A method and system for generating advertising content based on large models

Abstract

The present application relates to the technical field of advertisement delivery, and in particular to a method and system for generating advertisement delivery content based on a large model. The method comprises the following steps: obtaining industrial data of the city; using a large model to model the industrial data of the city to generate an industry-advertisement type mapping matrix; performing dynamic priority sorting on the industry-advertisement type mapping matrix to generate an advertisement type recommendation list for the city; obtaining geographic information data of the city; performing advertisement screen annotation on the geographic information data of the city to obtain annotated advertisement screen location information; and performing hierarchical output of candidate sites on the annotated advertisement screen location information to obtain an advertisement site priority list. The present application improves the accuracy and dynamics of advertisement delivery content generation by combining a large model, geographic information analysis, real-time crowd feature extraction, and closed-loop optimization.

Description

Technical Field

[0001] This invention relates to the field of advertising delivery technology, and in particular to a method and system for generating advertising delivery content based on a large model. Background Technology

[0002] The background technology of advertising content generation methods has undergone several stages of development. Initially, advertising content was manually written, with companies relying on advertising designers and copywriters to create content manually. This lacked personalization and automation, and resulted in low productivity. With the development of information technology, especially the rise of big data and machine learning, the advertising industry gradually began to use data analysis to optimize campaign performance. In the 21st century, breakthroughs in deep learning technology ushered in a new era for advertising content generation. Technologies such as Natural Language Processing (NLP) and Generative Adversarial Networks (GANs) began to be widely used for the automated generation of advertising content. In recent years, large-model-based advertising content generation methods have not only been applied to copywriting but have also made breakthroughs in areas such as advertising visual design and video advertising creation. However, current traditional advertising methods often neglect the impact of geographical location on advertising effectiveness, and the selection of advertising venues often relies on human experience. Furthermore, many traditional advertising systems fail to dynamically analyze real-time audience characteristics, leading to a mismatch between advertising content and the real-time needs of the audience, resulting in low accuracy and dynamism in advertising content generation. Summary of the Invention

[0003] Therefore, it is necessary to provide a method and system for generating advertising content based on a large model to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a method for generating advertising content based on a large model is provided, the method comprising the following steps:

[0005] Step S1: Obtain industry data for the city; use a large model to model the industry correlation of the city's industry data and generate an industry-advertising type mapping matrix; dynamically prioritize advertising types in the industry-advertising type mapping matrix and generate a recommended list of advertising types for the city.

[0006] Step S2: Obtain the geographic information data of the city; mark the advertising screens on the geographic information data of the city to obtain the marked advertising screen location information; output the candidate site hierarchy based on the marked advertising screen location information to obtain the advertising site priority list.

[0007] Step S3: Extract real-time crowd features based on the geographic information data of the city, and generate an attention heatmap for the marked advertising screen location information according to the real-time crowd features to obtain a heatmap of the screen area attention distribution; inject physical constraints into the heatmap of the screen area attention distribution; construct a dynamic location optimizer according to the injected conditions to generate advertising element coordinate suggestion data;

[0008] Step S4: Based on the recommended list of ad types, the list of venue priorities, and the suggested coordinates of ad elements, combine ad content to generate an ad image and text combination scheme; perform a closed-loop effect evaluation and content weight adjustment on the ad image and text combination scheme to execute the ad content generation optimization operation.

[0009] This invention, through comprehensive analysis of industry data, ad type recommendations, and real-time audience characteristics, can accurately deliver ad content at appropriate times and locations, improving ad reach and conversion rates. Dynamic optimization of ad screen locations and prioritization of ad types not only make ad delivery more precise but also rationally optimize ad budgets, avoiding resource waste. Closed-loop effect evaluation and content weight adjustment allow for real-time adjustments to ad strategies, ensuring optimal ad delivery performance. Because ad delivery is more personalized and meets user needs, ad content and presentation are more attractive, enhancing user acceptance and engagement. By utilizing large-scale models to model the correlation of industry data, deep-seated patterns and regularities can be extracted from massive amounts of data, making ad delivery more scientific and intelligent. Therefore, this invention, by combining large-scale models, geographic information analysis, real-time audience feature extraction, and closed-loop optimization, improves the accuracy and dynamism of ad content generation.

[0010] Preferably, step S1 includes the following steps:

[0011] Step S11: Obtain industry data for the city;

[0012] Step S12: Use a large model to extract industry keywords and consumption scenarios from the industry data of the city, and model the industry relevance of industry keywords and consumption scenarios to generate an industry-advertising type mapping matrix;

[0013] Step S13: Perform seasonal demand analysis on the industry-advertising type mapping matrix to generate seasonal demand analysis data;

[0014] Step S14: Use seasonal demand analysis data to dynamically prioritize advertising types in the industry-advertising type mapping matrix and generate a recommended list of advertising types for the city.

[0015] This invention generates an industry-advertising type mapping matrix based on the correlation modeling of industry keywords and consumption scenarios. This precise correlation model helps advertising placement better match industry characteristics and target audience needs, avoiding mismatched advertising content. Seasonal demand analysis provides data support for the dynamic prioritization of advertising types, ensuring that advertising content is adjusted in real time according to different seasons and demand fluctuations. This maximizes advertising effectiveness and avoids wasting advertising budget during periods of low demand. By considering the impact of seasonal demand, advertising placement is not only based on static data analysis but also keeps up with market dynamics. The advertising type recommendation list is adjusted with seasonal changes, making advertising content more aligned with audience needs, improving user engagement and advertising effectiveness. Utilizing large-scale model analysis of industry data, combined with seasonal demand forecasting, makes advertising placement more scientific and predictable. Improved advertising effectiveness not only depends on the current market situation but also on the ability to predict future trends based on historical data, allowing for proactive adjustments.

[0016] Preferably, step S12 includes the following steps:

[0017] Step S121: Collect consumption scenario data from the industry data of the city to obtain consumption scenario data of the city;

[0018] Step S122: Extract industry keywords from the city's industry data based on the city's consumption scenario data to obtain the core industry keywords;

[0019] Step S123: Cluster the core keywords of the industry to generate the main industry categories of the city, and use text analysis technology to perform descriptive analysis on the consumption scenario data based on the main industry categories of the city to generate the core characteristics of the consumption scenario and consumer demand data;

[0020] Step S124: Utilize the large model to perform correlation calculations on the core features of consumption scenarios and consumer demand data to generate industry-consumption scenario related data; perform data matrixing on the industry-consumption scenario related data to generate an industry-advertising type mapping matrix.

[0021] This invention helps accurately identify and describe the main industrial characteristics of a city and their corresponding consumer demands by extracting core industry keywords and analyzing the core features of consumption scenarios. Through text analysis technology, it deeply analyzes the relationship between industries and consumption scenarios, providing more accurate foundational data for advertising placement. The generation of industry-consumption scenario correlation data uses a large-scale model for correlation calculation, making the relationship between industries and consumption scenarios more scientific. This in-depth analysis helps advertisers identify which ad types are more suitable for matching which consumption scenarios, improving ad relevance and targeting. The industry-ad type mapping matrix generated through data matrixing makes the advertising process more intelligent and automated. This matrix can serve as the core framework for advertising strategies, helping advertisers optimize placement based on industry characteristics, consumer needs, and scenario relevance. The core features of industries and consumption scenarios, along with consumer demand data, provide a clear basis for prioritizing ad types and selecting placement locations. By optimizing the precise matching of ads, ad content can better meet the actual needs of target users, improving ad engagement and conversion rates. Descriptive and cluster analyses of consumption scenario data help advertisers better understand consumer behavior and preferences in the city. By delving into the core characteristics of consumption scenarios, advertisers can design advertising content that better meets the needs of their audience, thereby enhancing the personalization and appeal of their ads.

[0022] Preferably, step S2 includes the following steps:

[0023] Step S21: Obtain geographic information data of the city;

[0024] Step S22: Label the location information of the advertising screens in the geographic information data of the city, and use the CV model to perform scene semantic modeling on the geographic information data of the city based on the labeled location information of the advertising screens, and generate a scene label system.

[0025] Step S23: Perform multi-dimensional matching calculations on the city's recommended ad type list and scene tag system, eliminate conflicting scenarios, and generate ad scene matching data;

[0026] Step S24: Based on the advertising scene matching data, perform candidate site classification and output the marked advertising screen location information to obtain the advertising site priority list.

[0027] This invention, through the annotation of advertising screen location information and semantic modeling of scene using CV models, can accurately associate geographic information data with scene features, providing a more detailed and intelligent spatial understanding for advertising placement. This modeling ensures that the location selection and displayed content of the advertising screen are highly compatible with the surrounding environment, thereby enhancing the visual appeal and audience attention of the advertisement. Multi-dimensional matching calculation matches the recommended list of advertising types with the scene tag system, automatically eliminating scenes that do not match or are unsuitable for the advertising type. This scene filtering mechanism effectively avoids mismatch of advertising content, ensuring the accuracy and efficiency of advertising placement. By tiering and outputting candidate sites based on the annotated advertising screen location information, a priority list of advertising sites is generated, allowing advertisers to rationally allocate advertising resources based on the importance of the site, exposure, and the degree of matching with the advertising type. This priority list not only improves advertising effectiveness but also optimizes the allocation of advertising budgets. The combination of the scene tag system and advertising type enables more precise design of advertising content based on different geographical regions and scene characteristics. For example, displaying promotional advertisements in commercial areas and placing transportation vehicle advertisements in transportation hubs increases the acceptance and conversion rate of the advertisements due to the high degree of compatibility between the advertisements and the environment. Through multi-dimensional matching, personalized ads can be designed based on the needs and contextual characteristics of consumers in different geographical regions. Contextualized ad display not only enhances ad interactivity but also improves ad relevance and user engagement.

