Method and apparatus for processing advertisement conversion events

Abstract

The application discloses a processing method and device for an advertisement conversion event. The method comprises: in response to a received conversion event detection request, obtaining generated advertisement conversion data; based on the advertisement conversion data, determining a user behavior feature combination with a support degree greater than or equal to a first threshold in the advertisement conversion event, the support degree representing a ratio between a number of target conversion events containing the user behavior feature combination in the advertisement conversion event and a total number of the advertisement conversion events; determining a confidence of a candidate association mode created based on the user behavior feature combination, wherein the confidence represents a back transmission integrity ratio of the advertisement conversion event; and determining a back transmission anomaly detection result of the advertisement conversion event based on a back transmission confidence distribution of different user value levels determined based on the confidence. The application solves the technical problem of low accuracy in the back transmission anomaly detection process of the advertisement conversion data.

Description

Technical Field

[0001] This application relates to the field of internet advertising technology, and more specifically, to a method and apparatus for processing advertising conversion events. Background Technology

[0002] In current programmatic advertising on the internet, advertisers often engage in fraudulent practices such as partial traffic redirection (only redirecting ad conversion data) and selective redirection (redirecting only high-value users and filtering out low-value users) in order to reduce costs. To address this, advertising platforms typically rely on static rules or simple statistical indicators (such as overall redirection rate fluctuations, device fingerprints, etc.) to identify abnormal redirection behavior by advertisers, primarily filtering out fraudulent traffic, or using manually set redirection ratio thresholds for coarse-grained monitoring.

[0003] However, the above methods cannot identify advertisers' intentional, half-true, and half-false filtering and deduction behaviors of real advertising conversion data, and lack systematic verification of the completeness and unbiasedness of the returned data. In particular, they cannot identify covert fraud patterns such as abnormally high confidence levels in the returned data from high-value users, resulting in technical problems with low accuracy in the detection of abnormalities in the returned advertising conversion data.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This application provides a method and apparatus for processing advertising conversion events, so as to at least solve the technical problem of low accuracy in the process of detecting abnormalities in the return of advertising conversion data.

[0006] According to one aspect of the embodiments of this application, a method for processing advertising conversion events is provided, comprising: in response to a received conversion event detection request, acquiring generated advertising conversion data; based on the advertising conversion data, determining user behavior feature combinations in advertising conversion events with a support greater than or equal to a first threshold, wherein the support represents the ratio between the number of target conversion events containing the user behavior feature combinations in the advertising conversion event and the total number of advertising conversion events; determining the confidence level of candidate association patterns created based on the user behavior feature combinations, wherein the confidence level represents the return integrity ratio of the advertising conversion event under the user behavior feature combinations; and determining the return anomaly detection result of the advertising conversion event based on the return confidence level distribution of different user value levels determined by the confidence level.

[0007] According to another aspect of the embodiments of this application, an apparatus for processing advertising conversion events is also provided, comprising: a first acquisition unit, configured to acquire generated advertising conversion data in response to a received conversion event detection request; a first processing unit, configured to determine, based on the advertising conversion data, a combination of user behavior features in the advertising conversion event with a support greater than or equal to a first threshold, wherein the support represents the ratio between the number of target conversion events containing the combination of user behavior features in the advertising conversion event and the total number of advertising conversion events; a second processing unit, configured to determine the confidence level of a candidate association pattern created based on the combination of user behavior features, wherein the confidence level represents the return integrity ratio of the advertising conversion event under the combination of user behavior features; and a third processing unit, configured to determine the return anomaly detection result of the advertising conversion event based on the return confidence level distribution of different user value levels determined by the confidence level.

[0008] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, wherein a computer program is stored in the computer program, which is used to execute the above-described method for processing advertising conversion events when run by an electronic device.

[0009] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of the above-described method.

[0010] According to another aspect of the embodiments of this application, an electronic device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the processing method of the aforementioned advertising conversion event through the computer program.

[0011] Using the embodiments provided in this application, user behavior feature combinations (i.e., frequent itemsets) with support exceeding a first threshold in advertising conversion events are mined, and candidate association patterns are created based on these user behavior feature combinations during the feedback conversion event process. By quantifying the confidence level of each candidate association pattern, a basis for refined, data-driven detection of advertisers' selective feedback fraudulent behavior is obtained. Finally, the feedback confidence level distribution at different user value levels is calculated based on the confidence level, locating abnormal mutations in feedback behavior at user value breakpoints, thereby identifying advertisers' targeted deduction strategies based on user value. Compared to traditional solutions that can only detect overall feedback rate fluctuations, the technical solution of this application can achieve more granular, clearly attributable, and confidence-quantified anomaly detection results for abnormal feedback behavior in advertising conversion events. This improves the efficiency and accuracy of anomaly detection for selective feedback, ensures the authenticity of data fed back by the advertising platform, and guarantees the revenue of the advertising platform and the fairness of the algorithm. Attached Figure Description

[0012] The accompanying drawings, which are provided to further illustrate this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application.

[0013] Figure 1 This is a schematic diagram illustrating an application scenario of an optional method for processing advertising conversion events according to an embodiment of this application;

[0014] Figure 2 This is a flowchart of an optional method for processing advertising conversion events according to an embodiment of this application;

[0015] Figure 3 This is an overall flowchart of an optional method for processing advertising conversion events according to an embodiment of this application;

[0016] Figure 4 This is a schematic diagram of value breakpoint analysis based on the feedback confidence level of the user value division interval;

[0017] Figure 5 This is a diagram illustrating how to determine the return anomaly detection result by calculating the confidence level of the user based on the quantile of the return value;

[0018] Figure 6 This is a schematic diagram of the structure of an optional advertising conversion event processing device according to an embodiment of this application;

[0019] Figure 7 This is a schematic diagram of the structure of an optional electronic device according to an embodiment of this application. Detailed Implementation

[0020] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0021] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0022] The technical solutions in this application will comply with legal regulations during implementation. When operating according to the technical solutions in the embodiments, the data used will not involve user privacy, ensuring that the operation process is compliant and legal while guaranteeing data security. In addition, when the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use, and processing of related data must comply with the relevant regulations and standards of the relevant countries or regions.

[0023] According to one aspect of the embodiments of this application, a method for processing advertising conversion events is provided. As an optional implementation, the above-described method for processing advertising conversion events can be applied, but is not limited to, to applications such as... Figure 1 The application scenarios shown are as follows. In, for example... Figure 1 In the application scenario shown, the target terminal 102 can communicate with the server 106 via network 104, but is not limited to this. The server 106 can perform operations on the database 108, such as write or read data operations. The target terminal 102 may include, but is not limited to, a human-computer interaction screen, a processor, and a memory. The human-computer interaction screen may be used to display, but is not limited to, the feedback anomaly detection results obtained after processing using the technical solution of this application on the target terminal 102. The processor may be used, but is not limited to, to respond to the human-computer interaction operations, execute corresponding operations, or generate corresponding instructions and send the generated instructions to the server 106. The memory is used to store relevant processing data, such as advertising conversion data, user behavior feature combinations, and confidence levels.

[0024] Optionally, in this embodiment, the target terminal can be a terminal configured with a target client, which may include, but is not limited to, at least one of the following: mobile phone (such as Android phone, iOS phone, etc.), laptop computer, tablet computer, PDA, MID (Mobile Internet Devices), PAD, desktop computer, smart TV, etc. The target client may be a video client, instant messaging client, browser client, educational client, etc. The network may include, but is not limited to, wired network and wireless network, wherein the wired network includes: local area network, metropolitan area network and wide area network, and the wireless network includes: Bluetooth, WIFI and other networks that enable wireless communication. The server may be a single server, a server cluster composed of multiple servers, or a cloud server.

[0025] The technical solution of this application relates to the field of Internet advertising technology. Specifically, it belongs to the subfield of digital advertising placement risk control and conversion attribution verification, and further involves the application of conversion data feedback authenticity verification and anti-fraud technology in the entire advertising flow chain.

