Click behavior filtering method and device, equipment and storage medium
By dividing the exposure area into sub-regions, calculating the click percentage and Gini coefficient, filtering target sub-regions, and using behavioral feature information to filter induced clicks, the problem of high identification cost and low accuracy in existing technologies is solved, and efficient click behavior filtering is achieved.
Patent Information
- Application Number
- CN202510989441.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-17
- Publication Date
- 2025-11-14
AI Technical Summary
Existing technologies struggle to accurately identify and filter click fraud, and the cost of identification is high, impacting algorithm prediction model training and advertiser ROI.
The exposure area is divided into sub-regions, the click percentage of each sub-region is calculated, and the click concentration is assessed using the Gini coefficient. Target sub-regions are then selected, and the click probability is determined using behavioral feature information to automatically filter out induced clicks.
It achieves accurate identification and filtering of induced clicks, reduces identification costs, improves the accuracy of identification and filtering, and ensures the quality of training data for the algorithm model and the ROI of advertisers.
Smart Images
Figure CN120952874A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of click behavior recognition technology, and in particular to a method, apparatus, device and storage medium for filtering click behavior. Background Technology
[0002] Induced clicks refer to the act of enticing users to actively click on specific links, buttons, or areas by designing deceptive or tempting content. Examples include shrinking the size of an ad skip button, placing a fake skip button, or displaying enticing text like "Click here," all designed to generate more clicks. These clicks tend to cluster in certain locations. However, these clicks generated through abnormal methods do not reflect the user's true intentions and can negatively impact the training of algorithmic prediction models. Therefore, it is necessary to filter induced clicks to provide clean training data for the algorithm. One approach to identifying induced clicks is to use a media app on a real phone or emulator and collect screenshots during ad exposure. Manual or machine learning methods can then be used to identify the presence of induced content. However, this method is too costly. Another approach is to collect user click coordinates to construct a click heatmap and then check for coordinate clustering. If multiple induced clicks exist in the exposure area, or different materials are used to display different close buttons, this approach suffers from high identification costs and low accuracy.
[0003] Therefore, how to accurately identify and filter click fraudulent behavior and reduce identification costs is a technical problem that needs to be solved by those skilled in the art. Summary of the Invention
[0004] This application provides a method, apparatus, device, and storage medium for filtering click behavior, so as to accurately identify and filter induced click behavior and reduce identification costs.
[0005] Firstly, this application provides a method for filtering click behavior, including:
[0006] The exposure area is divided into sub-regions;
[0007] The click percentage of each sub-region is calculated based on the number of clicks in each sub-region; where the number of clicks in each sub-region is: the total number of click behaviors triggered by users in each sub-region;
[0008] The Gini coefficient of the exposure area is calculated based on the click percentage of each sub-region.
[0009] Determine whether the Gini coefficient is greater than a preset threshold;
[0010] If so, then the target sub-regions to be filtered are selected based on the click percentage of each sub-region, and the number of clicks to be filtered for the target sub-regions is determined based on the number of clicks in each sub-region.
[0011] Using the behavioral feature information of each click behavior within the target sub-region, the filtering probability of each click behavior within the target sub-region is determined;
[0012] Based on the filtering probability of each click behavior, filter the number of click behaviors from the target sub-region, and re-execute the step of calculating the click ratio of each sub-region based on the number of clicks in each sub-region until the Gini coefficient is not greater than the preset threshold.
[0013] Optionally, the target sub-regions to be filtered are selected based on the click percentage of each sub-region, and the number of clicks to be filtered for the target sub-regions is determined based on the number of clicks in each sub-region, including:
[0014] Select the sub-region with the highest click rate as the target sub-region; wherein, the sub-regions other than the target sub-region in the exposure area are other sub-regions;
[0015] Calculate the filtered click count based on the smoothing coefficient, the number of clicks in the target sub-region, the number of conversions in the target sub-region, the number of clicks in other sub-regions, and the number of conversions in other sub-regions.
[0016] Optionally, the step of calculating the filtered click count based on the smoothing coefficient, the number of clicks in the target sub-region, the number of conversions in the target sub-region, the number of clicks in other sub-regions, and the number of conversions in other sub-regions includes:
[0017] Identify at least one conversion target corresponding to the exposure area;
[0018] For each conversion goal, determine the corresponding conversion rate.
[0019] Based on the conversion rate and the number of clicks in the target sub-region, calculate the conversion number for each conversion target corresponding to the target sub-region;
[0020] Based on the conversion rate and the number of clicks in other sub-regions, calculate the conversion count for each conversion target in the other sub-regions;
[0021] The number of filtered clicks corresponding to each conversion target is calculated using the click count calculation rules, which are as follows:
[0022]
[0023] Among them, filter top To filter click counts, smooth is the smoothing factor, and clk is the clk value. top For the number of clicks in the target sub-region, conv top clk is the number of transformations in the target sub-region. other For the click count of other sub-regions, conv otherFor the transformation number of other sub-regions.
[0024] Optionally, after calculating the filtered click count corresponding to each conversion target using the click count calculation rules, the method further includes:
[0025] Determine if the number of filtered clicks for each conversion target is zero;
[0026] If not, filter out conversion targets with a click count greater than zero and output the filter target information; wherein, the filter target information includes: target sub-region, conversion target to be filtered, and the number of clicks corresponding to the conversion target to be filtered.
