Cross-platform advertisement intelligent matching and pushing system and method

Abstract

The application provides a cross-platform advertisement intelligent matching pushing system and method, comprising: extracting relevant user behavior logs from the matching detail set for the cause of the low-efficiency category of delivery effect; adopting a clustering algorithm to perform grouping processing on the user behavior logs to obtain a grouped behavior mode set; determining factors of mismatching with advertisement content according to the grouped behavior mode set; if a time period deviation in the factors exceeds a preset threshold, generating a corresponding time adjustment parameter to obtain a time optimization suggestion; and simulating adjustment effects in a cross-platform pushing scenario according to the time optimization suggestion and in combination with platform resource integration information in the matching process detail data to obtain an optimization suggestion set.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a cross-platform intelligent matching and push system and method for advertising. Background Technology

[0002] In the wave of digital marketing, the research and application of advertising push systems are particularly crucial, directly impacting advertising effectiveness and user experience optimization, and serving as a vital bridge connecting advertisers and target audiences. With the diversification of internet platforms, cross-platform intelligent advertising matching and push has gradually become a focus of industry attention. Its importance lies in breaking down platform barriers and achieving efficient resource integration and precise reach. However, the development of this field still faces many challenges and urgently requires innovative breakthroughs.

[0003] Currently, existing ad delivery methods in the market often lack transparency and feedback mechanisms in the matching process. Many systems focus solely on the final delivery results, neglecting to reveal details of the matching process. This leaves advertisers unable to understand why certain ads failed to reach their target audience or to obtain effective improvement strategies. Simultaneously, users often develop resistance to pushed content due to a lack of clear explanations of the reasons for ad recommendations. This information asymmetry not only affects advertisers' confidence but also weakens users' trust in the platform.

[0004] A deeper technical challenge lies in building a comprehensive matching feedback system and providing targeted information support to different roles. First, transparency in the matching process is crucial. The system needs to clearly present the correlation between user group characteristics and matching results behind complex calculations; otherwise, advertisers will struggle to understand the reasons for their campaign performance. Second, this transparency further demands higher accuracy in optimization suggestions. Simply revealing the problem is insufficient; the system must propose practical adjustments based on the analysis of matching results. For example, in some scenarios, ads may perform poorly due to inappropriate timing, but identifying and providing specific timing adjustments becomes a critical challenge. These two issues are closely related: insufficient transparency directly leads to a lack of optimization suggestions, while inaccurate suggestions, in turn, exacerbate advertisers' distrust of the matching process.

[0005] Therefore, building a feedback mechanism in cross-platform ad delivery that clearly reveals the details of the matching process and provides precise optimization suggestions has become a key issue in improving ad performance and user acceptance. Solving this problem will directly affect advertisers' reliance on the system, determine whether users are willing to accept pushed content, and thus impact the healthy development of the entire advertising ecosystem. Summary of the Invention

[0007] This invention provides a cross-platform intelligent advertising matching and push system and method, mainly including: The matching process details data are obtained from the cross-platform advertising push system. The matching process details data includes the association records between user group characteristics and advertising content. The matching process details data are processed by a classification algorithm to obtain a set of classified matching details. Based on the categorized set of matching details, the numerical distribution of the campaign performance metrics is obtained. If the click-through rate in the campaign performance metrics is lower than a preset threshold, the corresponding matching details are marked as inefficient categories, and the causes of the inefficient categories are determined. To address the causes of the poor performance of the aforementioned categories, relevant user behavior logs are extracted from the matching detail set, and a clustering algorithm is used to group the user behavior logs to obtain a set of grouped behavior patterns. Based on the grouped set of behavioral patterns, factors that do not match the advertising content are identified. If the time period deviation of the factors exceeds a preset threshold, corresponding time adjustment parameters are generated to obtain time optimization suggestions. Based on the time optimization suggestions and combined with the platform resource integration information in the matching process details data, the adjustment effect under the cross-platform push scenario is simulated to obtain a set of optimization suggestions. For the set of optimization suggestions, a feedback report is generated for the advertiser. The feedback report includes a transparent description of the matching process details and optimization suggestions. If the user trust index is lower than a preset threshold, the explanation of the matching process details is highlighted in the feedback report, resulting in the final feedback output.

