Accurate advertisement putting optimization method and system

Abstract

The invention discloses a precise advertisement putting optimization method and system, and relates to the technical field of advertisement putting, and the method comprises the following steps: collecting and recording key parameter information of different data sources in a merging process in real time, providing basic data for the subsequent anomaly detection, and providing an optimization result for guaranteeing the comprehensiveness of the data; recording each step of operation in all merging processes; and carrying out anomaly analysis on the key parameter information subjected to the standardized processing, and identifying abnormal data generated in the merging process by analyzing multi-dimensional information. According to the invention, by monitoring the cross-session persistence information and the platform situation consistency information in real time, the system can effectively prevent the problem of data merging errors or segmentation, improve the accuracy of user portraits, and accurately push advertisement contents, thereby improving the click rate and the conversion rate. A multi-level anomaly detection and grading processing mechanism ensures that risks are timely handled, the safety of data integration and the stability of advertisement putting are improved, and the data integrity is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of advertisement delivery, and in particular to a method and system for optimizing accurate advertisement delivery. Background Art

[0002] The method of optimizing precise advertising delivery refers to improving the accuracy and effectiveness of advertising delivery through technical means such as data analysis and user behavior tracking, so as to ensure that the advertisement is displayed to the target audience who is most likely to be interested at the right time and place. First, advertisers understand the characteristics and behavioral preferences of target users through data collection and analysis. This data may include information such as the user's age, gender, geographic location, interests and hobbies, and online behavior. With this information, advertisers can create detailed user portraits and develop personalized delivery strategies based on the characteristics of different audience groups. For example, the advertising content and delivery channels for young groups may be very different from the strategies for middle-aged and elderly users.

[0003] Secondly, precise advertising also involves the optimization of advertising content and form to increase user interaction and conversion rates. Through A / B testing, machine learning algorithms and automation tools, advertisers can adjust the display frequency, delivery time and advertising form of ads in real time to ensure the best advertising effect. In addition, continuous monitoring of advertising delivery data, such as click-through rate, conversion rate, etc., can help advertisers quickly identify problems and make timely adjustments, thereby further improving the effectiveness of advertising delivery. Through these means, advertisers can effectively reduce advertising costs and maximize the return on investment (ROI) of advertising.

[0004] The prior art has the following deficiencies:

[0005] The data obtained for advertising often comes from multiple sources, including websites, social media, mobile applications, etc. In the process of precise advertising, when merging data from different sources, if there is a merger anomaly, such as mistakenly merging the data of two different users into the same user, or dividing the data of the same user into multiple different users, it will cause the user portrait to be distorted, and the advertising system will not be able to accurately judge the user's needs and interests. As a result, the advertising content does not match the user's needs, resulting in a decline in advertising effectiveness, a deterioration in user experience, and even a waste of advertising budget and loss of business opportunities.

[0006] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0007] The purpose of the present invention is to provide a method and system for optimizing precise advertising delivery. By real-time monitoring and anomaly detection of cross-session persistence information and platform context consistency information during data merging, the system can effectively prevent the problem of incorrectly merging data of different users or segmenting data of the same user, thereby improving the accuracy of user portraits, ensuring that the advertising system can accurately push content that meets user needs, and improving advertising click-through rates and conversion rates. Through multi-level anomaly detection and hierarchical processing mechanisms, the system can promptly identify and respond to low, medium, and high-risk anomalies, avoid the adverse consequences of data integration errors, improve the security and controllability of the integration process, ensure the stability of advertising delivery and the integrity of data, so as to solve the problems in the above-mentioned background technology.

[0008] In order to achieve the above object, the present invention provides the following technical solution: a method for optimizing precise advertisement delivery, comprising the following steps:

[0009] Collect and record key parameter information of different data sources in the merging process in real time, provide basic data for subsequent anomaly detection, and record every step of the merging process to ensure the comprehensiveness of the data;

[0010] Perform abnormal analysis on the key parameter information after normalization, and identify abnormal data generated during the merging process by analyzing multi-dimensional information;

[0011] Based on the abnormal analysis results, the parameter information after abnormal analysis is input into the pre-trained machine learning model, and the potential abnormalities are identified through the machine learning model;

[0012] Further comprehensive analysis is conducted on the identified potential anomalies of different data sources to determine the risk level of potential anomalies in the data merger, and different response measures are taken for different potential anomaly risk levels.

[0013] Preferably, the parameter information generated by different data sources during the merging process includes cross-session persistence information and platform context consistency information. The cross-session persistence information refers to the consistency and coherence of user behavior between different sessions, and the platform context consistency information is used to measure whether the user's behavior patterns and interests on different platforms are consistent.

[0014] Preferably, after obtaining the cross-session persistence information and platform context consistency information generated by different data sources during the merging process, a cross-session persistence index is generated after performing an abnormal analysis on the cross-session persistence information, and a platform context consistency index is generated after performing an abnormal analysis on the platform context consistency information. The cross-session persistence index and the platform context consistency index are used to identify abnormal data generated by different data sources during the merging process.

[0015] Preferably, after obtaining the cross-session persistence index and platform context consistency index generated by different data sources during the merging process, the cross-session persistence index and platform context consistency index are input into a pre-trained machine learning model, and a merge anomaly coefficient is generated by the machine learning model. The merge anomaly coefficient is used to perform an intelligent evaluation of the merging process of different data sources.

[0016] Preferably, if the merge abnormality coefficient is greater than or equal to a preset merge abnormality coefficient reference threshold, an abnormal merge process signal is generated, indicating that different data sources are in an abnormal state during the merge process;

[0017] If the merge anomaly coefficient is less than a preset merge anomaly coefficient reference threshold, a normal merge process signal is generated, indicating that different data sources are in a stable state during the merge process.

[0018] Preferably, after an abnormal merging process signal is generated when different data are merged, several merged abnormality coefficients generated by different data sources in the merging process are continuously obtained to establish an analysis set, and the merged abnormality coefficient in the analysis set is compared with the first-level reference threshold, the second-level reference threshold and the merged abnormality coefficient reference threshold, wherein the second-level reference threshold is greater than the first-level reference threshold, and the first-level reference threshold is greater than the merged abnormality coefficient reference threshold, and the merged abnormality coefficient is compared and analyzed with the second-level reference threshold, the first-level reference threshold and the merged abnormality coefficient reference threshold, and the number of merged abnormality coefficients that are less than the first-level reference threshold and greater than or equal to the merged abnormality coefficient reference threshold is calibrated as Qa, the number of merged abnormality coefficients that are less than the second-level reference threshold and greater than or equal to the first-level reference threshold is calibrated as Qb, and the number of merged abnormality coefficients that are greater than or equal to the second-level reference threshold is calibrated as Qc.

[0019] Comprehensively analyze Qa, Qb, and Qc to generate the abnormal risk index ARI, based on the formula:

[0020] ,

[0021] In the formula, w 1 、w 2 、w 3 are the weight coefficients of Qa, Qb, and Qc respectively, and w 1 、w 2 、w 3 , are both greater than 0.

[0022] Preferably, the generated abnormal risk index is compared and analyzed with the preset first abnormal risk index reference threshold and the second abnormal risk index reference threshold to determine the potential abnormal risk level of the data merger. The analysis results are as follows:

[0023] If the abnormal risk index is less than the first abnormal risk index reference threshold, the data combined potential abnormal risk level is classified as low risk abnormality;

[0024] If the abnormal risk index is greater than or equal to the first abnormal risk index and less than or equal to the second abnormal risk index, the data merge potential abnormal risk level is classified as medium risk abnormality;

[0025] If the abnormal risk index is greater than or equal to the second abnormal risk index, the data merge potential abnormal risk level is classified as high risk abnormality.

