Image-text propaganda article intelligent design method based on large model and knowledge base

Abstract

The invention belongs to the technical field of data processing, and particularly relates to an image-text propaganda article intelligent design method based on a large model and a knowledge base. The front end receives user text description, and the user text description is preprocessed into a structured demand through grammar error correction and intention recognition; inputting an adaptive large model in the field of propaganda product design to generate an initial design parameter set (optimized by a genetic algorithm), and performing preference correction in combination with user historical record statistical parameter acceptance frequency; and finally, automatically adapting to a propaganda product type size template, generating a picture and outputting the picture to a front end, thereby realizing the unification of efficient, personalized and professional design.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of data processing, and particularly relates to a graphic propaganda material intelligent design method based on a large model and a knowledge base. BACKGROUND

[0002] Current graphic propaganda material design mainly relies on manual design or simple template tools: manual design has problems of low efficiency, high cost and high professional threshold, and is difficult to meet the rapid design needs of small and medium-sized enterprises or individual users; although the template tool can improve efficiency, it has the defect of template solidification and cannot flexibly generate a design scheme according to the personalized text needs of users, resulting in uneven quality of the output propaganda materials and difficulty in balancing personalization and professionalism. SUMMARY

[0003] The application proposes a graphic propaganda material intelligent design method based on a large model and a knowledge base to solve the technical problems in the above background art.

[0004] In order to achieve the above purpose, the technical scheme adopted by the application comprises the following steps:

[0005] The front end receives the text description of the graphic propaganda material input by the user, pre-processes the text description, and outputs a structured demand text, wherein the pre-processing includes grammar correction and intent recognition;

[0006] The grammar correction is used to correct ambiguities and wrong characters and determine the real requirements of the user;

[0007] The intent recognition is used to extract core demand types and clarify the intent information of the graphic propaganda material;

[0008] The structured demand text is input into a large model adapted to the field of propaganda material design, and an initial design parameter set is output; the initial design parameter set includes a color system, a layout structure, a font type and a list of visual elements;

[0009] The initial design parameter set is corrected according to the user's preferences, and a corrected design parameter set is output;

[0010] Based on the corrected design parameter set, a design scheme is generated and converted into a picture file, a size template is automatically adapted based on the type of the propaganda material, and the final picture is output to the front end.

[0011] As a preferred embodiment, the implementation of the grammar correction for correcting ambiguities and wrong characters and determining the real requirements of the user specifically comprises:

[0012] First, high-frequency error cases in the description of propaganda material requirements are collected to construct a field error expression set, a unique corresponding propaganda material design field standard expression is matched for each error expression, and a corpus is obtained by combining the field error expression set and the corresponding standard expression.

[0013] Based on the edit distance algorithm combined with the domain corpus, the surface errors of the requirement text are accurately identified and replaced, and the text to be corrected is defined. ,in is the number of text characters, and a set of candidate correct text segments is extracted from the domain corpus ,in is the number of candidate segments, ,in is the number of characters in the k-th candidate segment;

[0014] For text to be corrected With each candidate segment The minimum edit distance is , where k represents the kth candidate segment, and the minimum edit distance is calculated as follows:

[0015] ,in Indicates the first i characters of S and The minimum number of edits for the first j characters of the text to be corrected. When there are i characters to be corrected, but there are no characters to compare with the candidate correct text at this time, the first i characters of the text to be corrected need to be deleted, a total of i operations; when there are no characters in the text to be corrected, but there are j characters in the candidate correct text, the first j characters of the candidate correct text need to be inserted; when both texts have characters, it is necessary to select the one with the least number of operations from the three possibilities of deletion, insertion, and replacement / no operation. The deletion operation is to first calculate the distance between the first i-1 characters to be corrected and the first j candidate characters. , add a deletion operation; the insertion operation is that the text to be corrected lacks the jth character of the candidate text, which needs to be inserted and then matched, that is, first calculate the distance between the first i characters to be corrected and the first j-1 candidate characters , add an insert operation; replace / no operation is to determine whether the i-th character of the text to be corrected is consistent with the j-th character of the candidate text. If they are consistent, no operation is required. If they are inconsistent, replacement is required. Calculate the distance between the first i-1 characters to be corrected and the first j-1 candidate characters. , add 0 or 1 according to whether the characters are consistent, is the indicator function;

[0016] Set the edit distance threshold, select the edit distance less than the threshold, and get the smallest candidate segment ,right Replace the wrong part in the text to get the corrected text with grammatical correction .