[0027] Optionally, determining the filtering probability of each click behavior within the target sub-region using the behavioral feature information of each click behavior within the target sub-region includes:
[0028] Based on the filtering target information, determine the target click behavior to be filtered within the target sub-region that corresponds to the target to be filtered for conversion;
[0029] Determine the behavioral characteristic information corresponding to each target click behavior;
[0030] The filtering probability of each target click behavior is determined by utilizing the behavioral feature information of each target click behavior; wherein the filtering probability is determined by a machine learning model / filtering rule.
[0031] Optionally, dividing the exposure area into sub-regions includes:
[0032] Obtain the coordinate information of the exposure area and the click coordinates; wherein, the click coordinates are the click coordinates of each click action within the exposure area;
[0033] The click coordinates are normalized based on the coordinate information to generate normalized click coordinates, so as to count the number of clicks in each sub-region.
[0034] Optionally, calculating the Gini coefficient of the exposure area based on the click percentage of each sub-region includes:
[0035] Sort the click percentage of each sub-region according to a predetermined order and generate the sorting results;
[0036] Calculate the cumulative value for each sub-region based on the sorting results and the click percentage of each sub-region;
[0037] The Gini coefficient is calculated using the cumulative values of each sub-region and the division parameters; wherein, the division parameters are the parameters used to divide the exposure area into sub-regions.
[0038] Secondly, this application provides a click behavior filtering device, comprising:
[0039] The partitioning module is used to divide the exposure area into sub-regions;
[0040] The first calculation module is used to calculate the click percentage of each sub-region based on the number of clicks in each sub-region; wherein, the number of clicks in each sub-region is: the total number of click behaviors triggered by the user in each sub-region;
[0041] The second calculation module is used to calculate the Gini coefficient of the exposure area based on the click percentage of each sub-region;
[0042] The judgment module is used to determine whether the Gini coefficient is greater than a preset threshold; if so, the filtering module is triggered.
[0043] The filtering module is used to filter out the target sub-regions to be filtered based on the click ratio of each sub-region, and to determine the number of clicks to be filtered for the target sub-regions based on the number of clicks in each sub-region.
[0044] The probability determination module is used to determine the filtering probability of each click behavior in the target sub-region by utilizing the behavioral feature information of each click behavior in the target sub-region;
[0045] The filtering module is used to filter the filtered click behaviors from the target sub-region according to the filtering probability of each click behavior, and re-trigger the first calculation module until the Gini coefficient is not greater than a preset threshold.
[0046] Thirdly, this application provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;
[0047] Memory, used to store computer programs;
[0048] When the processor executes a program stored in memory, it implements the above-mentioned steps for filtering click behavior.
[0049] Fourthly, this application also provides a computer storage medium storing computer-executable instructions for performing the steps of the above-described click behavior filtering method.
[0050] Compared with the prior art, the technical solutions provided in this application have the following advantages: This application discloses a method, apparatus, device, and storage medium for filtering click behavior; This application divides the exposure area into sub-regions, calculates the Gini coefficient of the exposure area based on the number of clicks triggered by the user in each sub-region, and if the Gini coefficient is greater than a preset threshold, then the target sub-region to be filtered and the number of clicks to be filtered are determined. Based on the behavioral feature information of each click behavior in the target sub-region, the filtering probability of each click behavior in the target sub-region is determined, and the click behavior in the target sub-region is filtered in combination with the number of clicks to be filtered; It can be seen that this application evaluates whether there is induced click through the Gini coefficient. If there is induced click, the target sub-region where the induced click is located is automatically identified, and the induced click is accurately and effectively filtered based on the filtering probability of each click behavior in the target sub-region. The whole process does not require manual intervention, and can realize the automatic identification and filtering of induced click behavior, reduce identification costs, and improve the accuracy of identification and filtering. Attached Figure Description
[0051] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.
[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0053] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.
[0054] Figure 1 A schematic flowchart of a click behavior filtering method provided in an embodiment of this application;
[0055] Figure 2 A schematic diagram of a specific click behavior filtering method provided in this application embodiment;
[0056] Figure 3 A schematic diagram of a click behavior filtering device provided in an embodiment of this application;
[0057] Figure 4 This is a schematic diagram of an electronic device structure provided in an embodiment of this application. Detailed Implementation
[0058] Induced clicks refer to the act of designing deceptive or enticing content to induce users to actively click on specific links, buttons, or areas. The core purpose of this behavior is to manipulate user actions to achieve results contrary to the user's wishes, and it is often associated with cybersecurity threats or commercial fraud. For example, after an advertising network accesses traffic from a media outlet and runs ads, it shares revenue with the media outlet. Currently, revenue sharing methods include exposure bidding, revenue sharing, and fixed ECPM (Effective Cost Per Mille, Mille). In order to obtain more revenue sharing, the media outlet has an incentive to use abnormal methods (such as: reducing the size of the ad skip button, placing a fake skip button, or placing enticing text such as "Click here") to generate more clicks from users. These clicks will be concentrated in certain locations.
