An intelligent advertising customization service platform

Abstract

The invention discloses an advertisement customization service platform based on intelligence, and relates to the technical field of advertisement platforms. The advertisement customization service platform based on intelligence includes a data collection and preprocessing module, which is used to collect user feature data sets and preprocess the user feature data sets; a bias prediction and adjustment module, which is used to identify representative biases in user feature data sets and implement data bias adjustment measures according to representative biases; a fairness evaluation and optimization module, which is used to detect the fairness of different advertisement decisions to different groups and implement fairness optimization measures according to fairness detection results; an advertisement recommendation continuous monitoring module, which is used to generate a fairness inspection report according to fairness detection results and recommend and place advertisements according to the fairness inspection report. The problem that the existing advertisement platform inadvertently learns and amplifies biased data, resulting in unfair advertisement placement results, is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of advertising platforms, and in particular to an intelligent advertising customization service platform. Background Art

[0002] In the modern digital advertising industry, ad customization service platforms have developed rapidly, which use user data to optimize the targeting and efficiency of advertising. However, this big data and algorithm-based approach often faces data bias and fairness issues, especially when dealing with diverse and global user groups. Data bias may cause certain user groups to be over-represented or ignored, thus affecting the fairness and effectiveness of advertising.

[0003] Although existing technologies can customize ads based on user behavior and preferences, they often neglect the systematic evaluation and adjustment of the representativeness of data sets and fairness in the processing process. As a result, the advertising system may inadvertently exacerbate bias against certain groups, for example, unequal advertising display based on gender, age, race or geographic location.

[0004] In addition, traditional advertising platforms often lack effective mechanisms to monitor and adjust advertising algorithms in real time to cope with rapidly changing market and user dynamics. These platforms need a system that can dynamically adjust and optimize advertising strategies to ensure that all user groups can fairly receive advertising content that is valuable to them.

[0005] Therefore, in response to the above problems, there is an urgent need for an intelligent advertising customization service platform. Summary of the invention

[0006] Technical issues solved

[0007] In view of the deficiencies of the prior art, the present invention provides an intelligent advertising customization service platform, which solves the problem that the existing advertising platform inadvertently learns and amplifies biased data, resulting in unfair advertising delivery results.

[0008] Technical Solution

[0009] To achieve the above objectives, the present invention is implemented through the following technical solutions: an intelligent advertising customization service platform, including a data acquisition and preprocessing module, which is used to collect user feature data sets and preprocess the user feature data sets; a bias prediction and adjustment module, which is used to identify representative biases in user feature data sets and implement data bias adjustment measures based on representative biases; a fairness assessment and optimization module, which is used to detect the fairness of different advertising decisions to different groups and implement fairness optimization measures based on fairness detection results; an advertising recommendation continuous monitoring module, which is used to generate a fairness check report based on the fairness detection results and recommend and deliver advertisements based on the fairness detection report.

[0010] Furthermore, the specific process of collecting the user feature data set and preprocessing the user feature data set is as follows: collecting the user feature data set, the user feature data set includes race, gender, age and geographical location; performing data cleaning on the user feature data set and the advertisement response data, removing incomplete and erroneous records, calculating the proportions of the four groups of race, gender, age and geographical location in the user feature data set, recording them as actual characteristic proportions, and setting expected characteristic proportions corresponding to the four groups of race, gender, age and geographical location; obtaining the user advertisement click-through rates of the four groups of race, gender, age and geographical location, and calculating the average advertisement click-through rates of the four groups of race, gender, age and geographical location.

[0011] Furthermore, the specific process of identifying the representative deviation in the user feature data set and implementing data deviation adjustment measures based on the representative deviation is: obtaining the user feature data set, the actual feature ratio and the expected feature ratio, obtaining a bias detection value through comprehensive calculation of the user feature data set, the actual feature ratio and the expected feature ratio, and implementing data deviation adjustment measures based on the bias detection value.

[0012] Furthermore, the specific calculation formula of the bias detection value is:

[0013] Where BDV represents bias detection value, G represents user feature dataset, and p g represents the actual proportion of features of group g in the user feature dataset, q g Represents the expected proportion of features of group g in the user feature dataset.

