A multi-source data driven advertisement creative effect evaluation and optimization method

Abstract

The application discloses a multi-source data driven advertisement creative effect evaluation and optimization method, which comprises the following steps: multi-source advertisement data acquisition and preprocessing; advertisement creative multidimensional effect quantitative evaluation; creative feature and scene correlation analysis; advertisement creative targeting optimization generation; optimization effect verification and iterative output. The application has the beneficial effects that four types of data sources of the whole link of the advertisement are integrated, an evaluation system is constructed from four dimensions of exposure, conversion, cost and user feedback, the evaluation limitation of a single data source is broken through, the front end performance and the back end conversion of advertisement putting are comprehensively covered, and the comprehensive effect of the creative is truly reflected; the improved entropy weight method is adopted to automatically calculate the index weight based on sample data, the subjective deviation of manual weight assignment is avoided, the weight can be dynamically adjusted according to the putting scene and the industry characteristics, and the evaluation requirements of different types of advertisements are adapted; the iterative termination condition is automatically controlled through a convergence determination formula, the continuous iterative optimization of the advertisement creative effect is realized, and the ROI of advertisement putting is steadily improved.

Description

Technical Field

[0001] This invention relates to the fields of digital advertising technology and data processing technology, and in particular to a method for evaluating and optimizing the effectiveness of advertising creatives driven by multi-source data. Background Technology

[0002] With the rapid development of the digital advertising industry, ad creatives, as a core factor influencing campaign effectiveness, directly determine the return on investment (ROI) of advertising campaigns through their performance evaluation and optimization capabilities. Currently, the performance evaluation and optimization of ad creatives suffers from the following main shortcomings: 1. The evaluation data source is singular and the results are highly one-sided: Existing evaluations rely heavily on the exposure and click data of the advertising platform, which only covers the front-end performance of the advertisement. It cannot integrate multi-source data such as user conversion behavior, creative content attributes, and advertising scenarios, making it difficult to fully reflect the true effect of the creative throughout the entire process, resulting in significant bias in the evaluation results.

[0003] 2. The evaluation system is highly subjective and lacks quantitative standards: the weights of indicators rely heavily on human experience to set, which cannot adapt to the evaluation needs of different industries and different deployment scenarios, and is difficult to adjust dynamically, resulting in insufficient objectivity and adaptability of the evaluation results.

[0004] 3. Low efficiency and uncontrollable results in creative optimization: Existing creative optimization relies heavily on the experience and trial and error of operations personnel, lacks data-driven quantitative optimization models, has a vague optimization direction, long iteration cycle, unstable optimization results, and is difficult to adapt to the creative optimization needs of multiple channels and large volumes.

[0005] 4. The evaluation and optimization processes are fragmented and do not form a closed loop: In the existing technology, the effect evaluation and creative optimization are independent links, which cannot form a complete closed loop of "data collection-evaluation and analysis-optimization generation-verification and iteration", making it difficult to achieve continuous improvement of creative effects.

[0006] Therefore, it is necessary to propose a multi-source data-driven method for evaluating and optimizing the effectiveness of advertising creatives in order to address the aforementioned technical problems. Summary of the Invention

[0007] This invention provides a multi-source data-driven method for evaluating and optimizing the effectiveness of advertising creatives, which solves the problems of existing advertising creative evaluations being one-sided, highly subjective, inefficient, and lacking closed-loop iteration.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: A multi-source data-driven method for evaluating and optimizing the effectiveness of advertising creatives includes the following steps: S1. Multi-source advertising data collection and preprocessing: Collect multi-source raw data from the entire advertising delivery chain, clean, deduplicate, fill in missing values ​​and standardize the raw data to obtain a standardized dataset; S2. Multi-dimensional quantitative evaluation of advertising creative effects: Based on a standardized dataset, evaluation indicators are extracted from four dimensions: exposure effect, conversion effect, cost-effectiveness, and user feedback. A comprehensive effect evaluation model is constructed to calculate the comprehensive effect score of each advertising creative. S3. Creative Features and Scenario Correlation Analysis: Extract the content features and placement scenario features of advertising creatives, and combine them with the effect evaluation results to explore the core feature factors that affect the effect of creatives through correlation analysis. S4. Targeted Optimization and Generation of Ad Creatives: Based on the correlation between core feature factors and comprehensive performance scores, an iterative optimization algorithm is used to target and adjust the core features of low-scoring creatives to generate optimized candidate creative solutions. S5. Optimization effect verification and iterative output: Group the candidate creative solutions for comparative testing to verify the optimization effect. If the convergence condition is met, output the final optimized creative solution and evaluation report. If not, return to step S2 for a new round of iterative evaluation and optimization.

