Cross-language footwear design copywriting AI generation method

Abstract

The invention discloses a cross-language shoe design copywriting AI generation method, and relates to the technical field of shoe design copywriting intelligent generation, and the method comprises the steps: generating a primary design copywriting based on a trained multi-language field vertical large model; establishing a cross-language synonym graph, and correcting key terms of the primary design copywriting through the cross-language synonym graph and term constraint decoding to obtain a standard design copywriting; and correcting the cultural expression of the standard copywriting by adopting a style migration algorithm to obtain a final manuscript of the design copywriting. According to the method, a five-layer technical architecture of parameter analysis, knowledge fusion, multi-modal generation, semantic calibration and culture adaptation is constructed, and three industry pain points are innovatively solved; dynamic balance of copywriting between semantic accuracy and culture adaptability is realized through double engines of term atlas and style migration; aI generation efficiency is combined with creative control of human designers to form an enhanced innovation mode of AI proposal-manual screening-iterative optimization.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent generation of shoe design copy, and in particular to a cross-language AI generation method for shoe design copy. Background Art

[0002] The existing technology for generating shoe design copy has the following defects: Language uniformity: Mainstream AI copywriting tools (such as Jasper and Copy.ai) only support single-language generation and require manual translation of multiple language versions, which results in a 3-5-day extension in the cross-border e-commerce copywriting production cycle.

[0003] Poor industry adaptability: General NLP models (such as GPT-4) lack a professional terminology library for footwear (such as the Goodyear process for EVA midsoles), and the error rate of generated content reaches 12% (measured data).

[0004] Stylistic fragmentation: Manual translation leads to inconsistencies in the expression of selling points in Chinese and English copywriting. For example, the Chinese version emphasizes the national trend element, which is directly translated into "Chinese trend" in English, losing the cultural connotation.

[0005] Efficiency bottleneck: The traditional process requires designers to provide keywords → copywriters to write → translation companies to polish. The cost of producing copy for a single shoe model is very high. Summary of the Invention

[0006] In order to solve the technical problem of intelligent generation of shoe design copy, the present invention provides a cross-language shoe design copy AI generation method. The following technical solutions are adopted: A cross-language AI-generated method for footwear design copywriting includes the following steps: Step 1: Receive structured shoe design parameters and parse them into a machine-processable format to obtain required input parameters. The structured shoe design parameters include shoe type parameters, core technology parameters, and target market parameters. Step 2: Build a multimodal knowledge base for the footwear industry and retrieve structured and unstructured data from the footwear industry knowledge base based on the required input parameters: Structured data includes material parameters and process standards; Unstructured data includes multilingual popular copywriting templates and cultural image keywords in designer manuscripts; Step 3: Generate the initial design copy based on the trained multilingual vertical model; Step 4: Create a cross-language synonym graph, and use the cross-language synonym graph and term constraint decoding to revise the key terms of the initial design document to obtain the standard design document; Step 5: Use the style transfer algorithm to correct the cultural expression of the standard copy to obtain the final draft of the design copy.

[0007] By employing this technical solution, parameters such as shoe type (e.g., sneakers / leather shoes) and core technologies (e.g., air cushioning / 3D weaving) are converted into machine-readable JSON / XML formats. Combined with semantic associations within the footwear industry knowledge graph (e.g., mapping material parameters to process standards), the efficiency of analyzing design requirements is significantly improved. For example, the Vali platform can generate 500 renderings in 10 seconds by simply entering a keyword (e.g., "popcorn sole" + "cyberpunk style"), achieving over a hundredfold improvement in efficiency compared to traditional design methods.

[0008] Multilingual generation timeliness: Large multilingual models based on the Transformer architecture (such as the DeepSeek cross-lingual model) support simultaneous generation in multiple languages, significantly reducing time costs compared to traditional translation and localization processes.

[0009] By combining a cross-language synonym graph (e.g., a three-level mapping of "air cushion" → "aircushion" → "airecushion") with a footwear industry terminology database, we significantly reduce the error rate of specialized terminology. For example, in the Spanish-speaking market, we can automatically adapt the industry-standard term "mallarespirable" for mesh material.

[0010] The style transfer algorithm implicitly localizes the copy by analyzing the cultural imagery and keyword libraries of the target market (e.g., the conservative aesthetic of the Middle East or the minimalist style of Northern Europe). Test data shows that the transferred copy has significantly improved user acceptance in the target market.