[0014] Furthermore, the specific process of implementing data bias adjustment measures based on the bias detection value is: comparing the bias detection value with the first threshold in real time, and when the bias detection value is less than or equal to the first threshold, no adjustment is made; when the bias detection value is greater than the first threshold, resampling is performed, and for underrepresented groups, new samples are generated by random replication and the SMOTE synthesis method to increase the number of samples of these groups; for overrepresented groups, some samples are randomly deleted to reduce the proportion of these groups in the data set until the bias detection value is reduced below the first threshold.

[0015] Furthermore, the specific process of detecting the fairness of different advertising decisions to different groups and implementing fairness optimization measures based on the fairness detection results is: obtaining the user advertising click-through rate and the average advertising click-through rate, obtaining a fairness evaluation value through comprehensive calculation of the user advertising click-through rate and the average advertising click-through rate, and implementing fairness optimization measures based on the fairness evaluation value.

[0016] Furthermore, the specific calculation formula of the fairness evaluation value is: In the formula, FAV represents the fairness assessment value, a g,d Indicates the user's ad click rate, a d represents the average ad click-through rate, w g Represents the weight factor of group g.

[0017] Furthermore, the specific process of implementing fairness optimization measures based on the fairness evaluation value is as follows: comparing the fairness evaluation value with the second threshold in real time, and when the fairness evaluation value is greater than or equal to the second threshold, no processing is performed; when the fairness evaluation value is less than the second threshold, modifying the loss function, adding a fairness regularization term, adding a penalty factor related to group bias, adjusting the representativeness of each group in the data set, increasing the weight and number of minority group data, and reducing the proportion of the dominant group. For groups with insufficient performance, generating new samples by transforming existing data, and enhancing the generalization ability of these groups until the fairness evaluation value is increased above the second threshold.

[0018] Furthermore, the specific process of generating a fairness check report based on the fairness test results is: real-time detection of fairness evaluation values, collection of data on the impact of various advertising decisions on different groups, aggregation of the collected data to generate a fairness check report, and detailed display of which advertising decisions have unfair impacts on specific groups.

[0019] Furthermore, the specific process of recommending and placing advertisements according to the fairness detection report is as follows: according to the fairness report, the weights and parameters in the recommendation algorithm are adjusted to ensure equal advertisement display for each group; the adjusted recommendation algorithm is applied to actual advertisement placement to observe the effect after the adjustment.

[0020] Beneficial Effects

[0021] The present invention has the following beneficial effects:

[0022] (1) The present invention, through the implementation of the bias prediction and adjustment module, enables the platform to identify and adjust representative biases in data in real time, thereby improving the fairness of advertising delivery and effectively solving the problem of unfair advertising delivery caused by data bias in the prior art.

[0023] (2) The present invention optimizes the targeting and relevance of advertisements by accurately analyzing the user's behavior and feature data through the data collection and preprocessing module, thereby achieving a higher matching degree between the advertisement content and the user's needs, and effectively solving the problem of mismatch between the advertisement content and the user's preferences in the prior art.

[0024] (3) The present invention ensures that all groups can fairly receive appropriate advertisements by continuously monitoring the responses of different groups to advertisements through a fairness evaluation optimization module, thereby optimizing the allocation of advertising resources and effectively solving the problem of uneven resource allocation in the prior art.

[0025] (4) The present invention continuously adjusts the recommendation strategy according to market feedback through the continuous monitoring module of advertising recommendation, so that the platform can flexibly respond to market changes, thereby achieving higher market adaptability and service flexibility, and effectively solving the problem of slow response of advertising systems to market changes in the prior art.

[0026] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a structural diagram of an intelligent advertising customization service platform of the present invention. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0029] See also Figure 1 The embodiment of the present invention provides a technical solution: an intelligent advertising customization service platform, including a data collection and preprocessing module, which is used to collect user feature data sets and preprocess the user feature data sets; a bias prediction and adjustment module, which is used to identify representative biases in user feature data sets and implement data bias adjustment measures according to representative biases; a fairness assessment and optimization module, which is used to detect the fairness of different advertising decisions to different groups and implement fairness optimization measures according to fairness detection results; an advertising recommendation continuous monitoring module, which is used to generate a fairness inspection report according to the fairness detection results and recommend and deliver advertisements according to the fairness inspection report.