[0009] The multi-dimensional quantitative evaluation of the advertising creative's effectiveness in step S2 specifically involves constructing a multi-dimensional comprehensive effectiveness scoring model to calculate the comprehensive effectiveness score S of a single advertising creative. The calculation formula is as follows: ; in The click-through rate metric is normalized. The conversion rate metric is the normalized value. The normalized input-output ratio indicator. This is the normalized cost per thousand exposures metric. These are the weight coefficients of the corresponding indicators, and they satisfy... .

[0010] The exposure effect dimension also includes exposure completion rate and effective reach rate indicators; the conversion effect dimension also includes add-to-cart rate and lead retention rate indicators; the cost-effectiveness dimension also includes cost per conversion and cost per click indicators; and the user feedback dimension also includes interaction rate and negative feedback rate indicators. All indicators have been standardized by min-max and the value range is [0,1].

[0011] The weight coefficients of each evaluation indicator are dynamically calculated using an improved entropy weight method. The specific steps include: S41: Construct a judgment matrix with m creative samples and n evaluation indicators. ,in This represents the original value of the j-th indicator for the i-th creative idea; S42: Standardize the judgment matrix to obtain the standardized matrix. The standardized formula for the positive index is: The standardized formula for the negative index is: ; S43: Calculate the information entropy of the j-th indicator. The calculation formula is: ; in , ,when When, define ; S44: Calculate the weight coefficient of the j-th indicator. The calculation formula is: ; in The final weight of the j-th evaluation indicator; The information entropy of the j-th evaluation indicator is used to measure the degree of disorder and uncertainty of the indicator data. This refers to the total number of indicators in the evaluation system.

[0012] The multi-source raw data in step S1 includes four types of data sources: exposure and interaction data from the advertising platform, user conversion behavior data from the landing page / transaction terminal, content tag data from the advertising creative, and scenario context data from the advertising environment.

[0013] The creative content features extracted in step S3 include three categories: visual features, copywriting features, and audio features. The placement scenario features include three categories: placement time period, placement channel, and target audience tags. The correlation coefficient between each feature and the overall effect score is calculated through Pearson correlation analysis, and features with an absolute value of correlation coefficient greater than 0.3 are selected as core feature factors.

[0014] In step S4, the ad creative targeting optimization generation uses a gradient ascent iterative algorithm to adjust the core feature parameters of the creative. The specific iterative update formula is as follows: ; in Let f be the parameter value of the f-th core feature of the idea at the t-th iteration. These are the feature parameter values ​​after iterative updates; The learning rate is 0.01 to 0.1. The partial derivative of the overall score S with respect to this feature reflects the degree of influence of the feature on the overall score. This is the constraint adjustment coefficient for this feature, used to limit the reasonable range of feature adjustment, with a value range of [0.5, 1.5].

[0015] The group comparison test in step S5 adopts the A / B testing method. The advertising creatives before and after optimization are set as the control group and the experimental group. Under the same budget, audience and channel conditions, the variable comparison test is carried out, and the full-link data during the test period is collected to calculate the improvement rate of creative optimization effect.