[0011] A five-layer technical architecture consisting of parameter analysis → knowledge fusion → multimodal generation → semantic calibration → cultural adaptation has been constructed, innovatively solving three industry pain points: Through the dual engines of terminology graph and style transfer, a dynamic balance between semantic accuracy and cultural adaptability is achieved in copywriting. Combining the efficiency of AI generation with the creative control of human designers, this creates an enhanced innovation model of AI proposals, human screening, and iterative optimization. Establish a dynamic rule base for the global market to achieve pre-compliance in the design phase.

[0012] Optionally, step 6 is also included, in which the designer scores the generated final design document, updates the multilingual vertical large model parameters by using the PPO algorithm based on the scoring results, regenerates the first version of the design document based on the updated multilingual vertical large model, and regenerates the final design document based on steps 4 and 5.

[0013] By adopting the above technical solutions, the PPO algorithm controls the policy update amplitude within the range [1-ε, 1+ε] (ε is usually 0.2) by clipping the probability ratio and KL divergence constraints. This effectively avoids the generation quality fluctuation problem caused by policy mutation in traditional policy gradient methods (such as REINFORCE). Based on the RLHF framework, the system converts designer ratings (1-5 points) into reinforcement learning reward signals, and through multiple policy updates (typically 5-10 rounds), it achieves progressive optimization of the generation strategy. For example, in a sneaker copywriting task, after the designer's rating weight for the sense of technology dimension was increased from an initial 0.3 to 0.7, the frequency of air cushioning-related terms in the generated copy increased by 4.2 times.

[0014] Optionally, step 7 is also included, in which religious sensitive words and banned words in the corresponding national advertising law are automatically filtered out based on the compliance check algorithm.

[0015] By adopting the above technical solution, the risk of legal action and fines caused by copywriting violations is effectively reduced by automatically identifying and filtering banned words that violate the advertising laws of the target market, such as absolute terms prohibited in some countries (such as best and most comfortable), false slogans or misleading descriptions.

[0016] Automatically filter out religiously sensitive words and taboo words in specific cultural contexts, such as taboos on specific animals, colors, or patterns in some regions, to avoid negative public opinion and damage to brand image caused by cultural differences.

[0017] The automation of compliance checks that originally required manual review has significantly shortened the document review cycle and improved overall work efficiency.

[0018] Alternatively, a conditional generative model is used, with input parameters being structured data and unstructured data, to output a preliminary design copy in multiple languages. The generation formula is: ; Where x is the input parameter, l is the target language code, and s is the design style tag extracted from the multilingual popular copywriting template retrieved in step 2.

[0019] By adopting the above technical solution and introducing design style tags, the model can generate copy based on the cultural characteristics of the target market. For example, when generating copy for the Middle Eastern market, the model can automatically select appropriate expressions based on religious and cultural taboos to avoid cultural conflicts.

[0020] This cultural adaptability not only improves the market acceptance of the copy, but also reduces brand risks caused by cultural misunderstandings.

[0021] By constraining the design style tag(s), the model can ensure that the generated copy remains consistent in language, culture, and design style, thereby strengthening the brand image and enhancing consumers' awareness and trust in the product.

[0022] Optionally, natural language processing technology is used to extract high-frequency cultural keywords and style features from multilingual popular copywriting templates, build a style market mapping database, and combine it with the structured design parameters of shoes to predict the appropriate style labels through pre-trained classifiers.

[0023] Optionally, in step 4, the data source for the cross-language synonym graph is expert annotation and term pairs extracted from multilingual technical documents. The structure of the cross-language synonym graph is: graph nodes are terms, and edges are cross-language equivalence relationships. Term constraint decoding means that when generating standard design copy, if the current context requires the insertion of professional terms, dynamic masking is used to restrict candidate terms to only those corresponding terms in the cross-language synonym graph. The core formula for dynamic masking with term constraint decoding is: ; in is the result of key term correction, is a filter that handles synonyms, is the key term before the revision, is the corresponding term in the cross-language synonym graph, is a collection of key terms.

[0024] By adopting the above technical solution, Natural language processing technology extracts high-frequency cultural keywords and style characteristics from multilingual best-selling copywriting templates, constructs a style-market mapping database, and combines this with structural footwear design parameters to accurately predict appropriate style tags. This ensures the generated copywriting is highly culturally aligned with the target market, avoiding misunderstandings or discomfort caused by cultural differences.

