An ai creative design generation method based on cultural semantic knowledge graph

Abstract

The application discloses an AI cultural and creative design generation method based on a cultural semantic knowledge graph, and relates to the technical fields of artificial intelligence and cultural and creative design. The method collects multi-modal cultural data and pre-processes the same, extracts entities, relationships and attributes to generate a cultural element triple set, constructs a cultural semantic knowledge graph and performs graph embedding to obtain a cultural semantic vector representation; performs semantic analysis on user demand text, extracts design intention entities and constraint conditions; performs semantic matching on the design intention entities and the knowledge graph to retrieve a target cultural element set; inputs the same into a generation model as a condition to generate a cultural and creative design; and performs cultural semantic consistency verification on an initial design scheme, iteratively optimizes the same, and outputs a final scheme. The application solves problems such as shallow cultural semantic understanding, insufficient multi-modal data fusion and inaccurate generated content culture, and can be applied to fields such as cultural and creative product design and digital content creation.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and cultural and creative design technology, specifically an AI-based method for generating cultural and creative designs based on cultural semantic knowledge graphs. Background Technology

[0002] With the booming development of the digital economy, the cultural and creative industries have become a vital force driving economic growth and cultural dissemination. The rapid advancement of artificial intelligence (AI) technology has provided new technological pathways for the design of cultural and creative products, particularly in areas such as image generation, natural language processing, and knowledge reasoning, where AI has demonstrated enormous application potential. Meanwhile, China possesses abundant resources of outstanding traditional culture, encompassing diverse content such as cultural relics, intangible cultural heritage, folk customs, and classical texts, providing a profound cultural foundation for cultural and creative design. How to efficiently mine, organize, and transform these cultural resources using AI technology has become a core issue in the digital upgrade of the cultural and creative industries.

[0003] Currently, several AI-based solutions for generating cultural and creative designs exist. For example, some solutions acquire target cultural and creative element information, perform text encoding and visual feature extraction, and use multimodal embedding vectors as conditional input to guide the model in generating images. Other solutions parse user-inputted symbol lists and layout descriptions, retrieve symbol vector outlines and reference textures from a database, and establish a set of cultural and creative symbol layout constraints to generate cultural and creative images. Still others construct digital cultural knowledge graphs, combine creative input information to generate multidimensional constraints, and drive the generation model to synthesize content. Furthermore, data analysis methods for cultural and creative elements based on natural language processing technology have also been proposed, achieving quantitative analysis of cultural and creative elements through named entity recognition and semantic network topology analysis.

[0004] However, the aforementioned existing technologies still have significant shortcomings. First, most existing methods rely on single-modal data input (such as plain text or pure images), lacking deep integration and utilization of multimodal cultural data, making it difficult to comprehensively capture the semantic connotations of cultural elements. Second, existing methods are relatively superficial in their understanding and expression of cultural semantics, lacking a systematic cultural semantic modeling framework, which leads to cultural inaccuracies in the generated cultural and creative products, i.e., deviations between the generated content and the core semantic connotations of the target culture. Third, the knowledge graphs in existing methods are mostly statically constructed, lacking a dynamic update mechanism based on user interaction feedback, making it difficult to achieve continuous accumulation and optimization of cultural semantic knowledge. Fourth, existing methods lack semantic consistency constraints and verification in the design and transformation stage of cultural elements, resulting in insufficient coherence and systematicity in the cultural semantics of the generated cultural and creative design schemes. Summary of the Invention

[0005] This application provides an AI-based cultural and creative design generation method based on a cultural semantic knowledge graph. By constructing a cultural semantic knowledge graph that integrates multimodal cultural data and combining graph neural networks with a generative model, it achieves automatic generation of semantically driven cultural and creative designs. This method can effectively solve the problems of shallow cultural semantic understanding, insufficient multimodal data integration, static knowledge graph, and inaccurate cultural content in existing cultural and creative design generation methods in the background.