[0028] Preferably, based on the labeled advertising screen location information, the use of a CV model to perform scene semantic modeling of the geographic information data of the city includes:

[0029] Extract planning features from the geographic information data of the city, and divide the geographic information data of the city into spatial regions to generate spatial region hierarchical data;

[0030] Based on the labeled advertising screen location information, the spatial area hierarchical data is used to locate the advertising screen in space and generate advertising screen spatial location data; the images of the advertising screen spatial location data are retrieved using the area camera, and the images are used to perform target detection through a CV model to generate advertising scene visual features;

[0031] Extract scene elements with visual features of advertising scenes, and combine them with a region generation network to perform scene semantic modeling of images of advertising screen spatial positioning data, thereby generating advertising screen environment scene modeling data;

[0032] Semantic association tags are assigned to the advertising screen environment scene modeling data to generate environment tag association data; the environment tag association data and spatial region hierarchical data are used to construct a scene tag system for the advertising screen environment scene modeling data to obtain the scene tag system.

[0033] This invention enables accurate analysis of the location and surrounding geographical environment of advertising screens through the generation of spatial regional hierarchical data and advertising screen spatial positioning data. By extracting planning features and dividing spatial hierarchy into urban geographic information, advertisers can better understand the specific location of the advertising screen and its relative position in the urban space, thus providing a clear framework for advertising placement selection. The extraction of visual features of the advertising scene, combined with a target detection model, can identify specific environmental features around the advertising screen (such as traffic flow, building type, and pedestrian density), thereby providing detailed data on the effectiveness of advertising in that environment. This process ensures a high degree of fit between the advertising content and the surrounding environment, improving the relevance of the advertisement and audience attention. Through scene element extraction and region generation networks, the environmental features of the advertising screen are accurately transformed into data usable for decision-making, further establishing advertising screen environmental scene modeling data. This modeling not only considers geographical location but also covers the visual features of the scene and actual usage, making advertising placement decisions more comprehensive and accurate. By leveraging semantic association tagging, specific semantic labels (such as commercial areas, residential areas, transportation hubs, etc.) can be provided for advertising scenario modeling data. These labels not only help advertisers understand the characteristics of different scenarios but also enable targeted optimization of ad content and delivery strategies. Through the tagging system, advertisers can identify the specific needs of each ad screen's environment, further enhancing the personalization and customization of their ads. The construction of a scenario tagging system provides a powerful tool to help adapt ad content to different scenarios. For example, placing promotional ads in commercial areas and cultural ads in educational areas allows for precise matching, effectively improving ad acceptance and effectiveness while avoiding ineffective exposure.

[0034] Preferably, step S3, which involves extracting real-time crowd features based on the geographic information data of the city and generating an attention heatmap for the marked advertising screen location information based on these real-time crowd features, includes:

[0035] Retrieve area cameras based on geographic information data of the city;

[0036] The system filters environmental images from the area's cameras to obtain regional environmental images; it then identifies human silhouettes within these images and performs crowd attribute analysis on these silhouettes to generate real-time crowd characteristics.

[0037] Based on real-time crowd characteristics, the location information of the labeled advertising screen is analyzed for viewing distance to generate viewing distance data; a pre-trained attention prediction model is used to perform eye tracking prediction on the viewing distance data to generate eye tracking prediction data.

[0038] The eye-tracking prediction data is transformed into an attention heatmap to generate a heatmap of the screen area attention distribution.

[0039] This invention efficiently acquires crowd characteristic data by accessing area cameras and combining them with real-time image processing technology. Human contour recognition and crowd attribute analysis enable the system to obtain detailed information about the crowd in real time, such as number of people, age group, gender, and behavioral habits, thus providing advertisers with more accurate audience analysis. Real-time crowd characteristic data can directly influence the precise matching of advertising content. By analyzing advertising screen location information and crowd behavior, it ensures that the content of the advertisement is highly aligned with the needs of the surrounding audience. For example, advertisements can be personalized based on the age, interests, or consumption habits of the audience, improving the acceptance and effectiveness of the advertisement. Based on viewing distance analysis, it helps advertisers more accurately understand the exposure effect of different advertising screens in a specific area. This analysis reveals the distance, viewing angle, and viewing distance range between the advertising screen and the target audience, making the selection of advertising screen positions and the presentation of advertising content more consistent with the actual viewing distance of the audience, thereby avoiding advertising waste. Using an eye-tracking prediction model, advertisers can predict the area of ​​focus for the audience in a specific location and environment. Eye tracking helps identify viewer attention focal points, revealing the visual path and areas of focus when watching ads. This provides data support for ad content design, ensuring key ad elements are placed in areas where the audience is most likely to focus. Attention heatmaps transform eye tracking data into visual images, allowing advertisers to intuitively understand how viewers' attention is distributed on the ad screen. These heatmaps allow advertisers to see viewer attention hotspots, optimizing ad layout and visual design to maximize appeal and audience engagement.

[0040] Preferably, the physical constraint injection for the screen area attention distribution heatmap in step S3 includes:

[0041] The advertising screen location information is collected to obtain advertising screen information parameters; the screen resolution and pixel density are analyzed from the advertising screen information parameters to obtain the advertising screen resolution and pixel density.

[0042] The initial advertising boundary range is set based on the screen resolution and pixel density of the advertising screen, and the viewing distance range is calculated based on the set initial advertising boundary range to obtain the visible range of the advertising screen.

[0043] By applying a minimum font size physical constraint to the visible area of ​​the advertising screen using a heatmap of screen area attention distribution, the content layout safety boundary constraint is obtained. The calculation formula for the minimum font size physical constraint is shown below:

[0044]

[0045] In the formula, B(x) represents the content layout safety boundary constraint, and Aunsafe (x) represents the area of ​​the advertisement content that extends beyond the visible screen area, A total Let x be the total area of ​​the screen, and x be the input parameter.

[0046] This invention, through analysis of the resolution and pixel density of advertising screens, provides advertisers with a precise display range, ensuring that advertising content does not exceed the actual visible area of ​​the screen. This analysis prevents the loss of information due to the content exceeding the visible range, thereby optimizing the overall advertising effect. Based on the screen resolution and pixel density, a preliminary advertising boundary range can be set under actual display conditions. This setting helps advertisers avoid layout issues where content is too large or too small, ensuring consistency and standardization across different devices. Calculation of viewing distance and minimum font size limits ensure that advertising content is not only clearly visible within the visual range but also meets the requirements of actual viewing distance and screen size. By accurately calculating the minimum font size, the advertising content will not negatively impact the viewer's reading experience due to excessively small fonts, improving readability and attractiveness. The injection of physical constraints, especially the minimum font size limit, effectively prevents excessively small fonts, ensuring that viewers can clearly see the advertising information at normal viewing distances. By ensuring readability and visibility, the user experience of advertising is significantly improved, enhancing viewer attention to the advertising content. By calculating the parameters and layout safety boundaries of advertising screens, it is possible to ensure that the advertising content design across different advertising screens meets consistent standards and specifications. Regardless of screen size or display environment, advertising content can adhere to the same set of physical and design rules, thereby improving the uniformity of advertising delivery.

[0047] Preferably, the construction of the dynamic position optimizer based on the injection conditions in step S3 includes:

[0048] A multi-objective optimization parameter is constructed based on the content layout security boundary constraints. The multi-objective optimization parameter includes the minimum exposure rate and the minimum information interference value. The formula for calculating the minimum exposure rate is as follows:

[0049]

[0050] In the formula, E(x) is the minimum exposure rate, and A i Let A be the display area of ​​the advertisement in the i-th region. total V represents the total area of ​​the screen. x Let be the exposure of the advertisement in different viewpoints in the i-th region, n be the number of advertisement regions, and x be the input parameter;

[0051] The formula for calculating the minimum information interference value is as follows:

[0052]

[0053] In the formula, D(x) is the minimum information interference value, and L j (x) represents the brightness of the advertisement under ambient lighting condition j, C j (x) represents the visible contrast of information under ambient lighting condition j, L max and C max These are the maximum brightness and maximum contrast of the screen under given conditions, respectively, where m is the number of different environmental conditions and x is the input parameter.

[0054] A multi-objective optimization function is constructed based on the constraints of minimum exposure rate, minimum information interference value, and content layout safety boundary conditions. The formula of the multi-objective optimization function is shown below:

[0055] f(x)=w1·E(x)+w2·D(x)+w3·B(x);

[0056] In the formula, f(x) is the multi-objective optimization function, E(x) is the minimum exposure rate, D(x) is the minimum information interference value, B(x) is the content layout safety boundary constraint, w1 is the minimum exposure rate weight, w2 is the minimum information interference weight, w3 is the layout safety boundary constraint weight, and x is the input parameter.

[0057] By using a multi-objective optimization function to dynamically optimize the position of ad elements on the heatmap of screen area attention distribution, suggested coordinate data for ad elements is generated.