[0016] Step S3 also includes building a creative effect prediction model. The XGBoost regression model, which integrates multiple features, is used for training. The model is trained with core feature factors as input and comprehensive effect score as output to predict the expected effect score of candidate optimized creative ideas and screen high-potential optimization directions in advance.

[0017] The termination condition for iterative optimization is determined by the convergence criterion formula, which is as follows: ; in For the first The overall creative effect score after the second iteration of optimization The score represents the overall creative effect after the (t+1)th iteration optimization. This represents the relative rate of change of scores between two consecutive iterations. The preset convergence threshold ranges from 0.01 to 0.05; when the relative rate of change of the score... Less than or equal to the convergence threshold When the time is right, terminate the iterative optimization.

[0018] The final output includes an optimized package of advertising creative materials, a multi-dimensional performance evaluation report, a core impact feature analysis report, and recommendations for differentiated placement strategies.

[0019] Compared with the prior art, the present invention has the following advantages: 1. Integrate four types of data sources across the entire advertising chain, and build an evaluation system from four dimensions: exposure, conversion, cost, and user feedback. This breaks through the limitations of evaluation from a single data source, comprehensively covers the front-end performance and back-end conversion of advertising, and truly reflects the overall effect of creative content.

[0020] 2. An improved entropy weight method is adopted to automatically calculate the indicator weights based on sample data, avoiding the subjective bias of manual weighting. The weights can be dynamically adjusted according to the placement scenario and industry characteristics to adapt to the evaluation needs of different types of advertisements.

[0021] 3. By accurately identifying the core features affecting the creative effect through correlation analysis, and combining the gradient ascent iterative algorithm for targeted quantitative adjustment, the optimization direction is traceable and the effect is predictable, which greatly reduces the cost of manual trial and error and improves the efficiency and stability of creative optimization.

[0022] 4. Build a complete closed loop of "data collection - effect evaluation - feature analysis - creative optimization - verification and iteration", automatically control the iteration termination condition through the convergence judgment formula, realize continuous iterative optimization of advertising creative effect, and steadily improve the ROI of advertising delivery. Brief Description of the Drawings

[0023] Figure 1 is a flowchart of the advertising creative effect evaluation and optimization method of the present invention; Figure 2 is a convergence comparison curve diagram of multi-scenario advertising creative iterative optimization of the present invention. Detailed Description of the Embodiments

[0024] To enable those skilled in the art to better understand and implement the technical solution of the present invention, the present invention is further described below in conjunction with specific embodiments and the accompanying drawings, but the listed embodiments are not intended to limit the present invention.

[0025] As Figures 1 to 2 shown, a multi-source data-driven advertising creative effect evaluation and optimization method includes the following specific implementation steps: S1. Multi-source advertising data collection and preprocessing: Collect four types of multi-source raw data across the full link of advertising delivery, and build a unified data warehouse: Exposure and interaction data: collected from the advertising delivery platform, including front-end delivery data such as exposure volume, click volume, viewing duration, completion rate, and interaction volume; User conversion behavior data: collected from landing pages, e-commerce platforms, and CRM systems, including back-end conversion data such as jump volume, add-to-cart volume, order volume, conversion volume, and customer unit price; Creative content data: through content recognition and labeling processing, extract visual, copy, and audio feature labels of advertising creatives, including color saturation, copy keywords, video duration, dubbing style, etc.; Scenario context data: including environmental feature data such as delivery time period, delivery channel, target crowd portrait, and delivery region.

[0026] Preprocess the collected multi-source raw data: complete data cleaning through deduplication and outlier elimination, process missing values using the mean filling method, uniformly map all indicators to the [0,1] interval through the min-max normalization method, eliminate dimensional differences, and obtain a standardized data set.

[0027] S2. Multi-dimensional quantitative evaluation of advertising creative effect: Based on the standardized data set, build a comprehensive evaluation system covering four primary dimensions of exposure effect, conversion effect, cost benefit, and user feedback, with a total of 12 secondary indicators, calculate the comprehensive effect score S of a single creative through the comprehensive effect scoring model, and the calculation formula is: ; Among them, the higher the score of positive indicators (click-through rate, conversion rate, return on investment, etc.), the better the effect; the higher the score of negative indicators (cost per thousand impressions, cost per conversion, etc.), the worse the effect. Therefore, the calculation is performed by subtracting items.