[0006] To achieve the above objectives, this application provides the following technical solution: An AI-based method for generating cultural and creative designs based on cultural semantic knowledge graphs includes the following steps: S1. Collect multimodal cultural data, including cultural text data, cultural image data, and cultural audio data. Preprocess the multimodal cultural data to obtain standardized cultural data. S2. Entity extraction, relation extraction, and attribute extraction are performed on the standardized cultural data to generate a set of cultural element triples. The set of cultural element triples contains a structured knowledge representation of <head entity, relation, tail entity>. S3. Construct a cultural semantic knowledge graph based on the set of cultural element triples, and use a graph embedding algorithm to map the entities and relations in the cultural semantic knowledge graph to a low-dimensional vector space to generate a cultural semantic vector representation. S4. Obtain the cultural and creative design requirement text input by the user, perform semantic parsing on the cultural and creative design requirement text, and extract the design intent entity and semantic constraint conditions; S5. Perform semantic matching between the design intent entity and the cultural semantic knowledge graph. By calculating the semantic similarity between the vector representation of the design intent entity and the entity vector in the cultural semantic knowledge graph, retrieve the set of matching target cultural elements. S6. Based on the target cultural element set and the semantic constraints, construct a cultural and creative design generation constraint space, take the cultural semantic vector representation of the target cultural elements as the condition input, drive the preset generation model to generate cultural and creative designs, and obtain an initial cultural and creative design scheme. S7. Perform cultural semantic consistency verification on the initial cultural and creative design scheme, calculate the semantic alignment degree between the initial cultural and creative design scheme and the target cultural element set, and if the semantic alignment degree is lower than a preset threshold, optimize and adjust the condition input of the generation model based on the semantic alignment degree, iterate until the semantic alignment degree reaches the preset threshold, and output the final cultural and creative design scheme.

[0007] Furthermore, step S2, which involves entity extraction, relation extraction, and attribute extraction of the standardized cultural data, specifically includes: using a hybrid model based on a bidirectional encoder-representation converter-bidirectional long short-term memory network-conditional random field to perform fine-grained entity recognition on the cultural text data; using a graph neural network-based remote supervised relation extraction method to classify the relationships of the identified entities; and using a pre-trained language model-based attribute extraction method to extract the cultural attribute information of the entities. Through these multi-level extraction strategies, structured cultural knowledge can be extracted efficiently and accurately from unstructured cultural data, ensuring the integrity and accuracy of the cultural element triple set.

[0008] Furthermore, step S3, which involves using a graph embedding algorithm to map entities and relations in the cultural semantic knowledge graph to a low-dimensional vector space, specifically includes: using a graph embedding model based on translation distance (TransE) or a graph neural network model based on semantic matching (GCN, GAT) to vectorize the cultural semantic knowledge graph, generating a cultural semantic vector representation that integrates entity semantic information and relational structure information. By transforming the symbolic knowledge graph into computable cultural semantic vectors through graph embedding technology, a numerical semantic representation foundation is provided for the subsequent semantic matching and conditional input of the generation model.

[0009] Furthermore, the cultural semantic knowledge graph comprises three layers: a cultural entity layer, a cultural relationship layer, and a cultural attribute layer. The cultural entity layer contains cultural element entity nodes, which include at least one of the following: cultural relics, intangible cultural heritage, folk customs, historical figures, cultural symbols, and regional cultural entities. The cultural relationship layer contains semantic relationship edges between cultural element entities, which include at least one of the following: inheritance, influence, derivation, association, and spatiotemporal relationships. The cultural attribute layer contains attribute information of cultural element entities, which includes at least one of the following: cultural connotation descriptions, aesthetic feature tags, historical period tags, and geographical region tags. By constructing a three-layered cultural semantic knowledge graph, the multi-dimensional semantic information of cultural elements can be systematically organized and represented, providing rich and accurate cultural knowledge support for cultural and creative design.

[0010] Furthermore, step S4, which involves semantically parsing the cultural and creative design requirement text to extract design intent entities and semantic constraints, specifically includes: using a Transformer-based semantic parsing model to perform intent recognition and slot filling on the cultural and creative design requirement text; extracting design theme entities, design style entities, and target cultural element entities as design intent entities; and extracting color constraints, composition constraints, material constraints, and style constraints as semantic constraints. Through deep semantic parsing of user requirement text, the user's design intent and constraint requirements can be accurately captured, achieving a precise transformation from natural language requirements to structured design conditions.

