Advertisement content generation and optimization system based on brand identification and visual style analysis

Abstract

The invention relates to the technical field of advertisement generation, discloses an advertisement content generation and optimization system based on brand identification and visual style analysis, and establishes a closed-loop processing flow from content identification to multi-platform putting optimization around four core links of advertisement content feature collection, style modeling, content generation and platform linkage. Compared with an existing advertisement generation method depending on manual adjustment and experience judgment, the system performs multi-dimensional fusion analysis on advertisement visual styles and platform behaviors by introducing a deep learning model, so that the style consistency of advertisement contents and brand images, the adaptability of cross-platform propagation and the propagation efficiency are remarkably improved, and the system is suitable for popularization and application. Advertisement visual features and dynamic user behavior data are fused, and higher adaptability and discrimination precision are achieved. The introduction of the method not only solves the problems of inconsistent content styles and large fluctuation of the delivery effect in multi-platform delivery, but also remarkably improves the adaptability and intelligent recommendation ability of the advertisement system to the complex platform environment.

Description

Technical Field

[0001] The present invention relates to the technical field of advertisement generation, and in particular to an advertisement content generation and optimization system based on brand recognition and visual style analysis. Background Art

[0002] With the rapid development of artificial intelligence and digital media technologies, the intelligent generation and delivery optimization of advertising content has become a key research direction in the field of digital marketing. Deep learning-based content recognition and style modeling methods are gradually replacing traditional empirical creative processes and have become a key technology for improving advertising communication efficiency. In particular, in the multi-platform content dissemination environment, how to achieve consistency in brand visual style, accurately determine content adaptability, and coordinate and optimize cross-platform communication strategies have become urgent challenges for the industry.

[0003] Existing technologies often suffer from fragmented styles, uneven dissemination, and a lack of adaptation standards for multi-platform advertising content. Due to significant differences in user behavior, content presentation, and dissemination mechanisms across platforms, it's often difficult to achieve a consistent style and accurate adaptation for advertising content, which in turn impacts overall marketing effectiveness. Existing methods primarily rely on manual experience for content adjustment and style correction, lacking systematic data modeling capabilities and integrated analysis mechanisms, making them unable to meet the demands of intelligent multi-platform delivery.

[0004] Therefore, we proposed an advertising content generation and optimization system based on brand recognition and visual style analysis to solve the above problems. Summary of the Invention

[0005] The purpose of the present invention is to provide an advertising content generation and optimization system based on brand recognition and visual style analysis to solve the problems raised by the above background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solutions: an advertising content generation and optimization system based on brand recognition and visual style analysis, comprising a feature acquisition module, a cross-platform behavior aggregation module, a visual style analysis module, an advertising content generation module, an optimization module, and a feedback module;

[0007] The feature collection module is used to collect and pre-process advertisement features to generate a first data set and a second data set;

[0008] The cross-platform behavior aggregation module collects and pre-processes the platform information to generate a third data group and a fourth data group;

[0009] The visual style analysis module is used to integrate the first data group and the second data group and perform data analysis, and determine whether the current advertisement is usable based on the analysis results;

[0010] The advertising content generation module is used to integrate the first data group and the second data group, and perform re-analysis based on the analysis results of the visual style analysis module, and determine whether the advertising content is consistent with the brand requirements based on the analysis results;

[0011] The optimization module is used to extract and analyze the third data group and the fourth data group to determine whether the content between multiple platforms is consistent;

[0012] The feedback module is used to feed back various data to the visualization terminal.

[0013] Preferably, the feature acquisition module includes an information extraction unit and a first data preprocessing unit;

[0014] The information extraction unit collects advertising information data through collection software and sensor groups, including color adaptation SCS, copy keyword precision WAJ, layout complexity BJF, brand content visibility PKS, overall advertising time GKS, advertising dwell rate GZL, average viewing time PGK and delivery cost TFC;

[0015] The first data preprocessing unit preprocesses and dimensionlessly converts the collected multiple data, and reorganizes the processed multiple data into a first data group and a second data group;

[0016] The first data set includes color suitability (SCS), copy keyword precision (WAJ), layout complexity (BJF), and brand content visibility (PKS);

[0017] Color adaptation (SCS) is obtained by extracting the main color of the advertising image frame through a clustering algorithm. Copy keyword precision (WAJ) uses natural language processing technology to segment the advertising copy. Layout complexity (BJF) is obtained by performing edge detection on the image. Brand content visibility (PKS) uses an image recognition model to analyze the advertisement.

[0018] The second data group includes the total advertising time GKS, the advertising dwell rate GZL, the average viewing time PGK and the delivery cost TFC.

[0019] Preferably, the cross-platform behavior aggregation module includes a behavior extraction unit and a second data pre-processing unit;

[0020] The behavior extraction unit is used to collect platform information through collection software, including average interaction rate, complete viewing rate, viewing difference rate, average dwell time, bounce rate, average click-through rate, secondary click-through rate, average conversion rate, activity index, video bounce rate and cost difference;

[0021] The second data preprocessing unit is used to preprocess and dimensionlessly transform the collected multiple data, and reorganize them into a third data group and a fourth data group;

[0022] The third data group includes average interaction rate PHD, complete viewing rate WZG, multi-platform viewing difference rate DKC, average dwell time PTL, video jump rate TZL and average click-through rate PDJ;

[0023] The fourth data group includes bounce rate TCL, secondary click-through rate RDJ, average conversion rate PZH, activity index HYZ, and multi-platform exposure cost difference GPC.