[0028] The weighting coefficients of each indicator are dynamically calculated using an improved entropy weighting method. The specific process is as follows: Construct a judgment matrix with m creative samples and n evaluation indicators. ,in This represents the original value of the j-th indicator for the i-th creative idea; The judgment matrix is ​​standardized according to the positive and negative directions of the indicators to obtain the standardized matrix. The standardized formula for positive indicators is: The standardized formula for the negative index is: ; Calculate the information entropy of the j-th indicator. The information discrimination of the indicator is reflected by the following formula: ; In the formula Let be the proportion of the i-th sample under the j-th indicator. To adjust the coefficient, ensure that the information entropy ranges from [0,1]; when When, define 4. Calculate indicator weights based on information entropy. The lower the information entropy, the higher the indicator's discriminative power and the greater its weight. The weight calculation formula is as follows: ; S3. Analysis of the correlation between creative features and scenarios: Extract the content and scene features of the advertising creative. The content features include visual features (main color scheme, number of elements, image complexity, etc.), copywriting features (keyword type, copy length, emotional tone, etc.), and audio features (speech speed, background music style, voice-over timbre, etc.). The scene features include the time period of placement, channel type, age segmentation of the target audience, and geographic level.

[0029] By using Pearson correlation analysis to calculate the correlation coefficient between each feature and the overall effect score S, features with an absolute correlation coefficient greater than 0.3 are selected as core feature factors to identify key optimization directions that affect the creative effect.

[0030] Simultaneously, an XGBoost regression performance prediction model is constructed, with core feature factors as input variables and comprehensive performance score as output variables. It is trained and validated using historical creative samples to predict the expected performance of candidate optimization ideas and filter out low-potential optimization schemes in advance.

[0031] S4. Ad creative targeting optimization generation: Based on the correlation between core feature factors and comprehensive effect scores, a gradient ascent iterative algorithm is adopted to maximize the comprehensive effect score S, thereby adjusting the core features of low-scoring creative ideas and generating optimized candidate creative solutions.

[0032] The iterative update formula for the feature parameters is:

[0033] Calculated by a trained performance prediction model, it reflects the degree of influence of each unit change in this feature on the overall performance score. The learning rate is set to 0.05 in this embodiment to control the adjustment range of each iteration and avoid over-optimization. The feature constraint adjustment coefficient is set according to the design specifications of the creative materials to ensure that the feature adjustment is within a reasonable creative range. In this embodiment, the value is 1.0.

[0034] For different types of core features, corresponding creative generation rules are matched: for example, visual features are adjusted to optimize parameters such as color and composition, and copywriting features are adjusted to replace keywords and optimize length, ultimately generating candidate creative materials that can be directly deployed.

[0035] S5: Optimization Effect Verification and Iterative Output The optimized candidate creatives were set as the experimental group, and the original creatives were set as the control group. A / B testing was conducted under the same conditions of budget, target audience, and channels. The testing period was 3-7 days. Data from the entire process was collected during the testing period, and the overall performance score was recalculated to verify the optimization effect.

[0036] The convergence criterion is used to determine whether to terminate the iteration. The specific formula is as follows:

[0037] In this embodiment, the convergence threshold If the relative change rate of the comprehensive score between two adjacent iterations is less than or equal to 3%, the optimization effect is considered to be stable and the iteration is terminated. If the convergence condition is not met, the test data is added to the sample set and the process returns to step S2 for a new round of evaluation and optimization.

[0038] After the iteration ends, the final optimization results are output, including the optimized advertising creative material package, multi-dimensional effect evaluation report, core impact feature analysis report, and scenario-based differentiated delivery strategy suggestions.