[0058] This invention optimizes the display position of advertising elements through minimum exposure rate calculation, ensuring sufficient exposure for the advertisement in all areas. This process maximizes ad visibility, enhances ad dissemination, and ensures more viewers see the ad content, thereby increasing the ad's effectiveness and influence. The calculation of minimum information interference value helps analyze ad brightness and contrast under different environmental conditions, ensuring clear visibility under various lighting and contrast conditions. By reducing visual interference with the surrounding environment, the effectiveness of the ad content is improved, focusing viewer attention on the ad content itself rather than external distractions. By considering content layout safety boundary constraints, the invention ensures the ad does not exceed the screen's visible range, avoiding layout issues that affect the ad's display effect. This constraint ensures the ad is always correctly and clearly displayed on all devices and in all environments, improving the user's visual experience and the ad's appeal. The construction of a dynamic position optimizer allows ad content to adapt to different viewing angles and exposure conditions, ensuring optimal display effects under different environments and screen angles. Through this optimization, the ad can dynamically adapt to various viewing conditions, enhancing its performance in different scenarios. By using a multi-objective optimization function that comprehensively considers ad exposure, information interference, and layout security, it can generate precise ad element placement suggestions. This precise layout not only optimizes the ad display effect but also avoids invalid or overcrowded ad element arrangements, ensuring that the ad content displays its value to the greatest extent.

[0059] Preferably, step S4 includes the following steps:

[0060] Step S41: Based on the ad type recommendation list, venue priority list, and ad element coordinate suggestions, combine ad content to generate an ad image and text combination scheme;

[0061] Step S42: Conduct real-time click-through rate (CTR) testing on the ad image and text combination scheme to obtain the real-time CTR test results; perform cross-scenario style migration on the ad image and text combination scheme based on the real-time CTR test results to generate ad style migration data;

[0062] Step S43: Use advertising style migration data to conduct a closed-loop effect evaluation of the advertising image and text combination scheme, generate advertising performance evaluation data, and adjust the content weight of the advertising image and text combination scheme based on the advertising performance evaluation data to perform advertising content generation optimization.

[0063] This invention, by combining a recommendation list of ad types, a priority list of venues, and suggested coordinates for ad elements, enables more precise optimization of ad text and image combinations for different venues and target user groups. This process ensures a high degree of alignment between ad content and audience needs, enhancing ad appeal and relevance. Real-time click-through rate (CTR) testing provides immediate feedback for ad placement, helping advertisers adjust ad content promptly based on audience click behavior. This process makes ad placement strategies more flexible and dynamic, effectively responding to changes in user preferences and behaviors. Through cross-scenario style transfer of ad text and image combinations, ads can adapt to the display needs of different scenarios. Whether on outdoor screens, mobile devices, or other types of advertising media, ad content can adaptively adjust to environmental changes, thereby improving the universality and effectiveness of the ads. Ad style transfer optimizes the visual style of ads based on different display environments and audience needs. Through style transfer, ad content can retain core information while adjusting the layout, color, and style of text and images to better attract the attention of the target audience. Closed-loop performance evaluation provides precise feedback on advertising effectiveness by assessing the actual performance after ad campaigns are launched. This evaluation helps advertisers understand the actual conversion results of their ads, including key metrics such as click-through rate, engagement, and brand awareness, and further optimize the effectiveness of their ad campaigns.

[0064] This specification provides a system for generating ad content based on a large model, used to execute the aforementioned method for generating ad content based on a large model. The system for generating ad content based on a large model includes:

[0065] The ad type recommendation module is used to acquire industry data of the city; use a large model to model the industry correlation of the city's industry data to generate an industry-ad type mapping matrix; and dynamically prioritize ad types in the industry-ad type mapping matrix to generate an ad type recommendation list for the city.

[0066] The advertising venue recommendation module is used to obtain the geographic information data of the city; to mark the advertising screens in the geographic information data of the city, and to obtain the location information of the marked advertising screens; and to output the candidate venues in a hierarchical manner based on the location information of the marked advertising screens, and to obtain the priority list of advertising venues.

[0067] The ad content recommendation module is used to extract real-time audience features based on the geographic information data of the city, and generate an attention heatmap for the marked ad screen location information based on the real-time audience features to obtain a heatmap of the attention distribution of the screen area; inject physical constraints into the heatmap of the attention distribution of the screen area; and build a dynamic location optimizer based on the injected conditions to generate suggested coordinate data for ad elements.

[0068] The ad generation module is used to combine ad content based on the ad type recommendation list, the venue priority list, and the ad element coordinate suggestions to generate ad image and text combination schemes; and to perform closed-loop effect evaluation and content weight adjustment of the ad image and text combination schemes to perform ad content generation optimization work.

[0069] The beneficial effects of this invention lie in the following: The ad type recommendation module utilizes a large model to model industry relevance, accurately generating an industry-ad type mapping matrix based on the city's industry data. This process ensures a high degree of match between ad content and industry needs, improving ad relevance and effectiveness. Dynamic priority ranking of ad types can be adjusted in real-time based on seasonality or market demand changes, thereby enhancing ad placement effectiveness. The ad location recommendation module acquires city geographic information and marks ad screens to generate a location priority list, ensuring ad placement is in areas with the highest potential and exposure opportunities. This module provides advertisers with multiple location options through tiered candidate location output, helping to select the best placement location and maximize ad exposure and influence. The ad content recommendation module, through real-time audience feature extraction and attention heatmap generation, provides precise ad placement suggestions based on the audience's attention level at the ad screen's location. By combining audience characteristics and viewing distance analysis, the display position and method of ad content can be further optimized, ensuring that ads attract the target audience's attention to the maximum extent. The ad generation module automatically generates the optimal ad image and text combination scheme based on the ad type recommendation list, location priority list, and ad element coordinate suggestions. Through closed-loop effect evaluation and content weight adjustment, the system can dynamically adjust content based on real-time feedback data of advertisements, thereby continuously optimizing ad performance. By evaluating the real-time effects of ad image and text combinations, the system can promptly adjust ad content based on user click behavior, ad exposure, and interaction rates. This closed-loop optimization process ensures that ad content always maintains efficient and accurate performance, avoiding ineffective ad placement and resource waste. Therefore, this invention, by combining large-scale models, geographic information analysis, real-time audience feature extraction, and closed-loop optimization, improves the accuracy and dynamism of ad content generation. Attached Figure Description

[0070] Figure 1 A flowchart illustrating the steps of a method for generating ad content based on a large model;

[0071] Figure 2 for Figure 1 A detailed flowchart illustrating the implementation steps of step S2.

[0072] Figure 3 for Figure 1 A detailed flowchart illustrating the implementation steps of step S4.

[0073] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0074] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0075] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0076] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0077] To achieve the above objectives, please refer to Figures 1 to 3 A method for generating ad content based on a large model, the method comprising the following steps:

[0078] Step S1: Obtain industry data for the city; use a large model to model the industry correlation of the city's industry data and generate an industry-advertising type mapping matrix; dynamically prioritize advertising types in the industry-advertising type mapping matrix and generate a recommended list of advertising types for the city.

[0079] Step S2: Obtain the geographic information data of the city; mark the advertising screens on the geographic information data of the city to obtain the marked advertising screen location information; output the candidate site hierarchy based on the marked advertising screen location information to obtain the advertising site priority list.

[0080] Step S3: Extract real-time crowd features based on the geographic information data of the city, and generate an attention heatmap for the marked advertising screen location information according to the real-time crowd features to obtain a heatmap of the screen area attention distribution; inject physical constraints into the heatmap of the screen area attention distribution; construct a dynamic location optimizer according to the injected conditions to generate advertising element coordinate suggestion data;

[0081] Step S4: Based on the recommended list of ad types, the list of venue priorities, and the suggested coordinates of ad elements, combine ad content to generate an ad image and text combination scheme; perform a closed-loop effect evaluation and content weight adjustment on the ad image and text combination scheme to execute the ad content generation optimization operation.

[0082] This invention, through comprehensive analysis of industry data, ad type recommendations, and real-time audience characteristics, can accurately deliver ad content at appropriate times and locations, improving ad reach and conversion rates. Dynamic optimization of ad screen locations and prioritization of ad types not only make ad delivery more precise but also rationally optimize ad budgets, avoiding resource waste. Closed-loop effect evaluation and content weight adjustment allow for real-time adjustments to ad strategies, ensuring optimal ad delivery performance. Because ad delivery is more personalized and meets user needs, ad content and presentation are more attractive, enhancing user acceptance and engagement. By utilizing large-scale models to model the correlation of industry data, deep-seated patterns and regularities can be extracted from massive amounts of data, making ad delivery more scientific and intelligent. Therefore, this invention, by combining large-scale models, geographic information analysis, real-time audience feature extraction, and closed-loop optimization, improves the accuracy and dynamism of ad content generation.

[0083] In this embodiment of the invention, reference is made to Figure 1 The diagram illustrates the steps of a method for generating advertising content based on a large model according to the present invention. In this example, the method for generating advertising content based on a large model includes the following steps:

[0084] Step S1: Obtain industry data for the city; use a large model to model the industry correlation of the city's industry data and generate an industry-advertising type mapping matrix; dynamically prioritize advertising types in the industry-advertising type mapping matrix and generate a recommended list of advertising types for the city.

[0085] In this embodiment of the invention, various industry data of the city are obtained through urban economics, industrial parks, industry associations, etc. This data may include industry classification, industry size, major enterprises, development trends, etc. The collected data is processed to remove noise and standardize data from different sources to ensure consistent formatting for further analysis. A large pre-trained model suitable for analyzing industry correlations (e.g., deep learning-based graph neural networks, BERT, GPT, etc.) is selected and fine-tuned based on the industry data. The cleaned industry data is input, allowing the large model to learn the correlations between different industries; for example, whether the growth of one industry promotes the development of related industries (such as advertising, manufacturing, etc.). The model outputs the correlations between industries and quantifies the correlation of each pair of industries into numerical values. Advertising types that can be used for advertising are defined, such as video ads, social media ads, search engine ads, etc. Based on the industry correlation results, highly correlated advertising types are defined for each industry. An "industry-advertising type mapping matrix" is generated based on the industry-advertising type relationship, where each element represents the correlation between a certain industry and a certain advertising type. Based on performance evaluation metrics for ad types (such as click-through rate, conversion rate, and market demand), ad types are dynamically prioritized. Using dynamic ranking algorithms (such as weighted ranking and machine learning ranking algorithms), combined with real-time changes in industry and ad types, an ad type priority is generated that adjusts according to the characteristics of the city and current trends. Based on the mapping matrix between industry and ad types, and the dynamic priority ranking of ad types, a recommended list of ad types is generated using collaborative filtering or content recommendation algorithms.