[0026] Preferably, after performing anomaly analysis on the cross-session persistence information, the logic for generating the cross-session persistence index is as follows:

[0027] In the detection window, collect key behavior data of users in multiple devices or sessions. Key behavior data includes: , the start time of each session T s and end time T e , platform source identifier P i And session IP address information IP i , D i represents the device ID of the user of the i-th session, P i Indicates the platform source identifier of the ith session, IP i Indicates the IP address information of the i-th session;

[0028] The system establishes cross-session associations by acquiring key behavior data and matches each user behavior with a persistence metric to evaluate the matching degree of behaviors between different devices or sessions. The calculation expression is as follows: ij =f(D i , D j )+g(p i , p j )+h(IP i , IP j ), where M ij represents the matching degree between session i and session j, f(D i , D j ) is the consistency function of the device ID, which indicates the matching degree of the device ID between the i-th and j-th sessions, g(p i , p j ) is the platform source identifier consistency function, which represents the matching degree of the platform source identifier between the ith and ith sessions, h(IP i , IP j ) is the IP address information consistency function, which represents the matching degree of session IP address information between the i-th and j-th sessions;

[0029] The system applies a time weight to each session match. The time weight takes into account the time span of the session and the time gap between different sessions to evaluate the consistency of behavior across time periods. The calculation expression is as follows: Where W time (T s , T e ) represents the time weight, α is the time sensitivity parameter;

[0030] In order to identify abnormal situations, it is necessary to calculate the abnormal coefficient across sessions. The calculation expression is as follows: ij =1-W time (T s , T e )·M ij , where A ij represents the anomaly coefficient between the i-th and j-th sessions;

[0031] By comprehensively processing all abnormal coefficients, a quantitative persistence evaluation value is generated to generate the cross-session persistence index. The calculation expression is as follows: Where CSP represents the cross-session persistence index, n represents the total number of sessions, and ln(1+M ij ) is the logarithmic correction term of the matching degree.

[0032] Preferably, after performing an abnormal analysis on the platform context consistency information, the logic for generating the platform context consistency index is as follows:

[0033] In the detection window, collect user behavior data on different platforms, including behavior type B s and content category C s , where B s = {B 1 , B 2 , B 3 ,……,B z}, C s ={C 1 , C 2 , C 3 ,……,C z}, where z represents the total number of platforms and behavior type B s Indicates the user's behavior type on platform s, content classification C s Represents the content categories that users interact with on platform s;

[0034] Based on the user's behavior data on different platforms, the user's behavior consistency deviation coefficient on each platform is calculated. The behavior consistency deviation coefficient reflects the degree of difference in user behavior on different platforms. The calculation expression is as follows: D s =|u(B s ,C s)-u(B t ,C t )|, where D s represents the behavioral consistency deviation coefficient, B s and C s represent the user's behavior type and content category on platform s, B t and C t represents the user's behavior type and content category on platform t, and u(x) is a nonlinear mapping function used to quantify the behavior type and content category;

[0035] The influence weights of user behaviors on consistency on different platforms are different. Therefore, different weight coefficients are set for different types of platforms. The adjusted behavior consistency deviation coefficient is calculated. The calculation expression is as follows: W s =D s ×λ s , where W s is the adjusted behavioral consistency deviation coefficient, λ s is the weight coefficient of platform s, reflecting the weight of the impact of user behavior on consistency on the platform:

[0036] Considering the differences in user contexts on different platforms, the correlation context coefficient is introduced to adjust the deviation coefficient and further refine the consistency calculation. The calculation expression is as follows: s =W s γ(U s ,L s ), where A s represents the deviation coefficient after the contextual coefficient is adjusted, U s represents the temporal context of platform s, L s represents the geographical context of platform s, γ(U s ,L s ) represents the time and geographical context function, reflecting the impact of users’ usage habits in different time periods and locations on behavioral consistency;

[0037] The deviation coefficients of multiple platforms are integrated by nonlinear weighting to generate the platform context consistency index. The calculation expression is as follows:

[0038] ,

[0039] Where PCC represents the platform context consistency index, μ s Indicates the frequency of interaction between users and platform s.

[0040] An advertising precision delivery optimization system, characterized by comprising a data collection and recording module, an anomaly analysis module, a machine learning anomaly identification module, and an anomaly risk assessment and response module;

[0041] The data collection and recording module collects and records key parameter information of different data sources in the merging process in real time, provides basic data for subsequent anomaly detection, and records every step of the merging process to ensure the comprehensiveness of the data;

[0042] The abnormal analysis module performs abnormal analysis on the key parameter information after normalization, and identifies abnormal data generated during the merging process by analyzing multi-dimensional information;

[0043] The machine learning anomaly recognition module inputs the parameter information processed by the anomaly analysis into a pre-trained machine learning model based on the anomaly analysis results, and identifies potential anomalies through the machine learning model;

[0044] The abnormal risk assessment and response module further conducts comprehensive analysis on the potential abnormalities identified from different data sources, determines the potential abnormal risk level of the data merger, and takes different response measures for different potential abnormal risk levels.

[0045] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0046] The present invention conducts real-time monitoring and anomaly detection of cross-session persistence information and platform context consistency information during the data merging process. The system can effectively prevent common problems in data merging, such as mistakenly merging data of different users into the same user or mistakenly dividing the data of the same user into multiple users. This precise data integration process greatly improves the accuracy of user portraits, so that the advertising system can push advertising content that better meets their needs based on real user interests and behaviors. Accurate advertising delivery not only improves the click-through rate and conversion rate of advertisements, but also effectively reduces users' antipathy to insensitive advertisements and improves the overall user experience. In addition, through the introduction of automated anomaly detection and machine learning models, the system can promptly detect potential problems and take countermeasures to ensure the stable operation of the advertising system and reduce the waste of advertising budgets and loss of business opportunities due to data errors.

[0047] The present invention ensures that any potential problems arising in the data integration process can be handled in a timely manner by constructing a multi-level anomaly detection and processing mechanism. The hierarchical management of low-risk, medium-risk and high-risk anomalies enables the system to adopt different response strategies according to the severity of the anomaly. Low-risk anomalies can be easily resolved through automated monitoring and data correction, while for medium-risk and high-risk anomalies, the system can issue an alarm and terminate the data merging process in time to prevent the problem from spreading further. By monitoring the abnormal risk index, the data administrator can clearly understand the status of each link in the merging process and perform manual intervention when necessary. Such a hierarchical processing mechanism greatly improves the security and controllability of data integration and avoids the serious consequences caused by data integration errors. In addition, through data backup and log recording, the system can quickly recover in abnormal situations to ensure data security and integrity. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0049] Figure 1 The present invention is a method flow chart of a method for optimizing precise advertisement delivery.

[0050] Figure 2 A module schematic diagram of an advertisement precision delivery optimization system of the present invention. DETAILED DESCRIPTION

[0051] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0052] The present invention provides Figure 1 The method for optimizing precise advertisement delivery shown includes the following steps:

[0053] Collect and record key parameter information of different data sources in the merging process in real time, provide basic data for subsequent anomaly detection, and record every step of the merging process to ensure the comprehensiveness of the data;

[0054] The parameter information generated by different data sources during the merging process includes cross-session persistence information and platform context consistency information. Cross-session persistence information refers to the consistency and coherence of users' behaviors between different sessions. Platform context consistency information is used to measure whether users' behavior patterns and interests on different platforms are consistent.

[0055] During the data merging process, data from different channels (such as websites, social media, mobile applications, etc.) need to be matched, normalized, and merged. During this process, the system will generate a large amount of key parameter information, which is the core element to ensure the accuracy of data merging;

[0056] To ensure the accuracy and reliability of data merging, the system must fully record every step of the operation, including data collection, cleaning, deduplication, matching, classification and final merging. The key parameters generated in each link need to be recorded in detail so that the source of the problem can be traced when anomalies occur later.