[0026] The technical solution of this application can be widely applied to the conversion feedback stage of advertising (including but not limited to mobile application advertising, web advertising, video advertising, etc.). Targeting fraudulent behaviors such as deducting ad data from feedback and filtering user feedback, it uses behavioral feature analysis, data mining algorithms, and multi-dimensional data cross-validation to identify and intercept deductions and filtering of ad conversion data (such as paid conversions, app installations, user activations, etc.) during the feedback process. This ensures the authenticity of the data fed back by the advertising platform and allows for penalties such as traffic restrictions or suspension of advertising for fraudulent advertisers, achieving the technical effect of incentivizing advertisers to feed back genuine conversion data.

[0027] To address this issue, this application proposes a method for handling advertising conversion events. Figure 2 This is a flowchart of a method for processing advertising conversion events according to an embodiment of this application. The process includes the following steps S202 to S208.

[0028] It should be noted that the method for handling the advertising conversion event shown in steps S202 to S208 can be, but is not limited to, executed by an electronic device. The electronic device can be, but is not limited to, [the following]. Figure 1 The target terminal or server shown.

[0029] Step S202: In response to the received conversion event detection request, obtain the generated ad conversion data;

[0030] Step S204: Based on the advertising conversion data, determine the user behavior feature combinations in the advertising conversion events with a support greater than or equal to a first threshold, wherein the support represents the ratio between the number of target conversion events containing the user behavior feature combinations in the advertising conversion events and the total number of advertising conversion events;

[0031] Step S206: Determine the confidence level of the candidate association pattern created based on the user behavior feature combination, wherein the confidence level represents the return integrity ratio of the ad conversion event under the user behavior feature combination.

[0032] Step S208: Based on the feedback confidence distribution of different user value levels determined by the confidence level, determine the feedback anomaly detection result of the advertising conversion event.

[0033] Before explaining the technical solution of this application, we will first give a brief introduction to the professional terms or basic knowledge involved in the embodiments of this application.

[0034] In the current process of advertising traffic conversion feedback, there is a clear incentive misalignment and information asymmetry between advertisers and advertising platforms, which leads to cheating behaviors that damage the platform model and ecosystem health in the feedback process. These behaviors mainly include (1) deducting data feedback: In order to control advertising costs and optimize the ROI of traffic, advertisers intentionally only feedback part of the advertising conversion data, resulting in the advertising platform's statistical advertising conversion cost being too high and the "cost-conversion" relationship learned by the model being distorted; (2) selective feedback: Advertisers selectively feedback based on user value, such as only feedback high-value users and filtering low-value users, which leads to a systematic deviation in the advertising platform's traffic value assessment and a bias towards low-value traffic.

[0035] Both of these behaviors are essentially acts of cheating by "lying with data." Their direct result is the distortion of the advertising platform's delivery model, making it unable to accurately learn the true conversion effects of ads and accurately predict the value of traffic. This not only harms the advertising platform's revenue and the fairness of its algorithm but also increases costs for advertisers who genuinely submit data due to model distortion.

[0036] Existing methods for preventing data transfer fraud primarily focus on technical anti-fraud measures. One approach is based on a transfer ratio-based strategy, which determines anomalies by simply checking the transfer ratio. Specifically, fluctuations in the transfer ratio indicate whether data transfer is being penalized. However, this method fails if advertisers initially penalize transferred data from conversions. Another approach focuses on confirming whether a conversion event has occurred, but lacks effective verification and auditing mechanisms for the completeness, representativeness, and unbiasedness of the transferred data. If a third party selectively transfers data (e.g., only transferring data to high-value users), it cannot be detected.

[0037] It is evident that current anti-fraud strategies for ad conversion feedback are either too simplistic or have not yet verified and audited the completeness, representativeness, and unbiasedness of ad feedback data. Based on this, this application proposes an ad traffic conversion feedback deduction detection device based on the Apriori algorithm.

[0038] Among them, the Apriori algorithm is a classic association rule (or association pattern) data mining algorithm that can discover the association relationship between items (or user behavior attributes) in big data.

[0039] The Apriori algorithm includes two core metrics. The first measures the frequency with which an itemset appears in all transactions. The support is calculated as the ratio of transactions containing that itemset to the total number of transactions. An itemset includes, but is not limited to, two different items in a transaction, such as a desk and a textbook. In the context of commodity transactions, the purpose of calculating support is to identify frequently occurring item combinations.

[0040] The second is confidence score, which measures the conditional probability that item B will also occur when item A occurs, i.e., the reliability of rule A→B. The purpose of calculating confidence score is to mine meaningful association rules from frequent itemsets. Its meaning and function will be further described below with specific application scenarios and examples.

[0041] In this implementation, in response to an ad conversion event (hereinafter referred to as conversion event) detection request received by the ad platform, the generated ad conversion data (including conversion data that has been returned and that has not been returned) is obtained. An ad conversion event may refer to, but is not limited to, a specific action that has commercial value as defined by the advertiser after a user clicks on or views an ad. This action is recorded by the system and returned to the ad platform to measure the effectiveness of the ad campaign.

[0042] For example, after User 1 clicks on the target ad, downloads and installs the corresponding app, and then completes their first payment (128 yuan) within 30 minutes, the conversion event recorded by the system is this behavior encapsulated as a structured data record, such as {User Mechanism Segmentation: High Value, Feedback Status: Feedback Completed, Feedback Time Interval: 28 minutes, Feedback Device Type: Android, User Type: New User, Feedback Conversion Event Type: Paid, Feedback Channel Quality: High}.

[0043] As can be seen, ad conversion events not only indicate that the user has completed the core business objective of payment, but also carry rich contextual features for subsequent analysis using the Apriori algorithm to determine if there is any fraudulent behavior of only sending back high-value paying users. If the advertiser only sends back the 128 yuan order, but conceals the other 10 low-value orders of 5 yuan each, this conversion event becomes key evidence for identifying selective retransmission.

[0044] In other words, the acquired advertising conversion event (which can also be understood as a conversion feedback event) is treated as a transaction. It contains multiple items, including multi-dimensional behavioral attributes or user behavior data, such as user value segmentation, whether feedback is performed, feedback time interval, feedback device type, user type, feedback conversion event type, feedback channel quality, etc.

[0045] After constructing a large number of such transactions, the system uses the Apriori algorithm to mine frequently occurring combinations of user behavior features layer by layer. Taking a minimum support threshold of 0.6 as an example, the system will filter out combinations that appear at least 60% of all conversion events, such as {high value, already submitted}, {high value, Android, new user, submission time interval ≤ 0.5h}, etc. These frequent itemsets reflect the "high-frequency co-occurrence" user feature patterns in advertiser submission behavior, which is the basis for subsequent discovery of abnormal association patterns. For example, if the system finds that combination 1, "high value + new user", appears simultaneously in 90% of submission events, while combination 2, "low value + old user", almost never appears in the submission list, it indicates that there is a clear user selective bias in submission behavior.

[0046] After obtaining frequent itemsets, the system further calculates the confidence level of each candidate association rule. The confidence level measures the probability of whether the consequent "retransmission" occurs, given that the combination of antecedent features is satisfied. For example, the confidence level of the candidate association rule {low-value user} → {not transmitted} is as high as 98%, meaning that almost all low-value users were not transmitted; while the confidence level of the candidate association rule {high-value user} → {transmitted} is 100%, meaning that all high-value users were transmitted completely. By comparing the confidence level distribution of different user value levels (such as high-value, medium-value, and low-value based on payment amount), the system can identify value breakpoint anomalies. For example, when a user's payment amount exceeds 70 yuan, the transmission confidence level jumps sharply from 0.18 to 0.85, indicating that the advertiser has a clear "transmission threshold" strategy, that is, only transmitting ad conversion events or ad conversion data for users with amounts exceeding this threshold, i.e., implementing deducted transmission or selective transmission.

