Monitoring method and device for aggregated advertisement platform, equipment and medium

Abstract

The invention discloses a monitoring method, device and equipment for an aggregation advertisement platform and a medium, and relates to the technical field of Internet, the monitoring method comprises the following steps: collecting historical comprehensive data of n advertisements on # imgabs0 # advertisement platforms, n and # imgabs1 # being integers greater than or equal to 1; constructing a false identification model for identifying false data and a user behavior prediction model for predicting a platform skipping probability based on the historical comprehensive data; acquiring real-time integrated data of the n advertisements on # imgabs2 advertisement platforms, inputting the real-time integrated data into a false identification model, outputting false data, removing the false data from the real-time integrated data to obtain real integrated data, and calculating based on the real integrated data to obtain putting evaluation data; the authenticity of the real-time integrated data monitored by the aggregation advertisement platform is identified through the false identification model, the false data is filtered to obtain the real integrated data, the putting evaluation data is acquired based on the real integrated data, and the authenticity of the data monitored by the aggregation advertisement platform is improved.

Description

Technical Field

[0001] The present invention relates to the field of Internet technologies, and particularly to a monitoring method, device, equipment and medium for an aggregated advertising platform. Background Art

[0002] With the rapid development of the digital advertising industry, when advertising on multiple advertising platforms, in order to optimize the advertising placement effect, it is usually necessary to rely on the analysis tools and reports provided by the advertising platforms to monitor the placement effect; these platforms evaluate the advertising effect by monitoring basic data such as the click volume, exposure volume, and conversion rate of the advertisements.

[0003] For example, the patent application with the publication number CN107341682A discloses an advertising information monitoring and evaluation method and its device, which obtains user behavior information from multiple platforms and aggregates the user behavior information through user identification data; although it obtains user behavior information from multiple platforms, it cannot monitor the relevance of user behavior information between multiple advertising platforms, that is, it cannot predict the behavior path of tracking the same user between multiple platforms; and during the placement process of the aggregated advertising platform, it cannot identify and filter interference data in a timely manner, resulting in inaccurate advertising effect reports.

[0004] Therefore, the present invention provides a monitoring method, device, equipment and medium for an aggregated advertising platform. Summary of the Invention

[0005] The purpose of the present invention is to provide a monitoring method, device, equipment and medium for an aggregated advertising platform, so as to predict the behavior path of tracking the same user between multiple platforms, filter abnormal behaviors, and improve the monitoring accuracy of the aggregated advertising platform.

[0006] The purpose of the present invention is achieved by the following technical solutions. A monitoring method for an aggregated advertising platform, the monitoring method includes:

[0007] Collect historical comprehensive data of n types of advertisements on advertising platforms, where both n and are integers greater than or equal to 1;

[0008] Based on the historical comprehensive data, construct a false identification model for identifying false data and a user behavior prediction model for predicting the probability of jumping to a platform;

[0009] Obtain the real-time comprehensive data of n types of advertisements on advertising platforms, input the real-time comprehensive data into the false identification model, output false data, remove the false data from the real-time comprehensive data to obtain real comprehensive data, and calculate the placement evaluation data based on the real comprehensive data;

[0010] Input real comprehensive data into the user behavior prediction model to output the probability of jumping to the platform;

[0011] Based on the delivery evaluation data and the probability of jumping to the platform Advertising platform categories.

[0012] Preferably, based on the delivery evaluation data and the probability of jumping to the platform The methods of classifying advertising platforms include:

[0013] Step 1: Pre- The advertising platforms are divided into k advertising display categories, where k is an integer greater than 1;

[0014] Step 2: The real comprehensive data of advertising platforms and the probability of jumping to other platforms are converted into feature vectors, randomly select the feature vectors of h advertising platforms as the initial centroid, ;

[0015] Step 3: Calculate the Euclidean distance from each advertising platform to each centroid, and assign each advertising platform to the advertising display category closest to it;

[0016] Step 4: After each allocation is completed, repeat steps 2 and 3 to recalculate the centroid of each ad display category;

[0017] Step 5: Repeat steps 3 and 4 until the centroid of each ad display category is less than the preset centroid threshold. Clustering of advertising platforms.