[0025] In step 4, the cross-language synonym graph and term constraint decoding mechanism ensure the accuracy of the professional terminology used in the generated standard design copy. The dynamic masking formula ensures that the candidate professional terms during the generation process are limited to the corresponding terms in the cross-language synonym graph, thus avoiding the problem of incorrect or inconsistent terminology.

[0026] Optionally, step 5 includes the following sub-steps: Step 51: Collect historical hit copywriting in the target market, use the trained LDA model to extract the topic distribution of historical hit copywriting, and construct the target cultural distribution ; Step 52: Input the standard design document into the variational autoencoder, output the latent variable distribution parameters through the variational autoencoder backbone network BERT-base, and sample the latent variables; Step 53: distribute the target culture extracted by the LDA model Map to the latent space corresponding to the latent variables; Step 54: Use the style transfer loss function to force the latent variables to be distributed towards the target culture. Alignment; Step 54: Use Sentence-BERT to calculate the cosine similarity between the standard design text and the corrected standard design text. Step 55: Distribute the copywriting style towards the target culture by adjusting the latent variable formula Offset, get the adjusted latent variable; Step 56: Input the adjusted latent variables to the decoder and output the final draft of the design document after cultural correction.

[0027] Optionally, the formula for adjusting the latent variable in step 55 is: ; in is the mean of the original text latent variable, is to adjust the intensity, =0.3, z is the standard design text, It is the final draft of the design copy after cultural correction.

[0028] By adopting the above technical solution, the topic distribution of historical popular copywriting is extracted through the LDA model, and the target cultural distribution is constructed to ensure that the generated copywriting is highly consistent with the cultural preferences of the target market.

[0029] By using variational autoencoders and style transfer loss functions, the latent variables of standard design copy are aligned with the target cultural distribution, achieving accurate transfer of copy style.

[0030] By adjusting the latent variable formula, the copywriting style is shifted toward the target cultural distribution, ensuring that the generated copywriting is consistent in style with historical popular copywriting.

[0031] Sentence-BERT is used to calculate the cosine similarity between the standard design copy and the revised design copy to ensure that the semantic coherence of the copy is not affected during the style transfer process.

[0032] Through automated processes, design copy that meets the cultural preferences of the target market can be quickly generated, improving generation efficiency.

[0033] Combining multiple advanced technologies ensures that the generated copy meets high-quality standards in terms of cultural adaptability, style consistency, and semantic coherence.

[0034] Generate copywriting that is in line with the cultural preferences of the target market to enhance the product's appeal and competitiveness in the target market.

[0035] Through precise cultural positioning and style consistency, we can enhance brand image and strengthen consumers' identification and loyalty to the brand.

[0036] Help cross-border e-commerce companies quickly generate copywriting that meets the cultural preferences of different markets and increase product sales.

[0037] Provide fast and accurate copywriting services for the fast fashion industry to meet rapidly changing market demands.

[0038] Provide high-end brands with copywriting that has cultural depth and consistent style to enhance brand cultural connotation and market influence.

[0039] Combining multiple advanced technologies such as the LDA model, variational autoencoder, style transfer loss function, and Sentence-BERT, we achieve technological innovation in copywriting generation.

[0040] By clarifying technical steps and formulas, we protect the intellectual property rights of the methods and provide enterprises with technological competitive advantages.

[0041] A memory for storing a program designed using a cross-language AI method for generating shoe design copy.

[0042] A cross-language shoe design copy AI generation device comprises a memory, a processor, and a display, wherein the memory is used to store the shoe structured design parameters received in step 1, a shoe industry multimodal knowledge base, a cross-language synonym map, and a program designed using the cross-language shoe design copy AI generation method according to any one of claims 1 to 8; the processor is communicatively connected to the memory, runs the program to output a final design copy; and the display is communicatively connected to the processor, and the processor controls the display to display the final design copy.

[0043] In summary, the present invention includes at least one of the following beneficial technical effects: This invention can provide a cross-language AI method for generating footwear design copy, which combines shoe types and core technologies with the semantic associations of the footwear industry knowledge graph, greatly improving the efficiency of design requirement analysis.