[0024] Preferably, the visual style analysis module includes a visual model building unit and a visual model parsing unit;

[0025] The visual model building unit is used to extract data from the first data group and the second data group, including color adaptation SCS, copy keyword precision WAJ, layout complexity BJF, brand content visibility PKS, overall advertising time GKS, advertising dwell rate GZL, average viewing time PGK, and delivery cost TFC, and input the extracted multiple data into a pre-trained deep learning framework, perform feature fusion through a multi-layer neural network, and thus calculate and generate a basic visual style value;

[0026] The visual model parsing unit is used to analyze the basic value of the visual style and determine whether the current advertisement visual style is usable based on the analysis result. The specific analysis method is as follows:

[0027] When SJF≤0.5, it means that the current ad visual style cannot be used;

[0028] When SJF>0.5, it means that the current ad visual style can be used.

[0029] Preferably, the visual model building unit calculates and obtains the basic value of the visual style by the following formula:

[0030]

[0031] In the formula: SCS is color adaptation, WAJ is the precision of copy keywords, BJF is the complexity of layout, PKS is the visibility of brand content, GKS is the overall advertising time, GZL is the advertising dwell rate, PGK is the average viewing time, and TFC is the cost of delivery.

[0032] Preferably, the advertisement content comparison module includes an advertisement comparison model and an advertisement parsing unit;

[0033] The advertising comparison model is used to extract data from the first data group and the second data group, including the advertising dwell rate GZL, the average viewing time PGK, and the delivery cost TFC, and input the extracted multiple data together with the basic value of the visual style into a pre-trained deep learning framework, and perform feature fusion through a multi-layer neural network to calculate and generate the advertising consistency coefficient;

[0034] The advertisement analysis unit is used to analyze the calculated advertisement consistency coefficient and determine whether the advertisement content is consistent with the brand requirements based on the analysis result. The specific steps are as follows:

[0035] When GGH≤0.75, it means that the current advertising content is consistent with brand needs;

[0036] When GGH>0.75, it means that the current advertising content is inconsistent with brand needs.

[0037] Preferably, the advertisement comparison model calculates the advertisement consistency coefficient using the following formula:

[0038]

[0039] Where: SJF is the basic value of visual style, GZL is the advertising dwell rate, PGK is the average viewing time, and TFC is the delivery cost.

[0040] Preferably, the platform analysis module includes a platform basic information identification model and a platform basic information analysis unit;

[0041] The platform basic information identification model is used to extract data from the third data group and the fourth data group, including the average interaction rate PHD, the complete viewing rate WZG, the average dwell time PTL, the average conversion rate PZH, the activity index HYZ, and the bounce rate TCL, and input the extracted multiple data and the basic value of the visual style into a pre-trained deep learning framework, perform feature fusion through a multi-layer neural network, and thus calculate and generate the basic value of the platform style;

[0042]

[0043] Where: PHD is the average interaction rate, WZG is the complete viewing rate, PTL is the average dwell time, PZH is the average conversion rate, HYZ is the activity index, TCL is the bounce rate, and SJF is the basic value of the visual style;

[0044] The platform basic information analysis unit is used to analyze the calculated platform style basic value and determine whether the current platform information is available based on the analysis result. The specific analysis method is as follows:

[0045] When PSB≤0.65, it means that the current platform information is unavailable;

[0046] When PSB>0.65, the current platform information is available.

[0047] Preferably, the platform analysis module further includes a single platform comparison model and a single platform analysis unit;

[0048] The single-platform comparison model is used to extract data from the third data group and the fourth data group, including the complete viewing rate WZG, the platform style basic value PSB, the multi-platform viewing difference rate DKC, the average click-through rate PDJ, the secondary click-through rate RDJ, the average conversion rate PZH, and the multi-platform exposure cost difference GPR, and input the extracted multiple data together with the platform style basic value into a pre-trained deep learning framework, perform feature fusion through a multi-layer neural network, and thus calculate and generate a single-platform adaptation coefficient;

[0049]

[0050] Where: WZG is the complete viewing rate, PSB is the platform style basic value, DKC is the multi-platform viewing difference rate, PDJ is the average click-through rate, RDJ is the secondary click-through rate, PZH is the average conversion rate, GPR is the multi-platform exposure cost difference;

[0051] The single platform analysis unit is used to analyze and determine whether the single platform content is consistent with brand requirements based on the analysis results. The specific steps are as follows:

[0052] When DPT≤0.55, it means that the current single platform content is inconsistent with brand needs;

[0053] When DPT>0.55, the current single-platform content is consistent with brand needs.

[0054] Preferably, the platform analysis module further includes a multi-platform comparison model and a multi-platform analysis unit;

[0055] The multi-platform comparison model is used to extract data from the third data group and the fourth data group, including the multi-platform viewing difference rate DKC, average click-through rate PDJ, average dwell time PTL, secondary click-through rate RDJ, and multi-platform exposure cost difference GPC. The extracted data and the single-platform adaptation coefficient are input into the pre-trained deep learning framework, and feature fusion is performed through a multi-layer neural network to calculate and generate the multi-platform linkage coefficient. The specific calculation formula is as follows:

[0056]

[0057] Where: DKC is the multi-platform viewing difference rate, PSC is the single platform adaptation coefficient, PDJ is the average click-through rate, PTL is the average dwell time, RDJ is the secondary click-through rate, and GPC is the multi-platform exposure cost difference;

[0058] The multi-platform analysis unit is used to analyze the multi-platform linkage coefficient and determine the content consistency between the multiple platforms based on the analysis results. The specific analysis method is as follows:

[0059] When DPT ≤ 0.85, it indicates inconsistent content among multiple platforms;

[0060] When DPT>0.85, it means that the content is consistent across multiple platforms.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] 1. This advertising content generation and optimization system, based on brand recognition and visual style analysis, establishes a closed-loop process from content identification to multi-platform delivery optimization, centering on four core steps: feature collection, style modeling, content generation, and platform integration. Compared to existing ad generation methods that rely on manual adjustments and empirical judgment, this system incorporates deep learning models to conduct a multi-dimensional fusion analysis of ad visual style and platform behavior, significantly improving the consistency of ad content and brand image, as well as the adaptability and efficiency of cross-platform communication.