[0008] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses an intelligent optimization method for cross-platform ad delivery. It extracts the correlation records between user group characteristics and ad content from detailed matching process data, uses a classification algorithm to identify inefficient matching details with click-through rates below a threshold, and further analyzes the causes of their performance. For the causes of inefficiency, it uses a clustering algorithm to mine key behavioral patterns that do not match the ad content from relevant user behavior logs, discovering that time-period deviation is the main influencing factor, and generates corresponding time adjustment parameters to form optimization suggestions. Subsequently, it combines platform resource integration information to simulate the adjustment effect in a cross-platform push scenario, producing a reliable set of optimization suggestions. Finally, it generates a feedback report for advertisers, transparently presenting the matching process details and adaptively highlighting explanatory content based on user trust indicators. This invention effectively solves the integration problems of inefficient matching, opaque causes, lack of data support for optimization suggestions, and insufficient advertiser trust in cross-platform ad delivery. Through a closed-loop process encompassing matching detail classification, accurate diagnosis of inefficiency causes, behavioral pattern clustering, time deviation optimization, and effect simulation, it significantly improves the accuracy and timeliness of ad delivery, as well as advertisers' trust and acceptance of the system. Attached Figure Description

[0009] Figure 1 This is a flowchart of a cross-platform intelligent advertising matching and push system and method according to the present invention.

[0010] Figure 2 This is a schematic diagram of a cross-platform intelligent advertising matching and push system and method according to the present invention.

[0011] Figure 3 This is another schematic diagram of a cross-platform intelligent advertising matching and push system and method according to the present invention. Detailed Implementation

[0012] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] like Figures 1-3 This embodiment of a cross-platform intelligent advertising matching and push system and method may specifically include: S101. Obtain matching process detail data from the cross-platform advertising push system. The matching process detail data includes the association record between user group characteristics and advertising content. Process the matching process detail data using a classification algorithm to obtain a classified matching detail set.

[0014] Detailed data on the matching process is obtained through a cross-platform advertising push system. This data includes records of association between user group characteristics and advertising content. A classification algorithm is used to initially process the data, resulting in a categorized set of matching details. Based on this categorized set, the association records between user group characteristics and advertising content are extracted. Each group of association records is then grouped and organized to determine a set of grouped feature content. If duplicate association records exist in the grouped feature content sets, they are deduplicated. The feature content sets are updated with the deduplicated data to obtain a simplified set of associated features. For each group of features and advertising content matching details, a preset threshold is used for filtering. If the association strength of a matching detail is lower than the preset threshold, that detail is removed, resulting in a filtered subset of matching details. Core user group characteristics are extracted from the filtered subset of matching details. The correspondence between these core characteristics and advertising content is recorded and integrated to obtain an integrated set of matching maps. Using this integrated set of matching maps, potential association patterns between user group characteristics and advertising content are analyzed. A classification algorithm is used to perform secondary classification of these potential association patterns to determine the final classification association result. Based on the final classification and association results, cross-platform ad push matching optimization data is generated. The optimization data is then stored and processed to obtain a structured matching optimization dataset.

[0015] For example, in the business scenario of a cross-platform advertising push system, data acquisition and processing are core aspects. Suppose user data is collected from multiple advertising platforms, covering user group characteristics such as age, gender, and region, as well as information such as the type, theme, and timing of ad content. This data records the matching situation between certain ad content and specific user groups; for example, users aged 18-25 are more likely to view technology ads. Through classification algorithms, the data can be initially divided into different sets of matching details, such as classification by age group or interest tags, resulting in preliminary classification results such as "young users - technology ads".

[0016] Specifically, when extracting the correlation records between user group characteristics and advertising content, the data after each category is grouped and organized. For example, in the group "Young Users - Tech Ads," multiple records show that users aged 18-25 more frequently click on such ads between 8 PM and 10 PM. These records are summarized into a feature content set. If duplicate records are found, such as the same user clicking the same ad multiple times and being counted repeatedly, deduplication is performed, and the feature content set is updated to obtain a simplified set of correlated features. This step effectively reduces data redundancy and improves the accuracy of subsequent analysis.

[0017] For example, when filtering matching details, a correlation strength threshold is set, say 0.6. If a matching detail, such as "middle-aged users - food ads," has a correlation strength of only 0.3, that record is removed, resulting in a filtered subset of matching details. This process ensures the validity of the data and avoids low-relevance matches interfering with subsequent analysis. When extracting core features from the subset, assuming that the core feature of "young users" is "frequent use of social media," this feature is integrated with the correspondence of "technology ads" into a matching mapping set. This set provides a clear reference for subsequent analysis.

[0018] Specifically, when analyzing potential correlation patterns, a secondary classification algorithm is used to further uncover the deep connections between user behavior and advertising content.

[0019] For example, it was found that young users who frequently use social media not only have a high response rate to technology ads, but may also have a potential interest in fashion ads. This result is reflected in the classification and association results, generating optimization data, such as suggesting an increase in the proportion of fashion ads on social media platforms. Ultimately, this optimization data is structured and stored as a matching optimization dataset, facilitating system access and real-time adjustments.

[0020] For example, in the storage and processing stage, the dataset is optimized by splitting it into tables according to user groups and ad types to ensure data query efficiency and support subsequent adjustments to ad push strategies. This complete process not only improves the accuracy of ad matching but also reduces the system's computational burden through data simplification and filtering, significantly optimizing the effectiveness of cross-platform ad push.