[0030] Specifically, the user feature data set is collected and preprocessed as follows: First, the platform system collects user feature data sets including race, gender, age and geographic location. These data come from the user's registration information, behavior logs and interaction history, which are crucial to ensure the effectiveness of subsequent analysis and accurate advertising push.

[0031] Next, to ensure data quality, the system conducts a thorough data cleansing of these user feature data sets and data related to advertising responses. This includes removing incomplete and erroneous records in the database, such as correcting incorrect data input, deleting duplicate records, etc., to ensure the accuracy and availability of the data. On this basis, the platform calculates the proportion of the four groups of race, gender, age and geographic location in the user feature data set, and records it as the actual feature proportion. In addition, the platform also sets the expected feature proportions corresponding to these four groups. These expected proportions are usually based on broader market research or demographic data to ensure that advertising can fully cover and balance various groups.

[0032] Finally, the platform obtains the ad click-through rates of users in four groups: race, gender, age, and geographic location. This data reflects the reaction and interaction of different user groups to the ad content. At the same time, the system also calculates the average ad click-through rate of these groups to provide data support for ad optimization and user experience improvement. This series of data collection and preprocessing activities provides strong data support for the platform, making advertising more accurate and efficient, while enhancing the relevance of ad content and user satisfaction.

[0033] In this implementation scheme, the process of collecting and preprocessing user feature data sets not only optimizes the quality and availability of data, but also improves the accuracy and efficiency of advertising push. Through this process, the platform can ensure that the collected data, such as race, gender, age, and geographic location, are accurate, and incomplete and erroneous records in the data set are removed, thereby ensuring the stability of the basis for data processing and subsequent analysis. In addition, the actual proportion of each group is calculated and recorded with the expected proportion set, so that advertising can be balanced among different groups, effectively avoiding excessive or insufficient exposure to a certain group. At the same time, the acquired advertising click-through rate data further enables the platform to adjust and optimize advertising content according to actual user responses, enhancing the personalization and target accuracy of advertising. Overall, this step greatly enhances the operational efficiency of the platform and improves the user experience.

[0034] Specifically, the representative bias in the user feature data set is identified, and data bias adjustment measures are implemented according to the representative bias. The specific process is as follows: First, the platform system obtains a complete user feature data set, which includes data on dimensions such as race, gender, age, and geographic location, as well as the actual proportion of each dimension. These actual proportions reflect the distribution of each group in the overall user data. Next, the system ensures that the representativeness of each group matches social and market realities based on the pre-set expected proportions, which are usually based on demographic data and market research. The actual proportion of the feature is compared with the expected proportion of the feature, and the bias detection value is obtained through comprehensive calculation. Once the bias detection value indicates that there is a significant representative bias, the platform will implement data bias adjustment measures based on this indicator. These measures include upsampling underrepresented groups, that is, increasing the number of data samples for these groups; downsampling overrepresented groups, that is, reducing the number of data samples for these groups. In this way, the platform ensures the balanced representation of each group in the data set, thus laying a solid foundation for achieving fairer and more personalized advertising recommendations.

[0035] In this embodiment, in the advertising customization service platform of the present invention, the process of identifying and adjusting representative biases in the user feature data set is crucial. By comprehensively analyzing the actual proportion of users and the preset expected proportion, and calculating the bias detection value, the platform can accurately identify which groups are over-represented or under-represented in the data set. According to this bias detection value, targeted data bias adjustment measures, such as upsampling or downsampling, are implemented, which not only ensures the balanced representation of each group in the data set, but also improves the overall quality of the data and the fairness of advertising recommendations. This method effectively alleviates the problem of unfair advertising delivery caused by data bias in the prior art, thereby achieving highly personalized advertising content and precise targeting.