[0063] 2. Compared to traditional analysis methods based on single behavioral indicators or static platform scores, the platform analysis module proposed in this paper integrates ad visual features with dynamic user behavior data, offering greater adaptability and discriminative accuracy. Its introduction not only addresses the pain points of inconsistent content styles and fluctuating delivery results across multiple platforms, but also significantly improves the advertising system's adaptability and intelligent recommendation capabilities in complex platform environments, providing solid data support and technical assurance for brands to achieve higher returns on advertising and more precise communication results. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 It is a system step diagram of the present invention.

[0065] In the figure: 1. Feature acquisition module; 11. Advertising information extraction unit; 12. First data preprocessing unit; 2. Cross-platform behavior aggregation module; 21. Platform information extraction unit; 22. Second data preprocessing unit; 3. Basic information analysis module; 31. Basic information identification model; 32. Basic information parsing unit; 4. Advertising content comparison module; 41. Advertising comparison model; 42. Advertising parsing unit; 5. Platform analysis module; 51. Platform basic information identification model; 52. Platform basic information analysis unit; 53. Single platform comparison model; 54. Single platform analysis unit; 55. Multi-platform comparison model; 56. Multi-platform analysis unit; 6. Feedback module. DETAILED DESCRIPTION

[0066] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0067] Example 1: Please refer to Figure 1 , an advertising content generation and optimization system based on brand recognition and visual style analysis, including a feature collection module 1, a cross-platform behavior aggregation module 2, a visual style analysis module 3, an advertising content generation module 4, an optimization module 5, and a feedback module 6;

[0068] The feature collection module 1 is used to collect and pre-process advertisement features to generate a first data set and a second data set;

[0069] The cross-platform behavior aggregation module 2 collects and pre-processes the platform information to generate a third data group and a fourth data group;

[0070] The visual style analysis module 3 is used to integrate the first data group and the second data group and perform data analysis, and determine whether the current advertisement is usable based on the analysis results;

[0071] The advertising content generation module 4 is used to integrate the first data group and the second data group, and perform further analysis based on the analysis results of the visual style analysis module 3, and determine whether the advertising content is consistent with the brand requirements based on the analysis results;

[0072] The optimization module 5 is used to extract and analyze the third data group and the fourth data group to determine whether the content is consistent across multiple platforms;

[0073] The feedback module 6 is used to feed back various data to the visualization terminal.

[0074] In this embodiment, feature acquisition module 1 is used to comprehensively collect key visual elements, textual information, structural layout, and other features of advertising content, and perform standardized preprocessing on the collected results to construct a first data set and a second data set. This module ensures the uniformity and high quality of advertising data input into the system, improves the accuracy of subsequent analysis and processing, and provides reliable support for multi-dimensional modeling of advertising content.

[0075] Cross-Platform Behavior Aggregation Module 2 collects and pre-processes user behavior data and platform performance information from different social platforms and media channels, generating third and fourth data sets. This module enables structured, unified management of multi-platform data, encompassing key metrics such as engagement rate, bounce rate, cost of impression, and platform viewing behavior. This provides the system with comprehensive, quantifiable user behavior information, facilitating platform compatibility and synergy analysis.

[0076] Based on the integrated analysis of the first and second data sets, Visual Style Analysis Module 3 uses a pre-trained deep learning model to model and identify the visual style elements of the advertisement, thereby determining whether the current ad content meets basic usability standards. This module's analysis and processing effectively identifies potential visual style deviations, missing elements, or layout anomalies within the content, providing a basis for style direction and structural corrections for subsequent ad generation and optimization, thereby improving the style consistency and communication quality of ad design.

[0077] Based on the output of Visual Style Analysis Module 3, Ad Content Generation Module 4 conducts an in-depth reanalysis of the first and second data sets, integrating advertising characteristics with brand recognition requirements. Using intelligent content reconstruction methods, it generates advertising content that aligns with the brand's style and visual logic. This module ensures the adaptive generation of advertising content, avoiding the biases that can occur in manual creation and improving the consistency of generated content with the brand's tone and its fit for market launch.

[0078] Optimization Module 5 primarily extracts and conducts in-depth analysis of key parameters from the third and fourth data sets. Using a single-platform adaptation coefficient and a multi-platform linkage coefficient model, it compares the advertising performance across platforms, assessing the consistency and efficiency of advertising content across these platforms. Based on the data output, this module intelligently identifies inconsistencies in content across multiple platforms and provides optimization recommendations, enabling collaborative management of advertising across multiple platforms and scientific adjustments to delivery strategies, ultimately improving overall effectiveness and resource efficiency.

[0079] Feedback Module 6 integrates and displays the intermediate data generated by each analysis step in the system with the final evaluation results, and then feeds them back to users or the operation platform via a visual terminal. This module ensures transparency and actionable output of system data results, allowing users to intuitively understand the structural evaluation, style adaptability, and multi-platform consistency analysis results of advertising content, thereby improving decision-making efficiency and data utilization.