[0021] S102. Based on the classified set of matching details, obtain the numerical distribution of the campaign performance indicators. If the click-through rate in the campaign performance indicators is lower than a preset threshold, mark the corresponding matching details as inefficient categories and determine the cause of the inefficient category's campaign performance.

[0022] The process begins by obtaining a categorized set of matching details and calculating the click-through rate (CTR) for each detail based on the distribution of performance metrics. The CTR of each detail is then compared to a preset threshold. If the CTR is lower than the threshold, the detail is marked as inefficient, resulting in a set of inefficient category matching details. Multiple dimensions of performance metrics are extracted from this set to obtain an inefficient category metric dataset. A decision tree algorithm is used to partition the dataset, creating multiple sub-nodes with corresponding metric combinations, thus identifying the primary causes of performance issues for each sub-node. For each sub-node's primary cause, the original attribute value ranges for the corresponding matching details are obtained, resulting in a set of causal attribute ranges. The frequency of occurrence of each causal attribute range in the inefficient category matching detail set is statistically analyzed to obtain a causal frequency distribution. Finally, the causal performance causes are ranked according to this frequency distribution, resulting in a priority sequence for inefficient category performance causes.

[0023] For example, in the business field of cross-platform advertising, when calculating and analyzing the click-through rate (CTR) value for the categorized set of matching details, the distribution of the campaign performance metrics can be used.

[0024] Specifically, after obtaining the categorized set of matching details, the system first calculates the click-through rate (CTR) for each matching detail based on the distribution of performance metrics. Taking an ad push scenario as an example, suppose an ad's matching details for different user groups include features such as age range, interest tags, and location. The system uses historical data to calculate the CTR for each matching detail. For instance, the CTR for a certain type of ad is 3.5% for users aged 18-24, while the CTR for users aged 35-44 is only 1.2%. After comparing this with a preset threshold of 2.0%, the matching details for users aged 18-24 are retained, while the matching details for users aged 35-44 are marked as inefficient.

[0025] For example, when extracting multi-dimensional data on campaign performance metrics from the inefficient category matching detail set, one can focus on metrics such as impressions, clicks, and conversion rates. Taking the 35-44 age group as an example, the data may show high impressions but low clicks and conversion rates, indicating that the ad content may not have attracted the interest of this group. The system integrates these metrics into an inefficient category metric dataset, including information such as 50,000 impressions, 600 clicks, and a conversion rate of 0.5%.

[0026] Specifically, when using the decision tree algorithm to divide the inefficient category indicator data set, the system may split the data layer by layer according to dimensions such as region and interest tags, generating multiple sub-nodes, each sub-node corresponding to a combination of indicators.

[0027] For example, a child node might show that the click-through rate is lowest for users whose region is "Northern Region" and whose interest tag is "Technology". The main reason for its poor performance is that the ad content does not match the interests and preferences of this group.

[0028] For example, when obtaining the set of causal attribute ranges, for the sub-node "Northern Region + Technology Interest," the system may extract its original attribute value range, such as user activity time being 8-10 PM and commonly used devices being high-end Android phones, forming a set of causal attribute ranges. By statistically analyzing the frequency of occurrence of each causal attribute range in the inefficient category matching detail set, the system may find that the "Northern Region + Technology Interest" combination occurs most frequently, reaching 35%, thus ranking it first in the priority sequence, indicating that this is the most critical factor affecting the campaign's effectiveness.

[0029] In one embodiment, the system can further analyze the underlying causes of the causes in the priority sequence.

[0030] For example, a low click-through rate among users in "Northern Regions + Tech Interests" might be related to the ad copy's youthful style, which doesn't align with the group's preference for a more serious approach. This kind of analysis helps in optimizing ad content design, improving matching accuracy, and ultimately increasing overall campaign efficiency.

[0031] S103. For the reasons of the poor performance of the inefficient category, relevant user behavior logs are extracted from the matching detail set, and the user behavior logs are grouped using a clustering algorithm to obtain a set of grouped behavior patterns.

[0032] For the set of behavioral patterns, common features within the patterns are obtained by analyzing the grouped data, resulting in preliminary behavioral classification. Based on the preliminary classification, statistical tools are used to calculate the frequency and duration of user behaviors within each category to determine key behavioral indicators. If key behavioral indicators are below a preset threshold, the log data within the corresponding category is deeply filtered to identify anomalous behavior fragments and potential causes of inefficiency. For anomalous behavior fragments, corresponding campaign environment data is extracted from matching details to analyze the correlation between the environment and behavior, identifying environmental influencing factors. Based on these environmental influencing factors, the set of behavioral patterns is regrouped and adjusted, and clustering algorithms are used to optimize the classification boundaries, determining the final behavioral pattern classification. Through the final behavioral pattern classification, the correspondence between each category and campaign performance is obtained, identifying the core issues of inefficient categories. For these core issues, behavioral records for relevant time periods are extracted from the log data to analyze the specific scenarios in which the problems occurred, obtaining a detailed distribution of the problems.