[0059] If the revenue sharing model is based on bid-per-impression, more clicks will lead the algorithm to predict a higher click-through rate, thus increasing the bid for exposure and allowing the media to earn more cash. If the model is based on revenue sharing, the network typically charges advertisers per click; more clicks mean more money for advertisers, and the media also earns more revenue from new user acquisition. If the model is based on a fixed ECPM (expected cost per mille), a higher predicted bid will result in more exposure opportunities, leading to more cash for the media. However, since clicks generated through these abnormal methods do not reflect the user's true intentions, the subsequent conversion rate is very low, affecting the advertiser's ROI (Return on Investment), which in turn affects the advertiser's willingness to advertise on the network and also impacts the training of the algorithm's prediction model. Therefore, it is necessary to filter out these induced clicks to provide clean training data for the algorithm model and to protect the interests of advertisers and the network.
[0060] Therefore, this application provides a method, apparatus, device, and storage medium for filtering click behavior, so as to accurately identify and filter induced click behavior and reduce identification costs.
[0061] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0062] The following disclosure provides numerous different embodiments or examples for implementing various structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. These are merely examples and are not intended to limit the scope of the invention. Furthermore, reference numerals and / or letters may be repeated in different examples. Such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.
[0063] See Figure 1 This is a schematic flowchart of a click behavior filtering method provided in an embodiment of this application. The filtering method includes the following steps:
[0064] S101. Divide the exposure area into sub-regions.
[0065] It should be noted that when implementing the click behavior filtering method in this application, data collection is first required to obtain the relevant data used in filtering click behaviors. The data collected in this application mainly includes the following three parts:
[0066] 1. Exposure data: includes exposure timestamp, user information, ad conversion goal, and ad behavior identifier impression_id;
[0067] Among them, exposure data is used to record basic information about when an ad is displayed. Exposure timestamp refers to the specific time when the ad is actually viewed by a user. User information consists of user identifiers and attributes. User identifiers can be device ID (IdentityDocument, a unique code), user ID, etc., and attributes can be region, interest tags, etc., used for audience targeting analysis. The conversion goal of the ad refers to the expected user behavior endpoint set by the advertiser, such as: activating the app (launching the application for the first time), waking up the app (reopening a dormant application), or making a payment (completing a transaction), etc., used to define the business logic of the conversion event. The ad behavior identifier impression_id is a string used to uniquely identify a single ad exposure, used to connect all behavioral information triggered by each user in this ad campaign, to achieve full-link tracking.
[0068] 2. Click data: includes click timestamp, the exposure area coordinates of the ad at the time of click, click coordinates, the estimated conversion rate (Post-Click Conversion Rate, PCVR) provided by the ad delivery service, and the ad behavior identifier impression_id;
[0069] The click data records details of user interaction with the ads. The click timestamp refers to the specific time a user clicks the ad. The exposure area coordinates indicate the ad's display position on the device screen, typically described as the bottom left corner (x1, y1) and top right corner (x2, y2). Click coordinates indicate the specific location (x, y) of the user's click action on the screen. Combining these with the exposure area coordinates allows for the analysis of a click heatmap. The estimated conversion rate is the probability that a user will complete a target action after clicking the ad. The ad action identifier, `impression_id`, is used to associate the corresponding exposure event, ensuring that the click action is bound to a specific ad display.
[0070] 3. Conversion data: Impression ID (identifier of advertising behavior) and Conversion Rate (CVR).
[0071] Among them, conversion data is used for performance evaluation, and the advertising behavior identifier impression_id is used to bind to the initial exposure event, so that the conversion behavior (such as payment) can be traced back to the original ad display; conversion rate is used to evaluate the effectiveness of advertising.
[0072] In this application, after collecting the aforementioned parameters, the specific click behavior filtering process can be executed. The size of the exposure area can be determined based on its coordinate information. After determining the exposure area, this application needs to divide it into sub-regions according to the division parameters. These division parameters are the parameters used to divide the exposure area into sub-regions. For example, if the exposure area is rectangular, the division parameters are set to a width division parameter *m* and a height division parameter *n*. That is, the width of the rectangular exposure area is divided into *m* equal parts, and the height is divided into *n* equal parts, resulting in *m**n small rectangles. These small rectangles are the sub-regions in this application. This application uses this method to divide the exposure area into multiple sub-regions, allowing for the location of target sub-regions in areas prone to inducing clicks through subsequent steps.
[0073] S102. Calculate the click percentage of each sub-region based on the number of clicks in each sub-region; where the number of clicks in each sub-region is: the total number of click behaviors triggered by the user in each sub-region.
[0074] In this application, based on the collected coordinate information of the exposure area and the click coordinates, the number of clicks for each sub-area can be calculated. This number of clicks represents the total number of clicks triggered by the user within each sub-area. After determining the number of clicks for each sub-area, the click percentage for each sub-area can be calculated.
[0075] The formula for calculating the click percentage is as follows:
[0076] In this formula, r iclk represents the click percentage of the i-th sub-region. i clk is the number of clicks in the i-th sub-region. total This represents the total number of clicks across all sub-regions.
[0077] S103. Calculate the Gini coefficient of the exposure area based on the click percentage of each sub-region.
[0078] In this application, the Gini coefficient is an indicator used to measure the inequality of click distribution. If the calculated Gini coefficient is higher than the preset threshold, it indicates that there is a concentration of clicks in the exposure area, which may be due to induced clicks. In this case, the subsequent steps are continued. If the calculated Gini coefficient is not higher than the preset threshold, the process is stopped and exited.