[0018] Preferably, the method for constructing the false recognition model includes:

[0019] Collect multiple sets of historical comprehensive data, and mark the authenticity of the historical comprehensive data when collecting the historical comprehensive data. If the authenticity of the historical comprehensive data is false, it is marked as 0, and if the authenticity of the historical comprehensive data is true, it is marked as 1;

[0020] Historical comprehensive data and the labels corresponding to the historical comprehensive data are used as a data set, and the data set is divided into a training set and a test set according to a preset ratio. The model is trained by finding the optimal model parameters by minimizing the loss function, and the predicted label data is output. The training is stopped after the accuracy of the predicted label data and the actual label data reaches the preset accuracy. The model obtained by training is used as a false recognition model, and the false recognition model is one of random forest, support vector machine and logistic regression.

[0021] Preferably, the specific method of removing false data from real-time comprehensive data to obtain real comprehensive data includes:

[0022] According to the output result of the false recognition model, filter out the data entries with the predicted label being 0.

[0023] Preferably, the training method of the user behavior prediction model includes:

[0024] Set a sliding step size, collect multiple groups of historical comprehensive data based on the sliding step size, construct a user action sequence, where the user action sequence includes continuously staying on the current platform or jumping to the next platform, and use the user action sequence as the second data set;

[0025] The second data set is divided into a training set and a test set according to a preset ratio. The best model parameters are found by minimizing the loss function to train the model, and the probability of predicting the jump platform is output. Training stops until the accuracy of the predicted jump platform probability and the actual jump platform probability reaches the preset accuracy, and the trained model is used as the user behavior prediction model. The user behavior prediction model is a hidden Markov model or a long short-term memory network.

[0026] Preferably, the historical comprehensive data and the real-time comprehensive data include placement data and the user behavior data corresponding to the placement data;

[0027] The placement data includes the content of the advertising platform, the number of displays, the type of advertisement, and the unit price;

[0028] The content of the advertising platform includes the displayed content and the content browsed by the user;

[0029] The type of advertisement includes the advertisement type ID, the number of advertisement displays, the number of advertisement clicks, the number of advertisement conversions, and the advertisement type characteristics.

[0030] Preferably, the user behavior data includes the number of user clicks, the user's page stay duration, the browsed content, and the next-time sequence action;

[0031] The next-time sequence action is to continuously stay on the current platform or jump to the next platform.

[0032] Preferably, the placement evaluation data is calculated based on the placement data and the user behavior data corresponding to the placement data;

[0033] The placement evaluation data includes the click-through rate, the conversion rate, the cost per click, and the cost per conversion.

[0034] A monitoring device for aggregating advertising platforms, the device includes:

[0035] A false recognition module for training a false recognition model;

[0036] A user behavior prediction module for training a user behavior prediction model;

[0037] A collection module for collecting the real-time comprehensive data of the aggregation platform;

[0038] A data processing module, configured to input real-time comprehensive data into a false recognition model and a user behavior prediction model respectively to output false data and the probability of jumping to a platform, and remove the false data from the real-time comprehensive data to obtain real comprehensive data;

[0039] A clustering module, configured to calculate placement evaluation data based on the real comprehensive data, and classify advertising platforms based on the placement evaluation data and the probability of jumping to a platform.

[0040] A monitoring device, comprising: a memory for storing non-transitory computer-readable instructions; and a processor for running the computer-readable instructions, so that when the computer-readable instructions are executed by the processor, the above-mentioned monitoring method for aggregating advertising platforms is implemented.

[0041] A computer storage medium, comprising computer instructions, when the computer instructions are run on a device, the device is caused to execute the above-mentioned monitoring method for aggregating advertising platforms.

[0042] It can be seen from the above technical solutions that the present application has the following beneficial effects:

[0043] 1: The authenticity of the real-time comprehensive data monitored from the aggregated advertising platform is identified through the false recognition model, and the false data is filtered to obtain real comprehensive data. The placement evaluation data is obtained based on the real comprehensive data, improving the authenticity of the data for monitoring the aggregated advertising platform.

[0044] 2: By analyzing user behavior, the correlation between multiple advertising platforms is predicted, and multiple advertising platforms are clustered through the placement evaluation data and the probability of jumping to a platform, which is conducive to the classified display of the aggregation platform.

[0045] The above description is only an overview of the technical solution of the present invention. In order to understand the technical means of the present invention more clearly, it can be implemented according to the content of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and understandable, the following specific preferred embodiments are given and described in detail in conjunction with the accompanying drawings as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.