[0080] The proposed advertising content generation and optimization system, based on brand recognition and visual style analysis, establishes a closed-loop process from content identification to multi-platform delivery optimization, centering on four core steps: feature acquisition, style modeling, content generation, and platform integration. Compared to existing ad generation methods that rely on manual adjustments and empirical judgment, this system incorporates a deep learning model to conduct a multi-dimensional fusion analysis of ad visual style and platform behavior, significantly improving the consistency of ad content and brand image, as well as the adaptability and efficiency of cross-platform communication.

[0081] Example 2: Please refer to Figure 1 , the feature acquisition module 1 includes an information extraction unit 11 and a first data preprocessing unit 12;

[0082] The information extraction unit 11 collects advertising information data through collection software and a sensor group, including color adaptation SCS, copy keyword precision WAJ, layout complexity BJF, brand content visibility PKS, overall advertising time GKS, advertising dwell rate GZL, average viewing time PGK, and delivery cost TFC;

[0083] The first data preprocessing unit 12 preprocesses and dimensionlessly converts the collected data into a first data group and a second data group.

[0084] The first data set includes color suitability (SCS), copy keyword precision (WAJ), layout complexity (BJF), and brand content visibility (PKS);

[0085] Color adaptation (SCS) is obtained by extracting the main color of the advertising image frame through a clustering algorithm. Copy keyword precision (WAJ) uses natural language processing technology to segment the advertising copy. Layout complexity (BJF) is obtained by performing edge detection on the image. Brand content visibility (PKS) uses an image recognition model to analyze the advertisement.

[0086] The second data group includes the total advertising time GKS, the advertising dwell rate GZL, the average viewing time PGK and the delivery cost TFC.

[0087] The cross-platform behavior aggregation module 2 includes a behavior extraction unit 21 and a second data pre-processing unit 22;

[0088] The behavior extraction unit 21 is used to collect platform information through collection software, including average interaction rate, complete viewing rate, viewing difference rate, average dwell time, bounce rate, average click-through rate, secondary click-through rate, average conversion rate, activity index, video bounce rate and cost difference;

[0089] The second data preprocessing unit 22 is used to preprocess and dimensionlessly transform the collected multiple data, and reorganize them into a third data group and a fourth data group;

[0090] The third data group includes average interaction rate PHD, complete viewing rate WZG, multi-platform viewing difference rate DKC, average dwell time PTL, video jump rate TZL and average click-through rate PDJ;

[0091] The fourth data group includes bounce rate TCL, secondary click-through rate RDJ, average conversion rate PZH, activity index HYZ, and multi-platform exposure cost difference GPC.

[0092] In this embodiment: The present invention realizes high-precision collection and standardized processing of advertising content features and platform user behavior data through the information extraction unit and data preprocessing unit respectively set in the feature collection module 1 and the cross-platform behavior aggregation module 2. Compared with the advertising data acquisition method that relies on manual experience or simple statistics in the prior art, the present invention adopts an automated data collection and preprocessing mechanism, which can construct structured advertising content data and platform behavior data while ensuring data accuracy. Among them, the first data group includes color adaptation SCS, copy keyword precision WAJ, layout complexity BJF and brand content visibility PKS, which realizes a comprehensive characterization of advertising content in multiple dimensions such as visual style, text expression and brand communication; the second data group includes the overall advertising time GKS, advertising dwell rate GZL, average viewing time PGK and delivery cost TFC, which further quantifies the core information of advertising in terms of playback effect and resource consumption.

[0093] In the cross-platform behavior aggregation module 2, user behavior data from multiple platforms is automatically collected through the behavior extraction unit 21, and the collected data is dimensionlessly and standardized through the second data preprocessing unit 22 to form a third data group and a fourth data group. Among them, the third data group covers the average interaction rate PHD, the complete viewing rate WZG, the multi-platform viewing difference rate DKC, the average dwell time PTL, the video jump rate TZL and the average click-through rate PDJ, which comprehensively reflects the interaction and retention of users on each platform on the advertising content; the fourth data group includes the bounce rate TCL, the secondary click-through rate RDJ, the average conversion rate PZH, the activity index HYZ and the multi-platform exposure cost difference GPC, which can effectively characterize the audience response and cost-effectiveness of advertising in multi-platform dissemination. Through the above-mentioned data collection and preprocessing mechanism, the present invention can not only eliminate the problem of inconsistent data collection standards between platforms, but also ensure the uniformity and accuracy of the data structure, providing a high-quality input basis for the subsequent calculation and processing of the visual style analysis module 3, the advertising content generation module 4 and the optimization module 5.

[0094] Example 3: Please refer to Figure 1 , the visual style analysis module 3 includes a visual model building unit 31 and a visual model parsing unit 32;

[0095] The visual model building unit 31 is used to extract data from the first data group and the second data group, including color adaptation SCS, copy keyword precision WAJ, layout complexity BJF, brand content visibility PKS, overall advertising time GKS, advertising dwell rate GZL, average viewing time PGK, and delivery cost TFC, and input the extracted multiple data into a pre-trained deep learning framework, perform feature fusion through a multi-layer neural network, and thus calculate and generate a basic visual style value;

[0096] The visual model parsing unit 32 is used to analyze the basic value of the visual style and determine whether the current advertisement visual style is usable based on the analysis result. The specific analysis method is as follows:

[0097] When SJF≤0.5, it means that the current ad visual style cannot be used;

[0098] When SJF>0.5, it means that the current ad visual style can be used.

[0099] The visual model building unit 31 calculates and obtains the basic value of the visual style by the following formula:

[0100]

[0101] In the formula: SCS is color adaptation, WAJ is the precision of copy keywords, BJF is the complexity of layout, PKS is the visibility of brand content, GKS is the overall advertising time, GZL is the advertising dwell rate, PGK is the average viewing time, and TFC is the cost of delivery.