[0039] Example 1: Creative Evaluation and Optimization of Beauty Ads in Short Video Feeds This embodiment is applied to the performance advertising of beauty products on a short video feed platform, and fully implements the entire process method of this invention. The specific implementation process is as follows: Data Collection and Preprocessing: We collected full-link data from 20 short video ads with different creative concepts from a beauty brand over 30 days. This data covered four categories of data sources: exposure and interaction data (impressions, clicks, completion rate, engagement rate), conversion behavior data (bounces, add-to-cart, order conversion rate, ROI), creative content characteristics (video length, main color scheme, hook type in the first 3 seconds, copywriting, background music style), and campaign scenario data (campaign time, age demographics, geographic tier). The raw data underwent deduplication, outlier removal, mean imputation, and min-max standardization to unify the units of measurement, resulting in a standardized dataset.

[0040] Multi-dimensional Quantitative Evaluation of Results: An evaluation system with 4 dimensions and 12 indicators was constructed. An improved entropy weighting method was used to calculate the weights of each indicator, with the final weight allocation as follows: exposure effect 0.28, conversion effect 0.35, cost-effectiveness 0.24, and user feedback 0.13. The comprehensive score for 20 creative campaigns was calculated using a comprehensive scoring formula, with scores ranging from 0.32 to 0.78.

[0041] Core Feature Correlation Analysis: Through Pearson correlation analysis, core feature factors with an absolute correlation coefficient > 0.3 were selected and ranked by their degree of influence as follows: video completion rate (0.52), copywriting pain point keywords (0.46), impact of the first 3 seconds of the video (0.41), and proportion of prime time slots for advertising (0.34).

[0042] Targeted optimization generation: Set the iterative learning rate α=0.05 and the convergence threshold ε=0.03. Use the gradient ascent algorithm to perform targeted adjustment of the core features of the bottom 30% of the low-scoring creative ideas, and generate 8 sets of optimized candidate creative ideas.

[0043] Verification of iterative output: Set up a control group (original idea) and an experimental group (optimized idea) to conduct 7-day A / B testing with equal budgets. After the test, the comprehensive score of the experimental group improved to 0.89, with a relative improvement rate of 14.1%. After 3 rounds of iteration, the convergence condition was met, and the final optimized idea and evaluation report were output.

[0044] Example 2: Evaluation and Optimization of Creative Content for 3C Digital Products Graphic Ads on E-commerce Express This embodiment is applied to the image and text advertising of 3C digital products in the e-commerce platform's direct traffic feature. The evaluation weights and optimization directions are adjusted to suit the conversion-oriented business characteristics of e-commerce. The specific implementation process is as follows: Data collection and preprocessing: Collect data on 35 sets of graphic and text creative campaigns from a digital brand over 45 days, including five types of data sources: exposure and click data, conversion data, main image features, copywriting selling point features, and search scenario features. Complete data cleaning and standardization processing.

[0045] Multi-dimensional quantitative evaluation of effectiveness: Combining the business goal of strong conversion in e-commerce advertising, the appropriate weights are dynamically calculated using the entropy weight method: conversion effectiveness 0.42, cost-effectiveness 0.31, exposure effectiveness 0.19, and user feedback 0.08. The calculated comprehensive score range for all creatives is 0.27~0.71.

[0046] Correlation analysis of core features: Based on correlation analysis, the core feature factors were determined to be: the proportion of products displayed in the main image (0.57), the accuracy of selling point keywords (0.49), the eye-catching nature of promotional information (0.43), and the matching degree between the placement time and search peak (0.37).

[0047] Targeted optimization generation: Set the iterative learning rate α=0.04 and the convergence threshold ε=0.025, and iteratively optimize the main image composition, selling point copy, and price tag layout for low-scoring creatives to generate 10 sets of candidate optimized creatives.