[0017] As a preferred method, intent recognition is used to extract core demand types and clarify the intent information of graphic and text promotional materials. The specific implementation includes:

[0018] First, build a core library, including promotional product type labels and type identification keywords;

[0019] Corrected text for grammatical errors Perform extraction operations, extract type identification keywords and perform matching calculations with type identification keywords in the core library, calculate the matching degree, and screen candidate types with matching degrees greater than the set threshold;

[0020] If the candidate type is unique, then this type is the core requirement type. If there are multiple candidate types, the core requirement type is selected according to the degree of matching. If there is no candidate type, a requirement supplement prompt is output.

[0021] As a preferred embodiment, the operation of performing matching calculation between the type identification keywords extracted from the revised text and the type identification keywords in the core library is:

[0022] Generate keyword vectors for type recognition keywords in the core library and keyword vectors for recognition keywords in user-corrected texts through the preset pre-trained BERT model;

[0023] Combine each type of keyword in the core library to obtain the keyword vector set of that type ,in is the number of types, For the number of keywords of each type, calculate the type center vector of each type ,in For keywords Importance weight for type q;

[0024] Keyword vector set extracted from the revised text Calculate the demand aggregation vector of user keywords ,in The weight of the frequency of users mentioning this keyword;

[0025] For each type q, first calculate and type center vector Cosine similarity of and Euclidean distance , the Euclidean distance Normalize the distance matching score ;

[0026] Finally, the cosine similarity Matching score with distance Perform weighted fusion to obtain the final matching degree.

[0027] As a preferred method, the structured requirement text is input into the promotional product design domain adaptation model, and the specific operation of outputting the initial design parameter set is as follows:

[0028] Input the pre-processed structured demand text into the text embedding layer of the large-scale model adapted for the promotional material design domain to generate the demand feature vector;

[0029] The adaptation model generates candidate sets based on the demand feature vector and calculates the probability of each candidate for each parameter type matching the demand: ,in is the yth candidate of a certain type of parameter, is the eigenvector, is the model output layer function;

[0030] For each type of parameter, the top three candidates with the highest probability are selected to form a parameter candidate set. The candidate set is then optimized with multiple objectives. The genetic algorithm is used to optimize the obtained promotional product design score, and the optimal parameters are selected and recorded as the initial design parameter set.

[0031] Preferably, the initial design parameter set is modified by user preference, and the modified design parameter set is outputted as follows:

[0032] Identify the historical generation records of the currently logged-in user, and count the user's acceptance frequency of the design parameters in all records. The acceptance frequency is the ratio of the number of times the user selects the parameter in the process of historical record generation to the total number of selections. If the acceptance frequency is greater than the preset threshold, the user preference is corrected, and the dimension in the initial parameter set is replaced with the dimension parameter after the user preference correction.

[0033] Compared with the existing technology, the advantages and positive effects of the present invention lie in that it optimizes the design process through multi-step innovation. In the front-end preprocessing stage, the error corpus in the field of promotional products is combined with the edit distance algorithm for precise error correction, and the pre-trained BERT model and the core library keyword matching calculation are used to realize intent recognition and ensure the accuracy of demand input. The structured demand is then input into the large model for adapting the field of promotional product design to generate parameter candidates. The genetic algorithm uses the design score as the target for multi-objective optimization to screen out the initial parameter set including color, layout, etc., taking into account both professionalism and diversity. The initial parameter set is also modified by statistically analyzing the parameter acceptance frequency in the user's historical records to enhance personalization. Finally, the promotional product type and size template is automatically adapted to generate images, achieving the unity of high efficiency, personalization and professional design, and meeting the rapid design needs of small and medium-sized enterprises and individuals. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0035] Figure 1 The figure is a structural flow chart of an intelligent design method for graphic and text promotional materials based on a large model and a knowledge base. DETAILED DESCRIPTION

[0036] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described below in conjunction with the accompanying drawings and embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0037] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0038] Embodiment: In the current environment of digital communication and new media operation, graphic and text promotional materials have become an important carrier for external display and information transmission of enterprises, government agencies and personal brands. However, the existing graphic and text promotional material design methods mainly rely on the experience of manual designers, or simple generation tools based on limited template libraries. Although manual design can provide a certain degree of professionalism, it has problems such as high cost, low efficiency, and high professional threshold for designers. It is not suitable for the rapid promotion needs of small and medium-sized enterprises or individual users. Although templated design tools have improved efficiency to a certain extent, it is difficult to achieve accurate response to the personalized needs of users, which easily leads to serious homogeneity of promotional materials and lack of innovation and pertinence. Therefore, the present invention proposes an intelligent design method for graphic and text promotional materials based on a large model and a knowledge base. The specific implementation process is as follows: Figure 1 shown.