[0036] Specifically, the specific calculation formula of the bias detection value is:

[0037] In the formula, BDV stands for bias detection, and the value is used to quantify whether the representation of a specific user group in the advertising dataset is consistent with expectations, G represents the user feature dataset, and p g represents the actual proportion of features of group g in the user feature dataset, q g represents the expected proportion of features of group g in the user feature dataset, and the expected proportion q g Usually based on demographic data or market research to ensure that each group is represented fairly and without bias. When calculating the bias detection value, the weighting factor can be adjusted based on the market importance or advertising sensitivity of each group. Parameter p gThe acquisition of q involves complex data collection and processing processes, including user online behavior analysis, historical advertising response data, etc. g The setting of is based on extensive market research and historical data trends. This calculation method enables the platform to effectively monitor and adjust its advertising strategy to ensure that all user groups are properly represented, thereby improving advertising effectiveness while reducing data bias.

[0038] In this implementation scheme, by implementing the calculation formula of the bias detection value, the advertising customization service platform can effectively monitor and quantify the representative bias in the user feature data set. This step uses a precise mathematical model to evaluate the deviation between the actual proportion of each user group in the data set and its expected proportion, thereby ensuring that all user groups are represented fairly and balanced in advertising. In this way, the platform can not only discover and correct potential biases in the data, but also prevent legal and market risks that may be caused by unequal representation through timely adjustments. In addition, this bias detection and adjustment mechanism also enhances the personalization and accuracy of advertising, further improves user satisfaction and the efficiency of advertising delivery, and achieves the optimization of the utilization efficiency of advertising resources while ensuring the fairness of advertising.

[0039] Specifically, the specific process of implementing data bias adjustment measures based on the bias detection value is: comparing the bias detection value with the set first threshold in real time. This threshold is pre-set based on historical data and market research to identify representative bias levels that need attention.

[0040] When the bias detection value is less than or equal to the first threshold, the current user feature data set is considered to be representative and balanced, so no adjustment is made. This indicates that the delivery of advertising content is relatively fair and the representation ratio of each group is consistent with market or demographic data.

[0041] However, if the bias detection value exceeds the first threshold, this indicates that the representation of certain groups in the dataset is significantly different from what is expected. At this point, the platform will take resampling measures to adjust the composition of the dataset. For those underrepresented groups, the platform uses random replication of existing data and the use of synthetic minority oversampling technology SMOTE to generate new data samples. This method increases the amount of data for minority groups by artificially synthesizing new sample points, thereby improving their representativeness in the dataset.

[0042] For groups with overrepresentation, their proportion in the dataset is reduced by randomly deleting some samples. This process continues until the bias detection value drops below the first threshold, ensuring that all groups are fairly represented in the dataset.

[0043] In this embodiment, by real-time monitoring of the bias detection value and adjusting the data deviation according to the set threshold, the advertising customization service platform of the present invention can effectively respond to and correct the representative bias in the user feature data set. When the bias detection value exceeds the first threshold, the resampling processing measures taken, including random replication and synthetic minority oversampling SMOTE for underrepresented groups, and random sample deletion of overrepresented groups, are all for the purpose of rebalancing the proportions of each group in the data set. This adjustment mechanism not only ensures the diversity and balance of the data set, but also enhances the fairness and accuracy of advertising delivery. In addition, this step has a direct positive impact on improving the relevance of advertising content and user satisfaction. By continuously adjusting and optimizing the data set, the platform can more accurately locate and serve the target user group, thereby improving advertising efficiency while also ensuring the fairness and transparency of the service.

[0044] Specifically, the fairness of different advertising decisions for different groups is tested, and fairness optimization measures are implemented based on the fairness test results. The specific process is as follows: First, the user ad click-through rate of each group is obtained. These data directly reflect the reaction and interaction of different users to the advertising content. At the same time, the platform calculates the average ad click-through rate of all user groups. This statistical data serves as a comparison benchmark to help identify any group preferences or exclusions that may exist.

