oral accounts of ethnic minorities.
[0039] The second aspect of the present invention provides: a system for intelligently generating cultural elements based on a large model, for implementing any of the above-mentioned methods for intelligently generating cultural elements based on a large model, comprising: The cultural element extraction module is used to collect historical and cultural data of the target area, pre-process the historical and cultural data, build a regional database, generate prompt words, and establish a cultural taboo knowledge graph to optimize the extraction results and obtain cultural elements; The cultural exhibition planning scheme generation module is used to collect historical and cultural exhibition planning schemes, establish a planning scheme database, combine prompt words and refined cultural elements to generate cultural exhibition planning schemes, and build a cultural exhibition planning agent and a multi-role agent cluster to iterate the cultural exhibition planning schemes multiple times, thereby updating the planning scheme database; The cultural product naming module is used for users to input cultural product design requirements, match the most relevant scheme from the planning scheme database based on the cultural product design requirements, and finally generate a cultural product naming scheme.
[0040] The third aspect of the present invention provides: a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are loaded and executed by a processor, any of the above-mentioned methods for intelligently generating cultural elements based on large models is implemented.
[0041] The foregoing description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein and should not be construed as excluding other embodiments. Rather, the present invention can be used in various other combinations, modifications, and environments and can be modified within the scope of the concept described herein through the above teachings or techniques or knowledge in the relevant field. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention are intended to be protected by the appended claims.
Claims
1. A method for intelligently generating cultural elements based on a large model, characterized by: The following steps are involved: In the cultural element extraction stage, historical and cultural data of the target area are collected, pre-processed, a regional database is constructed, prompt words are generated, and a cultural taboo knowledge graph is established to optimize the extraction results and obtain cultural elements; During the cultural exhibition planning scheme generation phase, historical and cultural exhibition planning schemes are collected, a planning scheme database is established, and a cultural exhibition planning scheme is generated by combining prompt words and refined cultural elements. A cultural exhibition planning agent and a multi-role agent cluster are constructed to iterate the cultural exhibition planning scheme multiple times, thereby updating the planning scheme database. During the cultural product naming stage, users input cultural product design requirements, and the most relevant scheme is matched from the planning scheme database based on the cultural product design requirements, and finally a cultural product naming scheme is generated.
2. The method for intelligently generating cultural elements based on a large model according to claim 1, characterized in that: The cultural elements extraction stage also includes the following steps: Delete erroneous data, delete irrelevant data, and complete missing data from historical and cultural data, and then convert them into document data; Divide the document data into multiple data blocks based on sentences, contextual semantics, and punctuation marks. Use a pre-trained text embedding model to vectorize each data block, build an index based on the Faiss database, and store the vectorized data blocks offline to form a regional database. When a user inputs elements to refine requirements and target area-related information, a text embedding model is used to convert the user input text into an embedding vector. A first preset number of relevant data blocks are then retrieved from a regional database. The user input is then concatenated with each retrieved relevant data block. The BGE-Reranker model is used to interactively analyze the correlation between the user input and the relevant data blocks, thereby reordering the relevant data blocks and selecting a second preset number of relevant data blocks as the context for the prompt word to generate the prompt word. The prompt words are used to refine the cultural elements of the target area in conjunction with the context.
3. The method for intelligently generating cultural elements based on a large model according to claim 1, characterized in that: The cultural taboo knowledge graph is established through the following steps: Collect legal and regulatory data, establish a legal and regulatory database, and collect cultural taboo data from various regions to form a cultural taboo knowledge map; During the extraction and optimization process, the retrieval enhancement generation technology RAG is used to allow the large language model to determine whether the extracted cultural elements are legal and compliant; then, based on the regional labels entered by the user, the cultural taboo knowledge graph is retrieved, and based on heuristic rules, it is determined whether the extracted cultural elements violate local customs; at the same time, real-time popular data from social media is collected to guide the large language model to determine whether the extracted cultural elements are in line with current hot topics.