[0044] Large multilingual models based on the Transformer architecture (such as the DeepSeek cross-lingual model) support simultaneous generation of multiple languages, significantly reducing time costs compared to traditional translation and localization processes.

[0045] By combining a cross-language synonym graph with a footwear industry terminology database, we significantly reduce the error rate of specialized terminology. For example, in the Spanish-speaking market, we can automatically adapt to industry standard expressions.

[0046] The style transfer algorithm implicitly localizes the copy by analyzing the target market's cultural imagery and keyword library. Test data shows that the transferred copy has significantly improved user acceptance in the target market.

[0047] A five-layer technical architecture consisting of parameter analysis → knowledge fusion → multimodal generation → semantic calibration → cultural adaptation has been constructed, innovatively solving three industry pain points: Through the dual engines of terminology graph and style transfer, a dynamic balance between semantic accuracy and cultural adaptability is achieved in copywriting. Combining the efficiency of AI generation with the creative control of human designers, this creates an enhanced innovation model of AI proposals, human screening, and iterative optimization. Establish a dynamic rule base for the global market to achieve pre-compliance in the design phase. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of a cross-language AI method for generating footwear design copy; DETAILED DESCRIPTION

[0049] The present invention will be further described in detail below with reference to the accompanying drawings.

[0050] An embodiment of the present invention discloses a cross-language AI method for generating footwear design copy.

[0051] Reference Figure 1 ,Example 1, a cross-language shoe design copy AI generation method, comprising the following steps: Step 1: Receive structural design parameters for footwear and parse them into a machine-processable format to obtain required input parameters. The structural design parameters for footwear include shoe type parameters, core technology parameters, and target market parameters. Step 2: Build a multimodal knowledge base for the footwear industry and retrieve structured and unstructured data from the footwear industry knowledge base based on the required input parameters: Structured data includes material parameters and process standards; Unstructured data includes multilingual popular copywriting templates and cultural image keywords in designer manuscripts; Step 3: Generate the initial design copy based on the trained multilingual vertical model; Step 4: Create a cross-language synonym graph, and use the cross-language synonym graph and term constraint decoding to revise the key terms of the initial design document to obtain the standard design document; Step 5: Use the style transfer algorithm to correct the cultural expression of the standard copy to obtain the final draft of the design copy.

[0052] By converting parameters such as shoe type (e.g., sneakers / leather shoes) and core technologies (e.g., air cushioning / 3D weaving) into machine-readable JSON / XML formats and integrating them with the semantic associations of the footwear industry knowledge graph (e.g., mapping material parameters to process standards), the Vali platform significantly improves the efficiency of analyzing design requirements. For example, by inputting keywords (e.g., "popcorn sole" + "cyberpunk style"), the Vali platform can generate 500 renderings in 10 seconds, achieving over a hundred times the efficiency of traditional design.

[0053] Multilingual generation timeliness: Large multilingual models based on the Transformer architecture (such as the DeepSeek cross-lingual model) support simultaneous generation in multiple languages, significantly reducing time costs compared to traditional translation and localization processes.

[0054] By combining a cross-language synonym graph (e.g., a three-level mapping of "air cushion" → "aircushion" → "airecushion") with a footwear industry terminology database, we significantly reduce the error rate of specialized terminology. For example, in the Spanish-speaking market, we can automatically adapt the industry-standard term "mallarespirable" for mesh material.

[0055] The style transfer algorithm implicitly localizes the copy by analyzing the cultural imagery and keyword libraries of the target market (e.g., the conservative aesthetic of the Middle East or the minimalist style of Northern Europe). Test data shows that the transferred copy has significantly improved user acceptance in the target market.

[0056] A five-layer technical architecture consisting of parameter analysis → knowledge fusion → multimodal generation → semantic calibration → cultural adaptation has been constructed, innovatively solving three industry pain points: Through the dual engines of terminology graph and style transfer, a dynamic balance between semantic accuracy and cultural adaptability is achieved in copywriting. Combining the efficiency of AI generation with the creative control of human designers, this creates an enhanced innovation model of AI proposals, human screening, and iterative optimization. Establish a dynamic rule base for the global market to achieve pre-compliance in the design phase.

[0057] Example 2 also includes step 6, in which the designer scores the generated final draft of the design document, updates the parameters of the multilingual vertical large model based on the scoring results using the PPO algorithm, regenerates the initial version of the design document based on the updated multilingual vertical large model, and regenerates the final draft of the design document based on steps 4 and 5.

