Multi-source traffic realization distribution system and method oriented to APP platform

Abstract

The invention relates to the technical field of flow realization, and discloses an APP platform-oriented multi-source flow realization distribution system and method, which are used for solving the problem of inaccurate user behavior data acquisition caused by an interactive interface when APP flow realization distribution is carried out according to user behavior data. The method comprises the following steps: acquiring click events of an APP user through log data, screening the click events to obtain actual click events, constructing a user portrait according to the actual click events, identifying equipment information and an operation environment of login APP equipment, using a machine learning model, and according to the user portrait and the equipment information and the operation environment of the equipment, identifying the equipment information and the operation environment of the login APP equipment. Value evaluation is performed on the traffic in the APP to obtain the traffic grade, the historical log of the advertisement resources is obtained, the advertisement filling rate and the unit traffic income are obtained according to the historical log, the advertisement resource putting strategy is obtained through evaluation, the accuracy of user behavior data obtaining is effectively improved, and the advertisement matching effect is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of traffic monetization, and more particularly to a multi-source traffic monetization distribution system and method for an APP platform. BACKGROUND

[0002] With the rapid popularization of mobile Internet and smart terminals, digital advertising and content monetization based on APP platforms have become an important way of traffic operation. Traffic monetization systems provide sustainable commercial revenue paths for APP developers through user behavior data collection, traffic value assessment, advertisement resource matching and revenue distribution, etc. Especially in the context of the gradual decline of user growth bonuses, building an efficient and accurate multi-source traffic monetization distribution system has become an indispensable key component in the mobile application service ecosystem.

[0003] To improve traffic utilization and monetization revenue, the existing technology usually adopts a traffic intelligent grading method based on user behavior collection, analyzes the behaviors of users in the APP such as clicking, sliding, staying, and conversion, combines device information, environmental characteristics, and historical operation trajectories, grades different users according to their values, and accordingly formulates differentiated advertising placement and content distribution strategies. This method can tap potential high-value users, realize fine scheduling of traffic resources, and maximize multi-source revenue.

[0004] However, the above-mentioned technology at least has the following technical problems:

[0005] In actual application scenarios, the behavior data of APP users is interfered by various factors, especially the layout and operation logic of the interactive interface, which can significantly affect the real click intention of the user. The current common misleading interface design includes forced pop-up windows, advertisement button disguises, and induced jump areas, etc. These UI structures can guide users to click on advertising content without autonomous intention, and the behavior data recorded by the collection system deviates due to the lack of real interactive intention. For example, after the user mistakenly clicks on the advertisement, he quickly exits the page or has no subsequent interactive behavior, but the system identifies it as "interest click" or "high active behavior", which leads to user portrait construction deviation, traffic grading distortion, and further affects the advertising matching effect and the execution result of the monetization strategy. SUMMARY

[0006] In order to overcome the above-mentioned defects of the prior art, the present application provides a multi-source traffic monetization distribution system and method for an APP platform to solve the problems existing in the background art.

[0007] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0008] The application discloses a multi-source traffic monetization distribution system for an APP platform, which comprises a user behavior collection module, a device and environment identification module, a traffic intelligent grading module, an advertisement scheduling module and a revenue monitoring and settlement module.

[0009] Preferably, the step of screening the click event to obtain the actual click event comprises the following steps: collecting click event data in a detection time period through log data, wherein the click event data comprises click delay data, page dwell time after clicking, click position and interaction track data and total click times of the page; and evaluating a forced click index according to the click event data; determining whether the click event in the user behavior collection data is a user voluntary click according to the forced click index; if it is determined that the click event is not a user voluntary click, the click event is removed; all click events that are not user voluntary clicks are removed by traversing all click events of the user in the detection time period, so as to obtain the actual click event.

[0010] Preferably, the forced click index obtaining step is: obtaining click delay data, the click delay data being the click time after the page loading is completed, obtaining the advertisement display duration, performing ratio calculation on the click delay data and the advertisement display duration to obtain a click delay coefficient; obtaining user click coordinates, advertisement region coordinate range and interaction trajectory data, and calculating the interaction voluntary coefficient according to the user click coordinates, the advertisement region coordinate range and the interaction trajectory data; obtaining the total number of clicks of the user on the page, denoted as the current page click number, obtaining the total number of clicks of the user on other pages in the APP, and calculating the average page click number according to the total number of clicks of the user on other pages in the APP; performing ratio calculation on the current page click number and the average page click number to obtain a click frequency abnormality coefficient; obtaining the page dwell time after the user clicks, and performing normalization processing on the page dwell time, the click delay coefficient, the interaction voluntary coefficient and the click frequency abnormality coefficient, and evaluating the forced click index according to the normalized page dwell time, the normalized click delay coefficient, the normalized interaction voluntary coefficient and the normalized click frequency abnormality coefficient, and the specific obtaining steps are: In the formula, FC represents the forced click index, ST' represents the normalized page dwell time, CD' represents the normalized click delay coefficient, VI' represents the normalized interaction voluntary coefficient, AF' represents the normalized click frequency abnormality coefficient, and a1, a2, a3 and a4 represent the weight coefficients of the normalized page dwell time, the normalized click delay coefficient, the normalized interaction voluntary coefficient and the normalized click frequency abnormality coefficient.