[0057] These detailed records not only ensure the correctness of data merging, but also provide basic data support for subsequent anomaly detection. For example, if a user shows multiple geographic location changes in a short period of time, the system can trace back these records to determine whether there is a data merging error. Every step of the operation recorded by the system can ensure the transparency of the data during the merging and provide an important basis for troubleshooting and correcting problems. In addition, the complete key parameter information can also be used for the training and optimization of machine learning models, further improving the system's anomaly perception and correction capabilities. Therefore, it is crucial to fully record each operation step and ensure the comprehensiveness of key parameter information in the data merging process.

[0058] Perform abnormal analysis on the key parameter information after normalization, and identify abnormal data generated during the merging process by analyzing multi-dimensional information;

[0059] After obtaining the cross-session persistence information and platform context consistency information generated by different data sources during the merging process, an abnormal analysis is performed on the cross-session persistence information to generate a cross-session persistence index, and an abnormal analysis is performed on the platform context consistency information to generate a platform context consistency index. The cross-session persistence index and the platform context consistency index are used to identify abnormal data generated when different data sources are merged;

[0060] When data from different sources are merged, changes in the cross-session persistence index may indeed lead to the data of two different users being mistakenly merged into the same user, or the data of the same user being split into multiple different users, resulting in a distorted user profile and a negative impact on the judgment ability of the advertising system. The cross-session persistence index measures the coherence and consistency of user behavior across different devices, sessions, or platforms. If the index fluctuates abnormally, the system may not be able to correctly identify the cross-device or cross-platform behavior of the same user, resulting in multiple sessions being mistakenly identified as coming from different users, or multiple users' sessions being mistakenly attributed to the same person. For example, if a user uses different browsers or identity identifiers on a mobile phone and a computer, changes in the cross-session persistence index will prevent the system from correctly associating these sessions, resulting in data integration errors. A distorted user profile will not only make it difficult for the advertising system to accurately judge the needs and interests of users, but will also lead to irrelevant content being pushed to ads, reduce the effectiveness of advertising, waste budget, and affect the user experience. Therefore, maintaining the stability of the cross-session persistence index is crucial for data integration, which helps to generate accurate user profiles and thus achieve precise advertising.

[0061] After performing anomaly analysis on the cross-session persistence information, the logic for generating the cross-session persistence index is as follows:

[0062] In the detection window, collect key behavior data of users in multiple devices or sessions. Key behavior data includes: , the start time of each session T s and end time T e , platform source identifier P i And session IP address information IP i The above information will be used as the basic data input in the subsequent steps. i represents the device ID of the user of the i-th session, P i Indicates the platform source identifier of the ith session, IP i Indicates the IP address information of the i-th session;

[0063] The system establishes cross-session associations by acquiring key behavior data and matches each user behavior with a persistence metric to evaluate the matching degree of behaviors between different devices or sessions. The calculation expression is as follows: ij =f(D i , D j )+g(p i , p j )+h(IP i , IP j ), where M ij represents the matching degree between session i and session j, f(D i , D j) is the consistency function of the device ID, which indicates the matching degree of the device ID between the i-th and j-th sessions, g(p i , p j ) is the platform source identifier consistency function, which represents the matching degree of the platform source identifier between the i-th and j-th sessions, h(IP i , IP j ) is the IP address information consistency function, which represents the matching degree of session IP address information between the i-th and j-th sessions. This step determines the association between sessions by combining the matching degrees of device, platform and IP;

[0064] It should be noted that:

[0065] In the correlation of behaviors between different devices or sessions, different devices refer to various hardware terminals used by users, such as mobile phones, computers, tablets, smart TVs, etc., and the user behaviors on these devices may be correlated. Different sessions refer to the activity records of users over a period of time, that is, the interaction process between users and the platform over a period of time. For example, a user browses products on a mobile phone in the morning and pays for shopping on a computer in the afternoon. These two operations belong to two different sessions. The correlation measures whether the behaviors between these devices or sessions are coherent, helping the system determine whether they belong to the same user.

[0066] The device ID consistency function is used to evaluate whether the device IDs of two sessions are consistent. If the device IDs are the same, it means that the two sessions are from the same device, and the matching function may return 1, indicating a perfect match; if the device IDs are different, the function will return 0, indicating a low match.

[0067] The platform source identifier consistency function evaluates whether the access platforms of two sessions are consistent. If the user operates on the same platform, the match is high; if the user switches between different platforms (such as from a mobile phone to a computer), the match is low but there may still be a correlation. Generally, it returns 1 when the platforms are consistent, and returns 0.5 when the platforms are different but there is a correlation.

[0068] The IP address information consistency function is used to evaluate whether the IP addresses of two sessions are consistent, or whether they are in similar network environments. The matching degree of IP addresses can help determine whether two sessions are from the same network or geographical location. Exactly the same IP address usually indicates that the sessions are from the same physical location or network environment, and the matching degree is 1. If the IP addresses are different, a matching value between 0 and 1 is returned based on the similarity of the IP addresses (such as in the same city or subnet).

[0069] The system applies a time weight to each session match. The time weight takes into account the time span of the session and the time gap between different sessions to evaluate the consistency of behavior across time periods. The calculation expression is as follows: Where W time (T s , T e ) represents the time weight, α is the time sensitivity parameter;

[0070] It should be noted that time weight refers to the process of assigning different weights to different data points or events according to different time points or time intervals in data analysis or models. Its main function is to adjust the importance of data or events according to the proximity of time. The core concept of time weight is that events or behaviors closer to the current time tend to better reflect the user's current state or preference, so they should be given higher weights; while events that occurred earlier usually have lower weights because they are more different from current behaviors. Time weight can help the system more effectively capture the changing trends of user behavior in cross-time analysis and improve the accuracy of predictions and decisions.

[0071] In order to identify abnormal situations, it is necessary to calculate the abnormal coefficient across sessions. The calculation expression is as follows: ij =1-W time (T s , T e )·M ij , where A ij represents the anomaly coefficient between the i-th and j-th sessions;

[0072] It should be noted that in the process of data integration, behavior analysis or system monitoring, identifying abnormal situations is an important part of ensuring data accuracy and improving system performance. Abnormal situations usually refer to events or data that deviate from normal behavior or patterns, which may reflect problems in data collection, system operation, user behavior, etc. The following are several common types of abnormal situations and their detailed explanations:

[0073] 1. Data matching anomaly

[0074] Data matching anomalies usually occur during the merging of data from different sources, mainly manifested as different data not being correctly matched or being over-merged. Specific anomalies include:

[0075] Device ID mismatch: Device IDs of different users (such as the hardware identifiers of mobile phones or computers) are mistakenly identified as the same device, resulting in incorrect merging of user behavior data.

[0076] User identifier conflicts: Different users use similar login information, cookies, or account IDs, and the system mistakenly identifies these users as the same person, resulting in distorted data integration.

[0077] Inconsistent merging of platforms: User operations on different platforms (such as mobile apps and desktop browsers) are mistakenly identified as multiple independent users, resulting in segmentation of user behavior.

[0078] 2. Time anomaly

[0079] Temporal anomalies occur during data integration or behavior tracking and involve abnormal performance in the time dimension, usually including:

[0080] Timestamp conflict: The timestamps obtained by the system show that two events occurred at the same time at impossible points in time, or the time sequence is unreasonable. For example, it is physically impossible for a user to be in two different geographical locations at the same time, but the timestamps show that the behavior occurred almost at the same time.

[0081] Abnormal time span: The time interval between multiple sessions of a user does not conform to a regular pattern. For example, a user has no activity for a long period of time, and then suddenly performs a large number of operations, which may reflect an unusual behavior pattern or data collection problem.

[0082] 3. Abnormal geographical location

[0083] Geolocation anomalies refer to abnormalities in the user's location data, which may be due to data collection errors or abnormal user behavior. Common anomalies include:

[0084] IP address cross-region anomaly: A user's IP address crosses multiple geographic regions in a short period of time, reflecting an impossible physical movement. For example, a user jumps from one country to another within a few minutes, which may be due to the IP address being misidentified or the user using a virtual private network (VPN) or proxy service.