[0047] Figure 1 It is a schematic flowchart of a monitoring method for aggregating advertising platforms according to an embodiment of the present invention;

[0048] Figure 2 is a schematic flow chart for classifying advertising platforms in an embodiment of the present invention;

[0049] Figure 3 is a schematic structural diagram of a monitoring device for aggregating advertising platforms in an embodiment of the present invention. Detailed implementation manners

[0050] To further elaborate on the technical means and effects adopted by the present invention to achieve the intended invention purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation manners, structures, features, and effects of the third-party system monitoring system, method, device, equipment, and storage medium proposed according to the present invention.

[0051] It should be noted that in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Additionally, the term "comprising", "including", or any other variation thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or device including a series of elements not only includes those elements but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article, or device. Without further limitation, elements defined by the statement "including..." do not exclude the existence of additional identical elements in the process, method, article, or device including the said elements.

[0052] Embodiment 1

[0053] Refer to Figure 1 , for a monitoring method for aggregating advertising platforms, the monitoring method includes:

[0054] Collect historical comprehensive data of n types of advertisements on advertising platforms, where both n and are integers greater than or equal to 1; the historical comprehensive data includes placement data and user behavior data corresponding to the placement data, and the placement data and the user behavior data corresponding to the placement data are obtained from the advertising platforms where the advertisements are placed and collected through the APIs, advertising SDKs, user behavior tracking tools, etc. of the advertising platforms;

[0055] Specifically, the placement data includes the content of the advertising platform, the number of impressions, the type of advertisement, and the unit price; in some embodiments, the placement data further includes the advertisement display position, that is, the specific display position of the advertisement on the page or in the application, and the advertisement format, which includes banner advertisements, video advertisements, native advertisements, etc.; the content of the advertising platform includes the display content and the content browsed by the user. In some embodiments, the display content includes the comprehensive content of the platform, such as news, entertainment, humanities, etc., and the content browsed by the user, such as the content that the user often browses on the platform; the advertisement type includes the advertisement type ID, the number of advertisement impressions, the number of advertisement clicks, the number of advertisement conversions, and the advertisement type characteristics.

[0056] The user behavior data includes the number of user clicks, that is, which advertisements the user clicks, the time and frequency of clicks, etc., the user's page stay duration, the browsed content, and the next sequential action. In some embodiments, the user behavior data further includes the user conversion data, that is, whether the user performs operations such as purchase, registration, download, etc.; the user attribute data, such as age, gender, geographical location, device type, etc.; the next sequential action is to continue staying on the current platform or jump to the next platform, that is, whether the user jumps to other platforms after browsing the advertisement.

[0057] It is worth mentioning that in order to ensure the reliability of the comparison of the historical comprehensive data of n advertisements on advertising platforms, the collected historical comprehensive data needs to be cleaned and standardized. The advertising data of each platform (different formats, time zones, units, etc.) is uniformly cleaned and standardized. For example, indicators such as the number of advertisement impressions, the number of advertisement clicks, and the number of advertisement conversions are converted into a unified format and standard to ensure cross-platform comparison. The specific conversion method can use methods such as weighted average and normalization to uniformly process the advertising data of each platform, so as to generate a comprehensive effect score, and the timeliness of the data needs to be analyzed to determine the update time of the advertising data of different platforms, whether there is data delay or missing data, to ensure the accuracy of the monitoring results.

[0058] Build a false recognition model for identifying false data and a user behavior prediction model for predicting the probability of jumping to the next platform based on the historical comprehensive data;

[0059] Specifically, the method for building the false recognition model includes:

[0060] Collect multiple sets of historical comprehensive data, and mark the authenticity of the historical comprehensive data when collecting it. The authenticity is marked by those skilled in the art. The authenticity determination methods include that the user frequently clicks on the same advertisement but stays on the advertisement page for too short a time; clicks on multiple advertisements within an extremely short period, or clicks on multiple different advertisements within a very short time; the user's device information, IP address, or browser information changes frequently within a short time; the user's geographical location shows an abnormal distribution when clicking on advertisements, etc., which are not specifically limited herein;

[0061] If the authenticity of the historical comprehensive data is false, it is marked as 0. If the authenticity of the historical comprehensive data is true, it is marked as 1. Take the historical comprehensive data and the corresponding mark of the historical comprehensive data as a data set. The data set is divided into a training set and a test set according to a preset ratio. For example, 70% of the data set is used as the training set, and 30% of the data set is used as the test set for model training, and prediction label data is output. Stop training until the accuracy of the prediction label data and the actual label data reaches the preset accuracy, and use the trained model as a false recognition model. The false recognition model is one of a random forest, a support vector machine, and a logistic regression;

[0062] Taking the random forest as an example, the training process includes: randomly select a subset from the training data set, and use this subset to train a decision tree. Repeat the above process N times to train N trees. The output of each tree is a label (0 or 1), and finally the output of the model is determined by majority voting;

[0063] The condition for stopping training is that the accuracy of the model on the test set reaches the preset target accuracy, such as 95%, or after a predetermined number of training rounds, such as 1000 iterations. If the performance of the model stagnates in a certain iteration, stop training.