[0011] Preferably, the interaction voluntary coefficient obtaining step is: obtaining the advertisement region coordinate range and the user click coordinates, the advertisement region coordinate range including the maximum X-axis coordinate, the minimum X-axis coordinate, the maximum Y-axis coordinate and the minimum Y-axis coordinate, and the user click coordinates including the X-axis click coordinate and the Y-axis click coordinate; obtaining the advertisement region center coordinate according to the advertisement region coordinate range, calculating the Euclidean distance from the user click coordinate to the advertisement region center coordinate, denoted as the click Euclidean distance; obtaining the half diagonal length of the advertisement region as the equivalent radius of the advertisement region according to the advertisement region coordinate range, and calculating the position rationality according to the click Euclidean distance and the equivalent radius of the advertisement region; setting a detection time period, obtaining the interaction trajectory data in the detection time period before the user clicks, the interaction trajectory data including the X-axis coordinate and the Y-axis coordinate of each timestamp in the detection time period; evaluating the comprehensive interaction naturalness according to the interaction trajectory data, and performing product calculation on the position rationality and the comprehensive interaction naturalness to obtain the interaction voluntary coefficient.

[0012] Preferably, the step of evaluating the comprehensive interaction naturalness according to the interaction trajectory data is: obtaining the direction change of adjacent trajectory points according to the interaction trajectory data in the detection period, calculating the standard deviation of the direction change in the detection period, denoted as the trajectory smoothness; setting the number of trajectory points, obtaining the direction angle corresponding to the last trajectory point number, and calculating the average moving direction; calculating the direction of the center coordinates of the advertising region relative to the trajectory starting point, and obtaining the trajectory directionality according to the average moving direction and the direction of the center coordinates of the advertising region relative to the trajectory starting point; obtaining the speed of each trajectory point, calculating the standard deviation of the speed of each trajectory point as the trajectory speed variation; calculating the mean of the trajectory smoothness, the trajectory directionality and the trajectory speed variation to obtain the comprehensive interaction naturalness.

[0013] Preferably, the step of determining whether the click event in the user behavior collection data is a user voluntary click according to the forced click index is: comparing the forced click index with the forced threshold value, if the forced click index is greater than or equal to the forced threshold value, it is determined that the user is not voluntary click; if the forced click index is less than the forced threshold value, it is determined that the user is voluntary click.

[0014] Preferably, the step of constructing the user portrait according to the actual click event is: extracting the key fields of each click record from the screened actual click event; classifying the actual click event according to the content type, function module and business path dimension; based on the classified click event, counting the intensity index of each type of behavior, identifying the main preference type of the user in each type of click behavior through aggregation analysis, extracting the typical operation mode as the behavior mode of the user individual characteristics; converting the intensity index of each type of behavior and the behavior mode into a calculable user portrait label.

[0015] Preferably, the step of evaluating the value of the traffic in the APP according to the user portrait, the device information and the running environment of the device to obtain the traffic classification is: uniformly encoding the user portrait features, device information and running environment information to obtain a structured feature vector; inputting the above constructed feature vector into a machine learning model, learning the mapping relationship between the click rate, conversion rate and unit traffic revenue target variable and the input features through historical data, and outputting a prediction value, denoted as the expected value score of the current traffic sample; according to the expected value score, combining the set score threshold, the traffic is divided into different levels.