[0079] S104. Determine whether the Gini coefficient is greater than a preset threshold.
[0080] If yes, then execute S105; otherwise, end the process.
[0081] S105. Filter out the target sub-regions to be filtered based on the click percentage of each sub-region, and determine the number of clicks to be filtered for the target sub-regions based on the number of clicks in each sub-region.
[0082] In this application, the click percentage of each sub-region, calculated from the click count of each sub-region and the total click count, serves as a normalized indicator of the click popularity distribution of each sub-region. If induced clicks exist, they will cause the click percentage of the target sub-region to be abnormally high; if induced clicks do not exist, the click percentage of each sub-region will be relatively uniform. Therefore, when the Gini coefficient is detected to be greater than a preset threshold, this application can filter out target sub-regions based on the click percentage of each sub-region. In this application, the number of target sub-regions can be customized, such as selecting one or more sub-regions with the highest click percentage as target sub-regions.
[0083] This application also needs to determine the number of clicks to be filtered in the target sub-region. The number of clicks to be filtered can be determined based on the number of clicks in all sub-regions. By the difference between the different number of clicks in different sub-regions, it is determined how many click behaviors need to be filtered from the target sub-region. This application refers to the number of click behaviors to be filtered as the number of clicks to be filtered.
[0084] S106. Utilize the behavioral feature information of each click behavior within the target sub-region to determine the filtering probability of each click behavior within the target sub-region.
[0085] After determining the target sub-region to be filtered and the number of clicks to be filtered within the target sub-region through the above process, this application needs to determine which specific click behaviors within the target sub-region need to be filtered. To improve the accuracy of filtering, this application can determine the filtering probability of each click behavior based on its behavioral characteristic information. This filtering probability refers to the probability that each click behavior, estimated based on the user's behavioral characteristic information, is not a genuine click. The higher the filtering probability, the higher the likelihood that the click behavior needs to be filtered.
[0086] The behavioral characteristic information includes: the time difference between the click timestamp and the exposure timestamp, the PCVR of the service, and whether a conversion occurred. The filtering probability of each click behavior can be determined by the above information. For example, the smaller the time difference, the higher the filtering probability; the smaller the PCVR, the higher the filtering probability; and the higher the filtering probability of no conversion.
[0087] S107. Based on the filtering probability of each click behavior, filter the number of click behaviors to be filtered from the target sub-region, and execute S102 until the Gini coefficient is not greater than the preset threshold.
[0088] In this application, after determining the filtering probability of each click behavior within the target sub-region, the click behaviors can be filtered based on the filtering probability. Specifically, this application sorts the click behaviors within the target sub-region in descending order of filtering probability, and filters out click behaviors that do not generate conversions, etc., from high to low filtering probability. The above process is repeated until the process ends if the Gini coefficient is detected to be less than a preset threshold, or if no clicks need to be filtered.
[0089] In summary, this application assesses the existence of induced clicks using the Gini coefficient. If induced clicks are found, the target sub-region where the induced clicks are located is automatically identified. Based on the filtering probability of each click behavior within the target sub-region, the induced clicks are accurately and effectively filtered, retaining only the click behaviors of users with genuine intent. The entire process requires no manual intervention, thus achieving automatic identification and filtering of induced click behaviors, reducing identification costs, and improving the accuracy of identification and filtering.
[0090] In another embodiment of this application, dividing the exposure area into sub-regions includes:
[0091] Obtain the coordinate information of the exposure area and the click coordinates; where the click coordinates are the click coordinates of each click behavior within the exposure area; normalize each click coordinate according to the coordinate information to generate normalized click coordinates, so as to count the number of clicks in each sub-area through the normalized click coordinates.
[0092] In this application, due to differences in the size of different advertising materials and the screen size of different mobile phones, it is impossible to directly compare the original coordinates of the click coordinates. Therefore, this application requires normalization processing of each click coordinate, normalizing each click coordinate to...
[0093] Between [0,1]. For example: if the coordinate information of the exposure area includes the coordinates of the lower left corner (x1,y1) and the upper right corner (x2,y2), and the click coordinates are (x,y), then the normalized click coordinates are: After the above normalization process is completed, the width and height of the entire exposure area are both 1.
[0094] In another embodiment of this application, the process of calculating the Gini coefficient of the exposure area based on the click percentage of each sub-region includes:
[0095] The click percentages of each sub-region are sorted according to a predetermined order to generate a sorting result; based on the sorting result and the click percentage of each sub-region, the cumulative value of each sub-region is calculated; the Gini coefficient is calculated using the cumulative value of each sub-region and the division parameter; wherein, the division parameter is the parameter used to divide the exposure area into sub-regions.