[0033] By analyzing grouped data within a set of behavioral patterns, we can reveal typical user interaction patterns in various advertising scenarios. For example...

[0034] In one embodiment, historical campaign logs are first preliminarily grouped, classifying similar click-and-browse sequences into one category to obtain preliminary behavioral classification results, such as highly active browsing, quick-jumping, and deeply interactive behaviors. For user behaviors within each category, statistical tools are used to calculate behavior frequency and duration to determine key behavioral indicators.

[0035] Specifically, for users who frequently leave a page, their average page dwell time may be only 8 seconds, and their click frequency is 0.3 times per minute, far lower than the 1.2 times per minute for normal users.

[0036] It should be noted that if key behavioral indicators are below the preset threshold, such as a dwell time threshold of 15 seconds, then the log data in the corresponding category will be deeply filtered to obtain abnormal behavior fragments.

[0037] One possible implementation involves filtering out segments that close the page immediately after three consecutive clicks; these segments typically correspond to inefficient ad delivery performance.

[0038] Preferably, corresponding campaign environment data, including device type, ad placement location, network type, etc., are extracted from the matching details to analyze the correlation between environment and behavior and obtain environmental influencing factors.

[0039] For example, it was found that 80% of abnormal bounce behaviors occurred on low-resolution mobile devices when the ad placement was at the bottom of the page, indicating that environmental factors significantly inhibited users' willingness to continue interacting. Based on the environmental influencing factors, the behavioral pattern set was regrouped and adjusted, and a clustering algorithm was used to optimize the classification boundary to determine the final behavioral pattern classification.

[0040] Understandably, after adjustments, the original "quick bounce" category was subdivided into two types: mobile bottom ad bounce and bounce in weak network environments, making the classification more targeted. Through the final behavioral pattern classification, the correlation between each category and campaign performance was obtained. For example, the average click-through rate of mobile bottom ad bounces was only 0.45%, significantly lower than the overall average of 1.8%. The core issue for this inefficient category was poor ad placement. To address this core issue, behavioral records for relevant time periods were extracted from log data to analyze the specific scenarios in which the problems occurred, yielding a detailed distribution of the problems.

[0041] For example, during peak traffic hours from 8 PM to 10 PM, this type of abnormal behavior accounted for 65%, compared to only 22% during the day, indicating that users' attention is more easily distracted at night, further confirming the negative impact of location factors. The beneficial effect of this approach is that it can accurately pinpoint the root causes of inefficient ad placement, providing actionable optimization directions from the perspective of user behavior and environmental interaction, avoiding the waste of resources caused by generalized adjustments, and providing data support for subsequent ad placement strategy iterations.

[0042] S104. Based on the grouped set of behavioral patterns, determine the factors that do not match the advertising content. If the time period deviation of the factors exceeds a preset threshold, generate the corresponding time adjustment parameters and obtain time optimization suggestions.

[0043] Obtain the grouped user behavior pattern set and the current ad delivery time information. For each time distribution segment in the user behavior pattern set, extract the active peak time point sequence. Compare the active peak time point sequence with the ad delivery time information segment by segment to determine the time period misalignment positions. If the number of time period misalignment positions reaches three or more, calculate the offset duration value of each misalignment position. Input each offset duration value into a preset threshold comparison module to determine whether the offset duration value exceeds the corresponding threshold. When the offset duration value exceeds the corresponding threshold, generate a reverse shift time adjustment parameter for that offset duration value. By summarizing all time adjustment parameters that exceed the threshold, a final time optimization suggestion list is formed.

[0044] By analyzing the grouped user behavior patterns and the current ad delivery time information, we can first obtain the time distribution characteristics of each behavior pattern.

[0045] Specifically, by statistically analyzing the frequency of user interactions at different times, a sequence of peak activity time points is formed.

[0046] For example, user behavior patterns in one inefficient category show peak activity between 8 pm and 10 pm, while another pattern focuses on activity between 1 am and 3 am.

[0047] One possible implementation involves comparing these peak activity time sequences with the actual ad delivery times. If ads are primarily placed between 9 AM and 12 PM and between 2 PM and 5 PM, while user activity peaks at night, it's easy to identify multiple time-related misalignments.

[0048] For example, when the comparison reveals that the morning campaign coincides with the period of low user activity, the midday campaign partially overlaps with the period of medium activity, and there is no campaign at night but it corresponds to the period of highest user activity, at least three misaligned locations can be identified.

[0049] It should be noted that when the number of misaligned positions reaches three or more, the offset duration of each misaligned position is further calculated.