[0016] Preferably, a multi-source traffic monetization distribution method for an APP platform comprises the following steps:

[0017] Step 1: collect the click events of APP users through log data, filter the click events, obtain actual click events, and construct a user portrait according to the actual click events; Step 2: identify the device information and running environment of the APP device logged in, the device information includes model, operating system and screen parameters, and the running environment includes network status, positioning and time period; Step 3: use a machine learning model to evaluate the value of the traffic in the APP according to the user portrait and the device information and running environment of the device, and obtain traffic classification; Step 4: obtain the connected advertising resources, and obtain the historical logs of the advertising resources, obtain the ad fill rate and unit traffic revenue according to the historical logs, dynamically evaluate the advertising resource placement strategy according to the traffic classification, the ad fill rate and the unit traffic revenue, and display the ads according to the advertising resource placement strategy; Step 5: collect data such as ad display, click and conversion in real time, and calculate the estimated revenue and actual settlement revenue.

[0018] Technical effects and advantages of the present application:

[0019] Through log data collection, the click events of APP users are collected, the click events are filtered, actual click events are obtained, a user portrait is constructed according to the actual click events, the device information and running environment of the APP device logged in are identified, a machine learning model is used, the value of the traffic in the APP is evaluated according to the user portrait and the device information and running environment of the device, traffic classification is obtained, the historical logs of the advertising resources are obtained, the ad fill rate and unit traffic revenue are obtained according to the historical logs, and the advertising resource placement strategy is evaluated, thereby effectively improving the accuracy of user behavior data acquisition and improving the ad matching effect. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 A multi-source traffic monetization distribution system structure diagram for an APP platform is provided for the embodiments of the present application.

[0021] Figure 2 A multi-source traffic monetization distribution method flowchart for an APP platform is provided for the embodiments of the present application. DETAILED DESCRIPTION

[0022] The technical solutions in the present application will be described clearly and completely below with reference to the drawings in the present application, and the forms of each structure described in the following embodiments are only examples, and a multi-source traffic monetization distribution system and method for an APP platform are not limited to each structure described in the following embodiments, and all other embodiments obtained by those skilled in the art without creative labor belong to the scope of protection of the present application.

[0023] The present application provides a multi-source traffic monetization distribution system for an APP platform, as shown inFigure 1 As shown in the figure, the system comprises:

[0024] a user behavior collection module, configured to collect click events of APP users through log data, filter the click events, obtain actual click events, and construct a user portrait according to the actual click events;

[0025] In this embodiment, it needs to be specifically explained that the step of filtering the click events to obtain the actual click events is:

[0026] The user detection time period is collected through log data, and the click event data includes click delay data, page dwell time after clicking, click position and interaction track data, and total click times of the page. The forced click index is obtained according to the click event data;

[0027] According to the forced click index, it is determined whether the click event in the user behavior collection data is a user voluntary click;

[0028] If it is determined that it is not a user voluntary click, the click event is removed. All click events that are not user voluntary clicks are removed in the user detection time period, and the actual click events are obtained.

[0029] In this embodiment, it needs to be specifically explained that the step of obtaining the forced click index is:

[0030] The click delay data is obtained, the click delay data is the click time after the page is loaded, the advertisement display duration is obtained, the click delay data and the advertisement display duration are calculated by ratio, and the click delay coefficient is obtained. The higher the click delay coefficient is, the longer the user's browsing and thinking time is, and the more likely the click is a spontaneous behavior;

[0031] The user click coordinates, the advertisement area coordinate range and the interaction track data are obtained, and the interaction voluntary coefficient is calculated according to the user click coordinates, the advertisement area coordinate range and the interaction track data;

[0032] The total click times of the user on the page are obtained, which are recorded as the current page click times. The total click times of the user on other pages in the APP are obtained. The average page click times are calculated according to the total click times of the user on other pages in the APP. The current page click times and the average page click times are calculated by ratio to obtain the click frequency abnormal coefficient. If the user's click behavior in other parts is normal and only in the advertisement area there is an abnormal burst, attention should be paid to whether the UI design of this part exists the problem of guiding mis-click;

[0033] The page dwell time after the user clicks is obtained, the page dwell time, the click delay coefficient, the interaction voluntary coefficient, and the click frequency abnormality coefficient are normalized, and the forced click index is evaluated according to the normalized page dwell time, the click delay coefficient, the interaction voluntary coefficient, and the click frequency abnormality coefficient. The specific acquisition steps are as follows:

[0034]