[0064] The training method of the user behavior prediction model includes:

[0065] Set a sliding step size, that is, each time select the user behavior data within a time window as a training sample. The sliding step size determines the length of each data segment collected in the time series. Based on the sliding step size, collect multiple sets of historical comprehensive data and construct a user action sequence. The user action sequence includes staying on the current platform continuously or jumping to the next platform. For example, the user's behavior data at multiple time steps where, is the time step number. For each time step, the user's behavior can be one of the following:

[0066] Stay on the current platform: indicating that the user stays on the current platform;

[0067] Jump to the next platform: indicating that the user jumps from the current platform to another platform;

[0068] The user action sequence S can be expressed as: , where represents the user behavior at time , is the time step;

[0069] Taking the user action sequence as the second data set, the second data set is divided into a training set and a test set according to a preset ratio, for example, 70% for the training set and 30% for the test set. The division method can be random division or time series division. By minimizing the loss function, the best model parameters are found to train the model, and the probability of predicting the jump platform is output. Training stops until the accuracy of the predicted jump platform probability and the actual jump platform probability reaches the preset accuracy, and the model obtained by training is used as the user behavior prediction model;

[0070] The user behavior prediction model is a hidden Markov model or a long short-term memory network. For example, based on the hidden Markov model training method: The user's behavior is modeled as a transition between multiple hidden states, such as stay and jump; Define the state space, that is, the possible states of the user behavior. For example, state represents "staying on the platform", and state represents "jumping to the platform"; Define the observation space, that is, the observable behavior of the user, such as the platform type at a specific time point; Obtain the key parameters of the model, including the transition probability matrix A: representing the probability of transitioning from one state to another state:

[0071] , where represents the probability of transitioning from state to state , the emission probability matrix , representing the probability of observing a specific behavior in each state; The initial state probability , representing the probabilities of each state when the system is initialized;

[0072] During the training process, the Baum-Welch algorithm is used to estimate the transition probability matrix and the emission probability matrix, so that the model can minimize the log-likelihood function:

[0073] ;

[0074] where represents the set of model parameters, is the observed data, represents the set of parameters at time t observing the conditional probability of the observed data , represents the log-likelihood function of the model.

[0075] LSTM Training Method: Input the user's behavior sequence, which is usually composed of a series of tags (0 or 1) for platform stays and jumps; inside the LSTM, memory cells and hidden states are used to store the dependencies between time steps; output the predicted value, representing the probability that the user jumps to the next platform; train the LSTM model by minimizing the loss function, such as the cross-entropy loss function:

[0076] ;

[0077] where is the true label, is the probability of the jump platform predicted by the model. Update the model's parameters through the backpropagation algorithm. The training stops when the accuracy of the model on the test set reaches the preset target accuracy, such as 95%, or after a predetermined number of training epochs, such as 1000 iterations. If the performance of the model stagnates in a certain iteration, stop the training.

[0078] Obtain the real-time comprehensive data of n kinds of advertisements on advertising platforms, input the real-time comprehensive data into the false identification model, output the false data, and remove the false data from the real-time comprehensive data to obtain the real comprehensive data. Specifically, according to the output result of the false identification model, filter out the data entries with the predicted label of 0, and calculate the placement evaluation data based on the real comprehensive data;

[0079] It is worth mentioning that the types included in the real-time comprehensive data are the same as those of the historical comprehensive data, both of which are placement data and the user behavior data corresponding to the placement data.

[0080] Specifically, the placement evaluation data is calculated based on the placement data and the user behavior data corresponding to the placement data; the placement evaluation data includes click-through rate, conversion rate, click cost, conversion cost, and return on investment, which are respectively marked as , , , , . The conversion cost is the cost paid by the advertiser for each conversion.