[0058] The PPO algorithm controls the policy update amplitude within the range [1-ε, 1+ε] (ε is usually 0.2) by clipping the probability ratio and KL divergence constraints, effectively avoiding the generation quality fluctuation problem caused by policy mutation in traditional policy gradient methods (such as REINFORCE); Based on the RLHF framework, the system converts designer ratings (1-5 points) into reinforcement learning reward signals, and through multiple policy updates (typically 5-10 rounds), it achieves progressive optimization of the generation strategy. For example, in a sneaker copywriting task, after the designer's rating weight for the sense of technology dimension was increased from an initial 0.3 to 0.7, the frequency of air cushioning-related terms in the generated copy increased by 4.2 times.

[0059] Example 3 further includes step 7, automatically filtering religious sensitive words and banned words in the corresponding national advertising law based on the compliance check algorithm.

[0060] By automatically identifying and filtering banned words that violate advertising laws in target markets, such as absolute terms (such as best and most comfortable) prohibited in some countries, false advertising slogans, or misleading descriptions, the risk of legal action and fines due to copywriting violations is effectively reduced.

[0061] Automatically filter out religiously sensitive words and taboo words in specific cultural contexts, such as taboos on specific animals, colors, or patterns in some regions, to avoid negative public opinion and damage to brand image caused by cultural differences.

[0062] Automating compliance checks that originally required manual review has greatly shortened the document review cycle and improved overall work efficiency.

[0063] Example 4 uses a conditional generation model with structured and unstructured data as input parameters to output a multilingual preliminary design copy. The generation formula is: ; Where x is the input parameter, l is the target language code, and s is the design style tag extracted from the multilingual popular copywriting template retrieved in step 2.

[0064] The introduction of design style tags allows the model to generate copy based on the cultural characteristics of the target market. For example, when generating copy for the Middle Eastern market, the model can automatically select appropriate expressions based on religious and cultural taboos to avoid cultural conflicts.

[0065] This cultural adaptability not only improves the market acceptance of the copy, but also reduces brand risks caused by cultural misunderstandings.

[0066] By constraining the design style tag(s), the model can ensure that the generated copy remains consistent in language, culture, and design style, thereby strengthening the brand image and enhancing consumers' awareness and trust in the product.

[0067] Example 5: Natural language processing technology is used to extract high-frequency cultural keywords and style features from multilingual popular copywriting templates, a style market mapping database is constructed, and the pre-trained classifier is used to predict the adapted style labels in combination with the structured design parameters of the shoes.

[0068] In Example 6, in step 4, the data source of the cross-language synonym graph is expert annotation and term pairs extracted from multilingual technical documents. The structure of the cross-language synonym graph is: the graph nodes are terms, and the edges are cross-language equivalence relations. Term constraint decoding means that when generating standard design copy, if the current context requires the insertion of professional terms, dynamic masking is used to limit the candidate terms to only the corresponding terms in the cross-language synonym graph. The core formula for dynamic masking with term constraint decoding is: ; in is the result of key term correction, is a filter that handles synonyms, is the key term before the revision, is the corresponding term in the cross-language synonym graph, is a collection of key terms.

[0069] Natural language processing technology extracts high-frequency cultural keywords and style characteristics from multilingual best-selling copywriting templates, constructs a style-market mapping database, and combines this with structural footwear design parameters to accurately predict appropriate style tags. This ensures the generated copywriting is highly culturally aligned with the target market, avoiding misunderstandings or discomfort caused by cultural differences.

[0070] In step 4, the cross-language synonym graph and term constraint decoding mechanism ensure the accuracy of the professional terminology used in the generated standard design copy. The dynamic masking formula ensures that the candidate professional terms during the generation process are limited to the corresponding terms in the cross-language synonym graph, thus avoiding the problem of incorrect or inconsistent terminology.

[0071] In Example 7, step 5 includes the following sub-steps: Step 51: Collect historical hit copywriting in the target market, use the trained LDA model to extract the topic distribution of historical hit copywriting, and construct the target cultural distribution ; Step 52: Input the standard design document into the variational autoencoder, output the latent variable distribution parameters through the variational autoencoder backbone network BERT-base, and sample the latent variables; Step 53: distribute the target culture extracted by the LDA model Map to the latent space corresponding to the latent variables; Step 54: Use the style transfer loss function to force the latent variables to be distributed towards the target culture. Alignment; Step 54: Use Sentence-BERT to calculate the cosine similarity between the standard design text and the corrected standard design text. Step 55: Distribute the copywriting style towards the target culture by adjusting the latent variable formula Offset, get the adjusted latent variable; Step 56: Input the adjusted latent variables to the decoder and output the final draft of the design document after cultural correction.