[0102] In this embodiment, by providing a visual model building unit 31 and a visual model parsing unit 32 in the visual style analysis module 3, the automatic recognition and usability judgment capabilities of the visual style of advertising content are effectively improved. The visual model building unit 31 constructs a visual style calculation model based on deep learning technology. It can extract multiple key indicators from the first data group and the second data group, including color adaptation SCS, copy keyword precision WAJ, layout complexity BJF, brand content visibility PKS, overall advertising time GKS, advertising dwell rate GZL, average viewing time PGK, and delivery cost TFC, and input them into a pre-trained multi-layer neural network model for feature fusion processing, ultimately generating a basic visual style value with comprehensive discrimination capabilities. This calculation process not only achieves the efficient integration of data of different dimensions, but also ensures the stability and scalability of the style judgment model.

[0103] The visual model parsing unit 32 makes a quantitative judgment on the visual style availability of the advertisement based on the basic value of the visual style, and by setting a reasonable availability threshold, clearly distinguishes whether the current advertising content meets the brand communication standard. Compared with the traditional method that relies on manual subjective judgment, this module realizes the objective quantitative analysis of visual style consistency, greatly reduces human errors, and improves the automation and standardization level of style evaluation. Through the design of the above modules, the present invention can effectively solve the problems of insufficient visual style recognition ability, subjective judgment standards, and low efficiency of the evaluation process in the existing advertising content generation system, and ensure that the advertisement can maintain a consistent brand tone and aesthetic standards in the content generation and multi-platform adaptation links, thereby improving the visual quality and communication effect of the advertising content. This improvement has important application value and promotion significance for the current brand advertising practice where large-scale and multi-platform delivery is the norm.

[0104] Example 4: Please refer to Figure 1 , the advertisement content comparison module 4 includes an advertisement comparison model 41 and an advertisement parsing unit 42;

[0105] The advertising comparison model 41 is used to extract data from the first data group and the second data group, including the advertising dwell rate GZL, the average viewing time PGK, and the delivery cost TFC, and input the extracted multiple data together with the basic value of the visual style into a pre-trained deep learning framework. The multi-layer neural network is used to perform feature fusion to calculate and generate the advertising consistency coefficient;

[0106] The advertisement analysis unit 42 is used to analyze the calculated advertisement consistency coefficient and determine whether the advertisement content is consistent with the brand requirements based on the analysis result. The specific steps are as follows:

[0107] When GGH≤0.75, it means that the current advertising content is consistent with brand needs;

[0108] When GGH>0.75, it means that the current advertising content is inconsistent with brand needs.

[0109] The advertisement comparison model 41 calculates the advertisement consistency coefficient using the following formula:

[0110]

[0111] Where: SJF is the basic value of visual style, GZL is the advertising dwell rate, PGK is the average viewing time, and TFC is the delivery cost.

[0112] In this embodiment, the advertising content generation module 4 introduces an advertising comparison model 41 and an advertising analysis unit 42. By integrating visual style features with advertising performance parameters, an intelligent judgment of the consistency between advertising content and brand needs is achieved, further improving the brand adaptation capability and automation level of the advertising generation process. Based on the extraction of core parameters from the first and second data groups, including color adaptation, copy keyword precision, layout complexity, brand content visibility, advertising dwell rate, average viewing time, and delivery cost, the advertising comparison model 41 inputs the above parameters together with the visual style basic value SJF output by the visual model establishment unit 31 into the pre-trained deep neural network, performs multi-dimensional feature fusion analysis, and thus calculates and generates an advertising consistency coefficient. This coefficient is used to measure the degree of consistency between the advertising content and brand recognition requirements in the dual dimensions of visual style and audience behavior performance.

[0113] Ad parsing unit 42 evaluates the generated consistency coefficient, setting a scientifically sound threshold to clearly distinguish whether the current ad content meets brand consistency requirements. When the consistency coefficient is above the set threshold, the ad content is deemed consistent with the brand's tone, visual style, and communication presentation; otherwise, it indicates a stylistic deviation or insufficient audience acceptance. This module significantly overcomes the limitations of traditional manual review, such as its heavy reliance on subjective judgments of "brand consistency" and its single dimensionality, achieving quantitative assessment and automated judgment of ad content consistency.

[0114] By introducing the advertising content comparison model 41, the present invention can verify the brand compatibility of advertising materials during the content generation phase, fundamentally addressing issues such as brand style deviation and inconsistent aesthetics of delivered content, effectively improving the accuracy and intelligence of creative advertising generation. This mechanism is particularly well-suited to the core needs of brands, which currently strive to maintain consistent recognition and audience impression across multiple platforms and diverse content formats, and possesses significant practicality and commercial application value.

[0115] Example 5: Please refer to Figure 1 , the platform analysis module 5 includes a platform basic information identification model 51 and a platform basic information analysis unit 52;

[0116] The platform basic information identification model 51 is used to extract data from the third and fourth data groups, including the average interaction rate PHD, the complete viewing rate WZG, the average dwell time PTL, the average conversion rate PZH, the activity index HYZ, and the bounce rate TCL, and input the extracted data and the basic visual style value into the pre-trained deep learning framework, perform feature fusion through a multi-layer neural network, and thus calculate and generate the basic platform style value;

[0117]

[0118] Where: PHD is the average interaction rate, WZG is the complete viewing rate, PTL is the average dwell time, PZH is the average conversion rate, HYZ is the activity index, TCL is the bounce rate, and SJF is the basic value of the visual style;

[0119] The platform basic information analysis unit 52 is used to analyze the calculated platform style basic value and determine whether the current platform information is available based on the analysis result. The specific analysis method is as follows:

[0120] When PSB≤0.65, it means that the current platform information is unavailable;

[0121] When PSB>0.65, the current platform information is available.