[0035] In the formula, FC represents a forced click index, ST' represents a normalized page dwell time, and the longer the user stays on the landing page after clicking on the advertisement, the higher the degree of interest or willingness to participate in the advertisement content, and the more likely the click is a user-initiated true behavior; at this time, the forced click index should be reduced to avoid misjudging the behavior as a misleading click. On the contrary, if the user quickly exits the page after clicking, it usually means that the click is passive or caused by accidental touch, so the short page dwell time reflects the involuntary nature of the click behavior, and the forced click index should be increased accordingly. The relationship helps the system accurately distinguish between valid clicks and pseudo behaviors and improves the quality evaluation capability of advertisement delivery. CD' represents a normalized click delay coefficient, and when the click delay coefficient is high, the user is more likely to initiate a click behavior after fully browsing and thinking, which belongs to the interaction driven by true intention; at this time, the system should determine that the click is a voluntary click, and the forced click index is correspondingly lower. The relationship reflects the initiative of the user's click behavior: the longer the delay, the less likely it is caused by accidental touch or interface misdirection, thereby reducing the possibility of being identified as a "forced click", which helps to eliminate pseudo behavior data and improve the accuracy of user portrait construction. VI' represents a normalized interaction voluntary coefficient, and when the user's click position is concentrated in the position with clear advertisement identification in the interface, and the interaction trajectory shows natural sliding, pausing, and pointing operations, it means that the user is clicking the operation under clear awareness and active guidance, and the interaction behavior has high voluntary, so the forced click index should be lower. On the contrary, if the click position is close to the accidental touch high-frequency area (such as the edge, near the close button) and the trajectory is abrupt and has no continuous interaction action, it may be a non-voluntary click caused by interface induction or misdirection, the interaction voluntary coefficient is low, and the forced click index should be high. AF' represents a normalized click frequency anomaly coefficient, and when a page is clicked frequently in a short time, the number of clicks is much higher than the historical average level of similar pages, there may be misleading interface design or abnormal behavior of the user being forced to click multiple times, so the forced click index should be increased. In other words, the higher the click frequency anomaly coefficient, the more likely it reflects a non-natural or non-voluntary click scenario, such as repeated clicks caused by button camouflage, forced task chain, or interface obstruction, prompting the system to identify the behavior as low-quality traffic and filter it to enhance the accuracy and credibility of traffic grading. a1, a2, a3, and a4 represent the weight coefficients of the normalized page dwell time, the normalized click delay coefficient, the normalized interaction voluntary coefficient, and the normalized click frequency anomaly coefficient, respectively, and a1+a2+a3+a4=1. a1, a2, a3, and a4 are obtained by the analytic hierarchy process, which is a quantitative method for determining the relative weight of each factor in a multi-factor decision-making problem.The basic idea is to decompose a complex problem into multiple levels such as target layer, criterion layer and index layer, reflect the relative importance of subjective judgment by constructing a judgment matrix and comparing each factor with each other; then use consistency check and eigenvector calculation to obtain the weight coefficient of each factor. In this embodiment, the analytic hierarchy process is used to quantitatively compare the importance of the click delay coefficient, the interaction voluntary coefficient, the click abnormal coefficient and the dwell time and other factors according to the experience of field experts or historical evaluation results, so as to scientifically determine the weight of each parameter in the forced click index calculation, and ensure the rationality and interpretability of the evaluation model.

[0036] In this embodiment, it needs to be specifically pointed out that the interaction voluntary coefficient acquisition step is:

[0037] The advertisement region coordinate range and the user click coordinate are obtained, the advertisement region coordinate range includes the maximum X-axis coordinate, the minimum X-axis coordinate, the maximum Y-axis coordinate and the minimum Y-axis coordinate, and the user click coordinate includes the X-axis click coordinate and the Y-axis click coordinate;

[0038] The center coordinate of the advertisement region is obtained according to the advertisement region coordinate range, the Euclidean distance of the user click coordinate to the center coordinate of the advertisement region is calculated, which is recorded as the click Euclidean distance. The Euclidean distance is the most commonly used distance measurement method in geometry, which is used to calculate the straight line distance between two points. In the advertisement interaction scene, the click Euclidean distance represents the straight line deviation degree of the user click position and the advertisement center point. The smaller the distance, the closer the click to the core area of the advertisement, and the behavior may be more accurate. The larger the distance, the more likely it is a false touch or an edge click;

[0039] The half diagonal length of the advertisement region is obtained according to the advertisement region coordinate range as the equivalent radius of the advertisement region. The position rationality is calculated according to the click Euclidean distance and the equivalent radius of the advertisement region. The specific acquisition steps are:

[0040]

[0041] In the formula, PR represents the position rationality, PD dis represents the click Euclidean distance, R ad represents the equivalent radius of the advertisement region;

[0042] A detection time period is set, and interaction trajectory data in the detection time period before the user click is obtained, the interaction trajectory data including the X-axis coordinate and the Y-axis coordinate of each timestamp in the detection time period;

[0043] The comprehensive interaction naturalness is evaluated according to the interaction trajectory data, the position rationality and the comprehensive interaction naturalness are multiplied to obtain the interaction voluntary coefficient, and the interaction voluntary coefficient ∈ [0, 1]. The value is closer to 1, and the click behavior is more likely to be voluntary.