[0072] Optionally, the formula for adjusting the latent variable in step 55 is: ; in is the mean of the original text latent variable, is to adjust the intensity, =0.3, z is the standard design text, It is the final draft of the design copy after cultural correction.

[0073] The LDA model is used to extract the topic distribution of historical popular copywriting, construct the target cultural distribution, and ensure that the generated copywriting is highly consistent with the cultural preferences of the target market.

[0074] By using variational autoencoders and style transfer loss functions, the latent variables of standard design copy are aligned with the target cultural distribution, achieving accurate transfer of copy style.

[0075] By adjusting the latent variable formula, the copywriting style is shifted toward the target cultural distribution, ensuring that the generated copywriting is consistent in style with historical popular copywriting.

[0076] Sentence-BERT is used to calculate the cosine similarity between the standard design copy and the revised design copy to ensure that the semantic coherence of the copy is not affected during the style transfer process.

[0077] Through automated processes, design copy that meets the cultural preferences of the target market can be quickly generated, improving generation efficiency.

[0078] Combining multiple advanced technologies ensures that the generated copy meets high-quality standards in terms of cultural adaptability, style consistency, and semantic coherence.

[0079] Generate copywriting that is in line with the cultural preferences of the target market to enhance the product's appeal and competitiveness in the target market.

[0080] Through precise cultural positioning and style consistency, we can enhance brand image and strengthen consumers' identification and loyalty to the brand.

[0081] Help cross-border e-commerce companies quickly generate copywriting that meets the cultural preferences of different markets and increase product sales.

[0082] Provide fast and accurate copywriting services for the fast fashion industry to meet rapidly changing market demands.

[0083] Provide high-end brands with copywriting that has cultural depth and consistent style to enhance brand cultural connotation and market influence.

[0084] Combining multiple advanced technologies such as the LDA model, variational autoencoder, style transfer loss function, and Sentence-BERT, we achieve technological innovation in copywriting generation.

[0085] A memory for storing a program designed using a cross-language AI method for generating shoe design copy.

[0086] A cross-language shoe design copy AI generation device comprises a memory, a processor, and a display, wherein the memory is used to store the shoe structured design parameters received in step 1, a shoe industry multimodal knowledge base, a cross-language synonym map, and a program designed using the cross-language shoe design copy AI generation method according to any one of claims 1 to 8; the processor is communicatively connected to the memory, runs the program to output a final design copy; and the display is communicatively connected to the processor, and the processor controls the display to display the final design copy.

[0087] The following uses a specific example to illustrate the implementation principle of a cross-language shoe design copy AI generation method: Okay, based on the above complete solution, let’s use a specific implementation case to illustrate how to generate design copy for a pair of high-end women’s leather shoes for the French market.

[0088] Case: Generating design copy for high-end women's leather shoes for the French market Step 1: Receive and parse footwear structural design parameters Shoe type parameters: women's leather shoes Core technical parameters: full-grain cowhide, Goodyear welt technology, memory foam insole Target market parameters: France Required input parameters obtained after parsing (JSON format example): { shoe_type:Women'sLeatherShoes, core_technology:{ material:Full-grainCowhide, construction:GoodyearWelt, insole:MemoryFoam }, target_market:France } Step 2: Build a multimodal knowledge base for the footwear industry and retrieve data Structured data: Material database: properties of full-grain cowhide (such as breathability, durability, and premium feel) Process Database: Advantages of Goodyear welt processing (such as durability and repairability) Designer manuscript database: related women's leather shoe styles Unstructured data: Cultural imagery in designer manuscripts Keywords: elegance, classic, art, French style Multilingual Popular Copywriting Template Library: French High-end Women's Shoe Copywriting Template Step 3: Generate the initial design copy based on the trained multilingual domain model Using the conditional generative model, the input parameters include: x (structured data and unstructured data): See steps 1 and 2 l (target language code): fr (French) s (design style label): predicted by a pre-trained classifier, for example: Elegance, Classic, French Chic Generate formula: Text=ConditionGenModel(x,l,s) Hypothetical first version of the design copy generated (French): The beauty of the fleur-de-lis, the assemblage of Goodyear, the beauty of the pastime. Translated into Chinese: A full-grain calfskin upper with a Goodyear welt and a memory foam insole. Ultimate comfort, durability, and timeless elegance. Inspired by classic French art, these shoes embody the elegance and style of French style.