[0050] For example, the first misalignment occurs when the campaign ends 4 hours before user activity begins; the second occurs when the campaign begins 2.5 hours before the peak user activity; and the third occurs when there is no campaign activity during the entire nighttime active period, resulting in a 6-hour offset. These offset durations are then input into a preset threshold comparison module. The threshold is typically set to 2 hours, 3 hours, etc., and adjusted based on historical campaign data.

[0051] Preferably, when a certain offset duration value exceeds the corresponding threshold, such as an offset exceeding 3 hours, it is determined to be a significant misalignment, and a time adjustment parameter for reverse translation is generated for that offset.

[0052] Specifically, if the peak nighttime activity period is from 10 PM to 2 AM, and current campaigns are completely absent, adjustment parameters will be generated to shift a portion of the campaign budget to 9 PM to 1 AM. If morning campaigns are launched too early, parameters can be generated to compress or delete inefficient morning periods. By summarizing all cases exceeding the threshold, a final list of time optimization suggestions is formed, including specific suggestions such as "compress the original 9-12 AM campaigns to 10-11 AM and increase the budget weight," "add a 30% nighttime campaign period from 9 PM to 1 AM," and "remove campaigns with no activity in the early morning."

[0053] Understandably, this method, based on the misalignment analysis of peak activity periods and ad placement times, can accurately pinpoint one of the core causes of inefficient ad placement: insufficient time matching. Multiple validations have shown that adjusting the ad placement to better align with user activity periods typically leads to a 18% to 35% increase in click-through rate and a 12% to 28% decrease in conversion costs, thereby effectively improving overall ad placement efficiency and optimizing budget utilization.

[0054] S105. Based on the time optimization suggestions and combined with the platform resource integration information in the matching process details data, simulate the adjustment effect in the cross-platform push scenario to obtain a set of optimization suggestions.

[0055] Obtain the current timestamp and the time distribution sequence from historical push records. Determine the push activity interval to which the current time belongs based on the time distribution sequence to obtain the current active tag. Extract resource integration information for the corresponding platform from the matching process details data for the current active tag to obtain a platform resource vector. Use k-means clustering to group the platform resource vectors, obtaining resource cluster centers and the category to which each platform belongs. Select the category with the highest matching degree to the current active tag from the resource cluster centers to obtain the target resource category. Filter available resource combinations for cross-platform push scenarios using the target resource category to obtain a candidate push content list. Sort the candidate push content list according to the historical adjustment effects in the matching process details data to obtain an optimization suggestion set.

[0056] For example, in the business scenario of targeted advertising, the system first obtains the current system timestamp and extracts the exposure and click distribution sequence of each time period in the past week from the historical push records.

[0057] Specifically, suppose the sequence shows that click-through rate during the weekday morning peak (7:00-9:00) accounts for 28% of the total daily click-through rate, and during the evening peak (20:00-22:00) accounts for 35%, while on weekends it is concentrated between 14:00-17:00. By analyzing this sequence, we can determine which push activity period the current time belongs to, and thus assign the current activity label, such as "high activity during the weekday evening peak".

[0058] It should be noted that, for the currently active tags, resource integration information of the corresponding platform is extracted from the detailed data of the matching process to form a platform resource vector.

[0059] In one possible implementation, each platform resource vector includes dimensions such as the proportion of video content, the preference for image content duration, and the frequency of use of interactive components.

[0060] For example, the vectors on the Douyin platform may be biased towards a high proportion of 15-second short videos, while the Xiaohongshu platform tends to favor a combination of text and image notes and long images.

[0061] Preferably, the k-means clustering method is used to group all collected platform resource vectors. After clustering, several resource cluster centers are obtained, and the category to which each platform belongs is marked.

[0062] In one embodiment, clustering may form three types of centers: the first type is mainly short videos with high dynamic effects, the second type is mainly static text and images with light interaction, and the third type is a balanced type of mixed text, images and videos.

[0063] Specifically, from these resource cluster centers, the category with the highest matching degree with the currently active tags is selected.

[0064] For example, when the current active tag is "high activity during weekday evening rush hour", the system tends to select the first type of center because historical data shows that users during this period prefer fragmented and highly stimulating short video content, thus obtaining the target resource category.

[0065] In one embodiment, the available resource combinations for cross-platform push scenarios are further filtered by target resource category to generate a candidate push content list. Assuming the target category is short video-dominated, material combinations that match the characteristics of this category are selected from the resource pools of platforms such as Douyin and Kuaishou, such as a 15-second demonstration video of a personal care product paired with fun challenge stickers.

[0066] Understandably, the candidate push content list is sorted according to the historical adjustment effects in the detailed data of the matching process. The sorting is mainly based on indicators such as the increase in click-through rate and the percentage reduction in conversion cost in previous campaigns.