[0096] This application divides the exposure area into multiple sub-regions. Based on the click coordinates of each click action, the click count of each sub-region and the total click count of the exposure area can be determined. Then, the click percentage of each sub-region can be calculated based on the click count of each sub-region and the total click count. After calculating the click percentage of each sub-region, the click percentages of each sub-region can be sorted according to a predetermined order to generate a sorting result. This predetermined order can be ascending order, that is, sorting the click percentages of each sub-region in ascending order to generate a sorting result. Based on this sorting result, the cumulative value of each sub-region can be calculated. The formula for calculating the cumulative value is:
[0097]
[0098] Among them, R j For the j-th cumulative value, r i This represents the click percentage of the i-th sub-region in the sorting results. For example, in the sorting results, the cumulative value of the first sub-region is the click percentage of the first sub-region, the cumulative value of the second sub-region is the total click percentage of the first and second sub-regions, the cumulative value of the third sub-region is the total click percentage of the first, second, and third sub-regions, and so on.
[0099] After calculating the cumulative value of each sub-region, the Gini coefficient can be calculated using the cumulative value of each sub-region and the division parameters. The formula for calculating the Gini coefficient is as follows:
[0100]
[0101] Where gini is the Gini coefficient, m is the width division parameter, n is the height division parameter, i is the index number, and R i R is the cumulative value of the i-th sub-region. i+1 This is the cumulative value of the (i+1)th sub-region.
[0102] In summary, this application collects the coordinate information of the ad's exposure area and click coordinates, then normalizes the click coordinates to a rectangle with a width and height ranging from 0 to 1. The rectangle of the exposure area is divided into multiple smaller rectangles, and the click percentage of each smaller rectangle is calculated. Based on the click percentage distribution, the Gini coefficient can be calculated. This method maps each click coordinate to the relative space of the exposure area, unifying the scale and eliminating hardware differences, thus making the click positions of different devices and ad placements comparable. Furthermore, this application uses the Gini coefficient to assess whether click behavior in the exposure area is concentrated; if click concentration exists, subsequent click behavior filtering is performed.
[0103] In another embodiment of this application, the target sub-regions to be filtered are selected based on the click percentage of each sub-region, and the number of clicks to be filtered for the target sub-regions is determined based on the number of clicks in each sub-region, including:
[0104] Select the sub-region with the highest click percentage as the target sub-region; wherein, the sub-regions other than the target sub-region in the exposure area are other sub-regions; calculate the filtered click count based on the smoothing coefficient, the number of clicks in the target sub-region, the number of conversions in the target sub-region, the number of clicks in other sub-regions, and the number of conversions in other sub-regions.
[0105] In this application, at least one target sub-region can be selected for filtering based on the click percentage of each sub-region. In this embodiment, the number of target sub-regions is set to one, that is, the sub-region with the largest click percentage is selected as the target sub-region. Therefore, this application can divide the sub-region into two parts: one part is the target sub-region with the largest click percentage value, which can be referred to as the header rectangle S. top The other part is the sub-region S that is excluded from all sub-regions except the target sub-region. other .
[0106] After dividing the sub-region into the target sub-region and other sub-regions, the click count of the target sub-region and the click count of other sub-regions can be determined. Then, by comparing the ad conversion effects of the two parts, the number of clicks that need to be filtered is determined. This ad conversion effect is the conversion rate. Using the click count of the target sub-region, the click count of other sub-regions, and the conversion rate, the conversion count of the target sub-region and the conversion count of other sub-regions can be calculated. Then, based on the smoothing coefficient, the click count of the target sub-region, the conversion count of the target sub-region, the click count of other sub-regions, and the conversion count of other sub-regions, the number of clicks to be filtered is calculated.
[0107] In another embodiment of this application, the process of calculating the filtered click count based on the smoothing coefficient, the number of clicks in the target sub-region, the number of conversions in the target sub-region, the number of clicks in other sub-regions, and the number of conversions in other sub-regions includes:
[0108] Identify at least one conversion goal corresponding to the exposure area; for each conversion goal, determine the corresponding conversion rate; based on the conversion rate and the number of clicks in the target sub-region, calculate the number of conversions in the target sub-region corresponding to each conversion goal; based on the conversion rate and the number of clicks in other sub-regions, calculate the number of conversions in other sub-regions corresponding to each conversion goal.
[0109] The number of filtered clicks for each conversion goal is calculated using the following click-count calculation rules:
[0110]
[0111] Among them, filter top To filter click counts, smooth is the smoothing factor, and clk is the clk value. top For the number of clicks in the target sub-region, conv top clk is the number of transformations in the target sub-region. other For the click count of other sub-regions, conv other For the transformation number of other sub-regions.
[0112] In this application, since the conversion rates of advertisements with different conversion goals naturally differ, this application can group them according to the conversion goal. Specifically, for a certain conversion goal, S top The number of clicks for some parts is clk. top Based on the conversion rate of the conversion target and the number of clicks (clk) in the target sub-region. top The transformation number conv of the target subregion can then be calculated. top S other The number of clicks for some parts is clk. other Based on the conversion rate of the conversion target and the click count (clk) of other sub-regions. otherThis allows us to calculate the transformation number (conv) for other sub-regions. other Assuming the conversion rate is similar across different locations, the number of clicks that need to be filtered is filter. top The smoothing coefficient in the click count calculation rule is used to control the filtering intensity, and is usually set to around 1. The smaller the smoothing coefficient value, the stricter the filtering.
[0113] In another embodiment of this application, after calculating the filtered click count corresponding to each conversion target according to the click count calculation rules, the method further includes:
[0114] Determine if the number of filtered clicks for each conversion target is zero;
[0115] If yes, it means that no filtering of click behavior is required, and the process ends; if no, filter out conversion targets with a click count greater than zero and output the filter target information; wherein, the filter target information includes: target sub-region, conversion target to be filtered, and the number of clicks corresponding to the conversion target to be filtered.