[0047] Furthermore, the lift of each candidate association rule (which can also be understood as a candidate association pattern) is calculated to determine whether the association rule has actual business significance. If the lift of rule {high value, Android, new user} → {already uploaded} is 1.8, which is much greater than 1, it indicates that there is a strong positive correlation between the user behavior feature combination and the upload behavior, and it is not accidental; while the lift of rule {low value} → {already uploaded} is close to 0, indicating that the two are mutually exclusive. Combining confidence and lift, the system can accurately identify three typical cheating patterns: 1) selective uploads where high-value users upload all content and low-value users upload zero content; 2) threshold deductions for handling confidence mutations at a certain value breakpoint; 3) targeted uploads with abnormally high upload rates under specific device / channel combinations.

[0048] In this embodiment, the system first collects real advertising conversion feedback data to construct a transaction set with multi-dimensional features. Then, it uses the Apriori algorithm to mine frequently occurring combinations of user behavior features as a candidate set of potential abnormal patterns. Based on these combinations, it calculates the feedback confidence distribution at different user value levels and identifies abnormal patterns that significantly deviate from the normal distribution. This process enables intelligent reasoning from raw advertising conversion data to business rules, breaking through the static blind spots of traditional rule engines and providing advertising platforms with accurate, explainable, and traceable anti-fraud decision-making basis.

[0049] Using the embodiments provided in this application, user behavior feature combinations (i.e., frequent itemsets) with support exceeding a first threshold in advertising conversion events are mined, and candidate association patterns are created based on these user behavior feature combinations during the feedback conversion event process. By quantifying the confidence level of each candidate association pattern, a basis for refined, data-driven detection of advertisers' selective feedback fraudulent behavior is obtained. Finally, the feedback confidence level distribution at different user value levels is calculated based on the confidence level, locating abnormal mutations in feedback behavior at user value breakpoints, thereby identifying advertisers' targeted deduction strategies based on user value. Compared to traditional solutions that can only detect overall feedback rate fluctuations, the technical solution of this application can achieve more granular, clearly attributable, and confidence-quantified anomaly detection results for abnormal feedback behavior in advertising conversion events. This improves the efficiency and accuracy of anomaly detection for selective feedback, ensures the authenticity of data fed back by the advertising platform, and guarantees the revenue of the advertising platform and the fairness of the algorithm.

[0050] As an optional example, the above-mentioned determination of user behavior feature combinations with a support level greater than or equal to a first threshold in advertising conversion events based on the advertising conversion data includes:

[0051] Obtain multiple behavioral attribute fields contained in the advertising conversion feedback record corresponding to the advertising conversion event, wherein the multiple behavioral attribute fields include user value segmentation, feedback time interval and feedback device type, and the user value segmentation is used to quantitatively grade the degree of contribution of advertising conversion users to the revenue of the advertising entity.

[0052] Based on the multiple behavioral attribute fields, determine a combination of multiple attribute fields that include the at least one attribute field;

[0053] Determine the support degree corresponding to each attribute field combination in the plurality of attribute field combinations, and select a portion of the attribute field combinations whose support degree is greater than or equal to the first threshold from the plurality of attribute field combinations;

[0054] The combination of behavioral features corresponding to the combination of the aforementioned attribute fields is determined as the user behavior feature combination.

[0055] The user behavior feature combination includes behavioral features corresponding to at least one of the multiple behavioral attribute fields. User value segmentation can also be understood as a quantitative classification of users converted from advertising based on business objectives, used to assess the degree of revenue contribution of these users to the advertising entity.

[0056] In this embodiment, the generated ad conversion time records are first collected from the backend of the advertising platform. Each record consists of multiple structured behavioral attribute fields, which are set according to the advertiser's business objectives and the platform's risk control requirements. For example, user value segmentation is a quantitative classification based on the user's historical consumption behavior (such as cumulative payment amount, ARPU, and ECMP estimated value). For instance, users are divided into three categories: high value (payment ≥ 100 yuan), medium value (30~99 yuan), and low value (0~29 yuan). The core purpose is to identify the differences in revenue contribution of different user levels to the advertiser. The data return time interval reflects the timeliness of data return after the conversion, such as within 0.5 hours, within 1 hour, or within 24 hours. This is used to indicate whether the advertiser has a strategy of delayed return or "centralized return" of high-value events. The return device type, such as Android or web, is used to determine whether there is a preference for return under specific device environments.

[0057] Based on the aforementioned behavioral attribute fields, the system does not directly analyze individual fields, but rather randomly combines them to form attribute field combinations, i.e., the joint occurrence patterns of multiple features. For example, it generates combinations such as {high value, feedback time interval ≤ 0.5h}, {high value, Android, new user}, and {low value, feedback time interval > 24h}, which cover multiple dimensions of user value, time response, and device characteristics. Then, it calculates the frequency of these combinations in all advertising conversion events (which can also be understood as feedback conversion events or feedback conversion transactions), i.e., support.

[0058] For example, in 1000 postback conversion event records, the ratio between the number of ad conversion events containing the attribute field combination {high value, Android, new user} (e.g., the number of transactions containing this attribute field combination) and the total number of ad conversion events is the support score, such as 58%. Assuming the system sets a minimum support threshold of 60%, only attribute field combinations with a support score greater than or equal to 60% are retained, while attribute field combinations with a support score of only 15% (e.g., {low value, already posted back}) are removed. This filtering mechanism effectively filters out low-frequency, random combinations, focusing on high-frequency characteristic combinations that repeatedly appear in the advertiser's actual behavior patterns.

[0059] In other words, support can refer to, but is not limited to, the proportion of conversion events containing a certain combination of user behavior features among all returned ad conversion events (i.e., all ad conversion return transactions). When this proportion is greater than or equal to a first threshold, the combination of user behavior features is identified as a high-frequency behavior pattern. Furthermore, there is a correspondence between this combination of user behavior features under such a high-frequency behavior pattern and the frequent itemsets mined using the Apriori algorithm.

[0060] The system defines the selected high-support attribute field combinations as user behavior feature combinations. It's easy to understand that the multiple attribute field combinations selected through support filtering constitute the frequent itemsets with high occurrence frequency mined from all generated ad conversion events using the Apriori algorithm. A frequent item can contain one attribute field (such as high-value user or feedback interval), or it can contain two or more attribute fields (such as simultaneously containing high-value user and feedback interval). Multiple attribute field combinations correspond to multiple user behavior feature combinations.

[0061] The combination of the above multiple attribute fields forms the basis for subsequent association rule (which can also be understood as association pattern) mining (such as calculating confidence and lift), and is a key prerequisite for identifying the cheating pattern of "high-value users are almost 100% reported back, while low-value users are rarely reported back".

[0062] In this embodiment, key behavioral attribute fields are first defined through business logic. Then, multi-dimensional feature associations are constructed using random combinations. High-frequency behavioral patterns with statistical significance are filtered out through support, ultimately determining the core set of user behavior features for anomaly detection. This process automates the extraction of high-value feature patterns from raw data fields, providing a structured, quantifiable, and reusable analytical foundation for subsequent accurate identification and selective feedback, thus improving the generalization ability and business targeting of the anti-fraud model.

[0063] As an optional example, determining the confidence level of the candidate association patterns created based on the user behavior feature combinations includes:

[0064] Obtain the antecedent and consequent of each association pattern in the candidate association patterns, wherein the antecedent includes at least one behavioral attribute field corresponding to the user behavior feature combination, and the consequent includes a return identifier field indicating that the advertising conversion event has been returned.

[0065] Obtain a first number of ad conversion events that include both the antecedent and the consequent, and obtain a second number of ad conversion events that include only the antecedent.

[0066] The ratio between the first quantity and the second quantity is determined as the confidence level of each association pattern.

[0067] The return identifier field is set to 1, and the preceding information includes at least one behavior attribute field corresponding to at least one behavior feature in the user behavior feature combination.

[0068] First, each association pattern in the candidate association pattern refers to a candidate association rule or candidate association pattern extracted from the frequent itemset, including antecedent and consequent. The consequent can be, but is not limited to, the postback identifier field. When the postback identifier field is 1, it indicates that the advertiser did indeed post a conversion event (such as payment, installation, etc.) in that record, which is the goal of all rule analysis. The antecedent is a combination of user behavior characteristics, such as user value segmentation as high value, device type as webpage, and postback time interval within 0.5 hours. These fields together constitute a set of user behavior characteristic combinations. For example, if an association rule is {high value, Android, new user} → {postback=1}, it means that the system is concerned with whether advertisers tend to post their conversion behavior among user groups possessing these three characteristics.