[0044] In this embodiment, it needs to be specifically pointed out that the step of evaluating the comprehensive interaction naturalness according to the interaction trajectory data is:

[0045] According to the direction change of the adjacent trajectory points in the detection time period, the standard deviation of the direction change in the detection time period is calculated, denoted as the trajectory smoothness;

[0046] Set the number of trajectory points, for example, three trajectory points, obtain the direction angle corresponding to the last trajectory point number, and calculate the average moving direction;

[0047] The direction of the center coordinates of the advertising area relative to the starting point of the trajectory is calculated, and the specific obtaining steps are:

[0048]

[0049] In the formula, θ ad The direction of the center coordinates of the advertising area relative to the starting point of the trajectory is denoted as y center The Y-axis coordinate in the center coordinates of the advertising area is denoted as x center The X-axis coordinate in the center coordinates of the advertising area is denoted as y1, and x1 represents the Y-axis coordinate of the starting point of the trajectory.

[0050] According to the average moving direction and the direction of the center coordinates of the advertising area relative to the starting point of the trajectory, the trajectory directionality is calculated, and the specific obtaining steps are:

[0051]

[0052] In the formula, D represents the trajectory directionality, θ traj The average moving direction is denoted as θ ad The direction of the center coordinates of the advertising area relative to the starting point of the trajectory is denoted as;

[0053] The speed of each trajectory point is obtained, and the standard deviation of the speed of each trajectory point is calculated as the trajectory speed variation;

[0054] The trajectory smoothness, the trajectory directionality and the trajectory speed variation are mean calculated to obtain the comprehensive interaction naturalness.

[0055] In this embodiment, it needs to be specifically pointed out that the step of determining whether the click event in the user behavior collection data is a user voluntary click according to the forced click index is:

[0056] The forced click index is compared with a forced threshold value. If the forced click index is greater than or equal to the forced threshold value, it is determined that the user is involuntarily clicked. If the forced click index is less than the forced threshold value, it is determined that the user is voluntarily clicked. The forced threshold value is obtained by an adaptive threshold method. The adaptive threshold method is a method for dynamically determining a determination threshold value according to actual data distribution, and is used to avoid the determination deviation caused by a fixed threshold value in different scenarios. In the embodiment, the adaptive threshold method adjusts the forced threshold value by statistically analyzing the distribution characteristics of the forced click index in a certain time window or a certain number of user samples, and combining indicators such as quantile, standard deviation or clustering boundary. For example, the threshold value can be set as the 90% quantile value of the FCI in the current data set, or an abnormality detection formula based on the mean and standard deviation is used to dynamically update the threshold value, so that the user is adaptively identified as voluntarily clicked in different user groups, different APP versions or different interactive interfaces, and the misjudgment control ability and system adaptability are improved.

[0057] In the embodiment, it needs to be specifically explained that the user portrait is constructed according to the actual click event, and the steps are as follows:

[0058] The key fields of each click record are extracted from the screened actual click events, including user unique identifier, click timestamp, click content type, click position, target page category and the like. The information is used as the basic atomic data of user behavior characteristics, and provides input for subsequent feature classification and behavior modeling;

[0059] The actual click events are classified according to the content type, function module, business path and the like. For example, the product page click is classified into “consumption behavior”, the advertisement page click is classified into “commercial interest”, the setting page click is classified into “function familiarity” and the like, and the behavior type is divided in a structured manner to provide clear classification basis for the user portrait label;

[0060] Based on the classified click events, the intensity indicators of each type of behavior are counted, including frequency, active period (such as daily click number), click path depth (such as page jump level) and the like, which are used to reflect the activity, participation willingness and use preference of the user on various contents, and form the numerical basis of the portrait;

[0061] Through aggregation analysis, the main preference type of the user in each type of click behavior (such as preference for a certain type of product, active in a specific time period, strong response to advertisements and the like) is identified, and the typical operation mode (such as “entering the home page→clicking the recommendation→visiting the detail page”) is extracted as the behavior mode of the user individual characteristics;

[0062] The intensity indicators of each type of behavior and the behavior patterns are converted into computable user portrait labels such as "ad response positive", "morning active user", "preference for technology content", etc., and are encoded in vector form to form structured portrait data that can be used in downstream traffic classification, ad placement strategies and content recommendation logic.