[0086] Step S2: Obtain the geographic information data of the city; mark the advertising screens on the geographic information data of the city to obtain the marked advertising screen location information; output the candidate site hierarchy based on the marked advertising screen location information to obtain the advertising site priority list.

[0087] In this embodiment of the invention, detailed geographic information data of the city is acquired, including but not limited to city maps, geographic coordinates, building information, road networks, population density, and traffic flow. Data collected from different sources is formatted and standardized to ensure data consistency and usability. Location data of all advertising screens (including LED screens, digital billboards, bus stop advertising screens, etc.) in the city is collected. Advertising screens can be obtained from databases provided by city management systems or existing advertising platforms. Detailed annotations are made based on the physical location, size, shape, and display characteristics of each advertising screen. The location of each advertising screen can be described using latitude and longitude coordinates, surrounding geographic environment (such as transportation hubs, commercial areas, etc.), and other relevant information. Annotation rules are defined to ensure the accuracy of the advertising screen data. For example, a standard coordinate system is used for precise calibration of the advertising screens, while adding environmental information related to the advertising screen (such as whether it is located in a bustling commercial area, exposure during peak traffic hours, etc.). Data tables or databases are established to organize the annotation information of each advertising screen (such as advertising screen number, latitude and longitude, size, display range, surrounding traffic, environment, etc.) into a structured data format. Based on the labeled information, an analysis of the advertising screen location distribution is conducted to identify the density, distribution characteristics, and distance relationships of advertising screens in different areas to important geographical locations (such as shopping malls and transportation hubs). Selection criteria for candidate sites are defined, such as traffic flow, population density, commercial value, and time-of-day exposure. Based on these criteria, a multi-dimensional weighted scoring system (such as linear weighting or a machine learning-based scoring model) is used to score the location information of all advertising screens and classify them into different priority categories. For example, they can be divided into "high priority," "medium priority," and "low priority." A priority list of advertising sites is output based on the scoring results for use in advertising placement decisions. Candidate sites are prioritized based on the scoring and classification results of the advertising screens. Different advertising needs (such as brand exposure, promotional activities, etc.) can be considered to assign appropriate priorities to different sites.

[0088] Step S3: Extract real-time crowd features based on the geographic information data of the city, and generate an attention heatmap for the marked advertising screen location information according to the real-time crowd features to obtain a heatmap of the screen area attention distribution; inject physical constraints into the heatmap of the screen area attention distribution; construct a dynamic location optimizer according to the injected conditions to generate advertising element coordinate suggestion data;

[0089] In this embodiment of the invention, real-time crowd distribution is collected through smart surveillance cameras, public transportation systems, and location data (such as mobile phone positioning and Wi-Fi hotspot data) in the city. Geotag content on social media platforms (such as Twitter and Instagram geotags) is analyzed to aid in the extraction of crowd characteristics. Traffic flow information is obtained through traffic management systems, which can indirectly infer the characteristics of people in densely populated areas. Crowd density characteristics for different time periods and geographical areas are extracted based on the crowd distribution data. Regional crowd distribution maps are generated using algorithms such as heatmap analysis (e.g., Gaussian kernel-based density estimation). Crowd behavior patterns, including peak activity times, shopping hotspots, and workplace distribution, are analyzed by combining geographical data and sensor information. Demographic data (such as age, gender, and occupation) within the region are obtained by combining geographical information to further refine crowd characteristics. Changes in crowd distribution over a future period are predicted using time series analysis and predictive modeling methods to ensure that advertising decisions are based on real-time and future predictive data. Based on real-time crowd characteristic data and advertising screen location information, the crowd density and activity patterns in the area surrounding each advertising screen are weighted to generate an attention heatmap for the advertising screen. Considering factors such as crowd density, time of day, activity patterns, and traffic flow, a heatmap is generated for the coverage area of ​​each advertising screen, representing the distribution of crowd attention. Multi-dimensional temporal and spatial modeling of the heatmap is performed to create a time-based dynamic heatmap, adapting to different time periods and generating a multi-dimensional heatmap including information on crowd density, age distribution, gender distribution, etc., to analyze differences in attention to the advertising screens among different groups. Based on the physical characteristics and geographical location of the advertising screens, a series of constraints are designed: considering the viewing distance and angle of the advertising screens (e.g., some advertising screens are located at road intersections or building corners with limited viewing distance); combining the actual display size and positioning of the advertising screens to ensure that the generated heatmap matches the physical display area of ​​the screen, avoiding excessive attention to content outside the display area; and ensuring that the attention distribution reflected by the heatmap is consistent with the actual flow of people (e.g., preventing advertising screens in high-traffic areas from being ignored). These physical constraints are incorporated into the heatmap data to ensure that the heatmap not only accurately reflects crowd distribution but also takes into account the practical limitations of the advertising screens (such as viewing angle and distance limitations). Construct an optimization objective function to ensure that the display position of ad elements maximizes the distribution of audience attention. The objective function can consider factors such as audience density, time period, and attention heatmaps. Combine physical constraints with the optimization objective to form a comprehensive optimization model. This model will balance audience attention with the physical characteristics of the ad screen to maximize advertising effectiveness. Select a suitable algorithm based on the optimization objective, such as genetic algorithm, particle swarm optimization (PSO), or simulated annealing, to dynamically adjust the position of ad elements. Adjust the optimizer parameters based on real-time audience data feedback to adapt to real-time changes in ad delivery.For example, when the attention heatmap of a particular ad screen location changes, the optimizer can update the display position of the ad elements in real time. Multiple objectives are considered during position optimization, such as maximizing exposure, maximizing audience engagement, and ensuring that physical conditions (such as viewing distance and angle) are met. Based on the results of the optimization process, the optimizer outputs optimal ad element coordinate suggestions. This suggested data can be used as input to the ad delivery system to precisely guide the selection of ad content placement. The optimizer dynamically adjusts the ad element placement recommendations according to different scenarios (such as traffic changes at different times and behavioral characteristics in different regions) to ensure the best delivery results.

[0090] Step S4: Based on the recommended list of ad types, the list of venue priorities, and the suggested coordinates of ad elements, combine ad content to generate an ad image and text combination scheme; perform a closed-loop effect evaluation and content weight adjustment on the ad image and text combination scheme to execute the ad content generation optimization operation.

[0091] In this embodiment of the invention, by combining the ad types (such as promotions, brand promotions, event promotions, etc.) in the ad type recommendation list with the priority of the ad screens in the venue priority list, a suitable ad type is selected and matched with the corresponding ad screen. Different priorities are set for different types of ads, prioritizing brand promotion ads in high-exposure, high-traffic areas, while promotion ads can be placed in areas with high traffic during specific time periods. The priority of the ad screen is combined with optimized ad element coordinate suggestions to ensure that ad elements can be accurately placed in the corresponding ad screen positions. Considering the physical constraints of the ad screens (such as display size, viewing distance, etc.), the layout and size of the ad content are adjusted to adapt to the space of each ad screen and ensure optimal visual effects. Based on the matched ad type, venue priority, and coordinate suggestions, an ad graphic and text combination scheme is generated. This scheme includes: designing graphic and text content according to the theme of the ad (brand, promotion, etc.), and the design of the graphic and text must meet the display requirements of the ad screen (such as size, display duration, resolution, etc.). Determining the specific time for ad placement, including time period, frequency, and cycle, to optimize ad exposure timing. Designing the layout of the ad on the screen, such as the ratio of image to text, font size, and color scheme. Utilizing monitoring tools and sensor data, the system tracks ad performance in real time, monitoring the performance of ad image and text combinations through feedback mechanisms (such as viewer dwell time and triggered behaviors). A / B testing is conducted, deploying different ad image and text combinations to evaluate their performance and select the optimal one. Real-time data feedback (such as performance evaluation metrics) is fed back to the advertising system for adjustments and optimization. Based on the analysis results, the system can adjust ad delivery strategies and content to continuously improve ad effectiveness. An ad content weighting model is constructed based on factors such as ad type, ad screen location, time period, and audience. This model can dynamically adjust content weights based on machine learning algorithms (such as weighted regression and support vector machines) or rule engines. Initial weights are assigned to each element (image, text, animation, etc.) in the ad image and text combination. These weights are dynamically adjusted based on the audience characteristics of the ad screen and ad performance evaluation results. For example, visual elements of the ad content are given higher weights during high-traffic periods. The weights of the ad image and text content are fine-tuned based on the ad performance evaluation results. For example, if a certain type of ad image and text combination performs well during a specific time period, the weight of that type of content can be increased, and the design and delivery strategy of the ad content can be optimized.

[0092] Preferably, step S1 includes the following steps:

[0093] Step S11: Obtain industry data for the city;

[0094] Step S12: Use a large model to extract industry keywords and consumption scenarios from the industry data of the city, and model the industry relevance of industry keywords and consumption scenarios to generate an industry-advertising type mapping matrix;

[0095] Step S13: Perform seasonal demand analysis on the industry-advertising type mapping matrix to generate seasonal demand analysis data;

[0096] Step S14: Use seasonal demand analysis data to dynamically prioritize advertising types in the industry-advertising type mapping matrix and generate a recommended list of advertising types for the city.