[0069] The first quantity refers to the number of events among all ad conversion events that simultaneously meet the antecedent condition and the consequent is 1; that is, the total number of events that simultaneously meet the criteria of "high value + Android + new user" and are actually submitted back. The second quantity refers to the total number of events that only meet the antecedent condition, regardless of whether they are submitted back; that is, user conversion events of the "high value + Android + new user" type (regardless of whether the advertiser submits back). Assuming that out of 5000 conversion data points (one conversion data point corresponds to one conversion event), there are 1200 events of the "high value + Android + new user" type, of which 1150 are submitted back by the advertiser (the first quantity), then the confidence level is the ratio between 1150 and 1200, i.e., 95.8%. This high confidence level indicates that as long as the user possesses these three characteristics, the advertiser will almost certainly submit back, which is a typical signal of selective submission.

[0070] Using the ratio of the first quantity to the second quantity as the current confidence level essentially calculates the probability of an advertiser converting a user who already possesses a certain set of characteristics. The closer this value is to 1, the more targeted the advertiser's conversion behavior is towards this user group; conversely, if the user group is large but the conversion rate is extremely low (e.g., only 10%), it indicates that the advertiser may not be converting low-value users. By comparing the confidence levels under different antecedent combinations, abnormal conversion patterns can be identified. For example, if the confidence level for high-value users reaches 95%, while the confidence level for low-value users is only 2%, the system can determine that selective conversion behavior exists.

[0071] This embodiment presents a closed-loop analysis process, from defining the rule structure to data grouping and statistics, and finally to calculating conditional probabilities. It does not rely on the global feedback ratio but delves into the local correlation strength between user characteristics and feedback behavior. Through precise conditional probability modeling, it transforms previously ambiguous suspicions of abnormal feedback into quantifiable and comparable numerical indicators. This method overcomes the limitation of traditional feedback rate fluctuation detection, which fails during initial deductions, enabling advertising platforms to identify highly patterned fraudulent feedback behaviors hidden within massive amounts of data.

[0072] As an optional example, the above-mentioned distribution of postback confidence based on the confidence level of different user value levels, used to determine the postback anomaly detection result of the advertising conversion event, includes:

[0073] Based on the confidence level, the lift of each association pattern in the candidate association patterns is determined, wherein the lift represents the degree of deviation between the conditional probability of the ad conversion event being transmitted back and the proportion of the transmitted ad conversion events.

[0074] Based on the lift, abnormal association patterns are selected from the candidate association patterns, wherein the abnormal association patterns are used to indicate that the return of the advertising conversion event is the result of selective return filtering.

[0075] The anomaly detection results include the presence of the abnormal association pattern among the candidate association patterns. The lift can also be used to measure the deviation of the conditional probability of an ad conversion event being retransmitted, given the combination of user behavior features, from the proportion of retransmitted ad conversion events, where the proportion represents the ratio between the number of retransmitted ad conversion events and the total number of overall ad conversion events.

[0076] The lift is calculated by comparing the probability of a user returning to the platform under a specific combination of user characteristics with the proportion of returned conversions in the overall ad conversion events. In other words, it describes whether the probability of an advertiser returning to the platform is significantly higher than the average return rate among all users when the user is high-value, uses an Android device, or is a new user. For example, if the total return rate in all ad conversion events is 60%, the global return rate is 0.6. The confidence level of the rule {high-value user} → {return = 1} is 100%. In this case, the lift is the ratio between 100% and 60%, which is 1.67. This value is greater than 1, meaning that the presence of high-value users makes the likelihood of the advertiser returning to the platform 67% higher than random return, which is not accidental but a conscious selective return behavior.

[0077] Secondly, the physical significance of lift lies in eliminating interference from popular items. If a frequent itemset occurs with extremely high frequency (e.g., "new users" account for 80%), even if its confidence level with "returned items" is only 70%, it may appear normal due to its high background probability. However, if the confidence level for low-value users is only 5%, while the global return rate is 60%, its lift is the ratio between 5% and 60%, i.e., 0.08, far below 1. This means that advertisers deliberately do not return items to low-value users, which is also cheating, just in the opposite direction. Therefore, lift can not only identify positive cheating such as "only returning high-value items," but also capture negative cheating such as "only filtering low-value items," possessing bidirectional detection capabilities.

[0078] Furthermore, based on lift-rate filtering of abnormal association patterns, the judgment criterion is that rules with lift-rates significantly greater than 1 or significantly less than 1 are considered abnormal. For example, the lift-rate of the rule {high-value user, Android, new user} → {postback=1, postback interval=within 0.5h} is 1.67, which is much higher than 1. This indicates that advertisers not only postback to high-value users, but also tend to postback within a very short time. This combination of high value and immediate postback is a typical abnormal behavior characteristic of manual screening + automated postback.

[0079] As can be seen, the purpose of calculating confidence level is to filter out statistically significant anomalous association rules from numerous association patterns, that is, to filter out association patterns that are excessively reinforced in advertiser feedback behavior due to artificial selection of user value. In other words, lift is used to eliminate "randomness" and thus identify non-random, genuine, and exploitable association rules. When the lift is significantly greater than 1, it indicates that there is a strong positive dependency between the user behavior feature combination and feedback behavior, and this relationship cannot be naturally explained by the overall sample distribution. This reflects that the advertiser may have a strategic manipulation of feedback behavior to achieve cost optimization through fraudulent intent.

[0080] In this embodiment, the lift is used as the core criterion for detecting abnormalities in advertising conversion event feedback, realizing a technological shift from data relevance to the identification of behavioral manipulation intent.

[0081] If abnormal association patterns are filtered out from candidate association patterns using lift, it means that the current feedback behavior is abnormal. Therefore, the first detection method is to filter out abnormal association patterns as the result of the filtering.

[0082] In this embodiment, using the global callback ratio as a reference, the degree of non-random reinforcement between user characteristics and callback behavior is quantified by calculating the lift of each association rule, and abnormal association rules deviating from the normal distribution are identified based on this. This method breaks through the superficial logic of merely whether or not a callback occurs, delving into the level of "under what conditions a callback occurs," truly realizing the transformation from data anomaly detection to behavioral intent recognition. It is precisely through the lift that selective callback fraud, which appears to be normal behavior, can be accurately identified in massive conversion data, thus building an anti-fraud system with causal reasoning capabilities for advertising platforms.

[0083] As an optional implementation, the above method also includes:

[0084] Obtain a set of confidence levels for the abnormal association pattern;

[0085] Determine a first confidence distribution consisting of the set of confidence levels;

[0086] Determine the degree of dispersion of the first confidence distribution, and if the degree of dispersion is greater than or equal to a preset threshold, determine to perform complex selective backhaul on the advertising conversion event.

[0087] First, the aforementioned set of confidence scores for obtaining the abnormal association patterns can be, but is not limited to, the confidence scores of each abnormal association rule in the set of abnormal rules strongly correlated with user value segments after the Apriori algorithm has mined multiple frequent itemsets and generated association rules. For example, the confidence score of the rule {high-value user}→{return post} is 100%, while the confidence score of the rule {low-value user}→{return post} is 0%.

[0088] The aforementioned determination of the first confidence distribution, composed of the set of confidence levels, involves statistically modeling the aforementioned numerical set, treating it as a discrete confidence distribution. The degree of dispersion of the first confidence distribution is calculated, where dispersion may include, but is not limited to, variance or standard deviation. When the degree of dispersion is greater than or equal to a preset threshold, it is determined that the advertiser has performed a complex type of selective backfeeding behavior, the severity of which is higher than that of selective backfeeding behavior with a concentrated confidence distribution.

[0089] This is because the greater the degree of dispersion, the more differentiated the feedback strategy adopted by advertisers under different combinations of user behavior characteristics. Their cheating behavior is more refined and more covert, and it interferes more severely with the advertising platform's delivery model.