[0085] Frequent location switching: The user's location switches frequently within a short period of time, showing abnormal behavior. This usually indicates that there is a problem with the user's device or the network environment is unstable.

[0086] 4. Abnormal session duration

[0087] Abnormal session duration means that some user sessions are too long or too short, deviating from the normal range, which may reflect problems with the system or user behavior. Common abnormalities include:

[0088] Long session times: Some sessions last an abnormally long time (such as hours or days), which may indicate that the session was not closed properly or that the system did not correctly detect the end of the session.

[0089] Session time is too short: The session is suddenly interrupted or switched quickly, which may be caused by users frequently refreshing the page, system crashes, or malicious users performing a large number of rapid operations through automated scripts.

[0090] 5. Abnormal traffic or behavior

[0091] Traffic anomalies usually occur in user behavior analysis or system load monitoring, and are manifested as the system suddenly receiving a large amount of traffic or an abnormal surge in a certain behavior. These situations include:

[0092] Sudden traffic surge: A certain function of a website or application suddenly receives a large number of access requests, which may be caused by external attacks (such as DDoS attacks) or abnormal user behavior.

[0093] Inconsistent behavior patterns: User behavior that is significantly inconsistent with historical behavior. For example, if a user suddenly switches from buying low-priced products to buying large quantities of high-priced items, or if a user's click behavior on a website appears to be unusually frequent, this may indicate that the user's account has been compromised or interfered with by automated programs.

[0094] 6. Abnormal click-through rate or interaction rate

[0095] During ad delivery and user behavior monitoring, the system may detect abnormally high or low click-through rates or interaction rates. Common abnormalities include:

[0096] Abnormally high click-through rate: A sudden increase in the click-through rate of an ad or page may mean that there is an error in the advertising strategy or there is false click behavior (such as click fraud, robot traffic, etc.).

[0097] Abnormally low interaction rate: After the ad is released, the user interaction rate (such as clicks, shares, purchases) of the ad is much lower than expected, which may indicate that the target population does not match the ad content, or that there are technical problems with the ad display.

[0098] 7. Abnormal frequency behavior

[0099] Abnormal frequency behavior refers to the frequent execution of certain operations by users within a specific time period, which exceeds the normal behavior range. Typical abnormal frequency behaviors include:

[0100] Rapid, repeated logins: If a user logs in or attempts to log in multiple times within a short period of time, it may be a sign that their account is under malicious attack or brute force attempts.

[0101] Submitting a large number of requests in a short period of time: Users or programs send a large number of requests to the server in a short period of time, such as quickly submitting forms, refreshing pages, requesting API interfaces, etc. This behavior usually reflects automated script attacks or abnormal user behavior.

[0102] 8. Abnormal conversion rate

[0103] In e-commerce platforms or advertising, abnormal conversion rates (i.e. the proportion of users performing certain key behaviors, such as purchases, registrations, etc.) are situations that require special attention. Common abnormalities include:

[0104] A sudden increase in conversion rate: Users complete a large number of conversion behaviors (such as purchasing products and submitting orders) in a short period of time. This may be the result of a promotion, but it may also be a system failure or cheating.

[0105] Conversion rate is too low: The ad click rate is normal, but there is no corresponding conversion. This may be a problem with the advertising strategy, or there may be problems at the website technical level that prevent users from completing purchases or registrations.

[0106] 9. Abnormal login / registration behavior

[0107] Abnormal login or registration behavior is usually related to user account security or system performance issues. Common situations include:

[0108] Frequent login failures: If a user or IP address fails to log in multiple times in a short period of time, it may be that an attacker is trying to brute force the account password.

[0109] Batch registration anomaly: The system receives a large number of user registration requests in a short period of time, which may be caused by robot programs creating fake accounts in batches.

[0110] 10. Abnormal system performance

[0111] In addition to user behavior, system performance anomalies also require attention. This includes:

[0112] Slow response time: Some part of the system (e.g., database queries, page loads) is unusually slow to respond for a period of time, perhaps due to high server load or resource bottlenecks.

[0113] Surge in system errors: A large number of system errors (such as 500 errors and database connection errors) suddenly appear in the error log, reflecting a serious system failure or a problem in the data processing process.

[0114] Identifying anomalies is a key step to ensure the accuracy of system and user behavior data integration. Anomalies may come from multiple dimensions such as data collection, user behavior, geographic location, time, etc. Identifying these anomalies can help the system detect potential risks, prevent data deviations, cheating, and system operation failures, thereby ensuring the quality of data processing and user experience.

[0115] By comprehensively processing all abnormal coefficients, a quantitative persistence evaluation value is generated to generate the cross-session persistence index. The calculation expression is as follows:

[0116] ,

[0117] Where CSP represents the cross-session persistence index, n represents the total number of sessions, and ln(1+M ij ) is the logarithmic correction term of the matching degree, which is used to strengthen the influence of high matching degree;

[0118] It should be noted that the logarithmic correction term of the matching degree refers to a mathematical processing method that uses a logarithmic function to adjust the matching degree in order to smooth the data or reduce the impact of extreme values ​​on the results when calculating the matching degree. The purpose of this correction term is to convert the original matching degree (usually linear) into a nonlinear change through the characteristics of the logarithmic function, so as to more effectively handle extreme values ​​or increase the influence of certain specific ranges of data on the final results.

[0119] The role of the logarithmic correction term:

[0120] Smoothing extreme values: The logarithmic function grows slowly, so for the extremely large matching values, the impact of the values ​​will be reduced after logarithmic correction, avoiding excessive impact of extreme high values ​​on the final calculation of matching. For example, the difference between two data points with original matching degrees of 1 and 100 will be reduced after logarithmic correction.

[0121] Amplify differences in low-degree matches: The logarithmic correction term can also amplify differences in low-degree matches, making subtle differences between low-value matches more significant. In the lower-degree area, the system may pay more attention to slight differences, and the logarithmic correction can better distinguish different data points in this range.

[0122] Adjust data distribution: The original matching degree may show an uneven distribution (such as being highly concentrated in a certain range). Through logarithmic correction, the data distribution is smoother, which helps to improve the stability and accuracy of the calculation results.

[0123] From the cross-session persistence index calculation expression, we can see that the larger the cross-session persistence index performance value generated after the anomaly analysis of the cross-session persistence information, the greater the probability of anomalies in the data from different sources during the merging process. This is because the cross-session persistence index measures the consistency of behavior between different sessions or devices. If the index shows a large value, it means that the behavior between different sessions or devices is significantly different, and data from different sources may not be matched correctly, which will lead to anomalies during data integration, such as erroneously merging the data of two different users into one user, or erroneously dividing the data of the same user into multiple users. Conversely, the smaller the cross-session persistence index, the higher the consistency between different sessions, the higher the accuracy of data merging, and the lower the probability of anomalies.

[0124] When data from different sources are merged, the platform context consistency index changes, which may cause the data of two different users to be mistakenly merged into the same user, or the data of the same user to be divided into multiple users. This will directly lead to distortion of user portraits, and the advertising system will not be able to accurately judge the needs and interests of users. The platform context consistency index reflects the consistency of user behavior on different platforms. If the index is abnormal, it may mean that the system fails to correctly identify and associate the user's behavior on different platforms or devices. When the index fluctuates abnormally, the system may mistakenly believe that users on two different platforms are the same person, especially when some of their behaviors or device information are similar, which will cause the data of different users to be incorrectly merged. On the contrary, if the behavior of the same user on multiple platforms is not effectively associated by the system, it may cause their behavior data to be divided into multiple users, causing the same user to be identified as multiple independent users. These problems will cause the advertising system to be unable to build an accurate user portrait, and thus unable to accurately locate the interests and needs of users when pushing ads, ultimately affecting the conversion effect and accuracy of ads.