[0039] To overcome ambiguity, typos, and semantic ambiguity in user demand descriptions and ensure the input quality of subsequent large models, we employ grammatical error correction and intent recognition technologies. First, the front-end receives the user's input text description of the promotional product, preprocesses it, and outputs a structured demand text. This preprocessing includes grammatical error correction and intent recognition.

[0040] The grammatical error correction is used to correct ambiguity and typos and determine the user's true requirements. Specifically, first collect the frequently occurring error cases in the description of promotional product requirements and construct a domain error statement set. Each error statement is matched with a unique corresponding promotional product design domain standard statement. The domain error statement set and the corresponding standard statement are combined to obtain a corpus. Based on the edit distance algorithm combined with the domain corpus, the surface errors of the requirement text are accurately identified and replaced, and the text to be corrected is defined. ,in is the number of text characters, and a set of candidate correct text segments is extracted from the domain corpus ,in is the number of candidate segments, ,in is the number of characters in the kth candidate segment; for the text to be corrected With each candidate segment The minimum edit distance is , where k represents the kth candidate segment, and the minimum edit distance is calculated as follows:

[0041] in, Indicates the first i characters of S and The minimum number of edits for the first j characters of the text to be corrected. When there are i characters to be corrected, but there are no characters to compare with the candidate correct text at this time, the first i characters of the text to be corrected need to be deleted, a total of i operations; when there are no characters in the text to be corrected, but there are j characters in the candidate correct text, the first j characters of the candidate correct text need to be inserted; when both texts have characters, it is necessary to select the one with the least number of operations from the three possibilities of deletion, insertion, and replacement / no operation. The deletion operation is to first calculate the distance between the first i-1 characters to be corrected and the first j candidate characters. , add a deletion operation; the insertion operation is that the text to be corrected lacks the jth character of the candidate text, which needs to be inserted and then matched, that is, first calculate the distance between the first i characters to be corrected and the first j-1 candidate characters , add an insert operation; replace / no operation is to determine whether the i-th character of the text to be corrected is consistent with the j-th character of the candidate text. If they are consistent, no operation is required. If they are inconsistent, replacement is required. Calculate the distance between the first i-1 characters to be corrected and the first j-1 candidate characters. , add 0 or 1 according to whether the characters are consistent, is the indicator function; set the edit distance threshold, select the edit distance less than the threshold, and get the smallest candidate segment ,right Replace the wrong part in the text to get the corrected text with grammatical correction Finally, candidate segments with a minimum edit distance less than a preset threshold are selected as optimal replacements to correct the erroneous parts of the text to be corrected, resulting in a semantically complete and unambiguous correction text. This process significantly improves the accuracy of requirement input and prevents typos or ambiguous expressions from interfering with subsequent intent recognition and design generation.