[0081] Optionally, The calculation expression of is:

[0082] ;

[0083] where is the click-through rate obtained by placing the th type of advertisement on the th platform, is the number of clicks obtained by placing the th type of advertisement on the th platform, is the number of impressions obtained by placing the th type of advertisement on the class of platforms. The number of impressions is the total exposure volume on the x platform, obtained from the advertising platform API, where x represents different advertising platforms, and the exposure volume is an integer. Each time an advertisement is displayed, it counts as 1 exposure. The click-through rate measures the ratio of the number of clicks on an advertisement to the number of ad impressions, and is used to evaluate the attractiveness of the advertisement.

[0084] Optionally, the calculation expression of

[0085] ;

[0086] where is the conversion rate obtained by placing the th type of advertisement on the class of platforms. is the number of conversions obtained by placing the th type of advertisement on the class of platforms. The conversion rate measures the ratio of the number of conversions of an advertisement to the number of clicks, and is used to evaluate the actual effect of the advertisement.

[0087] Optionally, the calculation expression of

[0088] ;

[0089] where is the click cost of the th type of advertisement placed on the class of platforms. The click cost refers to the fee paid by the advertiser for each click on the advertisement. is the cost of the th type of advertisement placed on the class of platforms.

[0090] Optionally, the calculation expression of

[0091] ;

[0092] where is the conversion cost of the th type of advertisement placed on the class of platforms. The conversion cost refers to the fee paid by the advertiser for each conversion, and is used to measure the effectiveness of the advertisement placement.

[0093] Optionally, the calculation expression of

[0094] ;

[0095] Among them, is the return rate of the th type of platform for the th kind of advertisement. is the net return of the th type of platform for the th kind of advertisement. The return on investment is a key indicator for evaluating the effectiveness of advertisement placement, representing the ratio of the return of advertisement placement to the cost.

[0096] Optionally, in some embodiments, based on , , , and calculate the comprehensive score of the th type of platform for the th kind of advertisement, marked as , The calculation expression is:

[0097] ;

[0098] Among them, , , , and are preset weighting coefficients, and , , , and sum up to one. The weighting coefficients are adjusted according to the priorities of advertisers. These weights reflect the attention degrees of advertisers to different indicators. For example, advertisers may pay more attention to conversion rates and return rates, so higher weights will be given. The comprehensive score is used to evaluate the effectiveness of advertisement placement, can comprehensively reflect the advertisement effect, and helps advertisers make decisions.

[0099] Input the real comprehensive data into the user behavior prediction model to output the probability of jumping to the platform; The purpose is that the jumping behavior on the advertisement platform is usually closely related to the interests, needs of users and the relevance of advertisements. By predicting the probability of jumping to the platform, advertisers can identify potential user loss risks and timely adjust advertisement content or placement methods to prevent users from jumping to competing platforms; Optimize advertisement placement strategies, budget allocation, advertisement content personalization and platform selection through accurate prediction of user behavior. At the same time, the prediction results can also help advertisers identify potential risks, improve advertisement effects and user experiences, so as to achieve a higher return on advertisement investment.

[0100] Classify advertisement platforms based on the placement evaluation data and the probability of jumping to the platform.

[0101] SeeFigure 2 , specifically, the method for classifying advertising platforms based on delivery evaluation data and jump platform probability includes:

[0102] Step 1: Pre-divide advertising platforms into k advertising display categories, where k is an integer greater than 1, and the number of k represents dividing advertising platforms into several clusters. k is the predefined number of advertising display categories. For example, it is divided into three advertising display categories: high quality, standard quality, and low quality. The optimal value of k is determined by the experience of those skilled in the art or actual needs, and no specific limitation is made here;

[0103] Step 2: Convert the true comprehensive data and jump platform probability of advertising platforms into feature vectors, , where, is the feature vector of the i-th advertising platform, is the value of the -th advertising platform on the -th feature dimension. The feature dimensions include the number of displays , click-through rate , conversion rate , jump probability , etc.; randomly select the feature vectors of h advertising platforms as the initial centroids , , where, is the h-th initial centroid, is the centroid of the -th advertising display category;

[0104] Step 3: Calculate the Euclidean distance from each advertising platform to each centroid, and assign each advertising platform to the advertising display category with the closest distance. The calculation expression of the Euclidean distance is:

[0105] ;

[0106] where, is the Euclidean distance from the -th advertising platform to the centroid of the -th advertising display category, is the value of the -th advertising platform on the -th feature dimension, such as the number of displays, click-through rate, or conversion rate, etc., is the value of the centroid of the -th advertising display category on the -th feature dimension, is the total dimension number of the feature vector.