[0125] After performing anomaly analysis on the platform context consistency information, the logic for generating the platform context consistency index is as follows:

[0126] In the detection window, collect user behavior data on different platforms, including behavior type B s and content category C s , where B s = {B 1 , B 2 , B 3 ,……,B z}, C s ={C 1 , C 2 , C 3 ,……,C z}, where z represents the total number of platforms and behavior type B s represents the user's behavior type on platform s (such as purchase, browsing, sharing, commenting, etc.), content category C s Indicates the content categories that users interact with on platform s (e.g., technology, fashion, education, etc.);

[0127] It should be noted that the role of collecting behavior types and content categories is very important in the process of precise advertising and data integration, and mainly plays a role in the following aspects:

[0128] 1. Describe user behavior patterns and interest preferences

[0129] Behavior types (such as browsing, purchasing, liking, commenting, etc.) and content categories (such as technology, fashion, entertainment, etc.) together reflect how users interact and what they are interested in on the platform. This information can help the system fully understand the user's behavior patterns. For example, if a user frequently makes purchases on an e-commerce platform and focuses on the technology product category, it means that he has a strong interest in buying technology products. Conversely, if a user interacts more with entertainment content on a social platform, it may indicate that he pays more attention to entertainment content during his leisure time. This detailed classification of behavior and interests can effectively build a personalized portrait of the user.

[0130] 2. Improve the personalization of targeted advertising

[0131] The core of precise advertising is to push relevant advertising content to users based on their behaviors and interests. By collecting behavior types, we can know whether users are browsing, comparing products, or ready to buy, so as to push more suitable ads. For example, when users are browsing, push eye-catching product introductions; and when users have shown a tendency to buy, push promotional ads or discount information will be more convertible. Similarly, content categories help to lock in the subject areas that users are interested in, ensure that the advertising content is highly relevant to user preferences, avoid pushing irrelevant ads, and improve advertising effectiveness.

[0132] 3.Support multi-platform data integration

[0133] The types of user behaviors and content categories vary on different platforms. Collecting this information helps ensure the correlation of user behaviors across platforms when integrating multi-platform data. For example, on one platform, users mainly browse fashion content, while on another platform they mainly purchase technology products. Without collecting this detailed information, the system may not be able to correctly associate these behaviors, resulting in the incorrect merging of different users' behaviors into one person, or the splitting of the same user's data into multiple independent individuals. By collecting and analyzing this behavior and content data, such errors can be effectively avoided, thereby better integrating data and maintaining the integrity of user portraits.

[0134] 4. Improve the accuracy of data analysis and anomaly detection

[0135] Collecting behavior types and content categories can also help the system identify changes and anomalies in user behavior. For example, if a user suddenly switches from technology purchases to a lot of entertainment browsing, this may mean a change in user interests or needs. By analyzing these behavioral changes, new consumer trends can be discovered. In addition, abnormal behavior detection also relies on the collection of user behavior types, especially in multi-platform or cross-device situations. If user behavior types and content categories are completely inconsistent on different platforms, it may mean that there are problems in the data integration process or that the data quality is abnormal. Timely detection of these anomalies is crucial for advertising delivery and data accuracy optimization.

[0136] 5. Used for machine learning and model training

[0137] Behavior types and content categories are important feature data for training advertising delivery models and user portrait prediction models. By learning the behavior types and content categories of a large number of users on different platforms, the system can better predict users' future behavior or consumption tendencies. In this way, machine learning models can predict users' future needs based on their historical behaviors and interests, thereby pushing appropriate advertising content in advance and maximizing the conversion rate of advertising.

[0138] The purpose of collecting behavior types and content categories is to depict the personalized portrait of users, improve the precise delivery of advertisements, support multi-platform data integration, enhance the accuracy of data analysis and anomaly detection, and provide necessary feature data for machine learning and model training. This information is the basis for the operation of data-driven advertising systems, helping advertisers to understand users more effectively and formulate personalized delivery strategies, thereby achieving better delivery results.

[0139] Based on the user's behavior data on different platforms, the user's behavior consistency deviation coefficient on each platform is calculated. The behavior consistency deviation coefficient reflects the degree of difference in user behavior on different platforms. The calculation expression is as follows: D s =|u(B s ,C s )-u(B t ,C t )|, where D s represents the behavioral consistency deviation coefficient, B s and C s represent the user's behavior type and content category on platform s, B t and C t represents the user's behavior type and content category on platform t, and u(x) is a nonlinear mapping function used to quantify the behavior type and content category;

[0140] It should be noted that a nonlinear mapping function refers to a function that converts input data into output values ​​in a nonlinear way. This mapping relationship does not follow a simple linear proportional relationship, but converts input data into output through more complex calculations (such as square, exponential, logarithm, sine, etc.). Unlike linear functions, nonlinear mapping can handle more complex relationships because it is not limited to linear proportional changes between input and output, but can amplify, reduce or distort the effects of different inputs and capture complex interactions and changes in multidimensional data.

[0141] In the data integration and analysis process of precise advertising delivery, nonlinear mapping functions play an important role in mapping user behavior types and content categories into quantifiable values. Specifically, the roles of nonlinear mapping functions in this scenario include:

[0142] 1. Capture the complex relationship between behavior and content

[0143] The relationship between user behavior types (such as browsing, purchasing) and content categories (such as technology, fashion) is often not a simple linear one. Different combinations of behaviors and content may have different effects on user interests and needs. For example, the impact of a user's purchasing behavior on technology content may not be consistent with that of their browsing behavior. By introducing a nonlinear mapping function, the system can map this complex relationship between behavior and content into a numerical value. Nonlinear functions can accurately capture the complexity of the interaction between behavior and content, and reflect the true impact of these behaviors on user portraits.

[0144] 2. Amplify or minimize the impact of specific actions or content

[0145] The nonlinear mapping function can amplify or reduce the impact of certain behaviors or content categories on the consistency of user behavior. For example, if a certain behavior (such as purchase behavior) has a greater impact on the results of advertising, the nonlinear mapping function can amplify the impact of the purchase behavior through exponential, square, etc. For behaviors with less impact (such as simple browsing or liking), the function can appropriately reduce their impact. This adjustment allows the system to more accurately measure the weights of different behaviors, thereby better understanding user behavioral preferences.

[0146] 3. Complexity of dealing with multidimensional data

[0147] User behaviors and content categories are usually multi-dimensional and not limited to a specific platform or device. Non-linear mapping functions can handle complex interactions in multi-dimensional data. For example, when users show different behavior types and content preferences on multiple platforms or devices, linear mapping may not be able to capture these complex multi-dimensional interactions. Non-linear mapping functions can better reflect the consistency and differences of cross-platform and cross-device behaviors, making data integration more accurate.

[0148] 4. Improve the sensitivity of anomaly detection

[0149] Nonlinear mapping functions can also enhance the system's ability to detect abnormal behavior. When users' behavior types and content categories on different platforms are abnormal (for example, browsing a lot of fashion content on social platforms and purchasing technology products on e-commerce platforms), nonlinear functions can quickly detect these deviations and amplify these abnormal changes through complex mapping relationships, thereby helping the system to promptly detect potential merging errors or changes in user behavior. This increased sensitivity of anomaly detection is critical for accurate advertising delivery and can ensure the accuracy of data integration and delivery strategies.

[0150] 5. Provide richer data features for model use

[0151] The nonlinear mapping function converts complex behavior and content relationships into quantifiable values, which can be used as important feature data in machine learning and ad recommendation models. Compared with simple linear processing, nonlinear mapping captures the subtle changes between behavior and content, allowing the model to more deeply understand user behavior patterns, improve prediction accuracy and delivery effectiveness.