[0042] Intent recognition is used to extract the core demand type and clarify the intention information of graphic and text promotional materials. First, build a core library, including promotional material type labels and type identification keywords; correct the text for grammatical errors An extraction operation is performed, where the extracted type identification keywords are matched against those in the core library. The matching degree is calculated and candidate types with a matching degree exceeding a set threshold are selected. If a candidate type is unique, it is considered the core requirement type. If there are multiple candidate types, the core requirement type is selected based on the matching degree. If no candidate types exist, a supplementary requirement prompt is output. Specifically, a core library is first constructed, containing promotional product type labels (such as posters, leaflets, and social media covers) and their corresponding type identification keywords. After receiving the grammatically corrected text, the system performs word segmentation and feature extraction, extracting possible type identification keywords and matching them against the keywords in the core library one by one. This matching method uses a text similarity metric based on cosine similarity or the Jaccard coefficient to calculate the matching degree between the extracted keywords and the keywords in the library. If the matching degree exceeds a set threshold, the corresponding promotional product type is selected as a candidate type. When the candidate type is unique, the type is directly determined as the user's core requirement type; when there are multiple candidate types, they are sorted according to the degree of match, and the type with the highest degree of match is selected as the final requirement type; if no candidate type is matched, the system outputs a prompt message, requiring the user to supplement their design requirements. In this way, the system can effectively avoid demand classification errors while maintaining robustness, ensuring that the subsequent design generation link uses accurate type parameters as input, improving the overall design effect and user satisfaction. The matching calculation operation is: using the preset pre-trained BERT model to generate keyword vectors for type identification keywords in the core library and identification keyword vectors in the user's corrected text; combining the keywords of each type in the core library to obtain the keyword vector set of that type ,in is the number of types, For the number of keywords of each type, calculate the type center vector of each type ,in For keywords Importance weight for type q; for the keyword vector set extracted from the revised text Calculate the demand aggregation vector of user keywords ,in is the frequency weight of users mentioning this keyword; for each type q, first calculate and type center vector Cosine similarity of and Euclidean distance , the Euclidean distance Normalize the distance matching score ; Finally, the cosine similarity Matching score with distance The final matching degree is obtained by weighted fusion. The method takes into account the semantic relevance and spatial distance difference at the same time, can effectively improve the accuracy of demand type recognition, avoid the deviation caused by single similarity measurement, and thus improve the robustness and accuracy of user real demand type determination.

[0043] In order to convert the natural language demand into a form that can be processed by the model, the structured demand text is input into the propaganda product design field adaptive large model to output an initial design parameter set. The initial design parameter set includes a color system, a layout structure, a font type, and a list of visual elements. The specific operation of inputting the structured demand text into the propaganda product design field adaptive large model to output the initial design parameter set is to input the preprocessed structured demand text into the text embedding layer of the propaganda product design field adaptive large model to generate a demand feature vector. The adaptive large model generates a candidate set based on the demand feature vector, and calculates the adaptation probability of each candidate of each parameter to the demand: wherein is the yth candidate of a certain parameter, is the feature vector, is the model's output layer function. For each parameter category, the top three candidates with the highest probability are selected to form a parameter candidate set. This candidate set is then subjected to multi-objective optimization using a genetic algorithm, optimizing the resulting promotional product design score. The optimal parameters are then selected and recorded as the initial design parameter set. Specifically, the structured requirement text, obtained through grammatical correction, type identification, and requirement extraction, is first input into a large-scale model adapted for the promotional product design domain. During the input phase, the structured requirement text enters the model's text embedding layer. This layer, based on word embedding and context modeling, converts the requirement description into a unified high-dimensional requirement feature vector, ensuring that requirements expressed in different ways can be accurately represented in the semantic space. Subsequently, based on this requirement feature vector, the large-scale model generates candidate sets for four parameter categories: color scheme, layout structure, font type, and visual elements. For each parameter category, the large-scale model calculates its probability of matching the requirement feature vector using the output layer function. This calculation determines the degree of match between each parameter candidate and the user's requirement. During the candidate screening phase, the system ranks each parameter category according to its probability of matching and selects the top three candidates with the highest probability to form the parameter candidate set. This process ensures that a diverse set of parameters that best matches the requirements is retained, avoiding design rigidity caused by a single candidate. It also provides a richer search space for subsequent combinatorial optimization. After obtaining candidate sets for various parameters, the system further introduces a multi-objective optimization mechanism. The overall promotional product design score serves as the objective function. This optimization utilizes a genetic algorithm, whose core principle is to mimic the natural selection and genetic mechanisms of biological evolution to find the optimal solution within the vast candidate parameter space. The implementation process is as follows: First, each candidate parameter combination is encoded as a chromosome individual, and an initial population is constructed. The population size is preset based on computing resources and accuracy requirements. Next, the objective function is calculated for each individual, representing the individual's fitness. During the evolutionary process, the algorithm prioritizes individuals with higher fitness through selection. Crossover operations exchange gene fragments between two individuals to generate new solution combinations. Mutation operations randomly alter individual genes to increase population diversity and avoid local optima. After several iterations, individuals in the population gradually converge on the optimal solution. Ultimately, the individual with the highest fitness becomes the initial optimized design parameter set. The advantage of this method lies in its ability to perform a global search under complex multi-objective conditions, ensuring the coordination and innovation of parameter combinations. Ultimately, after several iterations, the genetic algorithm outputs the optimal solution, which becomes the initial design parameter set. This parameter set not only ensures a high degree of match with user requirements but also achieves global coordination among design parameters within the constraints of multi-objective optimization, providing stable and reliable input for subsequent detailed design generation.