4. The method for intelligently generating cultural elements based on a large model according to claim 1, characterized in that: The cultural exhibition planning scheme generation stage also includes the following steps: Use the pre-trained text embedding model to vectorize the extracted cultural elements; Retrieving a third preset number of historical and cultural exhibition planning schemes from the planning scheme database based on the quantized cultural elements, concatenating the cultural elements with the retrieved historical and cultural exhibition planning schemes, and then using the BGE-Reranker model to determine relevance, and retaining a fourth preset number of relevant schemes; Based on the prompt words, a large language model is used to understand and optimize relevant plans from multiple dimensions of the target area to obtain a cultural exhibition planning plan.
5. The method for intelligently generating cultural elements based on a large model according to claim 1 is characterized in that: The cultural product naming stage also includes the following steps: Based on the prompt words and taking cultural products as the benchmark, a large language model is used to split the most relevant solutions. After splitting, a pre-trained text embedding model is used to vectorize and cluster them to generate a cluster of cultural product naming solutions. The most relevant scheme is decomposed into sub-themes. After each sub-theme is combined with each naming scheme in the cultural product naming scheme cluster, the correlation between them is determined using the BGE-Reranker model, and the fifth preset number of highly correlated naming schemes is selected. Based on the prompt words, a large language model is used to analyze the common design mechanisms between sub-themes and highly relevant naming schemes, and corresponding cultural product naming schemes are generated for the sub-themes.
6. The method for intelligently generating cultural elements based on a large model according to any one of claims 1 to 5, characterized in that: The construction of the cultural exhibition planning agent and the multi-role agent cluster to iterate the cultural exhibition planning scheme multiple times includes the following steps: Initialize the cultural exhibition planning agent. Based on the target area's cultural elements and planning scheme database, design role prompts to define its planning functions. Integrate the search enhancement generation module to call the planning scheme database to optimize scheme generation, and set up an introspection mechanism to verify the feasibility of the draft. Build a multi-role intelligent agent cluster, including three types of intelligent agents: cultural compliance experts, art design consultants, and market analysts. Each intelligent agent uses unique prompt words to define evaluation dimensions, and review the compliance, artistic quality, and market appeal of the plan respectively; An iterative optimization mechanism is established. After the cultural exhibition planning agent generates a draft, the multi-role agent cluster reviews it in parallel and provides feedback on structured defect reports. The cultural planning agent performs targeted optimization based on the structured defect reports and resubmits them, repeating the cycle until the preset number of iterations or scoring threshold is met.
7. The method for intelligently generating cultural elements based on a large model according to any one of claims 1 to 5, characterized in that: The cultural exhibition is a lantern exhibition, and the cultural product is a lantern set.
8. The method for intelligently generating cultural elements based on a large model according to any one of claims 1 to 5, characterized in that: The historical and cultural data mentioned include local chronicles, intangible cultural heritage archives and oral accounts of ethnic minorities.
9. A system for intelligently generating cultural elements based on a large model, characterized by: A method for intelligently generating cultural elements based on a large model according to any one of claims 1 to 8, comprising: The cultural element extraction module is used to collect historical and cultural data of the target area, pre-process the historical and cultural data, build a regional database, generate prompt words, and establish a cultural taboo knowledge graph to optimize the extraction results and obtain cultural elements; The cultural exhibition planning scheme generation module is used to collect historical and cultural exhibition planning schemes, establish a planning scheme database, combine prompt words and refined cultural elements to generate cultural exhibition planning schemes, and build a cultural exhibition planning agent and a multi-role agent cluster to iterate the cultural exhibition planning schemes multiple times, thereby updating the planning scheme database; The cultural product naming module is used for users to input cultural product design requirements, match the most relevant scheme from the planning scheme database based on the cultural product design requirements, and finally generate a cultural product naming scheme.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions. When the computer-executable instructions are loaded and executed by the processor, the method for intelligently generating cultural elements based on a large model as described in any one of claims 1 to 8 is implemented.