[0045] The fairness evaluation value is calculated by combining the user ad click-through rate with the average ad click-through rate. This evaluation value is calculated using advanced algorithms and aims to quantify the fairness between groups in the advertising process. If the fairness evaluation value shows that certain groups are subject to biased or unfair ad display, the platform will implement specific fairness optimization measures.

[0046] In this implementation scheme, by implementing fairness assessment and optimization measures in the advertising customization service platform, the platform effectively improves the fairness and efficiency of advertising delivery. The specific process includes obtaining and analyzing the click-through rates of advertisements of different groups, comparing them with the average click-through rates of advertisements, and calculating fairness assessment values ​​based on these data. When it is found that certain groups have significant deviations from the average level, the platform ensures fairness for all user groups when receiving advertisements by adjusting the advertising algorithms and content. This approach not only solves the problem of group bias that traditional advertising systems may inadvertently ignore, but also enhances the relevance and attractiveness of advertising content, thereby increasing user engagement and satisfaction. Overall, this step significantly improves the user experience and brand reputation of the platform by ensuring the fairness and accuracy of advertising delivery, while also optimizing the investment effect of the advertising budget.

[0047] Specifically, the specific calculation formula for the fairness evaluation value is: In the formula, FAV represents the fairness assessment value, a g,dIndicates the user's ad click rate, a d represents the average ad click-through rate, w g Represents the weight factor of group g, which is obtained from the database. The weight factor w g The value range is usually from 0 to 1. This range allows the platform to adjust the influence of different groups according to their market importance or sensitivity to specific advertising content. In this way, the calculation of the fairness evaluation value not only provides a quantitative indicator to evaluate whether the advertising content is fair to different groups, but also allows advertising operators to dynamically adjust advertising strategies based on actual data. This calculation method can ensure fairness for all user groups when receiving advertisements, thereby helping the platform optimize advertising delivery effects, improve user satisfaction, and reduce potential risks caused by bias.

[0048] In this embodiment, by implementing the calculation formula of the fairness evaluation value, the advertising customization service platform of the present invention can effectively monitor and evaluate the impact of advertising delivery strategies on different user groups, ensuring the fair distribution of advertising content. The fairness evaluation value formula provides a quantitative indicator to measure the fairness of advertising decisions by comprehensively considering the advertising click-through rate and corresponding weight factors of each group. This step not only enhances the platform's control over advertising effects and ensures that advertising delivery will not be biased towards a specific group, but also helps advertisers optimize their marketing strategies through accurate data analysis, improve the overall advertising effect and user satisfaction. In addition, by dynamically adjusting advertising strategies to adapt to changing market and user needs, the platform improves the transparency and trust of advertising delivery.

[0049] Specifically, the specific process of implementing fairness optimization measures based on the fairness evaluation value is as follows: First, the platform will monitor and compare the current fairness evaluation value with the preset second threshold in real time. This second threshold is set based on thorough market analysis and previous advertising performance data to ensure fairness for all user groups when receiving advertisements.

[0050] When the fairness evaluation value is greater than or equal to the second threshold, it means that the current advertising strategy has reached the expected fairness standard, so no further adjustment is needed. This shows that the distribution of advertising content is balanced among various groups and the effect of advertising delivery meets the fairness requirements.

[0051] However, if the fairness assessment value is lower than the second threshold, this indicates that certain groups may not receive fair advertising exposure and need to take measures to adjust. At this point, the platform will modify the loss function of ad recommendation and add specially designed fairness regularization terms. These regularization terms include penalty factors related to group bias, and their purpose is to reduce the tendency of the model to favor certain groups. The platform will also adjust the representativeness of each group in the data set, especially increasing the data weight and quantity of groups that have been underperformed or underrepresented in previous advertising. At the same time, for these groups, the platform may adopt data enhancement techniques, such as generating new samples by transforming existing data, to enhance the generalization ability of these groups.

[0052] These adjustments are made until the fairness evaluation value rises above the second threshold, thus ensuring that fairness in advertising reception for all groups is significantly improved. This process not only improves the overall acceptance and effectiveness of advertising, but also enhances users’ trust in the platform, providing advertisers with a more fair and efficient advertising solution.