[0011] Furthermore, step S5, which involves semantically matching the design intent entity with the cultural semantic knowledge graph, specifically includes: calculating the cosine similarity or Euclidean distance between the vector representation of the design intent entity and the vectors of each entity in the cultural semantic knowledge graph; sorting the entities by similarity from high to low; and selecting the top N entities as the target cultural element set. The method also includes performing multi-hop reasoning queries in the cultural semantic knowledge graph based on the semantic relationships between the design intent entities to obtain an extended set of cultural elements with indirect semantic connections to the design intent entities. By combining semantic vector matching with multi-hop reasoning from the knowledge graph, not only can the cultural elements corresponding to the user's direct intent be accurately retrieved, but also potential cultural association needs of the user can be discovered, expanding the creative space of cultural and creative design.

[0012] Furthermore, the preset generation model mentioned in step S6 is an image generation network based on a diffusion model or an image generation network based on a generative adversarial network. The cultural semantic vector represents the generation process injected into the generation model through a cross-attention mechanism. By injecting the cultural semantic vector as a conditional input into the generation model, the generation process can be guided by cultural semantic knowledge, ensuring that the generated cultural and creative design schemes maintain semantic consistency in both visual form and cultural connotation.

[0013] Furthermore, step S7, which involves verifying the cultural semantic consistency of the initial cultural and creative design scheme, specifically includes: extracting the visual feature vector of the initial cultural and creative design scheme; performing cross-modal semantic alignment calculation between the visual feature vector and the cultural semantic vector representation of the target cultural element set; and calculating the semantic alignment degree using a cross-modal similarity measurement method under a contrastive learning framework. Through cross-modal semantic consistency verification, the quality of the generated scheme can be evaluated from the perspective of cultural connotation, avoiding the problem of cultural inaccuracies where the form is similar but the essence is not.

[0014] Furthermore, the method also includes: collecting user feedback data on the final cultural and creative design scheme, and dynamically updating the cultural semantic knowledge graph based on the feedback data. The dynamic update includes at least one operation among adding cultural element entities, updating the semantic relationship weights between entities, and adjusting the cultural semantic vector representation of entities. By introducing a user feedback-driven dynamic update mechanism for the knowledge graph, the cultural semantic knowledge graph can be continuously accumulated and optimized during use, thereby continuously improving the cultural accuracy and user satisfaction of the generated cultural and creative designs.

[0015] Furthermore, the preprocessing of the multimodal cultural data includes: segmenting, removing stop words, and tagging parts of speech for cultural text data; normalizing the size, correcting the color, and augmenting the data for cultural image data; and converting the audio data to text and extracting audio features. This standardized preprocessing of multimodal cultural data provides high-quality input data for subsequent entity extraction and knowledge graph construction, ensuring the quality of the constructed cultural semantic knowledge graph.

[0016] Compared with the prior art, the beneficial effects of this application are: 1. This application constructs a cultural semantic knowledge graph that integrates multimodal data such as text, images, and audio, thereby achieving a systematic organization and representation of multidimensional semantic information of cultural elements. This overcomes the shortcomings of existing technologies where single-modal data cannot fully capture the semantic connotation of culture, and provides a rich and accurate cultural knowledge foundation for cultural and creative design.

[0017] 2. This application uses a graph embedding algorithm to map entities and relations in a cultural semantic knowledge graph to a low-dimensional vector space, realizing the computable expression of cultural knowledge. By matching the semantic vectors of the design intent entity with the knowledge graph, it can accurately retrieve the set of target cultural elements that are semantically related to user needs, effectively solving the problems of relying on keyword matching and shallow semantic understanding in the existing technology for cultural element retrieval.

[0018] 3. This application uses cultural semantic vector representation as a conditional input to drive the generation model for cultural and creative design generation. Through the cross-attention mechanism, cultural semantics guides the entire generation process, ensuring that the generated cultural and creative design schemes maintain semantic consistency in visual form and cultural connotation, fundamentally solving the problem of cultural inaccuracy in generated content in existing technologies.

[0019] 4. This application introduces a user feedback-driven dynamic update mechanism for cultural semantic knowledge graphs, which enables the continuous accumulation of cultural knowledge and iterative optimization of semantic representation. This allows the cultural and creative design generation system to have self-learning capabilities and continuously improve the accuracy of cultural semantic understanding and the quality of design generation during use. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall process of the AI-based cultural and creative design generation method based on cultural semantic knowledge graphs in this application.