[0044] The initial design parameter set is then subjected to user preference correction, outputting a revised design parameter set. The system identifies the currently logged-in user's historical generation records and calculates the user's acceptance frequency for each design parameter across all records. The acceptance frequency is the ratio of the number of times the user selected that parameter in the historical generation records to the total number of times the parameter was selected. If the acceptance frequency exceeds a preset threshold, a user preference correction is performed, replacing the dimension in the initial parameter set with the corrected user preference dimension. Specifically, the system automatically identifies the currently logged-in user's historical generation records and extracts the design parameters involved. Subsequently, the user acceptance frequency is calculated for each parameter type. The acceptance frequency is defined as the ratio of the number of times the user selected a parameter in the historical records to the total number of times that parameter type was selected. For example, if a user selected a certain font type seven out of ten generation attempts, the acceptance frequency for that font type is 0.7. The system presets a threshold for the acceptance frequency. When the acceptance frequency for a parameter exceeds this threshold, it indicates that the parameter fully represents the user's stable preference. At this point, the system replaces the dimension of that parameter type in the initial design parameter set with the user preference parameter, thereby generating a revised design parameter set. This process ensures that the generated results not only retain the global rationality of the large model and optimization algorithm, but also incorporate the user's long-term personalized preferences, achieving an organic combination of model output and user aesthetic habits, and improving the practicality and satisfaction of the final design solution.

[0045] Finally, based on the revised design parameter set, the generated design is converted into an image file. A size template is automatically adapted based on the promotional product type, and the final image is output to the front-end. Specifically, after obtaining the revised design parameter set, the system uses it as the core input to drive the automatic generation of promotional product design plans. First, based on the design parameters, the system invokes the design generation engine to convert the abstract parameters into a concrete design plan. This engine automatically renders the color scheme, layout distribution, and element arrangement by combining rule templates with a graphics rendering model, forming a complete visual composition for the promotional product. Subsequently, based on the promotional product type identified in the user's request (such as posters, flyers, social media images, etc.), the system automatically adapts the corresponding size template to ensure that the generated design meets the usage specifications for that type in terms of resolution, proportions, and margins. Finally, the system renders the generated design into a standardized image file (such as PNG or JPEG format) and outputs it to the front-end interface for direct user preview and download. This process forms an automated closed-loop from parameter to image, ensuring both efficient design generation and the practicality and ready-to-use of the final promotional product.

[0046] The above description is merely a preferred embodiment of the present invention and does not constitute any other form of limitation to the present invention. Any person skilled in the art may utilize the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes for application in other fields. However, any simple modification, equivalent change, and modification of the above embodiments made in accordance with the technical essence of the present invention without departing from the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. An intelligent design method for graphic and text promotional materials based on a large model and a knowledge base, characterized by: The following steps are involved: The front end receives the text description of the graphic promotional material input by the user, pre-processes the text description, and outputs a structured demand text. The pre-processing includes grammatical error correction and intent recognition. Grammar correction is used to correct ambiguities and typos and determine the user's true requirements; Intent recognition is used to extract core demand types and clarify the intent information of graphic and text promotional materials; Input the structured requirement text into the promotional product design domain adaptation model and output the initial design parameter set; The initial design parameter set includes a color system, layout structure, font type, and a list of visual elements; Performing user preference correction on the initial design parameter set and outputting a corrected design parameter set; Based on the modified design parameter set, the generated design solution is converted into an image file, the size template is automatically adapted based on the type of promotional product, and the final image is output to the front end.