[0053] In this embodiment, by real-time monitoring and adjusting the fairness evaluation value relative to the set second threshold, the advertising customization service platform of the present invention effectively ensures the fairness of advertising delivery. When the fairness evaluation value is lower than the threshold, the optimization measures taken, including modifying the loss function to introduce fairness regularization terms and penalty factors related to group bias, adjusting the representativeness of each group in the data set, and strengthening the performance of minority groups through data enhancement technology, all help to improve the fairness and accuracy of advertising content. This dynamic adjustment mechanism not only makes the advertising delivery process more transparent and fair.

[0054] Specifically, the specific process of generating a fairness check report based on the fairness test results is: real-time detection of fairness evaluation values, collection of data on the impact of various advertising decisions on different groups, especially data on the impact of various advertising decisions such as advertising display frequency and click-through rate on different groups. These data are used to evaluate each group's response to advertising and help identify possible biases or unfairness. The collected data is aggregated to generate a fairness check report, which shows in detail which advertising decisions have an unfair impact on specific groups. This report not only contains quantitative statistics, but also shows in detail which advertising decisions may have an unfair impact on specific groups, such as certain groups being over-advertised or certain advertising content failing to effectively reach specific groups. This detailed data analysis and reporting mechanism enables the platform to better understand the fairness issues in the distribution of advertising content and provides a specific basis for improving advertising strategies.

[0055] In this embodiment, by implementing real-time monitoring of fairness evaluation values ​​and generating detailed fairness detection reports, the advertising customization service platform of the present invention significantly improves the transparency and fairness of advertising delivery. This process enables the platform to continuously collect and analyze the specific impact of different advertising decisions on various groups, effectively identify and resolve potential bias issues, and ensure a balanced distribution of advertising content. By showing in detail which advertising decisions may cause specific groups to be treated unfairly, the platform not only enhances its own responsiveness and the pertinence of its adjustment strategies, but also establishes a trust relationship with users and advertisers.

[0056] Specifically, the specific process of ad recommendation and delivery based on the fairness detection report is as follows: Based on the generated fairness detection report, specific measures are taken to adjust the weights and parameters in the recommendation algorithm. This step ensures fair distribution of advertising content, especially among different user groups. Adjustments include modifying the weights in the algorithm used to determine which ads are displayed to which user groups, and adjusting related parameters to reduce or eliminate bias against certain groups. For example, if the report shows that a certain group is less considered in ad display, we may increase the weight of this group to ensure that it receives more advertising content.

[0057] These adjusted recommendation algorithms are then applied to actual ad placements. By deploying these modifications in a real market environment, we can observe the effects of the adjustments, including key indicators such as user feedback, ad click-through rates, and ad conversion rates. This not only helps us evaluate the effectiveness of the adjustments, but also further optimizes the algorithm to ensure that all user groups have equal exposure opportunities in future ad placements.

[0058] In this embodiment, by adjusting the weights and parameters of the recommendation algorithm based on the fairness detection report, the advertising customization service platform of the present invention effectively improves the fairness and accuracy of advertising delivery. This measure ensures that each user group has equal opportunities in advertising display, thereby correcting potential biases and optimizing user experience. After applying the adjusted recommendation algorithm to actual advertising delivery, the platform can observe and evaluate the effects of the adjustments in real time, such as the click-through rate and user engagement of the advertisement. This continuous optimization process not only enhances the target accuracy of the advertisement, but also improves the overall advertising effect and user satisfaction.

[0059] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0060] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. An intelligent advertising customization service platform, characterized in that: include: A data collection and preprocessing module is used to collect user feature data sets and preprocess the user feature data sets; A bias prediction and adjustment module is used to identify representative biases in user feature data sets and implement data bias adjustment measures based on the representative biases; The fairness evaluation and optimization module is used to detect the fairness of different advertising decisions to different groups and implement fairness optimization measures based on the fairness test results; The advertising recommendation continuous monitoring module is used to generate a fairness check report based on the fairness test results, and to make advertising recommendations and deliver them based on the fairness test report.