[0067] For example, a video with challenge stickers saw a 42% increase in historical click-through rate during peak evening hours, while a purely explanatory video only saw an 18% increase. Therefore, the former is prioritized in the optimization suggestion set. This approach ensures that the optimization suggestion set not only includes the most suitable resource combinations for the current active period but also effectively reduces wasted advertising due to mismatches between platform characteristics and user activity. For instance, when the system provides three suggestions that are all short videos combined with interactive components, advertisers can choose them directly, thereby improving the efficiency of the exposure-to-conversion process. This typically results in an overall click-through rate increase of over 20%, significantly improving the utilization of the advertising budget.

[0068] S106. For the set of optimization suggestions, generate a feedback report for the advertiser. The feedback report includes a transparent description of the matching process details and optimization suggestions. If the user trust index is lower than a preset threshold, the explanation of the matching process details is highlighted in the feedback report to obtain the final feedback output.

[0069] Obtain the set of optimization suggestions and matching process log data. Extract the execution order and key decision points of each stage of the matching process from the log data to obtain a detailed sequence of the matching process. Calculate the contribution weight of each stage to the final matching result for the detailed sequence of the matching process, resulting in a contribution weight list. If the user trust index is below a preset threshold, select the top three stages from the contribution weight list, extract the detailed execution logs of the corresponding stages, and obtain highlighted explanations. Combine and arrange the detailed sequence of the matching process, the highlighted explanations, and the set of optimization suggestions to obtain the main structure data of the feedback report. Map the main structure data of the feedback report to a predefined report template using template filling to obtain the complete feedback report text. Output the complete feedback report text as the final feedback output.

[0070] For example, in the business area of ​​optimizing push content, the process of obtaining optimization suggestions and recording matching data can be understood as a systematic information processing process. The core is extracting valuable information from a large amount of historical push data.

[0071] Specifically, suppose a platform has pushed out 1,000 content notifications in the past 30 days, with detailed records for each notification, including notification time, user feedback, and content category. By organizing these records, each step of the matching process can be clearly identified, such as the priority rules for content filtering and the matching logic for user profiles. The advantage of this approach is that it provides a comprehensive data foundation for subsequent analysis, ensuring that optimization suggestions are targeted and effective.

[0072] For example, when extracting the detailed sequence of the matching process, key decision nodes can be identified from the recorded data. Suppose that in a single push notification, the system first filters content based on the user's active time, then further narrows the scope based on interest tags, and finally determines the final push content by combining historical click-through rates. These three nodes correspond to the three dimensions of time, interest, and historical behavior, respectively. Extracting the operational sequence and logic of these nodes helps to understand the overall picture of the matching process. Such a detailed sequence not only clearly shows the system's decision-making path but also provides a basis for subsequent weight analysis, thereby improving the transparency and interpretability of the push strategy.

[0073] For example, calculating the contribution weight list allows analysis of the impact of each step in the matching process on the final result. In the push notification example above, suppose active time filtering contributes 40% to the final user click-through rate, interest tag matching contributes 35%, and historical behavior analysis contributes 25%. This weight distribution clearly shows which steps have a greater impact on the result, providing direction for optimization. This analytical approach helps platforms focus on high-impact steps and improve resource allocation efficiency.

[0074] For example, in scenarios where the user trust index is below a threshold, extracting the execution logs of the top three weighted steps as highlighted explanatory content can enhance the user's understanding of the push notification results. Assuming the trust threshold is 70% and the current user trust index is 60%, the system will extract detailed logs of active time filtering, interest tag matching, and historical behavior analysis, such as which specific time periods were filtered, which tags were matched, and which historical data were referenced. This content can intuitively demonstrate the push notification logic to the user, enhancing trust and providing direct feedback for the platform to improve the user experience.

[0075] For example, when combining the detailed sequence of the matching process, highlighted explanations, and a set of optimization suggestions into the main body of a feedback report, a structured approach can ensure information integrity. Assume the report body is divided into three parts: first, an overview of the entire matching process; second, detailed explanations of key steps; and finally, a list of optimization suggestions based on current data. This structured arrangement allows users to quickly grasp the key points and provides a clear framework for subsequent template filling, thus improving report readability.

[0076] For example, when generating a complete feedback report using a template-based approach, a standardized report template can be pre-designed, including fixed titles, paragraph distribution, and data insertion points. Assuming the template includes a space for an overview of the matching process, the system will automatically fill in the extracted detailed sequences and insert an optimized list of content in the suggestions section. The advantages of this approach are high report generation efficiency, consistent formatting, easy access to necessary information for users, and reduced manual intervention costs.

[0077] For example, when outputting a complete feedback report, multiple channels can be used to ensure users receive the information. Once the report is generated, it can be pushed to users via platform messages or sent via email. This flexible output method improves user reach and provides the platform with more possibilities for collecting user feedback, thereby continuously optimizing the effectiveness of the push strategy.