[0122] In this embodiment: a platform basic information identification model 51 and a platform basic information analysis unit 52 are set in the platform analysis module 5. To address the problems of delivery adaptability differences and large performance fluctuations in the process of advertising content dissemination among multiple platforms, an automatic analysis mechanism based on deep learning and multi-dimensional data fusion is provided, thereby improving the consistency and controllability of the cross-platform dissemination of advertising content.

[0123] The platform basic information identification model 51 extracts and integrates key platform behavior indicators from the third data group and the fourth data group, jointly inputs the basic visual style value SJF, and uses a pre-trained multi-layer neural network to perform feature fusion calculations on multi-source heterogeneous data to generate the platform style basic value PSB.

[0124] The platform basic information analysis unit 52 determines platform suitability based on the PSB value, setting a reasonable threshold. When PSB ≤ 0.65, the platform is deemed unsuitable for advertising. When PSB > 0.65, the platform is deemed suitable for advertising. This mechanism enables quantitative matching between a platform's content carrying capacity and a brand's visual style, effectively assisting advertisers in strategic placement optimization and platform selection decisions.

[0125] Compared to traditional analysis methods based on single behavioral indicators or static platform scores, the platform analysis module 5 proposed in this invention integrates ad visual features with dynamic user behavior data, offering greater adaptability and discriminative accuracy. Its introduction not only addresses the pain points of inconsistent content styles and fluctuating delivery results across multiple platforms, but also significantly improves the advertising system's adaptability and intelligent recommendation capabilities in complex platform environments, providing solid data support and technical assurance for brands to achieve higher returns on advertising and more precise communication results.

[0126] Example 6: Please refer to Figure 1 , the platform analysis module 5 also includes a single platform comparison model 53 and a single platform analysis unit 54;

[0127] The single-platform comparison model 53 is used to extract data from the third data group and the fourth data group, including the complete viewing rate WZG, the platform style basic value PSB, the multi-platform viewing difference rate DKC, the average click-through rate PDJ, the secondary click-through rate RDJ, the average conversion rate PZH, and the multi-platform exposure cost difference GPR, and input the extracted multiple data together with the platform style basic value into a pre-trained deep learning framework, perform feature fusion through a multi-layer neural network, and thus calculate and generate the single-platform adaptation coefficient;

[0128]

[0129] Where: WZG is the complete viewing rate, PSB is the platform style basic value, DKC is the multi-platform viewing difference rate, PDJ is the average click-through rate, RDJ is the secondary click-through rate, PZH is the average conversion rate, GPR is the multi-platform exposure cost difference;

[0130] The single platform analysis unit 54 is used to analyze and determine whether the single platform content is consistent with the brand requirements based on the analysis results. The specific steps are as follows:

[0131] When DPT≤0.55, it means that the current single platform content is inconsistent with brand needs;

[0132] When DPT>0.55, the current single-platform content is consistent with brand needs.

[0133] In this embodiment, the platform analysis module 5 introduces a single-platform comparison model 53 and a single-platform analysis unit 54 to enable more refined analysis and judgment of the suitability of advertising content across each independent delivery platform. This module, from the dual perspectives of brand consistency and platform performance, effectively addresses the common issues of "content style adaptation bias" and "uncertain platform-specific response" in existing multi-platform advertising, significantly improving the delivery accuracy and management flexibility of the advertising system.

[0134] The single-platform comparison model 53 extracts key indicators reflecting the actual interactive performance of the platform from the third data group and the fourth data group, such as the complete viewing rate WZG, the multi-platform viewing difference rate DKC, the average click-through rate PDJ, the secondary click-through rate RDJ, the average conversion rate PZH, and the multi-platform exposure cost difference GPR, and combines them with the platform style basic value PSB calculated previously, and inputs them into the pre-trained deep neural network for feature fusion, and finally outputs the single-platform adaptation coefficient DPT for evaluating the degree of adaptation of advertising content to the platform.

[0135] The single-platform analysis unit 54 makes an availability judgment based on the calculated DPT value. When DPT≤0.55, the system believes that there is a large deviation between the advertising content on the platform and the brand demand, and it is not recommended to be placed. When DPT>0.55, it indicates that the advertising content is highly consistent with the visual style and behavioral feedback of the platform, and it can be recommended as a key placement platform.

[0136] This mechanism effectively transcends the limitations of traditional advertising strategies guided by "average platform performance," enabling individual analysis and strategy optimization between ad content and each target platform, providing brand owners with more refined support for placement decisions. In the current context of highly fragmented content distribution platforms and differentiated user behavior, this invention significantly improves the system's dynamic adaptability to single-platform placement strategies and the efficiency of resource allocation, offering significant practicality and promotional value.

[0137] Example 7: Please refer to Figure 1 , the platform analysis module 5 also includes a multi-platform comparison model 55 and a multi-platform analysis unit 56;

[0138] The multi-platform comparison model 55 is used to extract data from the third and fourth data groups, including the multi-platform viewing difference rate DKC, average click-through rate PDJ, average dwell time PTL, secondary click-through rate RDJ, and multi-platform exposure cost difference GPC. The extracted data and the single-platform adaptation coefficient are input into the pre-trained deep learning framework, and the multi-platform linkage coefficient is calculated by feature fusion through a multi-layer neural network. The specific calculation formula is as follows:

[0139]

[0140] Where: DKC is the multi-platform viewing difference rate, PSC is the single platform adaptation coefficient, PDJ is the average click-through rate, PTL is the average dwell time, RDJ is the secondary click-through rate, and GPC is the multi-platform exposure cost difference;

[0141] The multi-platform analysis unit 56 is used to analyze the multi-platform linkage coefficient and determine the content consistency between the multiple platforms based on the analysis results. The specific analysis method is as follows:

[0142] When DPT ≤ 0.85, it indicates inconsistent content among multiple platforms;

[0143] When DPT>0.85, it means that the content is consistent across multiple platforms.