[0089] Step 4: Build a cross-language synonym graph and perform term constraint decoding Cross-language synonym graph data source: Expert annotation: For example, GoodyearWelt=Goodyear, MemoryFoam=Mousseàmémoiredeforme Multilingual technical document extraction: For example, extracting professional terminology from French shoemaking technical documents Cross-language synonym graph structure: Node examples: Goodyear Welt, Memory Foam, Full-grain Cowhide; Edge example: GoodyearWelt--(equivalence relation)-->Goodyear; Term constraint decoding: Dynamic mask formula: term_corrected=filter_terms(term_original,graph_terms,term_set) in: term_corrected is the corrected term; filter_terms is a filter that handles synonyms; term_original is the term before revision; graph_terms is the corresponding term in the cross-language synonym graph; term_set is a set of key terms; When generating standard design copy, GoodyearWelt was replaced with its French equivalent, Goodyear, and MemoryFoam was replaced with Mousseàmémoiredeforme.

[0090] Step 5: Use style transfer algorithm to correct cultural expression Step 51: Construct target cultural distribution Collect historically popular women's shoe copywriting from the French market, use the LDA model to extract topic distributions, and construct the target cultural distribution. Assume that the extracted topics include Luxury, Craftsmanship, and Parisian Style.

[0091] Step 52: Coding Standard Design Document Input the standard design copy into the variational autoencoder (BERT-base as the backbone network), output the latent variable distribution parameters, and sample the latent variables to obtain z_standard.

[0092] Step 53: Map the target culture distribution to the latent space Map the target culture distribution extracted by the LDA model to the latent space corresponding to the latent variable to obtain z_culture.

[0093] Step 54: Style transfer loss function: Use style transfer loss function to force z_standard to align with z_culture.

[0094] Step 55: Adjust the latent variables: Sentence-BERT is used to calculate the cosine similarity between the standard design text and the revised standard design text to ensure semantic coherence.

[0095] Adjust the latent variable formula: z_adjusted=z_standard+α*(z_culture-z_standard) z_adjusted is the adjusted latent variable z_standard is the mean of the original latent variable α is the adjustment strength, for example 0.3 z_culture is the latent variable corresponding to the target culture distribution Step 56: Decode and get the final design document Input z_adjusted into the decoder, and output the final design with cultural corrections.

[0096] Assume that the final design document generated (in French): The beauty of the pleine fleur, the assemblage of Goodyear, the semelleintérieure of the mousse. Translated into Chinese: Full-grain calfskin upper with Goodyear welt and memory foam insole. Ultimate comfort, durability, and timeless elegance. Inspired by classic French art deco, this shoe embodies French elegance, luxury, and Parisian craftsmanship.

[0097] Step 6: Designer Rating and Model Update (Optional) The designer scores the final draft (e.g. 1-5 points).

[0098] Based on the scoring results, the PPO algorithm is used to update the parameters of the multilingual domain large model.

[0099] Regenerate the first version of the design document and repeat steps 4 and 5 to obtain the new final design document.

[0100] Step 7: Compliance Check Based on the compliance check algorithm, religious sensitive words and banned words in French advertising law are automatically filtered out.

[0101] For example, check whether there is exaggerated publicity, comparative advertising violations, etc.

[0102] The result: The final design document aligns with the cultural preferences of the French market, complies with local advertising regulations, and maintains accurate terminology. This document can be used for product promotion, website descriptions, and advertising, helping companies expand into the French market.