[0116] Specifically, after calculating the corresponding number of filtered clicks for each conversion target, this application needs to determine whether each number of filtered clicks is greater than zero. If the number of filtered clicks is greater than zero, the corresponding filtered target information is output for use in subsequent steps. If none of the conversion targets meet the condition after iterating through all conversion targets, that is, all conversion targets are zero, it means that no filtering of clicks is required, and the process ends.
[0117] In this application, the target information to be filtered is (top, conversion, filter) top ), where top represents the target sub-region, conversion represents the conversion targets to be filtered for clicks with a greater than zero, and filter top This indicates the number of filter clicks for the target sub-region.
[0118] In summary, in this application, if the Gini coefficient exceeds the set threshold, it indicates that clicks are concentrated, which may indicate induced clicks. In this case, the sub-region with the highest click percentage can be designated as the target sub-region to be filtered. Based on the conversion rate of different conversion goals, the conversion counts of the target sub-region and other sub-regions are determined. Combined with the click counts of the target sub-region and other sub-regions, the number of clicks to be filtered in the target sub-region is determined. If the number of clicks to be filtered is not zero, subsequent filtering operations can be performed. This method allows for the determination of different number of clicks to be filtered for different conversion goals, thereby improving the accuracy of the filtered click count and ultimately enhancing the filtering effect and accuracy.
[0119] In another embodiment of this application, determining the filtering probability of each click behavior within the target sub-region using behavioral feature information of each click behavior within the target sub-region includes:
[0120] Based on the filtering target information, determine the target click behavior to be filtered corresponding to the target to be filtered conversion within the target sub-region; determine the behavioral feature information corresponding to each target click behavior; use the behavioral feature information of each target click behavior to determine the filtering probability of each target click behavior; wherein, the filtering probability is determined by a machine learning model / filtering rule.
[0121] This application outputs the filtering target information (top, conversion, filter) through the above process. top After that, you can see the target sub-region (top), conversion target (conversion), and click count (filter) that need to be filtered. top Therefore, in this embodiment, click filtering needs to be performed based on the target information. First, based on the target information, click behaviors in the top block with a conversion goal of "conversion" need to be selected; this application refers to these click behaviors as target click behaviors. Then, based on the user's current advertising behavior information, a set of feature information is generated for each target click behavior; this application refers to this set as behavioral feature information. Then, according to filtering rules or a machine learning model, the filtering probability of each target click behavior being a click not intended by the user is obtained. The filtering probabilities are then sorted in descending order, and the top-ranked filters are selected. top Filter the data and output the filter label.
[0122] The advertising behavior information in this application may include: the time difference between the click timestamp and the exposure timestamp, the PCVR of the delivery service, and whether a conversion occurred. The filtering rules in this application may be set as follows: clicks with conversions are worse than those with no conversions, low PCVRs are worse than high PCVRs, and short click-exposure time differences are worse than long click-exposure time differences. That is, clicks that do not generate subsequent conversions are more likely to be induced clicks than clicks that generate conversions, so the filtering probability of clicks that do not generate conversions is higher; clicks with lower PCVRs are more suspicious than clicks with higher PCVRs, so the filtering probability of clicks with lower PCVRs is higher; clicks with shorter time differences are more suspicious than clicks with longer time differences, so the filtering probability of clicks with shorter time differences is higher.
[0123] See Figure 2 The figure shows a schematic flowchart of a specific click behavior filtering method provided in this application embodiment:
[0124] After data collection, this application needs to normalize the click coordinates to a rectangle with a width and height range of 0-1, and then divide the rectangle into multiple smaller rectangles. The click percentage for each smaller rectangle is calculated, and the Gini coefficient is calculated based on the click percentage distribution. If the Gini coefficient exceeds a set threshold, it indicates that clicks are concentrated, which may indicate induced clicks. In this case, the smaller rectangle with the highest click percentage needs to be identified, and the number of filtered clicks for that rectangle needs to be calculated. If the number of filtered clicks is not greater than zero, the process ends; if the number of filtered clicks is greater than zero, click filtering is performed, and the above operation is repeated. The process ends when the Gini coefficient is less than the threshold, or when no click filtering occurs.
[0125] As can be seen, this application uses a click heatmap and the Gini coefficient to assess whether induced clicks exist and where they are located. The Gini coefficient ranges from 0 to 1; a higher Gini coefficient indicates a greater likelihood of concentrated induced clicks. For identified areas with induced clicks, the probability of non-genuine clicks is obtained based on user characteristics through rules or machine learning, thus filtering out clicks. This solution is simple to implement, low-cost, and can effectively detect clicks even when multiple areas are clustered. This click filtering method cleanses the training data for the algorithm model, ensuring customer ROI and reducing invalid affiliate revenue sharing.