[0078] If the technical solution of this application involves the collection, processing, or application of personal information, the relevant products have, before implementing any personal information processing activities, fully and clearly informed individuals of the processing rules in accordance with the "Personal Information Protection Law of the People's Republic of China" and other current laws and regulations, and obtained their voluntary and explicit consent. If sensitive personal information is involved, the product has obtained the individual's separate consent before processing, and such consent is given in an explicit manner. For example, prominent signs are set up in the area where information collection devices such as cameras are located, clearly indicating "Entering is considered as consent to the collection of personal information"; or through pop-ups, checkboxes, user-initiated uploads, etc., under the premise of clearly listing the processor's identity, processing purpose, processing method, and information type, the user actively completes the authorization operation. The above mechanisms ensure that all personal information processing activities are based on legal authorization and fully comply with national compliance requirements regarding personal information protection.

[0079] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A cross-platform intelligent advertising matching and push system and method, characterized in that, The method includes: The matching process details data are obtained from the cross-platform advertising push system. The matching process details data includes the association records between user group characteristics and advertising content. The matching process details data are processed by a classification algorithm to obtain a set of classified matching details. Based on the categorized set of matching details, the numerical distribution of the campaign performance metrics is obtained. If the click-through rate in the campaign performance metrics is lower than a preset threshold, the corresponding matching details are marked as inefficient categories, and the causes of the inefficient categories are determined. To address the causes of the poor performance of the aforementioned categories, relevant user behavior logs are extracted from the matching detail set, and a clustering algorithm is used to group the user behavior logs to obtain a set of grouped behavior patterns. Based on the grouped set of behavioral patterns, factors that do not match the advertising content are identified. If the time period deviation of the factors exceeds a preset threshold, corresponding time adjustment parameters are generated to obtain time optimization suggestions. Based on the time optimization suggestions and combined with the platform resource integration information in the matching process details data, the adjustment effect under the cross-platform push scenario is simulated to obtain a set of optimization suggestions. For the set of optimization suggestions, a feedback report is generated for the advertiser. The feedback report includes a transparent description of the matching process details and optimization suggestions. If the user trust index is lower than a preset threshold, the explanation of the matching process details is highlighted in the feedback report, resulting in the final feedback output.

2. The cross-platform intelligent advertising matching and push system and method according to claim 1, characterized in that, The process involves obtaining matching process detail data from a cross-platform advertising push system. This matching process detail data includes records of the association between user group characteristics and advertising content. A classification algorithm is then used to process the matching process detail data to obtain a categorized set of matching details, including: Detailed data of the matching process is obtained through a cross-platform advertising push system. The data includes records of the association between user group characteristics and advertising content. The data is preliminarily processed using a classification algorithm to obtain a set of classified matching details. Based on the categorized matching details set, extract the association records between user group characteristics and advertising content, group and organize each group of association records, and determine the grouped feature content set. If there are duplicate related records in the grouped feature content set, the duplicate records are deduplicated, and the feature content set is updated with the deduplicated data to obtain the simplified related feature set. Obtain a simplified set of related features. For each set of features and the matching details of the advertisement content, use a preset threshold to filter them. If the correlation strength of the matching details is lower than the preset threshold, remove the matching details and determine the filtered subset of matching details. The core features of the user group are extracted from the filtered matching details subset, and the correspondence between the core features and the advertising content is recorded and integrated to obtain the integrated matching mapping set. By integrating the matching mapping set, we analyze the potential association patterns between user group characteristics and advertising content, and use a classification algorithm to perform secondary classification of the potential association patterns to determine the final classification association results. Based on the final classification and association results, cross-platform ad push matching optimization data is generated. The optimization data is then stored and processed to obtain a structured matching optimization dataset.

3. The cross-platform intelligent advertising matching and push system and method according to claim 1, characterized in that, The step involves obtaining the numerical distribution of campaign performance metrics based on the categorized set of matching details. If the click-through rate (CTR) among the campaign performance metrics is lower than a preset threshold, the corresponding matching detail is marked as an inefficient category. The cause of the inefficient category's campaign performance is then determined, including: Obtain the set of matching details after categorization, and calculate the click-through rate of each matching detail by the distribution of performance metrics. The click-through rate (CTR) value for each matching detail is compared with a preset threshold. If the CTR is lower than the preset threshold, the matching detail is marked as an inefficient category, resulting in a set of inefficient category matching details. Extract multiple dimensions of campaign performance metrics from the inefficient category matching detail set to obtain the inefficient category metric data set; The decision tree algorithm is used to divide the data set of inefficient category indicators, obtain the indicator combination corresponding to multiple sub-nodes, and determine the main causes of the delivery effect of each sub-node; For each child node, the original attribute value range of the corresponding matching details is obtained for the main reasons for the delivery effect, resulting in a set of causal attribute ranges; The frequency distribution of causes was obtained by statistically analyzing the occurrence frequency of each causal attribute range in the detail set of inefficient category matching. The causes of advertising effectiveness are ranked according to the frequency distribution of causes, resulting in a priority sequence of causes for inefficient advertising effectiveness.