[0097] In the embodiments of the present invention, industrial data of the city is collected by using public data sources, industry reports, economic data released by departments, etc. These data include information such as the production scale, growth trend, market demand, and industrial distribution of various industries. Preprocessing operations such as cleaning, deduplication, and missing value filling are performed on the collected industrial data to ensure the accuracy and integrity of the data. Data processing tools (such as Pandas in Python, SQL databases, etc.) are used for data cleaning and standardization to ensure the unified format of various data. Through natural language processing techniques (such as TF-IDF, Word2Vec or BERT models), keyword extraction is performed on text data related to industries (such as industry reports, news, social media comments, etc.) to extract representative industrial keywords. These industrial keywords can include industry names, product categories, consumption trends, market hotspots, etc. Based on information such as consumer behavior, purchase patterns, and consumption places in the industrial data, combined with scenario modeling techniques (such as scenario perception models, user behavior analysis), consumption scenarios are extracted. These scenarios include online / offline shopping, holiday consumption, shopping behavior in specific regions, etc. A large model (such as a deep learning model, clustering algorithm, etc.) is used to model the correlation between industrial keywords and consumption scenarios. By calculating the correlation between each industrial keyword and different consumption scenarios, an industrial-consumption scenario correlation matrix is generated. The model can be trained based on statistical methods (such as collaborative filtering, similarity analysis) or deep learning methods (such as neural networks, graph neural networks, etc.) to further improve the accuracy of the model. According to the results of the extraction of industrial keywords and consumption scenarios, combined with the advertising types (such as brand advertising, promotional advertising, event advertising, etc.), an industrial-advertising type mapping matrix is generated. This matrix reflects the correlation between different industries and advertising types, and can help understand which advertising types have a stronger match with which industries or consumption scenarios. Combining factors such as historical sales data, seasonal commodity demand data, and climate change, the seasonal demand characteristics of advertising types are analyzed. For example, certain advertising types have higher demand during holidays or seasonal changes. Time series analysis (such as ARIMA, LSTM, etc.) or seasonal decomposition methods (such as STL decomposition) are used for seasonal demand analysis. Machine learning algorithms (such as clustering analysis, trend analysis) are used to identify the seasonal demand patterns of different industrial-advertising types. Through data mining techniques, the demand fluctuation rules of each advertising type throughout the year are found. Based on the seasonal demand analysis, specific seasonal demand analysis data is generated. This data will include the demand intensity and fluctuations of each advertising type in different seasons, helping decision-makers understand the best placement timing of different advertising types. Based on the seasonal demand analysis data, the priorities of advertising types in the industrial-advertising type mapping matrix are adjusted. For advertising types with strong demand, higher priorities are assigned; for advertising types with weak demand, their priorities are reduced.A weighted algorithm can be used to dynamically prioritize each ad type based on factors such as seasonal demand intensity and the match between ad type and industry. Based on the adjusted ad type priorities, a recommended ad type list is generated for the city. This list will then dynamically generate the most suitable ad type delivery order based on different seasons, industry needs, and ad performance.

[0098] Preferably, step S12 includes the following steps:

[0099] Step S121: Collect consumption scenario data from the industry data of the city to obtain consumption scenario data of the city;

[0100] Step S122: Extract industry keywords from the city's industry data based on the city's consumption scenario data to obtain the core industry keywords;

[0101] Step S123: Cluster the core keywords of the industry to generate the main industry categories of the city, and use text analysis technology to perform descriptive analysis on the consumption scenario data based on the main industry categories of the city to generate the core characteristics of the consumption scenario and consumer demand data;

[0102] Step S124: Utilize the large model to perform correlation calculations on the core features of consumption scenarios and consumer demand data to generate industry-consumption scenario related data; perform data matrixing on the industry-consumption scenario related data to generate an industry-advertising type mapping matrix.

[0103] In this embodiment of the invention, data related to consumption scenarios is collected through e-commerce platforms (such as Taobao and JD.com), offline shopping mall data, industry reports, and social media (such as Weibo and Douyin). This data includes consumer shopping behavior, activity participation, and holiday spending. Consumption scenarios can be categorized into multiple types, such as online shopping, offline shopping, and specific events (such as festival promotions and Black Friday shopping festivals). Based on the characteristics of different industries, the content and characteristics of each consumption scenario are clearly defined. Text analysis tools (such as natural language processing technology) are used to annotate the collected consumption scenario data, extracting industry-related consumption scenarios. Consumption scenarios are classified, establishing a standard classification system. For example: e-commerce shopping, offline experiences, festival promotions, etc. The collected and annotated consumption scenario data is organized into a structured data format for subsequent analysis and processing. Data processing tools (such as Pandas and SQL databases) are used to clean, deduplicate, and standardize the data to ensure data quality. Text analysis is performed on the city's industry data, using Natural Language Processing (NLP) techniques (such as TF-IDF, Word2Vec, BERT, etc.) to extract keywords from the industry data. These keywords can include industry names, product types, market demands, etc. Based on the city's industry background and market dynamics, keywords representing the core characteristics of the industries are extracted. These keywords involve industry sectors, technological trends, and consumption trends. Clustering algorithms (such as K-means, hierarchical clustering, etc.) are used to cluster the extracted keywords, filtering out the core keywords for each industry sector. The extracted keywords are then refined and optimized to ensure that only the most representative and highly relevant keywords are retained. A list of core industry keywords is created, which will serve as the basis for subsequent analysis. Clustering algorithms (such as K-means, DBSCAN, hierarchical clustering, etc.) are used to perform cluster analysis on the core industry keywords to identify the main industry categories. Based on the clustering results, multiple industry categories are categorized, such as retail, catering, automotive, and electronics products, and the main representative keywords for each industry category are determined. Descriptive analysis of the main industry categories in the city is then performed using text analysis techniques. By analyzing the correlation between industry data and consumption scenario data, the consumption scenarios corresponding to each industry category are identified. Using data mining techniques, consumer demand characteristics, such as purchasing behavior, purchase frequency, and preferred products, are extracted for each industry category. The analysis results are quantified to generate core feature data and consumer demand data for each consumption scenario. Core feature data includes consumption time period, purchase method, and product type; consumer demand data includes price sensitivity, purchasing power, and promotional responsiveness. The core feature data and consumer demand data for each consumption scenario are organized to ensure consistent data format and undergo preprocessing such as missing value handling and normalization. Tools such as Pandas are used to integrate this data into a clear dataset for easy model training.Large-scale models (such as deep learning models and graph neural networks) are used to train the core features of consumption scenarios and consumer demand data to calculate the correlation between various industries and consumption scenarios. Through multi-layer neural networks and attention mechanisms, the potential complex relationships between industries and consumption scenarios are captured. Based on the model output, correlation data between industries and consumption scenarios is generated. This data reflects the strength of the correlation between different industry categories and different consumption scenarios, helping to understand which industries are more suitable for which consumption scenarios. The industry-consumption scenario correlation data is transformed into a matrix form. In this matrix, each row represents an industry category, each column represents a consumption scenario, and each value in the matrix represents the correlation between that industry category and that consumption scenario. Based on the correlation between industries and consumption scenarios, combined with the definition of advertising types (such as brand advertising, promotional advertising, event advertising, etc.), appropriate advertising types are assigned to the consumption scenarios corresponding to each industry category. For example, if an industry has a high correlation in a specific consumption scenario, the corresponding advertising type will be prioritized. The results of the above steps are integrated to generate the final industry-advertising type mapping matrix. This matrix not only reflects the correlation between industries and consumption scenarios but also provides the most suitable advertising type recommendation for each industry category.

[0104] As an example of the present invention, reference is made to... Figure 2 As shown, in this example, step S2 includes:

[0105] Step S21: Obtain geographic information data of the city;

[0106] Step S22: Label the location information of the advertising screens in the geographic information data of the city, and use the CV model to perform scene semantic modeling on the geographic information data of the city based on the labeled location information of the advertising screens, and generate a scene label system.

[0107] Step S23: Perform multi-dimensional matching calculations on the city's recommended ad type list and scene tag system, eliminate conflicting scenarios, and generate ad scene matching data;

[0108] Step S24: Based on the advertising scene matching data, perform candidate site classification and output the marked advertising screen location information to obtain the advertising site priority list.

[0109] In this embodiment of the invention, detailed geographic information data of the city is obtained by utilizing open data platforms (such as OpenStreetMap and government public data platforms) and third-party data providers (such as Gaode Maps and Baidu Maps). This includes data on city block divisions, building locations, road networks, public facilities, commercial areas, and traffic flow. Accurate location marking of advertising screens in the city is performed using GPS positioning and map annotation tools (such as ArcGIS and QGIS). The marked data includes the latitude and longitude of the advertising screen, its location within the block, its surrounding environment, the street it faces, and traffic flow information. The accuracy of the marking can be improved by combining manual marking with automated tools. Computer vision (CV) models are used to perform scene recognition on the marked advertising screen location information. First, high-resolution satellite images or street view data are used to process images of different areas of the city. Deep learning models (such as Convolutional Neural Networks (CNN), YOLO object detection models, or Faster R-CNN) are used to extract features from the images and identify different scene types (such as commercial areas, residential areas, transportation hubs, etc.). Combined with Geographic Information System (GIS) data, spatial information fusion is performed to accurately locate the scene type of each advertising screen. Based on scene recognition results and geographic information data, a scene tagging system is constructed. For example, scene tags can be divided into multiple dimensions, such as "high-traffic areas," "shopping mall surroundings," and "tourist attraction areas." One or more tags are assigned to each advertising screen to indicate the specific scene type in which the advertising screen is located. The advertising type recommendation list generated in step S12 is matched with the scene tagging system generated in step S22 in a multi-dimensional manner. Each advertising type will be matched with different scene tags based on its characteristics. Multiple factors are considered during the calculation, such as the target audience of the advertisement, the timeliness of the advertisement, and the consumer attributes of the scene. A weighted scoring algorithm is used to score the matching degree between the advertising type and the scene tag. The higher the matching degree of the advertising type, the better the effect of the placement in that scene. Conflict detection is performed on the advertising types and scene tags in the matching results. Based on business needs, unsuitable scenes are excluded. For example, some advertising types are not suitable for appearing in specific scenes, such as the conflict between children's product advertisements and educational or medical scenes. Inappropriate scenes are filtered out using predefined rules (e.g., matching of advertising content, scene type, and consumer interests). Finally, an ad scenario matching dataset is generated, which contains pairing information for each ad type with its best scenario tag, and the matching degree of each pair is evaluated.