[0126] See Figure 3 , Figure 3 This application provides a schematic diagram of a click behavior filtering device, which specifically includes:
[0127] The segmentation module 11 is used to divide the exposure area into sub-regions;
[0128] The first calculation module 12 is used to calculate the click percentage of each sub-region based on the click count of each sub-region; wherein, the click count of each sub-region is: the total number of click behaviors triggered by the user in each sub-region;
[0129] The second calculation module 13 is used to calculate the Gini coefficient of the exposure area based on the click ratio of each sub-region;
[0130] The judgment module 14 is used to determine whether the Gini coefficient is greater than a preset threshold; if so, the filtering module is triggered.
[0131] The filtering module 15 is used to filter out the target sub-regions to be filtered based on the click ratio of each sub-region, and to determine the number of clicks to be filtered for the target sub-regions based on the number of clicks in each sub-region.
[0132] The probability determination module 16 is used to determine the filtering probability of each click behavior in the target sub-region by utilizing the behavioral feature information of each click behavior in the target sub-region;
[0133] The filtering module 17 is used to filter the filtered click behaviors from the target sub-region according to the filtering probability of each click behavior, and re-trigger the first calculation module until the Gini coefficient is not greater than a preset threshold.
[0134] As an optional embodiment, the screening module includes:
[0135] The selection unit is used to select the sub-region with the largest click percentage as the target sub-region; wherein, the sub-regions other than the target sub-region in the exposure area are other sub-regions;
[0136] The calculation unit is used to calculate the number of filtered clicks based on the smoothing coefficient, the number of clicks in the target sub-region, the number of conversions in the target sub-region, the number of clicks in other sub-regions, and the number of conversions in other sub-regions.
[0137] As an optional embodiment, the computing unit is specifically used for:
[0138] Identify at least one conversion target corresponding to the exposure area; for each conversion target, determine the conversion rate corresponding to each conversion target; calculate the conversion count of the target sub-region corresponding to each conversion target based on the conversion rate and the click count of the target sub-region; calculate the conversion count of other sub-regions corresponding to each conversion target based on the conversion rate and the click count of other sub-regions; calculate the filtered click count corresponding to each conversion target according to the click count calculation rules, wherein the click count calculation rules are as follows:
[0139]
[0140] Among them, filter top To filter click counts, smooth is the smoothing factor, and clk is the clk value. top For the number of clicks in the target sub-region, conv top clk is the number of transformations in the target sub-region. other For the click count of other sub-regions, conv other For the transformation number of other sub-regions.
[0141] As an optional embodiment, the computing unit is further configured to:
[0142] Determine whether the number of filtered clicks corresponding to each conversion target is zero; if not, filter conversion targets with a number of filtered clicks greater than zero and output the filter target information; wherein, the filter target information includes: target sub-region, conversion target to be filtered, and the number of filtered clicks corresponding to the conversion target to be filtered.
[0143] As an optional embodiment, probability determination is specifically used for:
[0144] Based on the filtering target information, determine the target click behavior to be filtered corresponding to the target to be filtered conversion within the target sub-region; determine the behavioral feature information corresponding to each target click behavior; use the behavioral feature information of each target click behavior to determine the filtering probability of each target click behavior; wherein, the filtering probability is determined by a machine learning model / filtering rule.
[0145] As an optional embodiment, the partitioning module is specifically used for:
[0146] Obtain the coordinate information of the exposure area and the click coordinates; wherein, the click coordinates are the click coordinates of each click behavior within the exposure area; normalize each click coordinate according to the coordinate information to generate normalized click coordinates, so as to count the number of clicks in each sub-region through the normalized click coordinates.
[0147] As an optional embodiment, the second computing module is specifically used for:
[0148] The click percentages of each sub-region are sorted according to a predetermined order to generate a sorting result; based on the sorting result and the click percentage of each sub-region, the cumulative value of each sub-region is calculated; the Gini coefficient is calculated using the cumulative value of each sub-region and the division parameters; wherein, the division parameters are the parameters for dividing the exposure area into sub-regions.
[0149] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.
[0150] See Figure 4 , Figure 4 A schematic diagram of an electronic device structure provided in this application embodiment includes a processor 21, a communication interface 22, a memory 23 and a communication bus 24, wherein the processor 21, the communication interface 22 and the memory 23 communicate with each other through the communication bus 24;
[0151] Memory 23 is used to store computer programs;
[0152] When the processor 21 executes the program stored in the memory 23, it implements the steps of the filtering method described in any of the above method embodiments, which will not be repeated here.
[0153] The communication bus mentioned in the above terminal can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 4 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.
[0154] The communication interface is used for communication between the aforementioned terminal and other devices.
[0155] The memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.
[0156] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0157] In another exemplary embodiment, a computer storage medium is also provided, wherein the program instructions, when executed by a processor, implement the steps of the filtering method described in any of the above method embodiments. The storage medium may include various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.
[0158] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.
[0159] It should be understood that the terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “described” as used herein may also include the plural forms. The terms “comprising,” “including,” “containing,” and “having” are inclusive and therefore indicate the presence of the stated features, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, elements, components, and / or combinations thereof. The method steps, processes, and operations described herein are not construed as requiring them to be performed in a particular order described or illustrated unless the order of performance is explicitly indicated. It should also be understood that additional or alternative steps may be used.
[0160] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.