[0063] A device and environment identification module is used to identify the device information and running environment of the logged-in APP device, the device information including model, operating system and screen parameters, and the running environment including network status, positioning and time period, etc.

[0064] By accurately obtaining hardware information such as device model, operating system, screen parameters, and environment data such as network status, positioning, and time period, dynamic and personalized decision-making basis can be provided for traffic monetization distribution. The advantages are: on the one hand, by adapting to device characteristics (such as screen size, system version) to optimize ad display form, improve user experience and click rate; on the other hand, combined with real-time environment (such as network bandwidth, geographic location, use period) to intelligently allocate high-value ad resources (such as video ads preferentially placed in Wi-Fi environment), thereby significantly improving ad fill rate and monetization efficiency, while reducing invalid traffic loss due to device compatibility or environment mismatch.

[0065] A traffic intelligent classification module is used to use a machine learning model to value APP traffic according to user portraits and device information and running environment of the device to obtain traffic classification;

[0066] Through machine learning model dynamic analysis of user portraits (such as interests, consumption ability), device information (such as model, system version) and running environment (such as network, geographic location, period), accurate value classification of APP traffic is performed. The advantages are: improving monetization efficiency, preferentially matching high-value ads (such as high-priced brand ads) to high-potential traffic (such as high-end model users, Wi-Fi environment); optimizing user experience, avoiding low-relevance or excessive ad interference with users; dynamically adapting to market changes, adjusting traffic classification strategies in real time to respond to user behavior fluctuations or changes in advertiser demand, thereby maximizing platform revenue while maintaining user activity.

[0067] The machine learning model is a multi-feature fusion supervised learning model for value scoring and classification of each user traffic sample in the APP based on user portrait features and device information and running environment, etc. The model is usually trained using a deep neural network, with historical click-through rate, conversion rate or unit traffic revenue as labels for fitting, to realize value prediction and classification of unknown traffic, thereby guiding ad distribution strategies and resource priority allocation.

[0068] In this embodiment, it needs to be specifically pointed out that according to the user portrait and the device information and the running environment of the device, the value evaluation of the traffic in the APP is obtained by the traffic grading step:

[0069] The user portrait features, device information, and running environment information are uniformly coded to obtain a structured feature vector, which is normalized and numerically converted to ensure that different dimensional data can be effectively fused and involved in training or reasoning in the machine learning model, avoiding model bias to a certain feature dimension;

[0070] The above constructed feature vector is input into the machine learning model, the mapping relationship between the target variables such as click rate, conversion rate, and unit traffic revenue and the input features is learned through historical data, and a prediction value is output, which is the expected value score of the current traffic sample;

[0071] According to the expected value score, combined with the set score threshold, the traffic is divided into different levels (such as high-value users, medium-value users, and low-value users), which is used as the decision basis for ad engine, content distribution, or traffic strategy, for accurate placement and resource priority configuration.

[0072] The advertisement scheduling module is used to obtain the connected advertisement resources and obtain the historical logs of the advertisement resources, obtain the advertisement filling rate and the unit traffic revenue according to the historical logs, dynamically evaluate the advertisement resource placement strategy according to the traffic grading, the advertisement filling rate and the unit traffic revenue, and display the advertisements according to the advertisement resource placement strategy;

[0073] Through real-time integration of traffic grading, advertisement filling rate and unit traffic revenue data, dynamic optimization of advertisement placement strategy, its core advantage lies in: revenue maximization, intelligent matching of high-value traffic and high-priced advertisement resources, and improvement of overall monetization efficiency; filling rate balance, ensuring advertisement exposure while avoiding low-quality advertisements to reduce user experience; dynamic adaptability, automatically adjusting the placement strategy according to real-time market changes (such as advertiser bidding fluctuations or user traffic quality changes), ensuring that the platform continuously obtains optimal advertisement revenue in a changing environment.

[0074] The revenue monitoring and settlement module is used to collect advertisement display, click, and conversion data in real time, and calculate the estimated revenue and actual settlement revenue.

[0075] Through real-time tracking of advertisement display, click and conversion data, and comparison of estimated revenue and actual settlement revenue, its core value lies in: financial transparency, accurate calculation of advertisement revenue, avoiding settlement errors or traffic fraud; dynamic optimization basis, providing real-time data support for adjusting traffic distribution strategy, improving monetization efficiency; advertiser trust guarantee, enhancing cooperation confidence through data auditability, promoting long-term business ecosystem health.