2. The intelligent design method for graphic and text promotional materials based on a large model and a knowledge base according to claim 1, characterized in that: Grammar correction is used to correct ambiguities and typos. The implementation of determining the user's true requirements specifically includes: First, we collect frequently occurring error cases in promotional product requirement descriptions to construct a domain error statement set. We then match each error statement with a unique corresponding standard statement in the promotional product design domain. Combining the domain error statement set with the corresponding standard statement creates a corpus. Based on the edit distance algorithm combined with the domain corpus, the surface errors of the requirement text are accurately identified and replaced, and the text to be corrected is defined. ,in is the number of text characters, and a set of candidate correct text segments is extracted from the domain corpus ,in is the number of candidate segments, ,in is the number of characters in the kth candidate segment; For text to be corrected With each candidate segment The minimum edit distance is , where k represents the kth candidate segment, and the minimum edit distance is calculated as follows: ,in Indicates the first i characters of S and The minimum number of edits for the first j characters of the text to be corrected. When there are i characters to be corrected, but there are no characters to compare with the candidate correct text at this time, the first i characters of the text to be corrected need to be deleted, a total of i operations; when there are no characters in the text to be corrected, but there are j characters in the candidate correct text, the first j characters of the candidate correct text need to be inserted; when both texts have characters, it is necessary to select the one with the least number of operations from the three possibilities of deletion, insertion, and replacement / no operation. The deletion operation is to first calculate the distance between the first i-1 characters to be corrected and the first j candidate characters. , add a deletion operation; the insertion operation is that the text to be corrected lacks the jth character of the candidate text, which needs to be inserted and then matched, that is, first calculate the distance between the first i characters to be corrected and the first j-1 candidate characters , add an insert operation; replace / no operation is to determine whether the i-th character of the text to be corrected is consistent with the j-th character of the candidate text. If they are consistent, no operation is required. If they are inconsistent, replacement is required. Calculate the distance between the first i-1 characters to be corrected and the first j-1 candidate characters. , add 0 or 1 according to whether the characters are consistent, is the indicator function; Set the edit distance threshold, select the edit distance less than the threshold, and get the smallest candidate segment ,right Replace the wrong part in the text to get the corrected text with grammatical correction .

3. The intelligent design method for graphic and text promotional materials based on a large model and a knowledge base according to claim 1, characterized in that: Intent recognition is used to extract core demand types and clarify the intent information of graphic and text promotional materials. The specific implementation includes: First, build a core library, including promotional product type labels and type identification keywords; Corrected text for grammatical errors Perform extraction operations, extract type identification keywords and perform matching calculations with type identification keywords in the core library, calculate the matching degree, and screen candidate types with matching degrees greater than the set threshold; If the candidate type is unique, then this type is the core requirement type. If there are multiple candidate types, the core requirement type is selected according to the degree of matching. If there is no candidate type, a requirement supplement prompt is output.

4. The intelligent design method for graphic and text promotional materials based on a large model and a knowledge base according to claim 3, characterized in that: The operation of matching the type identification keywords extracted from the revised text with the type identification keywords in the core library is as follows: Generate keyword vectors for type recognition keywords in the core library and keyword vectors for recognition keywords in user-corrected texts through the preset pre-trained BERT model; Combine each type of keyword in the core library to obtain the keyword vector set of that type ,in is the number of types, For the number of keywords of each type, calculate the type center vector of each type ,in Keywords The importance weight of type q; Keyword vector set extracted from the revised text Calculate the demand aggregation vector of user keywords ,in The weight of the frequency of users mentioning this keyword; For each type q, first calculate and type center vector Cosine similarity of and Euclidean distance , the Euclidean distance Normalize the distance matching score ; Finally, the cosine similarity Matching score with distance Perform weighted fusion to obtain the final matching degree.

5. The intelligent design method for graphic and text promotional materials based on a large model and a knowledge base according to claim 1, characterized in that: The specific operations of inputting the structured requirement text into the promotional product design domain adaptation model and outputting the initial design parameter set are as follows: Input the pre-processed structured demand text into the text embedding layer of the large-scale model adapted for the promotional material design domain to generate the demand feature vector; The adaptation model generates candidate sets based on the demand feature vector and calculates the probability of each candidate for each parameter type matching the demand: ,in is the yth candidate of a certain type of parameter, is the eigenvector, is the model output layer function; For each type of parameter, the top three candidates with the highest probability are selected to form a parameter candidate set. The candidate set is then optimized with multiple objectives. The genetic algorithm is used to optimize the obtained promotional product design score, and the optimal parameters are selected and recorded as the initial design parameter set.

6. The intelligent design method for graphic and text promotional materials based on a large model and a knowledge base according to claim 1, characterized in that: The initial design parameter set is modified by user preference, and the implementation of outputting the modified design parameter set is: Identify the historical generation records of the currently logged-in user, and count the user's acceptance frequency of the design parameters in all records. The acceptance frequency is the ratio of the number of times the user selects the parameter in the process of historical record generation to the total number of selections. If the acceptance frequency is greater than the preset threshold, the user preference is corrected, and the dimension in the initial parameter set is replaced with the dimension parameter after the user preference correction.

Images