[0144] In this embodiment, the multi-platform comparison model 55 is used to extract key parameters from the third and fourth data sets, including the multi-platform viewing discrepancy rate (DKC), average click-through rate (PDJ), average dwell time (PTL), secondary click-through rate (RDJ), and the multi-platform cost of impression difference (GPC). Simultaneously, these parameters, combined with the single-platform adaptation coefficient (PSC) calculated for each platform, are input into a pre-trained deep learning framework. A multi-layer neural network is used to fuse and model these features, generating a multi-platform linkage coefficient (MPC) that reflects the comprehensive adaptability of the platforms.

[0145] The multi-platform analysis unit 56 makes content consistency judgment based on the content similarity index DPT in the above-mentioned multi-platform linkage coefficient MPC. By proposing the multi-platform linkage coefficient MPC as a unified quantitative indicator for measuring cross-platform collaborative performance, it makes up for the lack of a unified evaluation standard in traditional analysis. It uses a deep learning model to perform nonlinear fusion of multiple behavioral features, improves the modeling ability of complex relationships between user behaviors, enhances the generalization performance of the model, establishes a content consistency judgment mechanism based on the DPT indicator, provides data support for multi-platform content planning and material version control, achieves consistency and accuracy in content distribution, introduces the multi-platform cost difference factor GPC, optimizes the budget scheduling and delivery efficiency evaluation method, and improves the overall delivery cost-effectiveness.

[0146] The contents not described in detail in this specification belong to the prior art known to those skilled in the art.

[0147] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An advertising content generation and optimization system based on brand recognition and visual style analysis, characterized by: It includes a feature collection module (1), a cross-platform behavior aggregation module (2), a visual style analysis module (3), an advertising content generation module (4), an optimization module (5), and a feedback module (6); The feature collection module (1) is used to collect and pre-process advertisement features, thereby generating a first data group and a second data group; The cross-platform behavior aggregation module (2) collects and pre-processes the platform information to generate a third data group and a fourth data group; The visual style analysis module (3) is used to integrate the first data group and the second data group and perform data analysis, and judge whether the current advertisement is usable based on the analysis results; The advertising content generation module (4) is used to integrate the first data group and the second data group, and perform re-analysis based on the analysis results of the visual style analysis module (3), and determine whether the advertising content is consistent with the brand demand based on the analysis results; The optimization module (5) is used to extract and analyze the third data group and the fourth data group to determine whether the content between the multiple platforms is consistent; The feedback module (6) is used to feed back various data to the visualization terminal.

2. The advertising content generation and optimization system based on brand recognition and visual style analysis according to claim 1, characterized in that: The feature acquisition module (1) comprises an information extraction unit (11) and a first data preprocessing unit (12); The information extraction unit (11) collects advertising information data through collection software and a sensor group, including color adaptation SCS, copy keyword precision WAJ, layout complexity BJF, brand content visibility PKS, overall advertising time GKS, advertising dwell rate GZL, average viewing time PGK and delivery cost TFC; The first data preprocessing unit (12) preprocesses and dimensionlessly converts the collected multiple data, and reorganizes the processed multiple data into a first data group and a second data group; The first data set includes color suitability (SCS), copy keyword precision (WAJ), layout complexity (BJF), and brand content visibility (PKS); Color adaptation (SCS) is obtained by extracting the main color of the advertising image frame through a clustering algorithm. Copy keyword precision (WAJ) uses natural language processing technology to segment the advertising copy. Layout complexity (BJF) is obtained by performing edge detection on the image. Brand content visibility (PKS) uses an image recognition model to analyze the advertisement. The second data group includes the total advertising time GKS, the advertising dwell rate GZL, the average viewing time PGK and the delivery cost TFC.

3. The advertising content generation and optimization system based on brand recognition and visual style analysis according to claim 2, characterized in that: The cross-platform behavior aggregation module (2) includes a behavior extraction unit (21) and a second data pre-processing unit (22); The behavior extraction unit (21) is used to collect platform information through collection software, including average interaction rate, complete viewing rate, viewing difference rate, average stay time, bounce rate, average click rate, secondary click rate, average conversion rate, activity index, video jump rate and cost difference; The second data preprocessing unit (22) is used to preprocess and dimensionlessly transform the collected multiple data, and reorganize them into a third data group and a fourth data group; The third data group includes average interaction rate PHD, complete viewing rate WZG, multi-platform viewing difference rate DKC, average dwell time PTL, video jump rate TZL and average click-through rate PDJ; The fourth data group includes bounce rate TCL, secondary click-through rate RDJ, average conversion rate PZH, activity index HYZ, and multi-platform exposure cost difference GPC.

4. The advertising content generation and optimization system based on brand recognition and visual style analysis according to claim 3, characterized in that: The visual style analysis module (3) includes a visual model building unit (31) and a visual model parsing unit (32); The visual model building unit (31) is used to extract data from the first data group and the second data group, including color adaptation SCS, copy keyword precision WAJ, layout complexity BJF, brand content visibility PKS, overall advertising time GKS, advertising dwell rate GZL, average viewing time PGK and delivery cost TFC, input the extracted multiple data into a pre-trained deep learning framework, perform feature fusion through a multi-layer neural network, and thus calculate and generate a basic value of the visual style; The visual model parsing unit (32) is used to analyze the basic value of the visual style and determine whether the current advertisement visual style is usable based on the analysis result. The specific analysis method is as follows: When SJF≤0.5, it means that the current ad visual style cannot be used; When SJF>0.5, it means that the current ad visual style can be used.