[0021] The diagram shows: 1. Multimodal cultural data acquisition module; 2. Data preprocessing module; 3. Entity extraction module; 4. Relationship extraction module; 5. Attribute extraction module; 6. Cultural semantic knowledge graph construction module; 7. Graph embedding module; 8. User requirement input module; 9. Semantic parsing module; 10. Semantic matching module; 11. Target cultural element set; 12. Generative model; 13. Initial cultural and creative design scheme; 14. Semantic consistency verification module; 15. Iterative optimization module; 16. Final cultural and creative design scheme; 17. User feedback collection module; 18. Knowledge graph dynamic update module. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] In the description of this application, any descriptions of orientation, such as up, down, front, back, left, right, etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are merely for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application. When a feature is referred to as being set, fixed, or connected to another feature, it can be directly set, fixed, or connected to the other feature, or it can be indirectly set, fixed, or connected to the other feature.

[0024] Please see Figure 1 This application provides the following technical solution: an AI-based method for generating cultural and creative designs based on cultural semantic knowledge graphs.

[0025] S1. Multimodal cultural data acquisition and preprocessing Multimodal cultural data is collected, including cultural text data (such as ancient books and documents, local chronicles, intangible cultural heritage records, and cultural research papers), cultural image data (such as photographs of cultural relics, architectural images, traditional patterns, and paintings), and cultural audio data (such as folk music, opera excerpts, and oral history recordings). The multimodal cultural data is preprocessed to obtain standardized cultural data.

[0026] The preprocessing of multimodal cultural data specifically includes: segmenting, removing stop words, and tagging parts of speech for cultural text data, which can be done using open-source segmentation tools such as Jieba segmentation combined with a custom cultural domain dictionary; normalizing the size, correcting the color, and augmenting the data for cultural image data, adjusting the images to a preset size and performing color correction operations such as histogram equalization; and converting the audio data from speech to text and extracting audio features, using automatic speech recognition technology to convert the audio into text and then incorporating it into the text data flow, while extracting acoustic features such as Mel spectrum of the audio as auxiliary semantic information.

[0027] S2, Cultural Element Triad Extraction The standardized cultural data is subjected to entity extraction, relation extraction, and attribute extraction to generate a set of cultural element triples, which contains a structured knowledge representation of <head entity, relation, tail entity>.

[0028] Entity extraction employs a hybrid model based on Bidirectional Encoder-Representation Transformer (BERT), Bidirectional Long Short-Term Memory (BiLSTM), and Conditional Random Field (CRF) for fine-grained entity recognition of cultural text data. This model first uses BERT to perform contextual semantic encoding of the text, then uses BiLSTM to capture sequence dependencies, and finally uses CRF for globally optimal sequence labeling, identifying various entities in the cultural text, including but not limited to names of cultural relics, intangible cultural heritage items, historical figures, regions, and cultural symbols.

[0029] Relation extraction employs a remotely supervised relation extraction method based on graph neural networks to classify the relationships between identified entities. Specifically, an existing cultural domain knowledge base is used as a supervisory signal to automatically label training data. The semantic relationship patterns between entities are learned through a graph neural network model, and the relationships between entities are classified into types such as inheritance relationships, influence relationships, derivative relationships, association relationships, and spatiotemporal relationships.

[0030] The attribute extraction method uses a pre-trained language model to extract the cultural attribute information of entities, including cultural connotation descriptions, aesthetic feature labels, historical period labels, and geographical region labels.

[0031] S3, Construction and Embedding of Cultural Semantic Knowledge Graphs A cultural semantic knowledge graph is constructed based on the aforementioned set of cultural element triples. This cultural semantic knowledge graph comprises three layers: a cultural entity layer, a cultural relationship layer, and a cultural attribute layer. The cultural entity layer contains cultural element entity nodes, including cultural relics, intangible cultural heritage, folk customs, historical figures, cultural symbols, and regional cultural entities. The cultural relationship layer contains semantic relationship edges between cultural element entities, including inheritance relationships, influence relationships, derivation relationships, association relationships, and spatiotemporal relationships. The cultural attribute layer contains the attribute information of cultural element entities.