[0076] In this embodiment, it needs to be specifically pointed out that, as shown in Figure 2 A multi-source traffic monetization distribution method for APP platform, comprising the following steps:

[0077] Step 1: Collect the click events of APP users through log data, and filter the click events to obtain actual click events, and construct user portraits according to the actual click events;

[0078] Step 2: Identify the device information and running environment of the APP device, including model, operating system and screen parameters, etc., and the running environment includes network status, positioning and time period, etc.;

[0079] Step 3: Use a machine learning model to evaluate the value of the traffic in the APP according to the user portrait and the device information and running environment of the device to obtain traffic classification;

[0080] Step 4: Obtain the connected advertising resources and the historical log of the advertising resources, obtain the ad fill rate and unit traffic revenue according to the historical log, dynamically evaluate the advertising resource placement strategy according to the traffic classification, ad fill rate and unit traffic revenue, and display the ads according to the advertising resource placement strategy;

[0081] Step 5: Real-time collection of advertising display, click and conversion data, and calculation of estimated revenue and actual settlement revenue.

[0082] Finally, the above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application should be included in the protection scope of the present application.

[0083] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A multi-source traffic monetization and distribution system for APP platforms, characterized by: The system comprises: The user behavior collection module is used to collect the click events of APP users through log data, filter the click events, obtain the actual click events, and build user profiles based on the actual click events; The device and environment identification module is used to identify the device information and operating environment of the device logging into the APP. The device information includes the model, operating system, and screen parameters. The operating environment includes the network status, location, and time period. The intelligent traffic classification module uses machine learning models to evaluate the value of traffic within the app based on user profiles, device information, and operating environment, and to obtain traffic classification. The advertising scheduling module is used to obtain the connected advertising resources and obtain the historical logs of advertising resources. Based on the historical logs, it obtains the advertising fill rate and unit traffic revenue. Based on the traffic classification, advertising fill rate and unit traffic revenue, it dynamically evaluates the advertising resource delivery strategy and displays advertisements according to the advertising resource delivery strategy. The revenue monitoring and settlement module is used to collect data such as advertising display, clicks, and conversions in real time, and calculate estimated revenue and actual settlement revenue.

2. A multi-source traffic monetization and distribution system for APP platforms according to claim 1, characterized in that: The steps of filtering the click events and obtaining the actual click events are as follows: The click event data within the user detection period is collected through log data. The click event data includes click delay data, page dwell time after click, click location and interaction trajectory data, and the total number of clicks on the page. The forced click index is obtained based on the click event data evaluation. Determine whether the click events in the user behavior collection data are voluntary clicks by the user based on the forced click index; If it is determined that the click is not voluntary by the user, the click event is eliminated, and all click events of the user within the detection time period are traversed, and all click events that are not voluntary by the user are eliminated to obtain the actual click event.

3. A multi-source traffic realization and distribution system for APP platforms according to claim 2, characterized in that: The steps for obtaining the forced click index are: Obtain click delay data, which is the click time after the page is loaded. Obtain the ad display duration and calculate the ratio of the click delay data to the ad display duration to obtain the click delay coefficient. Obtain the user click coordinates, the advertising area coordinate range, and the interaction trajectory data, and calculate the interaction willingness coefficient based on the user click coordinates, the advertising area coordinate range, and the interaction trajectory data; Get the total number of clicks on the page clicked by the user, record it as the number of clicks on the current page, get the total number of clicks on other pages in the app by the user, calculate the average number of page clicks based on the total number of clicks on other pages in the app, and calculate the ratio of the number of clicks on the current page to the average number of page clicks to obtain the click frequency anomaly coefficient; Obtain the page dwell time after the user clicks, normalize the page dwell time, click delay coefficient, interaction willingness coefficient, and click frequency abnormality coefficient, and evaluate the forced click index based on the normalized page dwell time, click delay coefficient, interaction willingness coefficient, and click frequency abnormality coefficient. The specific acquisition steps are as follows: Where FC represents the forced click index, ST' represents the normalized page dwell time, CD' represents the normalized click delay coefficient, VI' represents the normalized interaction willingness coefficient, AF' represents the normalized click frequency anomaly coefficient, a1, a2, a3, and a4 represent the weight coefficients of the normalized page dwell time, the normalized click delay coefficient, the normalized interaction willingness coefficient, and the normalized click frequency anomaly coefficient.