[0152] The influence weights of user behaviors on consistency on different platforms are different. Therefore, different weight coefficients are set for different types of platforms (such as e-commerce platforms and social platforms) to calculate the adjusted behavior consistency deviation coefficient. The calculation expression is as follows: W s =D s ·λ s , where W s is the adjusted behavioral consistency deviation coefficient, λ s is the weight coefficient of platform s, reflecting the weight of the impact of user behavior on consistency on the platform;

[0153] Taking into account the contextual differences of users on different platforms (such as different usage habits between working time and leisure time), the correlation context coefficient is introduced to adjust the deviation coefficient and further refine the consistency calculation. The calculation expression is as follows: s =W s γ(U s ,L s ), where A s represents the deviation coefficient after the contextual coefficient is adjusted, U s represents the time context of platform s (e.g., the time periods when users use the platform in the morning, afternoon, and evening), L s represents the geographical context of platform s (e.g., whether users operate in different geographical locations), γ(U s ,L s) represents the time and geographical context function, reflecting the impact of users’ usage habits in different time periods and locations on behavioral consistency;

[0154] Temporal context refers to the user's behavior patterns and usage habits in a specific time period. This context reflects the differences in user activities over time. For example, user behaviors and needs may be different in different time periods of the day, weekdays and weekends, seasonal activities, etc. Users may deal with more work-related content during working hours, while tend to relax, entertain, shop, etc. at night or on weekends. Therefore, understanding the temporal context can help the system push the most relevant content or ads at the right time, improving user experience and ad conversion rates.

[0155] Geographic context refers to the changes in user behavior and needs in different geographical locations. The user's geographical location will affect their activity types and preferences. For example, the user's behavior may be significantly different at home, in the office, in public places, or when traveling. Geographic context also includes macro-geographic factors such as the city and country where the user is located. Cultural differences, regional consumption habits, etc. will all have an impact on user preferences and behaviors. Understanding geographic context can help the system push localized content or services based on the user's specific location, such as nearby store promotions, restaurant recommendations, etc., thereby improving the accuracy and relevance of advertising.

[0156] Contextual differences refer to the fact that the same user's behaviors and preferences may be different at different times, places, devices or platforms. These differences are often caused by factors such as the user's specific environment, usage scenario, and psychological state. For example, a user may focus mainly on career-related content during working hours, but may browse entertainment or shopping websites during leisure time. Usage habits on different devices (such as mobile phones and computers) or geographical locations (such as home and public places) will also affect user behavior. Contextual differences reflect changes in user needs and behaviors in different contexts. These differences need to be considered when delivering targeted advertisements in order to provide the most relevant advertising content in the appropriate context.

[0157] The deviation coefficients of multiple platforms are integrated by nonlinear weighting to generate the platform context consistency index. The calculation expression is as follows:

[0158] ,

[0159] Where PCC represents the platform context consistency index, μ s Indicates the frequency of interaction between users and platform s, reflecting the importance of the platform in the user's behavior pattern;

[0160] The frequency of user interaction with a platform refers to the frequency with which users perform various actions (such as browsing, clicking, purchasing, liking, etc.) on a specific platform. It reflects the user's activity and usage habits on the platform, and is usually measured by the user's interaction data with the platform. A higher interaction frequency means that the user frequently participates in activities on the platform and may have a higher dependence or interest in the content or services on the platform; a lower interaction frequency indicates that the user's usage is sparse or occasional. This frequency is crucial in advertising and user behavior analysis because it can help the system understand how much importance users attach to a specific platform, thereby optimizing advertising strategies.

[0161] From the calculation expression of the platform context consistency index, it can be seen that the larger the performance value of the platform context consistency index generated after the abnormal analysis of the platform context consistency information, the higher the probability of abnormalities when the data from different sources are merged. The platform context consistency index reflects the consistency of user behavior on different platforms or situations. If the index value is large, it means that there are significant differences in user behavior and interaction patterns on different platforms, and it is difficult for the system to correctly associate these behaviors, thereby increasing the risk of data integration errors. Conversely, the smaller the performance value of the platform context consistency index, the more consistent the user's behavior on different platforms, the less likely it is to have abnormalities when merging data, and the higher the accuracy of the integration.

[0162] Based on the abnormal analysis results, the parameter information after abnormal analysis is input into the pre-trained machine learning model, and the potential abnormalities are identified through the machine learning model;

[0163] After obtaining the cross-session persistence index CSP and platform context consistency index PCC generated by different data sources during the merging process, the cross-session persistence index CSP and the platform context consistency index PCC are input into the pre-trained machine learning model, and the merge anomaly coefficient MA is generated by the machine learning model. The merge anomaly coefficient MA is used to perform intelligent evaluation on the merging process of different data sources;

[0164] The machine learning model is not limited here, and any machine learning model that can generate a combined anomaly coefficient MA after comprehensive analysis of the cross-session persistence index CSP and the platform context consistency index PCC is acceptable. In order to implement the technical solution of the present invention, the present invention provides a specific implementation method:

[0165] The calculation formula of combined abnormal coefficient MA is:

[0166] ,

[0167] In the formula, j 1 、j 2 are the preset proportional coefficients of the cross-session persistence index CSP and the platform context consistency index PCC, and j1 、j 2 Both are greater than 0.

[0168] It can be seen from the calculation expression of the merge anomaly coefficient that, within the detection window, the larger the cross-session persistence index performance value generated after the anomaly analysis of the cross-session persistence information, the larger the platform context consistency index performance value generated after the anomaly analysis of the platform context consistency information, the larger the generated merge anomaly coefficient performance value, indicating that the probability of anomalies occurring in the merging process of different data sources is greater, and vice versa, the probability of anomalies occurring in the merging process of different data sources is smaller;

[0169] The combined anomaly coefficients generated by different data sources during the merging process are compared and analyzed with the pre-set combined anomaly coefficient reference threshold. The analysis results are as follows:

[0170] If the merge abnormality coefficient is greater than or equal to a preset merge abnormality coefficient reference threshold, an abnormal merge process signal is generated, indicating that different data sources are in an abnormal state during the merge process;

[0171] If the merge anomaly coefficient is less than a preset merge anomaly coefficient reference threshold, a normal merge process signal is generated, indicating that different data sources are in a stable state during the merge process;

[0172] Further comprehensive analysis is conducted on the identified potential anomalies of different data sources to determine the risk level of potential anomalies in the data merger, and different response measures are taken for different potential anomaly risk levels;

[0173] After an abnormal merging process signal is generated when different data are merged, several merged abnormal coefficients generated by different data sources in the merging process are continuously obtained to establish an analysis set, and the merged abnormal coefficient in the analysis set is compared with the first-level reference threshold, the second-level reference threshold and the merged abnormal coefficient reference threshold, wherein the second-level reference threshold is greater than the first-level reference threshold, and the first-level reference threshold is greater than the merged abnormal coefficient reference threshold, and the merged abnormal coefficient is compared and analyzed with the second-level reference threshold, the first-level reference threshold and the merged abnormal coefficient reference threshold, and the number of merged abnormal coefficients that are less than the first-level reference threshold and greater than or equal to the merged abnormal coefficient reference threshold is calibrated as Qa, the number of merged abnormal coefficients that are less than the second-level reference threshold and greater than or equal to the first-level reference threshold is calibrated as Qb, and the number of merged abnormal coefficients that are greater than or equal to the second-level reference threshold is calibrated as Qc.

[0174] Comprehensively analyze Qa, Qb, and Qc to generate the abnormal risk index ARI, based on the formula:

[0175] ,

[0176] In the formula, w1 、w 2 、w 3 are the weight coefficients of Qa, Qb, and Qc respectively, and w 1 、w 2 、w 3 , are all greater than 0;

[0177] It can be seen from the calculation expression of the abnormal risk index that the larger the abnormal risk index performance value, the greater the probability of abnormalities when data from different sources are merged, and the smaller the abnormal risk index performance value, the smaller the probability of abnormalities when data from different sources are merged.