[0048] Verification of iterative output: A / B testing was conducted under the same budget. After optimization, the overall score of the creative improved to 0.83, with a relative improvement rate of 16.9%. After 4 rounds of iteration, the convergence condition was met, and the optimal creative solution and delivery strategy were output.

[0049] Example 3: Evaluation and Optimization of Creative Content for Fast-Moving Consumer Goods Brand Ads on App Splash Screens This embodiment is applied to the splash screen brand advertisements for fast-moving consumer goods (FMCG) beverages on a leading news app. The evaluation system and optimization priorities are adjusted based on brand exposure guidance. The specific implementation process is as follows: Data collection and preprocessing: Collect data on 15 sets of splash screen creatives from a beverage brand over 20 days, including exposure and reach data, brand interaction data, creative visual features, animation features, and placement scenario data, and complete data standardization processing.

[0050] Multi-dimensional Quantitative Evaluation of Effectiveness: For brand advertising targeting high exposure and strong user perception, the appropriate weights were calculated using the entropy weight method: Exposure effectiveness 0.41, User feedback 0.29, Cost-effectiveness 0.18, and Conversion effectiveness 0.12. The calculated overall creative score ranged from 0.35 to 0.74.

[0051] Correlation analysis of core features: Through correlation analysis, the core feature factors were determined to be: brand logo display position (0.48), color saturation of the image (0.42), animation duration (0.38), and percentage of active advertising time (0.33).

[0052] Targeted optimization generation: Set the iterative learning rate α=0.06 and the convergence threshold ε=0.035, and iteratively optimize the layout of visual elements, color parameters, and animation duration to generate 6 sets of candidate optimized creative ideas.

[0053] Verification of iterative output: Conduct A / B testing with equal exposure. After optimization, the overall creative score improved to 0.86, with a relative improvement rate of 16.2%. After two rounds of iteration, the convergence condition was met, and the final brand creative solution was output.

[0054] By comparing the overall effect of Example 1 above with that of traditional manual optimization methods, the following table is obtained:

[0055] Table 2 compares the multi-dimensional indicators of the three implementation examples across different scenarios.

[0056] Based on Tables 1 and 2 above, Figure 2 It can be concluded that: Convergence Comparison Curves for Multi-Scenario Iterative Optimization: This curve shows the trend of the overall creative performance score with iteration rounds in three advertising scenarios: short video feed, e-commerce image and text ads, and brand splash screen ads. The performance scores of all three scenarios gradually increase and tend to stabilize with the advancement of iterations, verifying the convergence characteristics of the gradient optimization algorithm of this method; among them, the brand splash screen scenario has the fastest convergence speed, while the e-commerce image and text ad scenario requires more iteration rounds to achieve the optimal effect.

[0057] like Figure 2 As shown, all three curves continuously rise with the increase of iteration rounds, and the magnitude of the rise narrows with each round, eventually stabilizing, verifying the convergence characteristics of the gradient ascent optimization algorithm, which is consistent with the convergence determination formula in the claims. Example 3 (brand splash screen) has the fastest convergence speed, approaching the convergence value in the second round, because the adjustment space of the core features of brand advertising is concentrated, and the marginal effect of optimization is significant. Example 2 (e-commerce graphic and text) has the most convergence rounds, reaching stability in the fourth round, because the e-commerce conversion link is longer and the feature influencing factors are more complex, requiring multiple iterations to approach the optimal solution. The final scores of all examples are stable in the range of 0.83 to 0.89, proving that the method has stable optimization effects in different advertising scenarios.

[0058] This invention compares the cumulative return on investment (ROI) of the optimized scheme of this invention with that of the traditional manual optimization scheme over a 30-day deployment period. Starting in the middle of the deployment, the ROI of this scheme improved significantly faster than the traditional method, ultimately reaching a cumulative ROI of 3.47, a 31.9% improvement over the traditional method, demonstrating the long-term gain effect of closed-loop iterative optimization.