[0090] The so-called complex selective feedback can refer to, but is not limited to, advertisers not randomly deducting users or engaging in single-behavior fraud, but rather systematically and strategically feeding back only high-value users, forming a value screening model, which is a highly concealed form of fraud.

[0091] By constructing the confidence distribution in this embodiment and quantifying its discreteness, automated and data-driven identification of cheating patterns based on non-uniform feedback decisions is achieved. This breaks through the limitations of traditional single-rule threshold judgment, elevating feedback anomaly detection from "whether to feedback" to the cognitive level of "how to selectively feedback," thereby enhancing the intelligence of the anti-cheating system.

[0092] As another optional implementation, the determination of the return anomaly detection result of the advertising conversion event based on the return confidence distribution of different user value levels determined by the aforementioned confidence level further includes:

[0093] Based on preset value indicators, the user value in the advertising conversion event is divided into multiple user value levels;

[0094] Determine the feedback confidence level corresponding to each user value tier;

[0095] Based on the second confidence distribution formed by the backhaul confidence, a first value breakpoint where the backhaul confidence changes abruptly is determined, wherein the first value breakpoint is the lower limit of the first value interval corresponding to the target user value segment where the change occurs.

[0096] Based on the first value breakpoint, a second threshold for selective backhaul is determined, wherein the backhaul anomaly detection result includes the first value breakpoint.

[0097] One user value level corresponds to a value range of a user value segment.

[0098] This embodiment, as a second method for determining the return anomaly detection result, divides user value (such as payment amount, advertising eCPM, and user ARPU) into multiple intervals. Then, it applies the Apriori algorithm to each value interval to obtain the return confidence level corresponding to each interval. Based on the second confidence level distribution formed by these return confidence levels, it analyzes the abrupt changes in return confidence at value breakpoints to obtain the return anomaly detection result.

[0099] Specifically, assuming as Figure 4 As shown, based on preset value indicators, user value is divided into several value ranges: 0-29 yuan, 30-49 yuan, 50-69 yuan, 70-99 yuan, and over 100 yuan. Each user value level represents a value range, thus transforming the originally vague subjective divisions of high and low value into quantifiable objective segments.

[0100] Statistically analyze the percentage of advertiser feedback conversions (i.e., confidence level), support level, and the strength of the correlation (i.e., improvement level) between the value interval and feedback behavior within each interval. For example... Figure 4 As shown, when the paid amount is below 70 yuan, the confidence level of the feedback is only 0.1 to 0.18. However, when it reaches 70 yuan, the confidence level of the feedback suddenly jumps to over 0.85, and the improvement rate also rises from 0.36 to 1.7. This means that the advertiser may have set a hidden rule of "only feedback when the paid amount is greater than or equal to 70 yuan". All conversion data corresponding to conversion events below this amount are withheld, causing a clear breakpoint in the feedback data at 70 yuan (i.e., the first value breakpoint), which is a typical case of artificial manipulation.

[0101] Based on this first value breakpoint, the second threshold for advertisers to selectively send back data can be directly derived, namely, the actual trigger value for sending back data used by advertisers is 70 yuan. This means that there are strategies that only send back data when the amount exceeds a certain threshold, indicating abnormal operation behavior such as deducting traffic during sending back data. This threshold is not a manually set detection standard, but rather an anchor point for the actual behavior of cheaters derived through data mining. Ultimately, this first value breakpoint and the second threshold together constitute the abnormal sending back data detection result, providing the platform with precise basis for penalties (such as traffic limiting and account freezing).

[0102] In this embodiment, the association rule confidence of the Apriori algorithm is upgraded from single-rule analysis to distribution trend analysis oriented towards continuous value distribution. By identifying nonlinear abrupt changes in data distribution, the implicit fraud threshold of advertisers can be accurately identified. Compared with only detecting superficial phenomena such as high return rates of high-value users, the technical solution of this application can directly locate the decision-making critical point of fraudulent behavior, achieving a qualitative leap from behavioral anomaly identification to decoding fraud motives.

[0103] As another optional implementation, the above-mentioned determination of the return postback anomaly detection result of the advertising conversion event based on the return postback confidence distribution of different user value levels determined by the aforementioned confidence level further includes:

[0104] Based on preset value indicators, the user value in the advertising conversion event is ranked to obtain the ranked user value;

[0105] The sorted user values ​​are divided into a preset number of percentile intervals;

[0106] Determine the returned confidence level corresponding to each percentile interval, and construct a third confidence level distribution based on the returned confidence level;

[0107] If the confidence change gradient between adjacent percentile intervals indicated by the third confidence distribution is greater than or equal to the third threshold, it is determined that the advertising conversion event has a feedback anomaly.

[0108] This embodiment, as a third method for determining the return anomaly detection results, divides the data according to the percentile of user value, such as... Figure 5 As shown, the confidence level of feedback was calculated for each quantile: 0-10%, 10%-30%, 30%-60%, and 60%-100%. The results showed that the top 10% of users had a feedback rate as high as 95%, while the bottom 60%-100% of users (i.e., the majority of ordinary users) had a feedback rate as low as 5%. This extreme distribution—almost all high-value users feedback and almost none low-value users feedback—violates the natural distribution pattern of normal advertising conversions, indicating that advertisers are systematically only reporting the high-revenue conversion events, deliberately concealing a large number of other genuine conversion events or conversion data.

[0109] Figure 5The user value quantiles shown (0-10%) represent all users ranked from highest to lowest based on their payment amount when converting to the target ad. The top 10% of users (those who paid 70 yuan or more) have a return confidence level as high as 95%, while low-value users (those who paid 0-29 yuan and are at the bottom of the queue, 60-100%) have a return confidence level of 5%. It is evident that the higher the user value, the higher the return confidence level, indicating that advertisers are selectively returning users, with a significantly higher return rate for high-value users than for low-value users. Therefore, there is a possibility of deductions from return metrics.

[0110] Figure 4 and Figure 5 The data in the two tables shown demonstrates that advertisers' data return behavior is not random or natural, but rather based on a defined value threshold rule. The confidence, support, and lift calculated using the Apriori algorithm accurately identify this hidden fraudulent pattern behind the data. This differs from traditional, superficial methods that only detect fluctuations in the total amount of data returns or device fingerprints.

[0111] Alternatively, the system can calculate the gradient of the return confidence level change between adjacent percentile intervals, i.e., the difference in confidence levels between two adjacent intervals. When a gradient value exceeds a preset third threshold (e.g., greater than or equal to 50%), the system determines that there is a return anomaly. Figure 5 The absolute value of the gradient from 60%-100% (0.05) to 0%-10% (0.95) is as high as 90%, which is far beyond the normal fluctuation range. The system then judges that there is a backhaul anomaly.

[0112] To more clearly understand the handling methods for the above-mentioned advertising conversion events, the following will combine... Figure 3 The overall flowchart shown below provides further explanation.

[0113] S302, construct the transformation and return transaction;

[0114] A conversion feedback transaction contains multiple items with different dimensions, such as user value segmentation, whether feedback is required, and the type of feedback device.

[0115] S304 uses the Apriori algorithm for frequent itemset mining;

[0116] For details, please refer to the description in the above embodiments, which will not be repeated here.

[0117] S306, by calculating the confidence, support, and lift of different association rules, achieves the generation and analysis of association rules;

[0118] For example, the lift of each association rule is calculated based on the confidence level, and abnormal association rules are identified from all candidate association rules based on the lift.

[0119] S308, Determine if there are any abnormal association rules;

[0120] If so, proceed to step S310; otherwise, terminate the process.

[0121] S310, extract abnormal association rules and determine the existence of abnormal operation behaviors such as selective back-up and back-up deduction.

[0122] The above methods enable the detection of anomalies such as deductions in conversion feedback, improving detection accuracy and refining detection dimensions.