[0032] A graph embedding algorithm is employed to map entities and relations in the cultural semantic knowledge graph to a low-dimensional vector space, generating a cultural semantic vector representation. Specifically, the TransE graph embedding model based on translation distance can be used, where the head entity vector h, relation vector r, and tail entity vector t in each triple <h,r,t> satisfy the constraint relationship h+rt; or a graph neural network model based on graph convolutional network (GCN) and graph attention network (GAT) can be used, which aggregates the semantic information of neighboring nodes through a message passing mechanism to generate a cultural semantic vector representation that integrates entity semantic information and relational structure information.

[0033] S4, User Requirement Semantic Analysis Obtain the user-inputted text of cultural and creative design requirements, perform semantic parsing on the text, and extract the design intent entity and semantic constraints.

[0034] Specifically, a Transformer-based semantic parsing model is used to perform intent recognition and slot filling on the cultural and creative design requirement text. Intent recognition is used to determine the user's design purpose type (such as cultural and creative product design, cultural pattern design, cultural IP image design, etc.), and slot filling is used to extract key information elements from the text. Design theme entities, design style entities, and target cultural element entities are extracted as design intent entities; color constraints (such as main color tone, color scheme, etc.), composition constraints (such as symmetry, balance, rhythm, etc.), material constraints (such as paper, wood, metal, etc.), and style constraints (such as traditional style, modern style, fusion style, etc.) are extracted as semantic constraints.

[0035] S5. Semantic Matching and Cultural Element Retrieval The design intent entity is semantically matched with the cultural semantic knowledge graph. The semantic similarity (cosine similarity or Euclidean distance) between the vector representation of the design intent entity and the entity vector in the cultural semantic knowledge graph is calculated. The entities are sorted from high to low according to the similarity and the top N entities are selected as the target cultural element set.

[0036] Simultaneously, based on the semantic relationships between the entities with design intent, multi-hop reasoning queries are performed in the cultural semantic knowledge graph to obtain an extended set of cultural elements that have indirect semantic connections with the entities with design intent. For example, when a user inputs Dunhuang flying apsaras as a design theme, not only is the Dunhuang flying apsaras entity itself retrieved, but also related cultural elements with influence, derivative, and associated relationships with Dunhuang flying apsaras are obtained through multi-hop reasoning in the knowledge graph, such as extended elements like Tang Dynasty murals, Buddhist art, ribbon patterns, and Western Region music and dance, thereby expanding the scope of creative materials for cultural and creative design.

[0037] S6, Cultural and Creative Design Generation Based on the target cultural element set and the semantic constraints, a cultural and creative design generation constraint space is constructed. The cultural semantic vector representation of the target cultural elements is used as the condition input to drive the preset generation model to generate cultural and creative designs and obtain an initial cultural and creative design scheme.

[0038] The preset generative model is either an image generation network based on a diffusion model or an image generation network based on a generative adversarial network. Taking the diffusion model as an example, the generation process is as follows: starting from standard Gaussian noise, an image is generated through a stepwise denoising process. In each denoising step, a cultural semantic vector representation is injected into the generative model through a cross-attention mechanism, enabling the denoising network to refer to cultural semantic information when predicting noise, thereby guiding the generation process towards a direction that conforms to cultural semantics. Semantic constraints (color, composition, texture, style, etc.) are injected into the generative model as additional conditional inputs through an adaptive normalization layer or a style modulation network to achieve fine control over the generated results.

[0039] S7. Cultural Semantic Consistency Verification and Iterative Optimization The initial cultural and creative design scheme is subjected to cultural semantic consistency verification, and the semantic alignment degree between the initial cultural and creative design scheme and the target cultural element set is calculated.

[0040] Specifically, the visual feature vector of the initial cultural and creative design scheme is extracted (this can be done using a pre-trained visual encoder such as the CLIP image encoder). The visual feature vector is then used to perform cross-modal semantic alignment calculations with the cultural semantic vector representation of the target cultural element set. A cross-modal similarity measurement method within a contrastive learning framework is used to calculate the semantic alignment degree. If the semantic alignment degree is lower than a preset threshold (e.g., 0.75), the conditional input of the generative model is optimized based on this degree. Specifically, the injection weights of the cultural semantic vectors can be adjusted, the conditional encoder parameters of the generative model can be fine-tuned, or a more matching target cultural element can be retrieved and iterated until the semantic alignment degree reaches the preset threshold, at which point the final cultural and creative design scheme is output.