4. The cross-platform intelligent advertising matching and push system and method according to claim 1, characterized in that, The reasons for the poor performance of the targeting category are determined by extracting relevant user behavior logs from the matching detail set, and then using a clustering algorithm to group the user behavior logs to obtain a set of grouped behavior patterns, including: For a set of behavioral patterns, by analyzing the grouped data, we can obtain the common features within the patterns and get preliminary behavioral classification results. Based on the preliminary behavior classification results, statistical tools are used to calculate the frequency and duration of user behaviors within each category, and key behavior indicators are determined. If the key behavioral indicators are lower than the preset threshold, the log data in the corresponding category will be deeply filtered to obtain abnormal behavior fragments and determine the potential causes of inefficiency. For abnormal behavior segments, the corresponding delivery environment data is extracted from the matching details, the correlation between the environment and the behavior is analyzed, and the environmental influencing factors are obtained; Based on environmental influencing factors, the behavioral pattern set is regrouped and adjusted, and a clustering algorithm is used to optimize the classification boundary to determine the final behavioral pattern classification. By classifying the final behavioral patterns, we can obtain the correspondence between each category and the campaign performance, and identify the core problems of inefficient categories. For the core issues, we extract behavioral records for relevant time periods from the log data, analyze the specific scenarios in which the problems occurred, and obtain a detailed distribution of the problems.

5. The cross-platform intelligent advertising matching and push system and method according to claim 1, characterized in that, The step involves determining factors that do not match the advertising content based on the grouped set of behavioral patterns. If the time period deviation among these factors exceeds a preset threshold, corresponding time adjustment parameters are generated to obtain time optimization suggestions, including: Obtain the set of user behavior patterns after grouping and the current ad delivery time information; For each time distribution segment in the set of user behavior patterns, extract the sequence of active peak time points; By comparing the active peak time sequence with the ad content delivery time information segment by segment, the misalignment position of the time period is determined. If the number of misaligned positions in a time period reaches three or more, then calculate the offset duration value of each misaligned position; Input each offset duration value into the preset threshold comparison module to determine whether the offset duration value exceeds the corresponding threshold. When the offset duration value exceeds the corresponding threshold, a time adjustment parameter for reverse translation is generated for that offset duration value; By summarizing all time adjustment parameters that exceed the threshold, a final list of time optimization suggestions is generated.

6. The cross-platform intelligent advertising matching and push system and method according to claim 1, characterized in that, Based on the time optimization suggestions and combined with the platform resource integration information in the matching process details, the adjustment effect in a cross-platform push scenario is simulated to obtain a set of optimization suggestions, including: Get the current timestamp and the time distribution sequence in historical push records; Determine the current active push interval based on the time distribution sequence to obtain the current active tag; For currently active tags, extract the resource integration information of the corresponding platform from the detailed data of the matching process to obtain the platform resource vector; k-means clustering is used to group platform resource vectors to obtain resource cluster centers and the category to which each platform belongs; The target resource category is obtained by selecting the category with the highest matching degree with the currently active tag from the resource cluster center; By filtering out available resource combinations for cross-platform push scenarios based on target resource categories, a list of candidate push content is obtained. The candidate push content list is sorted according to the historical adjustment effects in the matching process details data to obtain a set of optimization suggestions.

7. The cross-platform intelligent advertising matching and push system and method according to claim 1, characterized in that, For the set of optimization suggestions, a feedback report is generated for the advertiser. This feedback report includes a transparent description of the matching process details and optimization suggestions. If the user trust index is below a preset threshold, the feedback report highlights the explanation of the matching process details, resulting in the final feedback output, including: Obtain the set of optimization suggestions and the data recorded during the matching process; By recording the execution order and key decision points of each step in the matching process, a detailed sequence of the matching process can be obtained. The contribution weights of each step in the detailed sequence of the matching process to the final matching result are calculated, resulting in a list of contribution weights. If the user trust index is lower than the preset threshold, the top three steps in the contribution weight list are selected, and the detailed execution logs of the corresponding steps are extracted to obtain the highlighted explanation content. The detailed sequence of the matching process, the highlighted explanatory content, and the set of optimization suggestions are combined and arranged to obtain the main structure data of the feedback report; The main structure data of the feedback report is mapped to a predefined report template using a template-filling method to obtain the complete feedback report text. Output the complete feedback report text as the final feedback output.

Images