[0110] Preferably, based on the labeled advertising screen location information, the use of a CV model to perform scene semantic modeling of the geographic information data of the city includes:

[0111] Extract planning features from the geographic information data of the city, and divide the geographic information data of the city into spatial regions to generate spatial region hierarchical data;

[0112] Based on the labeled advertising screen location information, the spatial area hierarchical data is used to locate the advertising screen in space and generate advertising screen spatial location data; the images of the advertising screen spatial location data are retrieved using the area camera, and the images are used to perform target detection through a CV model to generate advertising scene visual features;

[0113] Extract scene elements with visual features of advertising scenes, and combine them with a region generation network to perform scene semantic modeling of images of advertising screen spatial positioning data, thereby generating advertising screen environment scene modeling data;

[0114] Semantic association tags are assigned to the advertising screen environment scene modeling data to generate environment tag association data; the environment tag association data and spatial region hierarchical data are used to construct a scene tag system for the advertising screen environment scene modeling data to obtain the scene tag system.

[0115] In this embodiment of the invention, urban spatial features are extracted by utilizing urban planning data, such as road networks, building distribution, green spaces, and public areas. These features help in understanding the city's infrastructure, traffic flow, and commercial area distribution. Spatial analysis is performed using GIS tools (such as ArcGIS and QGIS) to identify planning features of different areas, such as traffic-intensive areas, commercial centers, residential areas, and school areas. Based on the extracted planning features, the urban space is divided into different levels of areas. For example, it can be divided into multiple levels such as street level, community level, and commercial center level. For each area, different weights are assigned according to its characteristics (such as traffic flow and commercial activities), and corresponding spatial area hierarchy data is generated for each area. The area division can be automatic using hierarchical clustering algorithms (such as K-means and DBSCAN) or manually adjusted using expert knowledge to ensure that the hierarchy data accurately reflects the city's spatial structure. The spatial area where the advertising screen is located is located by combining the labeled advertising screen location information (latitude and longitude, street location, etc.). Geographic Information System (GIS) technology is used to map the labeled advertising screen coordinates to the corresponding area in the spatial area hierarchy data. Based on the location information of the advertising screens and combined with spatial area hierarchy data, specific spatial positioning data for each advertising screen is generated, clarifying its location within the city and its regional hierarchy. Real-time image or video data of the area where the advertising screens are located is acquired through cameras deployed throughout the city (such as traffic monitoring cameras and public facility cameras). Camera locations can be filtered based on spatial area hierarchy data, selecting cameras located around the advertising screens to acquire images. Object detection techniques in deep learning (such as YOLO and Faster R-CNN) are used to detect objects in the images, identifying key elements (such as pedestrians, traffic flow, buildings, and shops). Visual features around the advertising screens are extracted using the object detection model, including crowd distribution, traffic density, and building types. Based on the object detection results, visual features of the advertising scene are extracted, such as scene type (commercial area, residential area, transportation hub, etc.), crowd density, and traffic flow. These visual features are used to describe the current state of the environment in which the advertising screens are located, providing data support for subsequent advertising content optimization. The visual features of the advertising scene are segmented to extract key scene elements, such as building type, road width, crowd flow, and shop density. Image segmentation techniques from computer vision (such as Mask R-CNN) are used to perform more detailed region segmentation of the image, distinguishing different types of scene elements. The extracted scene elements are then combined with region generation networks (such as Convolutional Neural Networks (CNN) and Generative Adversarial Networks (GAN)) to perform semantic modeling of the image in the area where the advertising screen is located.The model assigns semantic labels to elements in images, such as "high traffic density," "commercial area," or "leisure area," generating semantic data for the scene where each advertising screen is located and mapping it to the corresponding advertising screen spatial positioning data. The modeling data of the advertising screen's environment scene is then tagged, combining regional features and scene element information to generate scene labels related to the advertising screen's location. Natural Language Processing (NLP) technology is used to classify scene elements, generating environmental labels such as "high pedestrian traffic," "commercial shopping area," and "transportation hub." These generated labels are then associated with the advertising screen's spatial positioning data to form a label relationship between the advertising screen and its environment. By analyzing the characteristics of the area where the advertising screen is located through label association data, personalized suggestions for advertising placement are provided. Based on the environmental label association data and spatial regional hierarchical data, a complete scene label system is constructed. The scene label system includes the specific environmental label of the advertising screen's location and various information related to that environment (such as pedestrian traffic, traffic density, and business district attributes).

[0116] Preferably, step S3, which involves extracting real-time crowd features based on the geographic information data of the city and generating an attention heatmap for the marked advertising screen location information based on these real-time crowd features, includes:

[0117] Retrieve area cameras based on geographic information data of the city;

[0118] The system filters environmental images from the area's cameras to obtain regional environmental images; it then identifies human silhouettes within these images and performs crowd attribute analysis on these silhouettes to generate real-time crowd characteristics.

[0119] Based on real-time crowd characteristics, the location information of the labeled advertising screen is analyzed for viewing distance to generate viewing distance data; a pre-trained attention prediction model is used to perform eye tracking prediction on the viewing distance data to generate eye tracking prediction data.

[0120] The eye-tracking prediction data is transformed into an attention heatmap to generate a heatmap of the screen area attention distribution.

[0121] In this embodiment of the invention, location information of cameras in the urban area is obtained through Geographic Information System (GIS) data. Cameras deployed within the area are accessed, primarily including those in public places, commercial areas, and transportation hubs. Video streams from these cameras are retrieved by integrating urban surveillance systems or other relevant data platforms. Images containing the environment surrounding the advertising screen are selected from the retrieved images. Image recognition methods from computer vision can be used to remove irrelevant backgrounds or other interfering factors, ensuring that the image content is relevant to the location of the advertising screen and the flow of people. Scene classification models (such as ResNet and VGG) can be used for image filtering to ensure that the images meet the requirements of the environment and the advertising screen's positioning. Human pose detection algorithms (such as OpenPose and HRNet) are used to process the regional environmental images to identify human contours in the images. This process identifies the body contours and postures of each person in the crowd, providing basic data for subsequent crowd analysis. Based on the identified human contours, crowd attribute analysis is performed, including headcount, gender, age group, and behavioral patterns (such as walking, staying, and gathering). Further image analysis using Convolutional Neural Networks (CNNs) or deep learning models (such as YOLO and Faster R-CNN) identifies and classifies crowd characteristics, such as crowd density and activity areas, generating real-time crowd feature data including the number, location, and activity status of the crowd. Based on the location of the advertising screen, its surrounding environment, and real-time crowd characteristics, the viewing distance of the advertising screen is calculated. Viewing distance analysis considers the impact of line-of-sight obstruction, environmental interference, and crowd behavior on the visible range of the advertising screen. 3D modeling and simulation (e.g., deep learning models based on scene reconstruction) are used to evaluate the visible distance, angle, and audience distance of the advertising screen. Viewing distance data is calculated and generated, describing the viewing distance relationship between the advertising screen and crowds in different areas. This viewing distance data provides a foundation for subsequent advertising placement and content optimization. A pre-trained eye-tracking prediction model (e.g., the EYE-Track model based on convolutional neural networks) is used to analyze the viewing distance data and predict the eye movement trajectories of the crowd within the advertising screen area. This model predicts which areas are more attractive to viewers based on viewing distance data and crowd behavior patterns, simulating user eye movement and identifying hotspots of interest within the advertising screen. The eye-tracking data output by the model includes the area the viewer gazes at at each time point and the duration of that gaze. This data can be quantified to describe the intensity of attention in different areas of the advertising screen. Based on the eye-tracking prediction data, the gaze distribution of the attention areas is transformed into a heatmap using heatmap generation algorithms (such as Gaussian smoothing or heatmap fitting methods). The color intensity in the heatmap represents the strength of attention in a region, typically using color gradients (e.g., red for high attention, blue for low attention). This transforms the eye-tracking data into a heatmap of screen area attention distribution, displaying the attention levels in different areas.

[0122] Preferably, the physical constraint injection for the screen area attention distribution heatmap in step S3 includes:

[0123] The advertising screen location information is collected to obtain advertising screen information parameters; the screen resolution and pixel density are analyzed from the advertising screen information parameters to obtain the advertising screen resolution and pixel density.

[0124] The initial advertising boundary range is set based on the screen resolution and pixel density of the advertising screen, and the viewing distance range is calculated based on the set initial advertising boundary range to obtain the visible range of the advertising screen.