[0123] The operation and anti-fraud measures in advertising conversion feedback are essentially a dynamic economic game between advertisers and advertising platforms. The core conflict lies in the fact that advertisers maximize their own interests through rebates and conversion filtering, while platforms seek out real data to optimize the ecosystem through anti-fraud measures and model adjustments. The embodiments in this application enable advertising platforms to limit traffic to fraudulent advertisers by identifying rebates and filtering conversion feedback, thereby incentivizing advertising platforms to provide genuine conversion data and increasing the overall traffic value of the advertising platform, thus possessing significant economic value.

[0124] In a specific embodiment, the first step is to assume that conversion feedback is treated as a transaction. This transaction contains multiple items, as described in the above embodiment: Transaction = {User Value Segment, Whether Feedback is Received, Feedback Time Interval, Feedback Device Type, User Type, Feedback Conversion Event Type, Feedback Channel Quality}. In this embodiment, a conversion event can also be understood as a conversion transaction. In an advertising conversion traffic delivery scenario, a conversion event can be used, but is not limited to, to describe detailed information about a conversion data point, or a conversion event can be used, but is not limited to, to describe detailed information about a conversion data point, whether the conversion data point was fed back, and the feeding back method, etc. A conversion data point corresponds to a user's paid transaction or a single view, etc.

[0125] The above definitions are as follows: User value segmentation: {High value, Medium value, Low value}; Feedback status: {Feedback achieved, No feedback achieved}; Feedback event interval: {Within 0.5 hours, Within 1 hour, Within 6 hours, Within 24 hours}; Device type: {Android, Web}; User type: {New user, Returning user, Churned user}; Feedback conversion event type: {Paid, ECPPM, ARPU}; Feedback channel quality: {High, Medium, Low}. Here, ECPPM is the ad impression rate, and ARPU is the average number of users attracted by the ad, typically expressed as the amount spent.

[0126] The second step involves using the Apriori algorithm for frequent itemset mining, specifically including:

[0127] (1) Discover 1-item frequent itemsets;

[0128] Support scores are calculated as follows: 3 / 5 (0.6) for {high-value users}; 3 / 5 (0.6) for {return uploads}; 3 / 5 (0.6) for {return upload intervals}; 4 / 5 (0.8) for {new users}; and 4 / 5 (0.8) for {paid users}.

[0129] Assuming the minimum support is 0.6, then all the above 1-itemsets satisfy the threshold condition.

[0130] The mining of the above-mentioned 1-item frequent itemsets refers to the ratio between the number of target conversion events that contain only one item (indicating a user behavior feature) among all generated advertising conversion events and the total number of advertising conversion events.

[0131] (2) Discover 2-item frequent itemsets;

[0132] Support scores are calculated as follows: {High-value users, data return} has a support score of 3 / 5, or 0.6; {High-value users, data return interval 0.5h} has a support score of 3 / 5, or 0.6; {High-value users, new users} has a support score of 3 / 5, or 0.6; {High-value users, Android} has a support score of 3 / 5, or 0.6; {Data return, data return interval 0.5h} has a support score of 3 / 5, or 0.6; {Data return, new users} has a support score of 3 / 5, or 0.6; {Low-value users, no data return} has a support score of 1 / 5, or 0.2; {Medium-value users, no data return} has a support score of 1 / 5, or 0.2.

[0133] The mining of the aforementioned 2-item frequent itemsets refers to the ratio between the number of target conversion events containing two items (indicating a combination of two user behavior features) among all generated ad conversion events, and the total number of ad conversion events. The two items are randomly selected from multiple items in a predefined transaction. Assuming a minimum support of 0.6, the last two 2-itemsets do not meet the condition.

[0134] (3) Discover 3-item frequent itemsets.

[0135] Support is calculated as follows: the support for {high-value user, backhaul, backhaul interval 0.5h} is 3 / 5, or 0.6; the support for {high-value user, backhaul, new user} is 3 / 5, or 0.6; the support for {high-value user, backhaul, Android} is 3 / 5, or 0.6; the support for {backhaul, backhaul interval within 0.5h, new user} is 3 / 5, or 0.6; the support for {backhaul, backhaul interval 0.5h, Android} is 3 / 5, or 0.6. These three items are randomly combined from multiple items in a predefined transaction. Assuming the minimum support is 0.6, all 3-itemsets ultimately satisfy the condition.

[0136] The third step is the generation and analysis of association rules, which specifically includes calculating the confidence of each association rule composed of the above-mentioned combinations of various user behavior features and identifying abnormal association rules.

[0137] (1) For the rule {high-value user} → {return message}, the support 1 of the user behavior feature combination {high-value user, return message} including both the antecedent and the consequent is 0.6, and the support 2 of the user behavior feature combination {high-value user} including only the antecedent is 0.6. Therefore, the confidence of this rule is equal to the ratio between support 1 and support 2, which is equal to 1 (meaning 100%). Among them, the antecedent can be, but is not limited to, all user behavior features in the rule, such as including both high-value user and return message. The consequent can be, but is not limited to, the user behavior feature that is the latter among the user behavior features corresponding to the rule, such as return message.

[0138] (2) For the rule {Medium-value user} → {Return}, the support of the user behavior feature combination {Medium-value user, Return} which includes both the antecedent and the consequent is 3, which is equal to 0. The support of the user behavior feature combination {Medium-value user} which only includes the antecedent is 4, which is equal to 0.2. Therefore, the confidence of this rule is equal to the ratio between support 3 and support 4, which is equal to 0.

[0139] (3) For the rule {low-value user} → {return}, the support of the user behavior feature combination {low-value user, return} which includes both the antecedent and the consequent is 5, which is equal to 0. The support of the user behavior feature combination {low-value user} which only includes the antecedent is 6, which is equal to 0.2. Therefore, the confidence of this rule is equal to the ratio between the support of 5 and the support of 6, which is equal to 0.

[0140] From the above process of calculating confidence levels based on different user values, it can be determined that the confidence level decreases as the user value decreases, with the confidence level of feedback from low-value users being 0. Based on this conclusion, the following abnormal association rules can be extracted from candidate association patterns or candidate association rules.

[0141] For example, the antecedent of the abnormal association rule 1 is {value = low value}, and the consequent is {return = no}. The confidence level is close to 0, which means that the conversion events corresponding to low-value users are almost never returned.

[0142] In addition, the lift of each association rule is calculated, as follows:

[0143] (1) For the rule {high-value user} → {postback}, the support of {postback} is 3 / 5 = 0.6, and the lift is equal to the ratio between the confidence (equal to 1) and the support of 0.6, i.e. 1.67;

[0144] (2) For the rule {high-value user, Android, new user} → {backhaul, backhaul time interval 0.5h}, the confidence level is 1. For the consequent {backhaul, backhaul time interval 0.5h}, the support level is 3 / 5 = 0.6. The lift is equal to the ratio between the confidence level and the support level, i.e. 1.67.

[0145] The above calculations show that the improvement of high-value backhaul is greater than 1, indicating a strong positive correlation between high-value users and backhaul.

[0146] So, based on the lift, abnormal association rules can also be identified. For example, the antecedent of an abnormal association rule is {value = high-value user, device = Android, user = new user}, and the consequent is {return = yes, return time interval = immediately}. The calculated confidence is close to 100%, the support is significantly higher than the normal level, and the lift is also greater than 1 (strong positive correlation).

[0147] In addition, confidence, support and improvement can be calculated for different combinations of user value. Under normal circumstances, the confidence of each user value segment should be relatively uniform. Under abnormal circumstances, there is a significant deviation in the confidence of each user value segment, thereby identifying whether the advertiser is selectively transmitting data, such as only transmitting data to high-value users.

[0148] The fourth step is to analyze the breakpoints in the user value feedback process based on the calculation results of the support, confidence, and lift mentioned above.

[0149] (1) Divide user value into, for example Figure 4 The multiple intervals shown;

[0150] User value includes, but is not limited to, the amount paid for advertising conversion, advertising eCPM, or user ARPU.

[0151] (2) Run the Apriori algorithm for each interval;

[0152] The support of user behavior feature combinations, the confidence of each candidate association pattern, and the lift are calculated sequentially in the above manner.

[0153] (3) Analyze the abrupt change in the returned confidence at the value breakpoint based on the calculation results.