[0059] Due to the accumulation of data and model training in the early stages, the ROI of this invention is only slightly ahead. In the mid-stage of the campaign (8-20 days), the ROI of this invention is significantly faster than that of traditional methods through closed-loop iteration and continuous optimization of creatives, and the gap between the two continues to widen. In the late stage of the campaign (21-30 days), both curves tend to stabilize, and the final ROI of this invention stabilizes at 3.47, which is 31.9% higher than that of traditional methods. The curves intuitively reflect the long-term value of the "data collection-evaluation-optimization-validation" closed-loop mechanism of this method, which can continuously improve the effect during the campaign, rather than a one-time static optimization.

[0060] Core Feature Optimization Contribution Change Curve: This curve shows the relationship between the adjustment range and the effectiveness improvement rate of three core features: copywriting content, visuals, and placement scenarios. All three features exhibit a diminishing marginal effect, with copywriting content contributing the most to optimization and serving as the core focus of creative optimization. Over-adjusting visual features can easily lead to slower returns, requiring constraint coefficients to limit a reasonable range.

[0061] The score improvement rates of all three feature types initially increased rapidly with the increase in adjustment magnitude, and then the rate of increase gradually slowed down, exhibiting a typical law of diminishing marginal effect, which verifies the necessity of setting the learning rate and constraint coefficient in gradient optimization. The optimization contribution of copywriting content features is the highest, achieving a score improvement of more than 10% when the adjustment magnitude is 20%, making it the core focus of creative optimization. The optimization upper limit of visual features is the highest, but the marginal effect decreases significantly after the adjustment magnitude exceeds 30%, and excessive adjustment can easily lead to creatives deviating from the brand tone, requiring the constraint coefficient to limit a reasonable range. The optimization contribution of placement scenario features is relatively stable, and can achieve synergistic effects when combined with content feature adjustments, verifying the technical value of multi-source data fusion and multi-feature joint optimization.

[0062] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application also intends to include these modifications and variations.

Claims

1. A multi-source data-driven method for evaluating and optimizing the effectiveness of advertising creatives, characterized in that: Includes the following steps: S1. Multi-source advertising data collection and preprocessing: Collect multi-source raw data from the entire advertising delivery chain, clean, deduplicate, fill in missing values ​​and standardize the raw data to obtain a standardized dataset; S2. Multi-dimensional quantitative evaluation of advertising creative effects: Based on a standardized dataset, evaluation indicators are extracted from four dimensions: exposure effect, conversion effect, cost-effectiveness, and user feedback. A comprehensive effect evaluation model is constructed to calculate the comprehensive effect score of each advertising creative. S3. Creative Features and Scenario Correlation Analysis: Extract the content features and placement scenario features of advertising creatives, and combine them with the effect evaluation results to explore the core feature factors that affect the effect of creatives through correlation analysis. S4. Targeted Optimization and Generation of Ad Creatives: Based on the correlation between core feature factors and comprehensive performance scores, an iterative optimization algorithm is used to target and adjust the core features of low-scoring creatives to generate optimized candidate creative solutions. S5. Optimization effect verification and iterative output: Group the candidate creative solutions for comparative testing to verify the optimization effect. If the convergence condition is met, output the final optimized creative solution and evaluation report. If not, return to step S2 for a new round of iterative evaluation and optimization.

2. The method for evaluating and optimizing advertising creative effects driven by multi-source data according to claim 1, characterized in that: The multi-dimensional quantitative evaluation of the advertising creative's effectiveness in step S2 specifically involves constructing a multi-dimensional comprehensive effectiveness scoring model to calculate the comprehensive effectiveness score S of a single advertising creative. The calculation formula is as follows: ； in The click-through rate metric is normalized. The conversion rate metric is the normalized value. The normalized input-output ratio indicator. This is the normalized cost per thousand exposures metric. These are the weight coefficients of the corresponding indicators, and they satisfy... .