[0125] By applying a minimum font size physical constraint to the visible area of ​​the advertising screen using a heatmap of screen area attention distribution, the content layout safety boundary constraint is obtained. The calculation formula for the minimum font size physical constraint is shown below:

[0126]

[0127] In the formula, B(x) represents the content layout safety boundary constraint, and A unsafe (x) represents the area of ​​the advertisement content that extends beyond the visible screen area, A total Let x be the total area of ​​the screen, and x be the input parameter.

[0128] In this embodiment of the invention, basic parameters of the advertising screen are collected, including the screen's size, display technology (such as LED or LCD), screen resolution, pixel density (PPI), and installation angle (relative to the ground or viewing area). To ensure accurate information sources, the actual installation environment, size, and configuration of the advertising screen need to be synchronized with a database or real-time monitoring system. The horizontal and vertical resolutions of the advertising screen are confirmed; common formats include 1920x1080 and 3840x2160. The pixel density (PPI) is calculated based on the advertising screen's physical size and resolution. Based on the advertising screen's size, resolution, and viewing distance analysis, the initial boundary range of the advertisement is determined. This boundary range defines the advertisement's "display area," meaning all advertising content should remain within this area to ensure full visibility within the viewer's line of sight. The visible range of the advertising screen is calculated through viewing distance analysis. Viewing distance calculation considers the distribution of the crowd, viewing angle, and the installation location of the advertising screen. Viewing distance models, such as those based on camera angle and scene depth, can be used to calculate the area where the crowd can clearly see the advertising content. Using the generated screen area attention distribution heatmap, the hotspots of viewer attention on the advertising screen are mapped to the display area of ​​the advertising screen, identifying areas of high and low attention. To ensure the readability and security of the advertising content, a minimum font size needs to be set to ensure that the advertising content is clearly visible when the viewer is far away or at a large angle. Calculate the area A of the portion of the advertising content that extends beyond the visible screen area.unsafe (x). If a portion of the ad content extends beyond the visible area (e.g., is not visible from a distance), the area of ​​that portion needs to be calculated and adjusted. Adjustments are made using physical constraints based on the minimum font size to ensure the ad content does not exceed the screen's visible range, as shown in the following formula: In the formula, B(x) represents the content layout safety boundary constraint, and A unsafe (x) represents the area of ​​the advertisement content that extends beyond the visible screen area, A total Let x be the total screen area and 'x' be the input parameter. Based on the calculation results, adjust the layout of the advertising content to ensure that all advertising content is within the visible range and meets the minimum font size limit. By redefining the position of the advertising content on the heatmap, ensure that important information is displayed in the viewer's focal area.

[0129] Preferably, the construction of the dynamic position optimizer based on the injection conditions in step S3 includes:

[0130] A multi-objective optimization parameter is constructed based on the content layout security boundary constraints. The multi-objective optimization parameter includes the minimum exposure rate and the minimum information interference value. The formula for calculating the minimum exposure rate is as follows:

[0131]

[0132] In the formula, E(x) is the minimum exposure rate, and A i Let A be the display area of ​​the advertisement in the i-th region. total V represents the total area of ​​the screen. x Let be the exposure of the advertisement in different viewpoints in the i-th region, n be the number of advertisement regions, and x be the input parameter;

[0133] The formula for calculating the minimum information interference value is as follows:

[0134]

[0135] In the formula, D(x) is the minimum information interference value, and L j (x) represents the brightness of the advertisement under ambient lighting condition j, C j (x) represents the visible contrast of information under ambient lighting condition j, L max and C max These are the maximum brightness and maximum contrast of the screen under given conditions, respectively, where m is the number of different environmental conditions and x is the input parameter.

[0136] A multi-objective optimization function is constructed based on the constraints of minimum exposure rate, minimum information interference value, and content layout safety boundary conditions. The formula of the multi-objective optimization function is shown below:

[0137] f(x)=w1·E(x)+w2·D(x)+w3·B(x);

[0138] In the formula, f(x) is the multi-objective optimization function, E(x) is the minimum exposure rate, D(x) is the minimum information interference value, B(x) is the content layout safety boundary constraint, w1 is the minimum exposure rate weight, w2 is the minimum information interference weight, w3 is the layout safety boundary constraint weight, and x is the input parameter.

[0139] By using a multi-objective optimization function to dynamically optimize the position of ad elements on the heatmap of screen area attention distribution, suggested coordinate data for ad elements is generated.

[0140] In this embodiment of the invention, the display area A of the advertisement in each region is determined by... i The exposure V is calculated based on the viewing distance and viewing angle in each area. x Exposure is evaluated using a perspective model or data captured by a camera. Finally, the exposure rates of each area are summed to obtain the minimum exposure rate E(x). The brightness and contrast of the advertising screen are monitored and calculated in real time, considering the impact of different ambient lighting conditions; lighting data is acquired and brightness L is calculated based on environmental changes. j (x) and contrast C j(x); Normalize these values ​​to the ratios of maximum brightness and maximum contrast, and sum them to obtain the minimum information interference value D(x). Through parameter tuning, determine the weights w1, w2, and w3 for each objective. These weights can be adjusted according to the actual advertising effect and priority. For example, if the advertising exposure rate is prioritized, w1 will be larger; if the screen visibility range and content security are prioritized, the weight of w3 will be increased accordingly; if the environmental impact is significant, w2 needs to be adjusted. By adjusting the weights and optimizing the balance of each objective, an optimization function suitable for the current advertising needs is constructed. The formula for the multi-objective optimization function is as follows: f(x) = w1·E(x) + w2·D(x) + w3·B(x); where f(x) is the multi-objective optimization function, E(x) is the minimum exposure rate, D(x) is the minimum information interference value, B(x) is the content layout safety boundary constraint, w1 is the minimum exposure rate weight, w2 is the minimum information interference weight, w3 is the layout safety boundary constraint weight, and x is the input parameter. Based on the generated content layout safety boundary constraint, minimum exposure rate, and minimum information interference value, the dynamic position optimization process begins. Algorithms based on optimization theory (such as genetic algorithms, particle swarm optimization, simulated annealing, etc.) are used to optimize the position of the advertising element on the screen. This optimization process ensures that the position of the advertising element maximizes exposure within the visible range while minimizing information interference and satisfying the layout safety boundary constraint. Once optimized, the resulting ad element coordinates represent the optimal display position of the ad on the screen. The generated ad element coordinate suggestions will be optimized based on factors such as the current ad screen's visibility range, exposure requirements, and environmental influences to ensure maximum ad effectiveness.

[0141] As an example of the present invention, reference is made to Figure 3 As shown, step S4 in this example includes:

[0142] Step S41: Based on the ad type recommendation list, venue priority list, and ad element coordinate suggestions, combine ad content to generate an ad image and text combination scheme;

[0143] Step S42: Conduct real-time click-through rate (CTR) testing on the ad image and text combination scheme to obtain the real-time CTR test results; perform cross-scenario style migration on the ad image and text combination scheme based on the real-time CTR test results to generate ad style migration data;

[0144] Step S43: Use advertising style migration data to conduct a closed-loop effect evaluation of the advertising image and text combination scheme, generate advertising performance evaluation data, and adjust the content weight of the advertising image and text combination scheme based on the advertising performance evaluation data to perform advertising content generation optimization.

[0145] In this embodiment of the invention, the advertising objective type is analyzed and a corresponding advertising type recommendation list is generated based on the advertising objectives (such as brand exposure, product promotion, interactive guidance, etc.). The recommendation list is dynamically adjusted considering factors such as audience behavior data, historical click data, and regional characteristics. A venue priority list is generated based on the actual situation of the advertising area (such as geographical location, audience density, exposure time, etc.). High-priority venues include densely populated areas, transportation hubs, and commercial centers. Combining advertising environment data (such as geographical information, venue characteristics, etc.) with advertising type recommendations, a dynamic optimizer generates suggested coordinates for advertising elements. A multi-objective optimization algorithm ensures that the advertising display location is consistent with venue priority and advertising type requirements. Advertising images, text, videos, and other elements are reasonably combined according to the coordinate suggestions to ensure that the advertising content visually matches the audience preferences of the venue. Content layout optimization ensures the best presentation of graphic and text ads in terms of visual and interactive effects. The graphic and text ad combination scheme is tested in specific deployment scenarios, and user click data is monitored in real time. A / B testing and other methods are used to compare the effects of different ad combinations, collecting key data such as click-through rate and conversion rate. The ad combination is adjusted based on the test results to optimize advertising effectiveness. Based on the visual characteristics and audience needs of different scenarios, style transfer is performed on the ad graphic and text combination. A pre-trained style transfer model is used to adjust the ad's image style, color tone, layout, and other elements to adapt to different scenarios. A deep learning model automatically adjusts the ad style to ensure consistent performance across scenarios. Style transfer generates adjustment data including ad color tone, design elements, and text layout. This data helps the advertising team further understand the performance of ad content in different scenarios. User feedback, click-through rate (CTR), and dwell time are used to evaluate the ad's visual appeal and message delivery effectiveness. Multi-dimensional analysis methods, including ad visual effects, user interaction, and conversion rate, are used to evaluate ad performance. If the style transfer does not meet expectations, the ad design and style transfer parameters are adjusted based on the evaluation results. Real-time CTR, conversion rate, user dwell time, and exposure data are summarized and analyzed to generate a detailed ad performance evaluation report. The performance of the ad in different venues, time periods, and audience groups is analyzed to provide guidance for campaign strategies. Based on the evaluation data, the element weights in the ad graphic and text combination are adjusted. For example, if a certain type of ad content (such as text ads or image ads) achieves a higher click-through rate, the weight of that element is increased. Weighting of different elements within the ad content (such as titles, images, and videos) is adjusted to ensure a more reasonable proportion within the ad mix. After adjusting the ad content weights, an optimized ad text and image combination is generated. The optimized ad strategy is then pushed to the ad delivery system in real time for further testing and evaluation, completing the closed-loop optimization process.