[0154] refer to Figure 4 It can be seen that the confidence level changes abruptly at 70 yuan, indicating that there may be a selective back-transmission threshold, that is, a strategy of only transmitting data above a certain threshold. The result is that there is an anomaly in the back-transmission deduction.

[0155] In addition to the above-mentioned determination methods, this application embodiment also provides another method for determining anomaly detection results. Specifically, the user's returned value can be divided into intervals according to the quantiles (which can also be understood as percentiles), and a confidence score can be calculated for each interval. The existence of an anomaly can be determined based on the calculation results. For details, please refer to... Figure 5 This indicates that the higher the user value, the higher the confidence level of the feedback, suggesting that advertisers are selectively sending feedback, with a significantly higher feedback rate from high-value users compared to low-value users. Therefore, there is a possibility of deductions from feedback metrics.

[0156] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0157] According to another aspect of the embodiments of this application, as follows is also provided Figure 6 An ad conversion event processing device is shown, the device comprising:

[0158] The first acquisition unit 602 is used to acquire generated advertising conversion data in response to a received conversion event detection request;

[0159] The first processing unit 604 is configured to determine, based on the advertising conversion data, combinations of user behavior features in advertising conversion events with a support greater than or equal to a first threshold, wherein the support represents the ratio between the number of target conversion events containing the user behavior feature combinations in the advertising conversion events and the total number of advertising conversion events;

[0160] The second processing unit 606 is used to determine the confidence level of the candidate association pattern created based on the combination of user behavior features, wherein the confidence level represents the return integrity ratio of the return of the advertising conversion event under the combination of user behavior features.

[0161] The third processing unit 608 is used to determine the return anomaly detection result of the advertising conversion event based on the return confidence distribution of different user value levels determined by the confidence level.

[0162] Optionally, the first processing unit 604 includes:

[0163] The first acquisition module is used to acquire multiple behavioral attribute fields contained in the advertising conversion feedback record corresponding to the advertising conversion event. The multiple behavioral attribute fields include user value segmentation, feedback time interval and feedback device type. The user value segmentation is used to quantitatively grade the degree of contribution of advertising conversion users to the revenue of the advertising placement entity.

[0164] The first processing module is used to determine a combination of multiple attribute fields that include at least one attribute field based on the multiple behavioral attribute fields;

[0165] The second processing module is used to determine the support degree corresponding to each attribute field combination in the plurality of attribute field combinations, and to select a portion of the attribute field combinations whose support degree is greater than or equal to the first threshold from the plurality of attribute field combinations.

[0166] The third processing module is used to determine the combination of behavioral features corresponding to the combination of the partial attribute fields as the combination of user behavioral features.

[0167] Optionally, the second processing unit 606 includes:

[0168] The second acquisition module is used to acquire the antecedent and consequent of each association pattern in the candidate association patterns, wherein the antecedent includes at least one behavioral attribute field corresponding to the user behavior feature combination, and the consequent includes a return identifier field indicating that the advertising conversion event has been returned.

[0169] The third acquisition module is used to acquire a first number of a first part of advertising conversion events that includes the antecedent and the consequent, and to acquire a second number of a second part of advertising conversion events that includes only the antecedent;

[0170] The fourth processing module is used to determine the ratio between the first quantity and the second quantity as the confidence level of each association pattern.

[0171] Optionally, the third processing unit 608 includes:

[0172] The fifth processing module is used to determine the lift of each association pattern in the candidate association patterns based on the confidence level, wherein the lift represents the degree of deviation between the conditional probability of the ad conversion event being transmitted back and the proportion of the transmitted ad conversion events.

[0173] The filtering module is used to filter out abnormal association patterns from the candidate association patterns based on the lift degree, wherein the abnormal association patterns are used to indicate that the feedback of the advertising conversion event is the result after selective feedback filtering.

[0174] Optionally, the above-mentioned device further includes:

[0175] The second acquisition unit is used to acquire a set of confidence levels for the abnormal association pattern;

[0176] The fourth processing unit is used to determine the first confidence distribution composed of the set of confidence levels;

[0177] The fifth processing unit is used to determine the degree of dispersion of the first confidence distribution, and if the degree of dispersion is greater than or equal to a preset threshold, to determine to perform complex selective backhaul of the advertising conversion event.

[0178] Optionally, the third processing unit 608 further includes:

[0179] The first segmentation module is used to divide the user value in the advertising conversion event into multiple user value levels based on preset value indicators;

[0180] The sixth processing module is used to determine the feedback confidence level corresponding to each user value level;

[0181] The seventh processing module is used to determine the first value breakpoint where the returned confidence level changes abruptly, based on the second confidence level distribution formed by the returned confidence level, wherein the first value breakpoint is the lower limit of the first value interval corresponding to the target user value segment where the change occurs.

[0182] The eighth processing module is used to determine a second threshold for selective backhaul based on the first value breakpoint, wherein the backhaul anomaly detection result includes the first value breakpoint.

[0183] Optionally, the third processing unit 608 further includes:

[0184] The sorting module is used to sort the user value in the advertising conversion event based on a preset value index, so as to obtain the sorted user value.

[0185] The second division module is used to divide the sorted user value into a preset number of percentile intervals;

[0186] The ninth processing module is used to determine the feedback confidence level corresponding to each percentile interval, and to construct a third confidence level distribution based on the feedback confidence level;

[0187] The tenth processing module is used to determine that there is a feedback anomaly in the advertising conversion event when the confidence change gradient of the adjacent percentile interval indicated by the third confidence distribution is greater than or equal to the third threshold.

[0188] It should be noted that the embodiments of the advertising conversion event processing device here can refer to the embodiments of the advertising conversion event processing method described above, and will not be repeated here.

[0189] According to another aspect of the embodiments of this application, an electronic device for implementing the above-described method for processing advertising conversion events is also provided. This electronic device may be... Figure 1 The target terminal or server is shown. This embodiment uses the electronic device as an example to illustrate the concept. Figure 7 As shown, the electronic device includes a memory 702 and a processor 704. The memory 702 stores a computer program, and the processor 704 is configured to execute the steps in any of the above method embodiments via the computer program.

[0190] Optionally, the aforementioned electronic device may be located in at least one of a plurality of network devices of the computer.

[0191] Optionally, the processor described above can be configured to perform the following steps via a computer program:

[0192] S1, in response to the received conversion event detection request, obtains the generated ad conversion data;

[0193] S2, based on the advertising conversion data, determine the user behavior feature combinations in the advertising conversion events with a support greater than or equal to a first threshold, wherein the support represents the ratio between the number of target conversion events containing the user behavior feature combinations in the advertising conversion events and the total number of advertising conversion events;

[0194] S3, determine the confidence level of the candidate association pattern created based on the user behavior feature combination, wherein the confidence level represents the return integrity ratio of the ad conversion event under the user behavior feature combination;

[0195] S4. Based on the return confidence distribution of different user value levels determined by the confidence level, determine the return anomaly detection result of the advertising conversion event.

[0196] Alternatively, as those skilled in the art will understand, Figure 7 The structure shown is for illustrative purposes only. Figure 7 This does not limit the structure of the aforementioned electronic devices or electronic equipment. For example, electronic devices or electronic equipment may also include components that are more... Figure 7 The more or fewer components shown (such as network interfaces, etc.), or having the same Figure 7 The different configurations shown.

[0197] The memory 702 can be used to store software programs and modules, such as the program instructions / modules corresponding to the advertising conversion event processing method and apparatus in this embodiment. The processor 704 executes various functional applications and data processing by running the software programs and modules stored in the memory 702, thereby realizing the aforementioned advertising conversion event processing method. The memory 702 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 702 may further include memory remotely located relative to the processor 704, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof. Specifically, the memory 702 may be used, but is not limited to, to store advertising conversion data, confidence levels, and feedback anomaly detection results. As an example, such as... Figure 7 As shown, the memory 702 may include, but is not limited to, the first acquisition unit 602, the first processing unit 604, the second processing unit 606, and the third processing unit 608 in the aforementioned advertising conversion event processing device. Furthermore, it may include, but is not limited to, other module units in the aforementioned advertising conversion event processing device, which will not be elaborated upon in this example.