4. A multi-source traffic realization and distribution system for APP platforms according to claim 3, characterized in that: The steps for obtaining the interaction willingness coefficient are: Get the coordinate range of the ad area and the user click coordinates. The ad area coordinate range includes the maximum X-axis coordinate, the minimum X-axis coordinate, the maximum Y-axis coordinate, and the minimum Y-axis coordinate. The user click coordinates include the X-axis click coordinate and the Y-axis click coordinate. Get the center coordinates of the ad area based on the ad area coordinate range, and calculate the Euclidean distance from the user's click coordinates to the center coordinates of the ad area, which is recorded as the click Euclidean distance; The semi-diagonal length of the advertising area is obtained according to the coordinate range of the advertising area, which is used as the equivalent radius of the advertising area. The rationality of the position is calculated based on the Euclidean distance of the click and the equivalent radius of the advertising area. Set a detection time period and obtain the interaction trajectory data within the detection period before the user clicks. The interaction trajectory data includes the X-axis coordinate and Y-axis coordinate of each timestamp within the detection period; The comprehensive interaction naturalness is obtained based on the interaction trajectory data evaluation, and the interaction voluntary coefficient is obtained by multiplying the location rationality and the comprehensive interaction naturalness.

5. The multi-source traffic monetization and distribution system for APP platforms according to claim 4 is characterized by: The step of evaluating the comprehensive interaction naturalness based on the interaction trajectory data is as follows: The direction changes of adjacent trajectory points are obtained based on the interactive trajectory data during the detection period, and the standard deviation of the direction changes during the detection period is calculated and recorded as the trajectory smoothness; Set the number of trajectory points, obtain the direction angle corresponding to the last trajectory point, and calculate the average moving direction; Calculate the direction of the center coordinates of the advertising area relative to the starting point of the trajectory, and calculate the trajectory directivity based on the average moving direction and the direction of the center coordinates of the advertising area relative to the starting point of the trajectory; Get the speed of each trajectory point and calculate the standard deviation of the speed of each trajectory point as the trajectory speed change; The trajectory smoothness, trajectory directionality, and trajectory speed change are averaged to obtain the comprehensive interaction naturalness.

6. The multi-source traffic monetization and distribution system for APP platforms according to claim 2 is characterized by: The step of determining whether a click event in the user behavior collection data is a voluntary click by the user according to the forced click index is as follows: The forced click index is compared with the forced threshold. If the forced click index is greater than or equal to the forced threshold, it is determined that the user clicked involuntarily; if the forced click index is less than the forced threshold, it is determined that the user clicked voluntarily.

7. The multi-source traffic monetization and distribution system for APP platforms according to claim 1 is characterized by: The steps for constructing a user portrait based on actual click events are: Extract the key fields of each click record from the filtered actual click events; Classify actual click events by content type, functional module, and business path dimensions; Based on the classified click events, we collect statistics on the intensity indicators of each type of behavior. Through aggregate analysis, we identify the main preference types of users in each type of click behavior and extract typical operation patterns as the behavioral patterns of individual user characteristics. Convert the intensity indicators and behavior patterns of each type of behavior into computable user profile labels.

8. The multi-source traffic monetization and distribution system for APP platforms according to claim 1 is characterized by: The steps for evaluating the value of traffic within the APP based on the user profile, device information, and operating environment to obtain traffic classification are as follows: User profile features, device information, and operating environment information are uniformly encoded to obtain a structured feature vector; The constructed feature vector is input into the machine learning model. The model learns the mapping relationship between the target variables of click-through rate, conversion rate, and unit traffic revenue and the input features through historical data, and outputs a predicted value, which is recorded as the expected value score of the current traffic sample. Traffic is divided into different levels based on the expected value score and the set score threshold.

9. A multi-source traffic monetization and distribution method for an APP platform, used to implement the multi-source traffic monetization and distribution system for an APP platform according to any one of claims 1 to 8, characterized in that: The following steps are involved: Step 1: Collect app users' click events through log data, filter the click events to obtain actual click events, and build user profiles based on the actual click events. Step 2: Identify the device information and operating environment of the device logging into the APP. The device information includes model, operating system, and screen parameters. The operating environment includes network status, location, and time period. Step 3: Use a machine learning model to evaluate the value of in-app traffic based on user profiles, device information, and operating environment to derive traffic tiers. Step 4: Obtain the connected advertising resources and historical logs of the advertising resources. Calculate the ad fill rate and unit traffic revenue based on the historical logs. Dynamically evaluate the traffic tier, ad fill rate, and unit traffic revenue to determine the advertising resource delivery strategy, and display ads based on the ad resource delivery strategy. Step 5: Collect data on ad impressions, clicks, and conversions in real time, and calculate estimated revenue and actual settlement revenue.

Images