[0103] The above are all preferred embodiments of the present invention, and are not intended to limit the scope of protection of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A cross-language AI method for generating footwear design copy, characterized by: The following steps are involved: Step 1: Receive structured shoe design parameters and parse them into a machine-processable format to obtain required input parameters. The structured shoe design parameters include shoe type parameters, core technology parameters, and target market parameters. Step 2: Build a multimodal knowledge base for the footwear industry and retrieve structured and unstructured data from the footwear industry knowledge base based on the required input parameters: Structured data includes material parameters and process standards; Unstructured data includes multilingual popular copywriting templates and cultural image keywords in designer manuscripts; Step 3: Generate the initial design copy based on the trained multilingual vertical model; Step 4: Create a cross-language synonym graph, and use the cross-language synonym graph and term constraint decoding to revise the key terms of the initial design document to obtain the standard design document; Step 5: Use the style transfer algorithm to correct the cultural expression of the standard copy to obtain the final draft of the design copy.

2. A cross-language shoe design copy AI generation method according to claim 1, characterized in that: It also includes step 6, in which the designer scores the generated final draft of the design document, updates the parameters of the multilingual vertical large model based on the scoring results using the PPO algorithm, regenerates the first version of the design document based on the updated multilingual vertical large model, and regenerates the final draft of the design document based on steps 4 and 5.

3. The cross-language AI method for generating footwear design copy according to claim 2, characterized in that: The method also includes step 7, which automatically filters religious sensitive words and banned words in the corresponding national advertising law based on the compliance check algorithm.

4. The cross-language AI method for generating footwear design copy according to claim 3, characterized in that: Using a conditional generative model, the input parameters are structured data and unstructured data, and the output is a multilingual preliminary design copy. The generation formula is: ； Where x is the input parameter, l is the target language code, and s is the design style tag extracted from the multilingual popular copywriting template retrieved in step 2.

5. The cross-language AI method for generating footwear design copy according to claim 4, characterized in that: Natural language processing technology is used to extract high-frequency cultural keywords and style features from multilingual popular copywriting templates, build a style market mapping database, combine it with the structured design parameters of shoes, and use pre-trained classifiers to predict the appropriate style labels.

6. The cross-language AI method for generating footwear design copy according to claim 5, characterized in that: In step 4, the data source for the cross-language synonym graph is expert annotation and term pairs extracted from multilingual technical documents. The structure of the cross-language synonym graph is: the graph nodes are terms, and the edges are cross-language equivalence relationships. Term constraint decoding means that when generating standard design copy, if the current context requires the insertion of professional terms, dynamic masking is used to restrict the candidate terms to only the corresponding terms in the cross-language synonym graph. The core formula for dynamic masking with term constraint decoding is: ； in is the result of key term correction, is a filter that handles synonyms, is the key term before the revision, is the corresponding term in the cross-language synonym graph, is a collection of key terms.

7. The cross-language AI method for generating footwear design copy according to claim 6, characterized in that: Step 5 includes the following sub-steps: Step 51: Collect historical hit copywriting in the target market, use the trained LDA model to extract the topic distribution of historical hit copywriting, and construct the target cultural distribution ; Step 52: Input the standard design document into the variational autoencoder, output the latent variable distribution parameters through the variational autoencoder backbone network BERT-base, and sample the latent variables; Step 53: distribute the target culture extracted by the LDA model Map to the latent space corresponding to the latent variables; Step 54: Use the style transfer loss function to force the latent variables to be distributed towards the target culture. Alignment; Step 54: Use Sentence-BERT to calculate the cosine similarity between the standard design text and the corrected standard design text. Step 55: Distribute the copywriting style towards the target culture by adjusting the latent variable formula Offset, get the adjusted latent variable; Step 56: Input the adjusted latent variables to the decoder and output the final draft of the design document after cultural correction.

8. The cross-language AI method for generating footwear design copy according to claim 7, characterized in that: The formula for adjusting the latent variable in step 55 is: ； in is the mean of the original text latent variable, is to adjust the intensity, =0.3, z is the standard design text, It is the final draft of the design copy after cultural correction.

9. A memory, characterized in that: A program is stored that is designed using the cross-language shoe design copy AI generation method according to any one of claims 1 to 8.

10. A cross-language AI-powered shoe design copywriting generation device, characterized by: The system comprises a memory, a processor and a display, wherein the memory is used to store the shoe structured design parameters received in step 1, the shoe industry multimodal knowledge base, the cross-language synonym map and the program designed using the cross-language shoe design copy AI generation method according to any one of claims 1 to 8, the processor is communicatively connected to the memory, runs the program to output the final draft of the design copy, and the display is communicatively connected to the processor, and the processor controls the display to display the final draft of the design copy.

Images