2. The intelligent advertising customization service platform according to claim 1, characterized in that: The specific process of collecting the user feature data set and preprocessing the user feature data set is as follows: Collect user characteristic data sets, including race, gender, age, and geographic location; Perform data cleaning on the user feature data set and the advertising response data, remove incomplete and erroneous records, calculate the proportions of the four groups of race, gender, age and geographic location in the user feature data set, record them as the actual proportions of the characteristics, and set the expected proportions of the characteristics corresponding to the four groups of race, gender, age and geographic location; Get the user ad click-through rate of four groups: race, gender, age, and geographic location, and calculate the average ad click-through rate of the four groups: race, gender, age, and geographic location.

3. The intelligent advertising customization service platform according to claim 1, characterized in that: The specific process of identifying representative deviations in the user feature data set and implementing data deviation adjustment measures based on the representative deviations is as follows: Obtain user feature data sets, actual feature ratios, and expected feature ratios, obtain bias detection values ​​through comprehensive calculation of the user feature data sets, actual feature ratios, and expected feature ratios, and implement data bias adjustment measures based on the bias detection values.

4. The intelligent advertising customization service platform according to claim 3, characterized in that: The specific calculation formula of the bias detection value is: Where BDV represents bias detection value, G represents user feature dataset, and p g represents the actual proportion of features of group g in the user feature dataset, q g Represents the expected proportion of features of group g in the user feature dataset.

5. The intelligent advertising customization service platform according to claim 3 is characterized by: The specific process of implementing the data deviation adjustment measures according to the bias detection value is as follows: comparing the bias detection value with the first threshold in real time, and when the bias detection value is less than or equal to the first threshold, no adjustment is made; When the bias detection value is greater than the first threshold, resampling is performed. For underrepresented groups, new samples are generated by random replication and the SMOTE synthesis method to increase the number of samples of these groups. For overrepresented groups, some samples are randomly deleted to reduce the proportion of these groups in the data set until the bias detection value drops below the first threshold.

6. The intelligent advertising customization service platform according to claim 1, characterized in that: The specific process of detecting the fairness of different advertising decisions to different groups and implementing fairness optimization measures according to the fairness detection results is as follows: Obtain the user's ad click-through rate and the average ad click-through rate, calculate the fairness evaluation value through the comprehensive calculation of the user's ad click-through rate and the average ad click-through rate, and implement fairness optimization measures based on the fairness evaluation value.

7. The intelligent advertising customization service platform according to claim 6, characterized in that: The specific calculation formula of the fairness evaluation value is: In the formula, FAV represents the fairness assessment value, a g,d Indicates the user's ad click rate, a d represents the average ad click-through rate, w g Represents the weight factor of group g.

8. The intelligent advertising customization service platform according to claim 6, characterized in that: The specific process of implementing fairness optimization measures according to the fairness evaluation value is as follows: Compare the fairness evaluation value with the second threshold in real time, and when the fairness evaluation value is greater than or equal to the second threshold, no processing is performed; When the fairness evaluation value is less than the second threshold, modify the loss function, add a fairness regularization term, add a penalty factor related to group bias, adjust the representativeness of each group in the data set, increase the weight and number of minority group data, and reduce the proportion of the dominant group. For groups with underperforming performance, generate new samples by transforming existing data to enhance the generalization ability of these groups until the fairness evaluation value is raised above the second threshold.

9. The intelligent advertising customization service platform according to claim 1, characterized in that: The specific process of generating a fairness check report based on the fairness test results is as follows: Detect fairness evaluation values ​​in real time, collect data on the impact of various advertising decisions on different groups, summarize the collected data to generate a fairness detection report, and show in detail which advertising decisions have unfair impacts on specific groups.

10. The intelligent advertising customization service platform according to claim 1, characterized in that: The specific process of recommending and placing advertisements based on the fairness detection report is as follows: Based on the fairness report, adjust the weights and parameters in the recommendation algorithm to ensure equal ad display for all groups; Apply the adjusted recommendation algorithm to actual advertising and observe the adjusted effect.

Images