[0107] Step 4: After each allocation, repeat Step 2 and Step 3 to recalculate the centroid of each ad display category. The new centroid is the mean of the feature vectors of all ad platforms in the ad display category. For each ad display category The new centroid , The calculation expression is as follows:

[0108] ;

[0109] where is the set of ad platforms allocated in the th ad display category, is the number of ad platforms in the set .

[0110] Step 5: Repeat Step 3 and Step 4 until the centroid of each ad display category is less than the preset centroid threshold, which is determined by those skilled in the art according to actual needs and is not specifically limited here. Complete the clustering division of the ad platforms.

[0111] The purpose of this embodiment is to conduct aggregated monitoring on aggregated ads, that is, after centrally analyzing the effects of different ads placed on different ad platforms, classify and display the obtained feedback effects, enabling advertisers to intuitively see the placement effects of different ads on all ad platforms, rather than just analyzing single data.

[0112] Embodiment 2

[0113] Referring to Figure 3 shown, a monitoring device for an aggregated ad platform, the device includes a false identification module, a user behavior prediction module, a collection module, a data processing module, and a clustering module, where each module is connected through a wired and / or wireless network;

[0114] The false identification module is used to train a false identification model. The data used for training the false identification model in this embodiment is obtained from the historical monitoring data or monitoring logs of each ad platform, and will not be elaborated here;

[0115] The user behavior prediction module is used to train a user behavior prediction model;

[0116] The collection module is used to collect the real-time comprehensive data of the aggregation platform. The collection module is docked with the API interfaces of multiple ad platforms to collect data in real time and perform preliminary preprocessing, such as format conversion, missing value filling, etc.;

[0117] A data processing module, configured to input real-time comprehensive data into a false identification model and a user behavior prediction model respectively, output false data and the probability of jumping to a platform, and remove the false data from the real-time comprehensive data to obtain real comprehensive data;

[0118] A clustering module, configured to calculate and obtain placement evaluation data based on the real comprehensive data, and classify advertising platforms based on the placement evaluation data and the probability of jumping to a platform, so as to optimize the advertising effect.

[0119] A monitoring device, comprising: a memory for storing non-transitory computer-readable instructions; and a processor for running the computer-readable instructions, so that when the computer-readable instructions are executed by the processor, the above-mentioned monitoring method for aggregating advertising platforms is implemented.

[0120] A computer storage medium, comprising computer instructions, when the computer instructions are run on a device, the device is caused to execute the above-mentioned monitoring method for aggregating advertising platforms.

[0121] An embodiment of the present invention further provides a computer program product, when the computer program product runs on a computer, the computer is caused to execute the above-related steps to implement the above-mentioned monitoring method for aggregating advertising platforms in the above embodiment.

[0122] Wherein, the device, computer storage medium, computer program product or chip provided by the present invention are all used to execute the corresponding method provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding method provided above, and will not be elaborated here.

[0123] The above is only a preferred embodiment of the present invention, and does not impose any form of limitation on the present invention. Although the present invention has been disclosed above with a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to the above-disclosed technical content to be equivalent variations of equivalent embodiments within the scope of the technical solution of the present invention. However, as long as it does not depart from the content of the technical solution of the present invention, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A monitoring method for an aggregated advertising platform, characterized in that: Monitoring methods include: Collect n kinds of advertisements in Historical comprehensive data of advertising platforms, n and All are integers greater than or equal to 1; Build a false identification model to identify false data and a user behavior prediction model to predict the probability of jumping to another platform based on historical comprehensive data; Get n types of ads in The real-time comprehensive data of each advertising platform is input into a false identification model, false data is output, and false data is removed from the real-time comprehensive data to obtain real comprehensive data, and the delivery evaluation data is calculated based on the real comprehensive data; Input real comprehensive data into the user behavior prediction model to output the probability of jumping to the platform; Based on the delivery evaluation data and the probability of jumping to the platform Advertising platform categories.