3. The method for evaluating and optimizing advertising creative effects driven by multi-source data according to claim 2, characterized in that: The exposure effect dimension also includes exposure completion rate and effective reach rate indicators; the conversion effect dimension also includes add-to-cart rate and lead retention rate indicators; the cost-effectiveness dimension also includes cost per conversion and cost per click indicators; and the user feedback dimension also includes interaction rate and negative feedback rate indicators. All indicators have been standardized by min-max and the value range is [0,1].

4. The method for evaluating and optimizing advertising creative effects driven by multi-source data according to claim 2, characterized in that: The weight coefficients of the evaluation indicators for each dimension are dynamically calculated using an improved entropy weight method. The specific steps include: S41: Construct a judgment matrix with m creative samples and n evaluation indicators. ,in This represents the original value of the j-th indicator for the i-th creative idea; S42: Standardize the judgment matrix to obtain the standardized matrix. The standardized formula for the positive index is: The standardized formula for the negative index is: ; S43: Calculate the information entropy of the j-th indicator. The calculation formula is: ； in , ,when When, define ; S44: Calculate the weight coefficient of the j-th indicator. The calculation formula is: ； in The final weight of the j-th evaluation indicator; The information entropy of the j-th evaluation indicator is used to measure the degree of disorder and uncertainty of the indicator data. This refers to the total number of indicators in the evaluation system.

5. The method for evaluating and optimizing advertising creative effects driven by multi-source data according to claim 1, characterized in that, The multi-source raw data in step S1 includes four types of data sources: exposure and interaction data from the advertising platform, user conversion behavior data from the landing page / transaction terminal, content tag data from the advertising creative, and scene context data from the advertising environment.

6. The method for evaluating and optimizing advertising creative effects driven by multi-source data according to claim 1, characterized in that: The creative content features extracted in step S3 include three categories: visual features, copywriting features, and audio features. The placement scenario features include three categories: placement time period, placement channel, and target audience tags. The correlation coefficient between each feature and the overall effect score is calculated through Pearson correlation analysis, and features with an absolute value of correlation coefficient greater than 0.3 are selected as core feature factors.

7. The method for evaluating and optimizing advertising creative effects driven by multi-source data according to claim 1, characterized in that, The advertising creative targeting optimization generation in step S4 uses a gradient ascent iterative algorithm to adjust the core feature parameters of the creative. The specific iterative update formula is as follows: ； in Let f be the parameter value of the f-th core feature of the idea at the t-th iteration. These are the feature parameter values ​​after iterative updates; The learning rate is 0.01 to 0.

1. The partial derivative of the overall score S with respect to this feature reflects the degree of influence of the feature on the overall score. This is the constraint adjustment coefficient for this feature, used to limit the reasonable range of feature adjustment, with a value range of [0.5, 1.5].

8. The method for evaluating and optimizing advertising creative effects driven by multi-source data according to claim 1, characterized in that: The group comparison test in step S5 adopts the A / B testing method. The advertising creatives before and after optimization are set as the control group and the experimental group. Under the same conditions of investment budget, target audience and channel, the variable comparison test is carried out, the full-link data during the test period is collected, and the improvement rate of creative optimization effect is calculated.

9. The method for evaluating and optimizing advertising creative effects driven by multi-source data according to claim 1, characterized in that: Step S3 also includes constructing a creative effect prediction model. The model uses an XGBoost regression model that integrates multiple features, with core feature factors as input and comprehensive effect score as output for training. This model predicts the expected effect score of candidate optimized creative ideas and screens high-potential optimization directions in advance.

10. The method for evaluating and optimizing advertising creative effects driven by multi-source data according to claim 8, characterized in that: The termination condition for the iterative optimization is determined by a convergence criterion formula, specifically: ； in For the first The overall creative effect score after the second iteration of optimization The score represents the overall creative effect after the (t+1)th iteration optimization. This represents the relative rate of change of scores between two consecutive iterations. The preset convergence threshold ranges from 0.01 to 0.05; when the relative rate of change of the score... Less than or equal to the convergence threshold When the time is right, terminate the iterative optimization.

Images