[0198] Optionally, the transmission device 706 described above is used to receive or send data via a network. Specific examples of the network described above may include wired networks and wireless networks. In one example, the transmission device 706 includes a Network Interface Controller (NIC), which can be connected to other network devices and a router via a network cable to communicate with the Internet or a local area network. In another example, the transmission device 706 is a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0199] In addition, the above-mentioned electronic device also includes: a display 708 for displaying the feedback anomaly detection results; and a connection bus 710 for connecting the various module components in the above-mentioned electronic device.

[0200] In other embodiments, the target terminal or server described above can be a node in a distributed system. This distributed system can be a blockchain system, formed by connecting multiple nodes through network communication. The nodes can form a point-to-point network, and any type of computing device, such as a server or target terminal, can become a node in the blockchain system by joining this point-to-point network.

[0201] According to another aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the advertising conversion event processing method provided in various optional implementations of the above-described server verification processing, wherein the computer program is configured to execute the steps in any of the above-described method embodiments at runtime.

[0202] Optionally, in this embodiment, the computer-readable storage medium described above may be configured to store a computer program for performing the following steps:

[0203] S1, in response to the received conversion event detection request, obtains the generated ad conversion data;

[0204] S2, based on the advertising conversion data, determine the user behavior feature combinations in the advertising conversion events with a support greater than or equal to a first threshold, wherein the support represents the ratio between the number of target conversion events containing the user behavior feature combinations in the advertising conversion events and the total number of advertising conversion events;

[0205] S3, determine the confidence level of the candidate association pattern created based on the user behavior feature combination, wherein the confidence level represents the return integrity ratio of the ad conversion event under the user behavior feature combination;

[0206] S4. Based on the return confidence distribution of different user value levels determined by the confidence level, determine the return anomaly detection result of the advertising conversion event.

[0207] Optionally, in embodiments of this application, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0208] Optionally, in this embodiment, those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the target terminal. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0209] The sequence numbers of the embodiments in this application are merely for description and do not represent the superiority or inferiority of the embodiments. If the integrated units in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in the aforementioned computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause one or more computer devices (which may be personal computers, servers, or network devices, etc.) to execute all or part of the steps of the methods in the various embodiments of this application.

[0210] In the above embodiments of this application, the descriptions of each embodiment have their own emphasis. Parts not described in detail in a certain embodiment can be referred to in the relevant descriptions of other embodiments. It should be understood that the disclosed client can be implemented in other ways in the several embodiments provided in this application. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division. In actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling or direct coupling or communication connection may be through some interfaces; the indirect coupling or communication connection of units or modules may be electrical or other forms.

[0211] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units.

[0212] The above are merely preferred embodiments of this application. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for handling advertising conversion events, characterized in that, include: In response to a received conversion event detection request, retrieve the generated ad conversion data; Based on the advertising conversion data, user behavior feature combinations with a support greater than or equal to a first threshold are determined in advertising conversion events, wherein the support represents the ratio between the number of target conversion events containing the user behavior feature combinations and the total number of advertising conversion events; Determine the confidence level of the candidate association pattern created based on the user behavior feature combination, wherein the confidence level represents the return integrity ratio of the ad conversion event under the user behavior feature combination; Based on the return confidence distribution of different user value levels determined by the aforementioned confidence level, the return anomaly detection result of the advertising conversion event is determined.

2. The method according to claim 1, characterized in that, The step of determining the combination of user behavior features with a support level greater than or equal to a first threshold in an ad conversion event based on the ad conversion data includes: Obtain multiple behavioral attribute fields contained in the advertising conversion feedback record corresponding to the advertising conversion event, wherein the multiple behavioral attribute fields include user value segmentation, feedback time interval and feedback device type, and the user value segmentation is used to quantitatively grade the degree of contribution of advertising conversion users to the revenue of the advertising entity. Based on the multiple behavioral attribute fields, determine a combination of multiple attribute fields that include at least one attribute field; Determine the support degree corresponding to each attribute field combination in the plurality of attribute field combinations, and select a portion of the attribute field combinations whose support degree is greater than or equal to the first threshold from the plurality of attribute field combinations; The combination of behavioral features corresponding to the combination of the aforementioned attribute fields is determined as the user behavior feature combination.

3. The method according to claim 1, characterized in that, Determining the confidence level of the candidate association patterns created based on the user behavior feature combination includes: Obtain the antecedent and consequent of each association pattern in the candidate association patterns, wherein the antecedent includes at least one behavioral attribute field corresponding to the user behavior feature combination, and the consequent includes a return identifier field indicating that the advertising conversion event has been returned. Obtain a first number of ad conversion events that include both the antecedent and the consequent, and obtain a second number of ad conversion events that include only the antecedent. The ratio between the first quantity and the second quantity is determined as the confidence level of each association pattern.

4. The method according to claim 1, characterized in that, The distribution of feedback confidence based on the confidence level of different user value levels determined by the confidence level, used to determine the feedback anomaly detection result of the advertising conversion event, includes: Based on the confidence level, the lift of each association pattern in the candidate association patterns is determined, wherein the lift represents the degree of deviation between the conditional probability of the ad conversion event being transmitted back and the proportion of the transmitted ad conversion events. Based on the lift, abnormal association patterns are selected from the candidate association patterns, wherein the abnormal association patterns are used to indicate that the return of the advertising conversion event is the result of selective return filtering.

5. The method according to claim 4, characterized in that, The method further includes: Obtain a set of confidence levels for the abnormal association pattern; Determine a first confidence distribution consisting of the set of confidence levels; Determine the degree of dispersion of the first confidence distribution, and if the degree of dispersion is greater than or equal to a preset threshold, determine to perform complex selective backhaul on the advertising conversion event.

6. The method according to claim 1, characterized in that, The method of determining the return confidence distribution based on the confidence level of different user value levels, and determining the return anomaly detection result of the advertising conversion event, further includes: Based on preset value indicators, the user value in the advertising conversion event is divided into multiple user value levels; Determine the feedback confidence level corresponding to each user value tier; Based on the second confidence distribution formed by the backhaul confidence, a first value breakpoint where the backhaul confidence changes abruptly is determined, wherein the first value breakpoint is the lower limit of the first value interval corresponding to the target user value segment where the change occurs. Based on the first value breakpoint, a second threshold for selective backhaul is determined, wherein the backhaul anomaly detection result includes the first value breakpoint.

7. The method according to claim 1, characterized in that, The method of determining the return confidence distribution based on the confidence level of different user value levels, and determining the return anomaly detection result of the advertising conversion event, further includes: Based on preset value indicators, the user value in the advertising conversion event is ranked to obtain the ranked user value; The sorted user values ​​are divided into a preset number of percentile intervals; Determine the returned confidence level corresponding to each percentile interval, and construct a third confidence level distribution based on the returned confidence level; If the confidence change gradient between adjacent percentile intervals indicated by the third confidence distribution is greater than or equal to the third threshold, it is determined that the advertising conversion event has a feedback anomaly.

8. A device for processing advertising conversion events, characterized in that, include: The first acquisition unit is used to acquire the generated advertising conversion data in response to the received conversion event detection request; The first processing unit is configured to determine, based on the advertising conversion data, combinations of user behavior features in advertising conversion events with a support greater than or equal to a first threshold, wherein the support represents the ratio between the number of target conversion events containing the user behavior feature combinations in the advertising conversion events and the total number of advertising conversion events; The second processing unit is used to determine the confidence level of the candidate association pattern created based on the user behavior feature combination, wherein the confidence level represents the return integrity ratio of the ad conversion event under the user behavior feature combination. The third processing unit is used to determine the return anomaly detection result of the advertising conversion event based on the return confidence distribution of different user value levels determined by the confidence level.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein the program can be executed by a terminal device or computer at runtime as described in any one of claims 1 to 7.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to perform the method as described in any one of claims 1 to 7 via the computer program.

Images