[0146] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0147] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for generating ad content based on a large model, characterized in that, Includes the following steps: Step S1: Obtain industry data for the city; use a large model to model the industry correlation of the city's industry data and generate an industry-advertising type mapping matrix; dynamically prioritize advertising types in the industry-advertising type mapping matrix and generate a recommended list of advertising types for the city. Step S2: Obtain the geographic information data of the city; mark the advertising screens on the geographic information data of the city to obtain the marked advertising screen location information; output the candidate site hierarchy based on the marked advertising screen location information to obtain the advertising site priority list. Step S3: Extract real-time crowd features based on the geographic information data of the city, and generate an attention heatmap for the marked advertising screen location information based on the real-time crowd features to obtain a heatmap of the screen area attention distribution. Physical constraints are injected into the heatmap of screen area attention distribution; a dynamic position optimizer is constructed based on the injected conditions to generate suggested coordinate data for ad elements. Step S4: Based on the recommended list of ad types, the list of venue priorities, and the suggested coordinates of ad elements, combine ad content to generate an ad image and text combination scheme; perform a closed-loop effect evaluation and content weight adjustment on the ad image and text combination scheme to execute the ad content generation optimization operation.

2. The method for generating advertising content based on a large model according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Obtain industry data for the city; Step S12: Use a large model to extract industry keywords and consumption scenarios from the industry data of the city, and model the industry relevance of industry keywords and consumption scenarios to generate an industry-advertising type mapping matrix; Step S13: Perform seasonal demand analysis on the industry-advertising type mapping matrix to generate seasonal demand analysis data; Step S14: Use seasonal demand analysis data to dynamically prioritize advertising types in the industry-advertising type mapping matrix and generate a recommended list of advertising types for the city.

3. The method for generating advertising content based on a large model according to claim 2, characterized in that, Step S12 includes the following steps: Step S121: Collect consumption scenario data from the industry data of the city to obtain consumption scenario data of the city; Step S122: Extract industry keywords from the city's industry data based on the city's consumption scenario data to obtain the core industry keywords; Step S123: Cluster the core keywords of the industry to generate the main industry categories of the city, and use text analysis technology to perform descriptive analysis on the consumption scenario data based on the main industry categories of the city to generate the core characteristics of the consumption scenario and consumer demand data; Step S124: Utilize the large model to perform correlation calculations on the core features of consumption scenarios and consumer demand data to generate industry-consumption scenario related data; perform data matrixing on the industry-consumption scenario related data to generate an industry-advertising type mapping matrix.

4. The method for generating advertising content based on a large model according to claim 1, characterized in that, Step S2 includes the following steps: Step S21: Obtain geographic information data of the city; Step S22: Label the location information of the advertising screens in the geographic information data of the city, and use the CV model to perform scene semantic modeling on the geographic information data of the city based on the labeled location information of the advertising screens, and generate a scene label system. Step S23: Perform multi-dimensional matching calculations on the city's recommended ad type list and scene tag system, eliminate conflicting scenarios, and generate ad scene matching data; Step S24: Based on the advertising scene matching data, perform candidate site classification and output the marked advertising screen location information to obtain the advertising site priority list.

5. The method for generating advertising content based on a large model according to claim 4, characterized in that, Based on the labeled location information of the advertising screens, semantic modeling of the scene using CV models on the geographic information data of the city includes: Extract planning features from the geographic information data of the city, and divide the geographic information data of the city into spatial regions to generate spatial region hierarchical data; Based on the labeled advertising screen location information, the spatial area hierarchical data is used to locate the advertising screen in space and generate advertising screen spatial location data; the images of the advertising screen spatial location data are retrieved using the area camera, and the images are used to perform target detection through a CV model to generate advertising scene visual features; Extract scene elements with visual features of advertising scenes, and combine them with a region generation network to perform scene semantic modeling of images of advertising screen spatial positioning data, thereby generating advertising screen environment scene modeling data; Semantic association tags are assigned to the advertising screen environment scene modeling data to generate environment tag association data; the environment tag association data and spatial region hierarchical data are used to construct a scene tag system for the advertising screen environment scene modeling data to obtain the scene tag system.

6. The method for generating advertising content based on a large model according to claim 1, characterized in that, Step S3, which involves extracting real-time crowd features based on the city's geographic information data and generating an attention heatmap for the labeled advertising screen locations based on these real-time crowd features, includes: Retrieve area cameras based on geographic information data of the city; The system filters environmental images from the area's cameras to obtain regional environmental images; it then identifies human silhouettes within these images and performs crowd attribute analysis on these silhouettes to generate real-time crowd characteristics. Based on real-time crowd characteristics, the location information of the labeled advertising screen is analyzed for viewing distance to generate viewing distance data; a pre-trained attention prediction model is used to perform eye tracking prediction on the viewing distance data to generate eye tracking prediction data. The eye-tracking prediction data is transformed into an attention heatmap to generate a heatmap of the screen area attention distribution.

7. The method for generating advertising content based on a large model according to claim 1, characterized in that, Step S3, which involves injecting physical constraints into the heatmap of attention distribution in the screen area, includes: The advertising screen location information is collected to obtain advertising screen information parameters; the screen resolution and pixel density are analyzed from the advertising screen information parameters to obtain the advertising screen resolution and pixel density. The initial advertising boundary range is set based on the screen resolution and pixel density of the advertising screen, and the viewing distance range is calculated based on the set initial advertising boundary range to obtain the visible range of the advertising screen. By applying a minimum font size physical constraint to the visible area of ​​the advertising screen using a heatmap of screen area attention distribution, the content layout safety boundary constraint is obtained. The calculation formula for the minimum font size physical constraint is shown below: ； In the formula, To establish security boundary constraints for content layout, This refers to the area of ​​the advertisement content that extends beyond the visible area of ​​the screen. The total area of ​​the screen. For input parameters.

8. The method for generating advertising content based on a large model according to claim 1, characterized in that, Step S3, which involves constructing the dynamic position optimizer based on the injection conditions, includes: A multi-objective optimization parameter is constructed based on the content layout security boundary constraints. The multi-objective optimization parameter includes the minimum exposure rate and the minimum information interference value. The formula for calculating the minimum exposure rate is as follows: ； In the formula, To minimize exposure, For the advertisement in the first The display area of ​​each region The total area of ​​the screen. For the advertisement in the first Exposure levels from different perspectives in different regions For the number of advertising areas, For input parameters; The formula for calculating the minimum information interference value is as follows: ； In the formula, To minimize information interference, In order to meet ambient lighting conditions The brightness of the advertisement below, In order to meet ambient lighting conditions The information below shows the contrast. and These are the screen's maximum brightness and maximum contrast ratio under given conditions. The number of different ambient lighting conditions. For input parameters; A multi-objective optimization function is constructed based on the constraints of minimum exposure rate, minimum information interference value, and content layout safety boundary conditions. The formula of the multi-objective optimization function is shown below: ； In the formula, For multi-objective optimization functions, To minimize exposure, To minimize information interference, To establish security boundary constraints for content layout, Weighted by minimum exposure rate The minimum information interference weight, To determine the weights for safety boundary constraints, For input parameters; By using a multi-objective optimization function to dynamically optimize the position of ad elements on the heatmap of screen area attention distribution, suggested coordinate data for ad elements is generated.

9. The method for generating advertising content based on a large model according to claim 1, characterized in that, Step S4 includes the following steps: Step S41: Based on the ad type recommendation list, venue priority list, and ad element coordinate suggestions, combine ad content to generate an ad image and text combination scheme; Step S42: Conduct real-time click-through rate (CTR) testing on the ad image and text combination scheme to obtain the real-time CTR test results; perform cross-scenario style migration on the ad image and text combination scheme based on the real-time CTR test results to generate ad style migration data; Step S43: Use advertising style migration data to conduct a closed-loop effect evaluation of the advertising image and text combination scheme, generate advertising performance evaluation data, and adjust the content weight of the advertising image and text combination scheme based on the advertising performance evaluation data to perform advertising content generation optimization.

10. A system for generating advertising content based on a large model, characterized in that, The system for performing the large-model-based ad delivery content generation method as described in claim 1, the system comprising: The ad type recommendation module is used to acquire industry data of the city; use a large model to model the industry correlation of the city's industry data to generate an industry-ad type mapping matrix; and dynamically prioritize ad types in the industry-ad type mapping matrix to generate an ad type recommendation list for the city. The advertising venue recommendation module is used to obtain the geographic information data of the city; to mark the advertising screens in the geographic information data of the city, and to obtain the location information of the marked advertising screens; and to output the candidate venues in a hierarchical manner based on the location information of the marked advertising screens, and to obtain the priority list of advertising venues. The ad content recommendation module is used to extract real-time audience features based on the geographic information data of the city, and generate an attention heatmap for the marked ad screen location information based on the real-time audience features to obtain a heatmap of the attention distribution of the screen area; inject physical constraints into the heatmap of the attention distribution of the screen area; and build a dynamic location optimizer based on the injected conditions to generate suggested coordinate data for ad elements. The ad generation module is used to combine ad content based on the ad type recommendation list, the venue priority list, and the ad element coordinate suggestions to generate ad image and text combination schemes; and to perform closed-loop effect evaluation and content weight adjustment of the ad image and text combination schemes to perform ad content generation optimization work.

Images