[0178] The generated abnormal risk index is compared and analyzed with the preset first abnormal risk index reference threshold and the second abnormal risk index reference threshold to determine the potential abnormal risk level of the data merger. The analysis results are as follows:

[0179] If the abnormal risk index is less than the first abnormal risk index reference threshold, the data combined potential abnormal risk level is classified as low risk abnormality;

[0180] If the abnormal risk index is greater than or equal to the first abnormal risk index and less than or equal to the second abnormal risk index, the data merge potential abnormal risk level is classified as medium risk abnormality;

[0181] If the abnormal risk index is greater than or equal to the second abnormal risk index, the data merge potential abnormal risk level is classified as high risk abnormality;

[0182] When the data merging potential abnormal risk level is classified as low risk abnormality, it means that the abnormal coefficient in the merging process is low, and the abnormal phenomenon may be just a slight fluctuation or a small inconsistency between the data. In this case, the system can take milder countermeasures, such as:

[0183] Automated monitoring and continuous tracking: The system can continue to monitor the subsequent behavior of the data merging process to ensure that the anomaly does not expand or develop further. Low risk usually does not require immediate action, but it is necessary to closely observe the subsequent data integration.

[0184] Data correction prompts: The system can generate warnings or reminders to notify data managers for review, but without immediate manual intervention. Minor data correction tools can be used to automate the cleaning and optimization of subtle errors in merges, such as normalization of data formats or adjustment of minor timestamp differences.

[0185] Recording logs: All detected low-risk anomalies should be recorded in the system log for later analysis. By accumulating these records, more context can be provided for potential problems in the future.

[0186] When the abnormal risk index falls into the medium-risk abnormal range, it means that more obvious abnormalities have occurred in the merging process, which may affect the accuracy or completeness of the data. In this case, the system should take more proactive measures, including:

[0187] Automated data correction: The system can start an automated data repair process to correct problems based on predefined rules. For example, it can automatically fix problems such as inconsistent timestamps and mismatched device IDs to ensure that data sources can be accurately merged.

[0188] Manual intervention prompt: At the same time, the system can send an alert to the data administrator or relevant personnel, indicating that manual intervention is required. The purpose of manual intervention is to manually review or verify the data when automatic correction fails to resolve or clarify the source of the anomaly.

[0189] Data isolation processing: For medium-risk anomalies, consider temporarily isolating the abnormal data and processing it independently or further analyzing it to avoid affecting the integrity and accuracy of the overall data consolidation. While isolating, other normal consolidation processes can still be continued to ensure the stability of data processing.

[0190] When the abnormal risk index is judged as a high-risk abnormality, it means that there are serious problems in the data merging process, which may lead to distortion of user portraits, advertising errors or system failures. In this case, serious measures must be taken immediately to deal with it, including:

[0191] Stop the data merge operation: The system should immediately suspend the current data merge process to prevent abnormal data from further affecting the overall data set. Stopping the operation helps prevent the problem from spreading or having a greater impact on other parts.

[0192] Deep anomaly analysis and repair: High-risk anomalies may require deep analysis of the data to find the root cause of the anomaly. It may require the intervention of data scientists or technical experts to identify and repair the problem through advanced data analysis methods (such as machine learning anomaly detection). Further repairs may include data rollback, manual adjustments, or re-execution of part of the data integration process.

[0193] Emergency handling and backup recovery: In high-risk situations, the system should initiate emergency plans, including restoring data from backups, to ensure that serious data merging errors do not lead to overall destruction of system data. At the same time, the system should record detailed exception reports for subsequent audits and system improvements.

[0194] The present invention conducts real-time monitoring and anomaly detection of cross-session persistence information and platform context consistency information during the data merging process. The system can effectively prevent common problems in data merging, such as mistakenly merging data of different users into the same user or mistakenly dividing the data of the same user into multiple users. This precise data integration process greatly improves the accuracy of user portraits, so that the advertising system can push advertising content that better meets their needs based on real user interests and behaviors. Accurate advertising delivery not only improves the click-through rate and conversion rate of advertisements, but also effectively reduces users' antipathy to insensitive advertisements and improves the overall user experience. In addition, through the introduction of automated anomaly detection and machine learning models, the system can promptly detect potential problems and take countermeasures to ensure the stable operation of the advertising system and reduce the waste of advertising budgets and loss of business opportunities due to data errors.

[0195] The present invention ensures that any potential problems arising in the data integration process can be handled in a timely manner by constructing a multi-level anomaly detection and processing mechanism. The hierarchical management of low-risk, medium-risk and high-risk anomalies enables the system to adopt different response strategies according to the severity of the anomaly. Low-risk anomalies can be easily resolved through automated monitoring and data correction, while for medium-risk and high-risk anomalies, the system can issue an alarm and terminate the data merging process in time to prevent the problem from spreading further. By monitoring the abnormal risk index, the data administrator can clearly understand the status of each link in the merging process and perform manual intervention when necessary. Such a hierarchical processing mechanism greatly improves the security and controllability of data integration and avoids the serious consequences caused by data integration errors. In addition, through data backup and log recording, the system can quickly recover in abnormal situations to ensure data security and integrity.

[0196] The present invention provides Figure 2 An advertising precision delivery optimization system shown includes a data collection and recording module, an anomaly analysis module, a machine learning anomaly identification module, and an anomaly risk assessment and response module;

[0197] The data collection and recording module collects and records key parameter information of different data sources in the merging process in real time, provides basic data for subsequent anomaly detection, and records every step of the merging process to ensure the comprehensiveness of the data;

[0198] The abnormal analysis module performs abnormal analysis on the key parameter information after normalization, and identifies abnormal data generated during the merging process by analyzing multi-dimensional information;

[0199] The machine learning anomaly recognition module inputs the parameter information processed by the anomaly analysis into a pre-trained machine learning model based on the anomaly analysis results, and identifies potential anomalies through the machine learning model;

[0200] The abnormal risk assessment and response module further conducts comprehensive analysis on the potential abnormalities identified from different data sources, determines the potential abnormal risk level of the data merger, and takes different response measures for different potential abnormal risk levels.

[0201] An advertising precision delivery optimization method provided in an embodiment of the present invention is implemented through the above-mentioned advertising precision delivery optimization system. The specific methods and processes of the advertising precision delivery optimization system are detailed in the embodiment of the above-mentioned advertising precision delivery optimization method, which will not be repeated here.

[0202] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for optimizing precise advertising delivery, characterized in that: The following steps are involved: Collect and record key parameter information of different data sources in the merging process in real time, provide basic data for subsequent anomaly detection, and record every step of the merging process to ensure the comprehensiveness of the data; Perform abnormal analysis on the key parameter information after normalization, and identify abnormal data generated during the merging process by analyzing multi-dimensional information; Based on the abnormal analysis results, the parameter information after abnormal analysis is input into the pre-trained machine learning model, and the potential abnormalities are identified through the machine learning model; Further comprehensive analysis is conducted on the identified potential anomalies of different data sources to determine the risk level of potential anomalies in the data merger, and different response measures are taken for different potential anomaly risk levels.

2. The method for optimizing precise advertisement delivery according to claim 1, characterized in that: The parameter information generated by different data sources during the merging process includes cross-session persistence information and platform context consistency information. Cross-session persistence information refers to the consistency and coherence of user behavior between different sessions. Platform context consistency information is used to measure whether the user's behavior patterns and interests on different platforms are consistent.

3. The method for optimizing precise advertisement delivery according to claim 2, characterized in that: After obtaining the cross-session persistence information and platform context consistency information generated by different data sources during the merging process, a cross-session persistence index is generated after an abnormal analysis of the cross-session persistence information, and a platform context consistency index is generated after an abnormal analysis of the platform context consistency information. The cross-session persistence index and the platform context consistency index are used to identify abnormal data generated when different data sources are merged.