Claims
1. A method for filtering click behavior, characterized in that, include: The exposure area is divided into sub-regions; The click percentage of each sub-region is calculated based on the number of clicks in each sub-region; where the number of clicks in each sub-region is: the total number of click behaviors triggered by users in each sub-region; The Gini coefficient of the exposure area is calculated based on the click percentage of each sub-region. Determine whether the Gini coefficient is greater than a preset threshold; If so, then the target sub-regions to be filtered are selected based on the click percentage of each sub-region, and the number of clicks to be filtered for the target sub-regions is determined based on the number of clicks in each sub-region. Using the behavioral feature information of each click behavior within the target sub-region, the filtering probability of each click behavior within the target sub-region is determined; Based on the filtering probability of each click behavior, filter the number of click behaviors from the target sub-region, and re-execute the step of calculating the click ratio of each sub-region based on the number of clicks in each sub-region until the Gini coefficient is not greater than the preset threshold.
2. The filtration method according to claim 1, characterized in that, The target sub-regions to be filtered are identified based on the click percentage of each sub-region, and the number of clicks to be filtered for each target sub-region is determined based on the click count of each sub-region, including: Select the sub-region with the highest click rate as the target sub-region; wherein, the sub-regions other than the target sub-region in the exposure area are other sub-regions; Calculate the filtered click count based on the smoothing coefficient, the number of clicks in the target sub-region, the number of conversions in the target sub-region, the number of clicks in other sub-regions, and the number of conversions in other sub-regions.
3. The filtration method according to claim 2, characterized in that, The calculation of the filtered click count based on the smoothing coefficient, the number of clicks in the target sub-region, the number of conversions in the target sub-region, the number of clicks in other sub-regions, and the number of conversions in other sub-regions includes: Identify at least one conversion target corresponding to the exposure area; For each conversion goal, determine the corresponding conversion rate. Based on the conversion rate and the number of clicks in the target sub-region, calculate the conversion number for each conversion target corresponding to the target sub-region; Based on the conversion rate and the number of clicks in other sub-regions, calculate the conversion count for each conversion target in the other sub-regions; The number of filtered clicks corresponding to each conversion target is calculated using the click count calculation rules, which are as follows: Among them, filter top To filter click counts, smooth is the smoothing factor, and clk is the clk value. top For the number of clicks in the target sub-region, conv top clk is the number of transformations in the target sub-region. other For the click count of other sub-regions, conv other For the transformation number of other sub-regions.
4. The filtration method according to claim 3, characterized in that, After calculating the filtered click count for each conversion target using the click count calculation rules, the method further includes: Determine if the number of filtered clicks for each conversion target is zero; If not, filter out conversion targets with a click count greater than zero and output the filter target information; wherein, the filter target information includes: target sub-region, conversion target to be filtered, and the number of clicks corresponding to the conversion target to be filtered.
5. The filtration method according to claim 4, characterized in that, Determining the filtering probability of each click behavior within the target sub-region using behavioral feature information of each click behavior within the target sub-region includes: Based on the filtering target information, determine the target click behavior to be filtered within the target sub-region that corresponds to the target to be filtered for conversion; Determine the behavioral characteristic information corresponding to each target click behavior; The filtering probability of each target click behavior is determined by utilizing the behavioral feature information of each target click behavior; wherein the filtering probability is determined by a machine learning model / filtering rule.
6. The filtering method according to any one of claims 1 to 5, characterized in that, The process of dividing the exposure area into sub-regions includes: Obtain the coordinate information of the exposure area and the click coordinates; wherein, the click coordinates are the click coordinates of each click action within the exposure area; The click coordinates are normalized based on the coordinate information to generate normalized click coordinates, so as to count the number of clicks in each sub-region.
7. The filtration method according to claim 6, characterized in that, The calculation of the Gini coefficient of the exposure area based on the click percentage of each sub-region includes: Sort the click percentage of each sub-region according to a predetermined order and generate the sorting results; Calculate the cumulative value for each sub-region based on the sorting results and the click percentage of each sub-region; The Gini coefficient is calculated using the cumulative values of each sub-region and the division parameters; wherein, the division parameters are the parameters used to divide the exposure area into sub-regions.
8. A click behavior filtering device, characterized in that, include: The partitioning module is used to divide the exposure area into sub-regions; The first calculation module is used to calculate the click percentage of each sub-region based on the number of clicks in each sub-region; wherein, the number of clicks in each sub-region is: the total number of click behaviors triggered by the user in each sub-region; The second calculation module is used to calculate the Gini coefficient of the exposure area based on the click percentage of each sub-region; The judgment module is used to determine whether the Gini coefficient is greater than a preset threshold; if so, the filtering module is triggered. The filtering module is used to filter out the target sub-regions to be filtered based on the click ratio of each sub-region, and to determine the number of clicks to be filtered for the target sub-regions based on the number of clicks in each sub-region. The probability determination module is used to determine the filtering probability of each click behavior in the target sub-region by utilizing the behavioral feature information of each click behavior in the target sub-region; The filtering module is used to filter the filtered click behaviors from the target sub-region according to the filtering probability of each click behavior, and re-trigger the first calculation module until the Gini coefficient is not greater than a preset threshold.
9. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a program stored in memory, it implements the steps of the click behavior filtering method as described in any one of claims 1 to 7.
10. A computer storage medium, characterized in that, The computer storage medium stores computer-executable instructions for performing the steps of the click behavior filtering method according to any one of claims 1 to 7.