2. The monitoring method for an aggregated advertising platform according to claim 1, characterized in that: Based on the delivery evaluation data and the probability of jumping to the platform The methods of classifying advertising platforms include: Step 1: Pre- The advertising platforms are divided into k advertising display categories, where k is an integer greater than 1; Step 2: The real comprehensive data of advertising platforms and the probability of jumping to other platforms are converted into feature vectors, randomly select the feature vectors of h advertising platforms as the initial centroid, ; Step 3: Calculate the Euclidean distance from each advertising platform to each centroid, and assign each advertising platform to the advertising display category closest to it; Step 4: After each allocation is completed, repeat steps 2 and 3 to recalculate the centroid of each ad display category; Step 5: Repeat steps 3 and 4 until the centroid of each ad display category is less than the preset centroid threshold. Clustering of advertising platforms.

3. The monitoring method for an aggregated advertising platform according to claim 1, characterized in that: The method for constructing the false recognition model includes: Collect multiple sets of historical comprehensive data, and mark the authenticity of the historical comprehensive data when collecting the historical comprehensive data. If the authenticity of the historical comprehensive data is false, it is marked as 0, and if the authenticity of the historical comprehensive data is true, it is marked as 1; Historical comprehensive data and the labels corresponding to the historical comprehensive data are used as a data set, and the data set is divided into a training set and a test set according to a preset ratio. The model is trained by finding the optimal model parameters by minimizing the loss function, and the predicted label data is output. The training is stopped after the accuracy of the predicted label data and the actual label data reaches the preset accuracy. The model obtained by training is used as a false recognition model, and the false recognition model is one of random forest, support vector machine and logistic regression.

4. The monitoring method for an aggregated advertising platform according to claim 3, characterized in that: The specific methods for removing false data from real-time comprehensive data to obtain real comprehensive data include: According to the output results of the false recognition model, the data entries with a predicted label of 0 are filtered out.

5. The monitoring method for an aggregated advertising platform according to claim 1, characterized in that: The training method of the user behavior prediction model includes: Set the sliding step length, collect multiple sets of historical comprehensive data based on the sliding step length, and construct a user action sequence. The user action sequence includes staying on the current platform or jumping to the next platform. The user action sequence is used as the second data set; The second data set is divided into a training set and a test set according to a preset ratio. The model is trained by finding the optimal model parameters by minimizing the loss function, and the predicted platform jump probability is output. The training is stopped until the accuracy of the predicted platform jump probability and the actual platform jump probability reaches the preset accuracy. The model obtained by training is used as the user behavior prediction model, which is a hidden Markov model or a long short-term memory network.

6. The monitoring method for an aggregated advertising platform according to claim 1, characterized in that: The historical comprehensive data and real-time comprehensive data include delivery data and user behavior data corresponding to the delivery data; The delivery data includes the advertising platform content, number of impressions, ad type, and unit price; Advertising platform content includes displayed content and user-browsed content; Ad type includes ad type ID, ad impressions, ad clicks, ad conversions, and ad type features.

7. The monitoring method for an aggregated advertising platform according to claim 6, characterized in that: The user behavior data includes the number of user clicks, the length of time the user stays on the page, the content browsed, and the next sequential action; The next sequential action is to continue staying on the current platform or jump to the next platform.

8. The monitoring method for an aggregated advertising platform according to claim 6, characterized in that: The delivery evaluation data is calculated based on the delivery data and user behavior data corresponding to the delivery data; The delivery evaluation data includes click-through rate, conversion rate, click cost and conversion cost.

9. A monitoring device for an aggregated advertising platform, which is used to implement the monitoring method for an aggregated advertising platform according to any one of claims 1 to 8, characterized in that: The device includes: False identification module, used to train false identification models; User behavior prediction module, used to train user behavior prediction models; The collection module is used to collect real-time comprehensive data from the aggregation platform; A data processing module is used to input the real-time comprehensive data into the false identification model and the user behavior prediction model respectively, output the false data and the probability of jumping to the platform respectively, and remove the false data from the real-time comprehensive data to obtain the real comprehensive data; The clustering module obtains delivery evaluation data based on real comprehensive data calculations, and classifies advertising platforms based on the delivery evaluation data and the probability of jumping to the platform.

10. A monitoring device, characterized in that: include: A memory for storing non-temporary computer-readable instructions; and a processor for running the computer-readable instructions, so that when the computer-readable instructions are executed by the processor, the monitoring method for an aggregate advertising platform described in any one of claims 1 to 8 is implemented.

11. A computer storage medium, characterized in that: The method comprises computer instructions, which, when executed on a device, cause the device to execute the monitoring method for an aggregate advertising platform as claimed in any one of claims 1 to 8.

Images