5. The advertising content generation and optimization system based on brand recognition and visual style analysis according to claim 4, characterized in that: The visual model building unit (31) calculates and obtains the basic value of the visual style by the following formula: In the formula: SCS is color adaptation, WAJ is the precision of copy keywords, BJF is the complexity of layout, PKS is the visibility of brand content, GKS is the overall advertising time, GZL is the advertising dwell rate, PGK is the average viewing time, and TFC is the cost of delivery.

6. The advertising content generation and optimization system based on brand recognition and visual style analysis according to claim 5, characterized in that: The advertisement content comparison module (4) includes an advertisement comparison model (41) and an advertisement parsing unit (42); The advertising comparison model (41) is used to extract data from the first data group and the second data group, including the advertising dwell rate GZL, the average viewing time PGK, and the delivery cost TFC, and input the extracted multiple data together with the basic value of the visual style into a pre-trained deep learning framework, and perform feature fusion through a multi-layer neural network to calculate and generate an advertising consistency coefficient; The advertisement analysis unit (42) is used to analyze the calculated advertisement consistency coefficient and determine whether the advertisement content is consistent with the brand demand based on the analysis result. The specific steps are as follows: When GGH≤0.75, it means that the current advertising content is consistent with brand needs; When GGH>0.75, it means that the current advertising content is inconsistent with brand needs.

7. The advertising content generation and optimization system based on brand recognition and visual style analysis according to claim 6, characterized in that: The advertisement comparison model (41) calculates the advertisement consistency coefficient using the following formula: Where: SJF is the basic value of visual style, GZL is the advertising dwell rate, PGK is the average viewing time, and TFC is the delivery cost.

8. The advertising content generation and optimization system based on brand recognition and visual style analysis according to claim 7, characterized in that: The platform analysis module (5) includes a platform basic information identification model (51) and a platform basic information analysis unit (52); The platform basic information identification model (51) is used to extract data from the third data group and the fourth data group, including the average interaction rate PHD, the complete viewing rate WZG, the average stay time PTL, the average conversion rate PZH, the activity index HYZ, and the bounce rate TCL, and input the extracted multiple data and the basic value of the visual style into the pre-trained deep learning framework, perform feature fusion through a multi-layer neural network, and thus calculate and generate the basic value of the platform style; Where: PHD is the average interaction rate, WZG is the complete viewing rate, PTL is the average dwell time, PZH is the average conversion rate, HYZ is the activity index, TCL is the bounce rate, and SJF is the basic value of the visual style; The platform basic information analysis unit (52) is used to analyze the calculated platform style basic value and determine whether the current platform information is available based on the analysis result. The specific analysis method is as follows: When PSB≤0.65, it means that the current platform information is unavailable; When PSB>0.65, the current platform information is available.

9. The advertising content generation and optimization system based on brand recognition and visual style analysis according to claim 8, characterized in that: The platform analysis module (5) further includes a single platform comparison model (53) and a single platform analysis unit (54); The single platform comparison model (53) is used to extract data from the third data group and the fourth data group, including the complete viewing rate WZG, the platform style basic value PSB, the multi-platform viewing difference rate DKC, the average click-through rate PDJ, the secondary click-through rate RDJ, the average conversion rate PZH, and the multi-platform exposure cost difference GPR, and input the extracted multiple data together with the platform style basic value into the pre-trained deep learning framework, perform feature fusion through a multi-layer neural network, and thus calculate and generate the single platform adaptation coefficient; Where: WZG is the complete viewing rate, PSB is the platform style basic value, DKC is the multi-platform viewing difference rate, PDJ is the average click-through rate, RDJ is the secondary click-through rate, PZH is the average conversion rate, GPR is the multi-platform exposure cost difference; The single platform analysis unit (54) is used to analyze and determine whether the single platform content is consistent with the brand requirements based on the analysis results. The specific steps are as follows: When DPT≤0.55, it means that the current single platform content is inconsistent with brand needs; When DPT>0.55, the current single-platform content is consistent with brand needs.

10. The advertising content generation and optimization system based on brand recognition and visual style analysis according to claim 9, characterized in that: The platform analysis module (5) further includes a multi-platform comparison model (55) and a multi-platform analysis unit (56); The multi-platform comparison model (55) is used to extract data from the third data group and the fourth data group, including the multi-platform viewing difference rate DKC, the average click-through rate PDJ, the average stay time PTL, the secondary click-through rate RDJ, and the multi-platform exposure cost difference GPC, and input the extracted multiple data together with the single platform adaptation coefficient into the pre-trained deep learning framework, and perform feature fusion through a multi-layer neural network to calculate and generate the multi-platform linkage coefficient. The specific calculation formula is as follows: Where: DKC is the multi-platform viewing difference rate, PSC is the single platform adaptation coefficient, PDJ is the average click-through rate, PTL is the average dwell time, RDJ is the secondary click-through rate, and GPC is the multi-platform exposure cost difference; The multi-platform analysis unit (56) is used to analyze the multi-platform linkage coefficient and determine the consistency of content between the multiple platforms based on the analysis results. The specific analysis method is as follows: When DPT ≤ 0.85, it indicates inconsistent content among multiple platforms; When DPT>0.85, it means that the content is consistent across multiple platforms.

Images