[0041] S8, User Feedback-Driven Dynamic Updates of the Knowledge Graph Feedback data from users on the final cultural and creative design scheme is collected, including satisfaction ratings, modification suggestions, and cultural element preferences. The cultural semantic knowledge graph is dynamically updated based on this feedback data. Dynamic updates include at least one of the following operations: adding new cultural element entities (incorporating new cultural elements mentioned in user feedback into the knowledge graph), updating the semantic relationship weights between entities (adjusting relationship strength based on user preferences for specific combinations of cultural elements), and adjusting the cultural semantic vector representations of entities (updating graph embedding results through incremental learning). Through this user feedback-driven dynamic update mechanism, the cultural semantic knowledge graph continuously accumulates and optimizes during use, constantly improving the cultural accuracy of generated cultural and creative designs and user satisfaction.

[0042] It is worth noting that the graph embedding model disclosed in this embodiment can be a classic model such as TransE, DistMult, or ComplEx, or a graph neural network model such as GCN or GAT, which can be flexibly configured according to the scale and density of the cultural semantic knowledge graph. The semantic parsing model can be a fine-tuned model based on BERT or an end-to-end semantic parsing model based on T5. The generative model can be an open-source diffusion model such as StableDiffusion or a generative adversarial network model such as StyleGAN, preferably a dedicated generative model that has been fine-tuned on a cultural image dataset. Cross-modal semantic alignment computation can use the visual-text alignment framework of the CLIP model. The control and scheduling of the above models can be implemented using commonly used deep learning frameworks in the prior art (such as PyTorch, TensorFlow, etc.), and the model training and inference can use conventional training strategies and inference methods in the prior art. The core chip of the control switch group can be a GPU (such as NVIDIA A100) or a dedicated AI acceleration chip (such as Huawei Ascend, Cambricon, etc.), which can be configured according to the computing requirements of the actual application scenario.

[0043] Although embodiments of this application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An AI-based method for generating cultural and creative designs based on cultural semantic knowledge graphs, characterized in that, Includes the following steps: S1. Collect multimodal cultural data, including cultural text data, cultural image data, and cultural audio data. Preprocess the multimodal cultural data to obtain standardized cultural data. S2. Entity extraction, relation extraction, and attribute extraction are performed on the standardized cultural data to generate a set of cultural element triples. The set of cultural element triples contains a structured knowledge representation of <head entity, relation, tail entity>. S3. Construct a cultural semantic knowledge graph based on the set of cultural element triples, and use a graph embedding algorithm to map the entities and relations in the cultural semantic knowledge graph to a low-dimensional vector space to generate a cultural semantic vector representation. S4. Obtain the cultural and creative design requirement text input by the user, perform semantic parsing on the cultural and creative design requirement text, and extract the design intent entity and semantic constraint conditions; S5. Perform semantic matching between the design intent entity and the cultural semantic knowledge graph. By calculating the semantic similarity between the vector representation of the design intent entity and the entity vector in the cultural semantic knowledge graph, retrieve the set of matching target cultural elements. S6. Based on the target cultural element set and the semantic constraints, construct a cultural and creative design generation constraint space, take the cultural semantic vector representation of the target cultural elements as the condition input, drive the preset generation model to generate cultural and creative designs, and obtain an initial cultural and creative design scheme. S7. Perform cultural semantic consistency verification on the initial cultural and creative design scheme, calculate the semantic alignment degree between the initial cultural and creative design scheme and the target cultural element set, and if the semantic alignment degree is lower than a preset threshold, optimize and adjust the condition input of the generation model based on the semantic alignment degree, iterate until the semantic alignment degree reaches the preset threshold, and output the final cultural and creative design scheme.

2. The AI-based cultural and creative design generation method based on cultural semantic knowledge graph as described in claim 1, characterized in that: The entity extraction, relation extraction, and attribute extraction of the standardized cultural data in step S2 specifically includes: using a hybrid model based on bidirectional encoder-representation converter-bidirectional long short-term memory network-conditional random field to perform fine-grained entity recognition on the cultural text data; using a remote supervised relation extraction method based on graph neural network to classify the relationships of the identified entities; and using an attribute extraction method based on pre-trained language model to extract the cultural attribute information of the entities.