4. The method for optimizing precise advertisement delivery according to claim 3, characterized in that: After obtaining the cross-session persistence index and platform context consistency index generated by different data sources during the merging process, the cross-session persistence index and platform context consistency index are input into a pre-trained machine learning model, and a merge anomaly coefficient is generated by the machine learning model. The merge process of different data sources is intelligently evaluated through the merge anomaly coefficient.

5. The method for optimizing precise advertisement delivery according to claim 4, characterized in that: The combined anomaly coefficients generated by different data sources during the merging process are compared and analyzed with the pre-set combined anomaly coefficient reference threshold. The analysis results are as follows: If the merge abnormality coefficient is greater than or equal to a preset merge abnormality coefficient reference threshold, an abnormal merge process signal is generated, indicating that different data sources are in an abnormal state during the merge process; If the merge anomaly coefficient is less than a preset merge anomaly coefficient reference threshold, a normal merge process signal is generated, indicating that different data sources are in a stable state during the merge process.

6. The method for optimizing accurate advertisement delivery according to claim 5, characterized in that: After an abnormal merging process signal is generated when different data are merged, a number of merged abnormal coefficients generated by different data sources in the merging process are continuously obtained to establish an analysis set, and the merged abnormal coefficient in the analysis set is compared with the first-level reference threshold, the second-level reference threshold, and the merged abnormal coefficient reference threshold, wherein the second-level reference threshold is greater than the first-level reference threshold, and the first-level reference threshold is greater than the merged abnormal coefficient reference threshold, and the merged abnormal coefficient is compared and analyzed with the second-level reference threshold, the first-level reference threshold, and the merged abnormal coefficient reference threshold, and the number of merged abnormal coefficients that are less than the first-level reference threshold and greater than or equal to the merged abnormal coefficient reference threshold is calibrated as Qa, the number of merged abnormal coefficients that are less than the second-level reference threshold and greater than or equal to the first-level reference threshold is calibrated as Qb, and the number of merged abnormal coefficients that are greater than or equal to the second-level reference threshold is calibrated as Qc; Comprehensively analyze Qa, Qb, and Qc to generate the abnormal risk index ARI, based on the formula: ， Wherein, w1, w2, w3 are the weight coefficients of Qa, Qb, Qc respectively, and w1, w2, w3 are all greater than 0.

7. The method for optimizing precise advertisement delivery according to claim 6, characterized in that: The generated abnormal risk index is compared and analyzed with the preset first abnormal risk index reference threshold and the second abnormal risk index reference threshold to determine the potential abnormal risk level of the data merger. The analysis results are as follows: If the abnormal risk index is less than the first abnormal risk index reference threshold, the data combined potential abnormal risk level is classified as low risk abnormality; If the abnormal risk index is greater than or equal to the first abnormal risk index and less than or equal to the second abnormal risk index, the data merge potential abnormal risk level is classified as medium risk abnormality; If the abnormal risk index is greater than or equal to the second abnormal risk index, the data merge potential abnormal risk level is classified as high risk abnormality.

8. The method for optimizing accurate advertisement delivery according to claim 3, characterized in that: After performing anomaly analysis on the cross-session persistence information, the logic for generating the cross-session persistence index is as follows: In the detection window, collect key behavior data of users in multiple devices or sessions. Key behavior data includes: The start time of each session is T s and end time T e , platform source identifier P i And session IP address information IP i , D i represents the device ID of the user of the i-th session, P i Indicates the platform source identifier of the ith session, IP i Indicates the IP address information of the i-th session; The system establishes cross-session associations by acquiring key behavior data and matches each user behavior with a persistence metric to evaluate the matching degree of behaviors between different devices or sessions. The calculation expression is as follows: ij =f(D i , D j )+g(p i , p j )+h(IP i , IP j ), where M ij represents the matching degree between session i and session j, f(D i , D j ) is the consistency function of the device ID, which indicates the matching degree of the device ID between the i-th and j-th sessions, g(p i , p j ) is the platform source identifier consistency function, which represents the matching degree of the platform source identifier between the ith and ith sessions, h(IP i , IP j ) is the IP address information consistency function, which represents the matching degree of session IP address information between the i-th and j-th sessions; The system applies a time weight to each session match. The time weight takes into account the time span of the session and the time gap between different sessions to evaluate the consistency of behavior across time periods. The calculation expression is as follows: Where W time (T s , T e ) represents the time weight, α is the time sensitivity parameter; In order to identify abnormal situations, it is necessary to calculate the abnormal coefficient across sessions. The calculation expression is as follows: ij =1-W time (T s , T e )·M ij , where A ij represents the anomaly coefficient between the i-th and j-th sessions; By comprehensively processing all abnormal coefficients, a quantitative persistence evaluation value is generated to generate the cross-session persistence index. The calculation expression is as follows: ， Where CSP represents the cross-session persistence index, n represents the total number of sessions, and ln(1+M ij ) is the logarithmic correction term of the matching degree.

9. The method for optimizing accurate advertisement delivery according to claim 3, characterized in that: After performing anomaly analysis on the platform context consistency information, the logic for generating the platform context consistency index is as follows: In the detection window, collect user behavior data on different platforms, including behavior type B s and content category C s , where B s = {B1, B2, B3, ..., B z }, C s ={C1, C2, C3, ..., C z }, where z represents the total number of platforms and behavior type B s Indicates the user's behavior type on platform s, content classification C s Represents the content categories that users interact with on platform s; Based on the user's behavior data on different platforms, the user's behavior consistency deviation coefficient on each platform is calculated. The behavior consistency deviation coefficient reflects the degree of difference in user behavior on different platforms. The calculation expression is as follows: D s =|u(B s ,C s )-u(B t ,C t )|, where D s represents the behavioral consistency deviation coefficient, B s and C s represent the user's behavior type and content category on platform s, B t and C t represents the user's behavior type and content category on platform t, and u(x) is a nonlinear mapping function used to quantify the behavior type and content category; The influence weights of user behaviors on consistency on different platforms are different. Therefore, different weight coefficients are set for different types of platforms. The adjusted behavior consistency deviation coefficient is calculated. The calculation expression is as follows: W s =D s ×λ s , where W s is the adjusted behavioral consistency deviation coefficient, λ s is the weight coefficient of platform s, reflecting the weight of the impact of user behavior on consistency on the platform: Considering the differences in user contexts on different platforms, the correlation context coefficient is introduced to adjust the deviation coefficient and further refine the consistency calculation. The calculation expression is as follows: s =W s γ(U s ,L s ), where A s represents the deviation coefficient after the contextual coefficient is adjusted, U s represents the temporal context of platform s, L s represents the geographical context of platform s, γ(U s ,L s ) represents the time and geographical context function, reflecting the impact of users’ usage habits in different time periods and locations on behavioral consistency; The deviation coefficients of multiple platforms are integrated by nonlinear weighting to generate the platform context consistency index. The calculation expression is as follows: ， Where PCC represents the platform context consistency index, μ s Indicates the frequency of user interaction with platform s.

10. An advertisement precision delivery optimization system, used to implement an advertisement precision delivery optimization method as described in any one of claims 1 to 9, characterized in that: It includes data collection and recording module, anomaly analysis module, machine learning anomaly identification module, and anomaly risk assessment and response module; The data collection and recording module collects and records key parameter information of different data sources in the merging process in real time, provides basic data for subsequent anomaly detection, and records every step of the merging process to ensure the comprehensiveness of the data; The abnormal analysis module performs abnormal analysis on the key parameter information after normalization, and identifies abnormal data generated during the merging process by analyzing multi-dimensional information; The machine learning anomaly recognition module inputs the parameter information processed by the anomaly analysis into a pre-trained machine learning model based on the anomaly analysis results, and identifies potential anomalies through the machine learning model; The abnormal risk assessment and response module further conducts comprehensive analysis on the potential abnormalities identified from different data sources, determines the potential abnormal risk level of the data merger, and takes different response measures for different potential abnormal risk levels.

Images