3. The AI-based cultural and creative design generation method based on cultural semantic knowledge graph as described in claim 1, characterized in that: The step S3, which involves using a graph embedding algorithm to map entities and relations in the cultural semantic knowledge graph to a low-dimensional vector space, specifically includes: using a graph embedding model based on translation distance or a graph neural network model based on semantic matching to vectorize the cultural semantic knowledge graph, thereby generating a cultural semantic vector representation that integrates entity semantic information and relational structure information.

4. The AI-based cultural and creative design generation method based on cultural semantic knowledge graph as described in claim 1, characterized in that: The cultural semantic knowledge graph comprises three layers: a cultural entity layer, a cultural relationship layer, and a cultural attribute layer. The cultural entity layer contains cultural element entity nodes, which include at least one of the following: cultural relics entities, intangible cultural heritage entities, folk customs entities, historical figures entities, cultural symbols entities, and regional cultural entities. The cultural relationship layer contains semantic relationship edges between cultural element entities, which include at least one of the following: inheritance relationship, influence relationship, derivation relationship, association relationship, and spatiotemporal relationship. The cultural attribute layer contains attribute information of cultural element entities, which includes at least one of the following: cultural connotation description, aesthetic feature tags, historical period tags, and geographical region tags.

5. The AI-based cultural and creative design generation method based on cultural semantic knowledge graph according to claim 1, characterized in that: Step S4, which involves semantic parsing of the cultural and creative design requirement text to extract design intent entities and semantic constraints, specifically includes: using a Transformer-based semantic parsing model to perform intent recognition and slot filling on the cultural and creative design requirement text, extracting design theme entities, design style entities, and target cultural element entities as design intent entities; and extracting color constraints, composition constraints, material constraints, and style constraints as semantic constraints.

6. The AI-based cultural and creative design generation method based on cultural semantic knowledge graph according to claim 1, characterized in that: Step S5, which involves semantically matching the design intent entity with the cultural semantic knowledge graph, specifically includes: calculating the cosine similarity or Euclidean distance between the vector representation of the design intent entity and the vectors of each entity in the cultural semantic knowledge graph; sorting the entities by similarity from high to low; and selecting the top N entities as the target cultural element set. The method also includes performing multi-hop reasoning queries in the cultural semantic knowledge graph based on the semantic relationships between the design intent entities to obtain an extended cultural element set that has indirect semantic association with the design intent entity.

7. The AI-based cultural and creative design generation method based on cultural semantic knowledge graph according to claim 1, characterized in that: The preset generative model mentioned in step S6 is an image generation network based on a diffusion model or an image generation network based on a generative adversarial network. The cultural semantic vector represents the generation process of the generative model injected through a cross-attention mechanism.

8. The AI-based cultural and creative design generation method based on cultural semantic knowledge graph according to claim 1, characterized in that: The step S7, which involves verifying the cultural semantic consistency of the initial cultural and creative design scheme, specifically includes: extracting the visual feature vector of the initial cultural and creative design scheme; performing cross-modal semantic alignment calculation between the visual feature vector and the cultural semantic vector representation of the target cultural element set; and calculating the semantic alignment degree using a cross-modal similarity measurement method under a contrastive learning framework.

9. The AI-based cultural and creative design generation method based on cultural semantic knowledge graph according to claim 1, characterized in that: The method further includes: collecting user feedback data on the final cultural and creative design scheme, and dynamically updating the cultural semantic knowledge graph based on the feedback data. The dynamic update includes at least one of the following operations: adding cultural element entities, updating the semantic relationship weights between entities, and adjusting the cultural semantic vector representation of entities.

10. The AI-based cultural and creative design generation method based on cultural semantic knowledge graph according to claim 1, characterized in that: The preprocessing of the multimodal cultural data includes: word segmentation, stop word removal, and part-of-speech tagging of cultural text data; size normalization, color correction, and data enhancement of cultural image data; and speech-to-text conversion and audio feature extraction of cultural audio data.

Images