Mobile programmed advertisement traffic optimization method and system

Abstract

The invention provides a mobile programmed advertisement traffic optimization method and system, and relates to the technical field of traffic optimization, and the method comprises the steps: judging whether there is an advertisement message pushed by an application program at the current moment; if yes, acquiring use state information, current position information and current moment information of the mobile equipment; the method comprises the steps of obtaining reading behavior information of a user for a plurality of historical advertisement messages in a first preset time period before a current moment, and further determining a push evaluation index and a push suppression coefficient to judge whether to push the advertisement messages or not. And if the advertisement messages are not pushed, temporarily storing the advertisement messages, and determining a traffic optimization scheme for the application program according to the number of the advertisement messages temporarily stored by the application program and the pushing suppression coefficient. According to the method and the device, the necessity of pushing the current advertisement message can be automatically judged, the advertisement message is temporarily stored when the advertisement message does not need to be pushed, and the advertisement message is not pushed immediately, so that the disturbance to a user is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic optimization, and in particular to a method and system for optimizing mobile programmatic advertising traffic. Background Art

[0002] When pushing advertising messages, applications usually push them directly and issue prompts, such as sound prompts or vibration prompts, but do not consider whether it is convenient for the user when pushing. For example, at night, when the user is resting, the application may directly push advertising messages and issue prompts, which may disturb the user's rest, or when the user is in a meeting in a conference room, the application may directly push advertising messages and issue prompts, which may disturb the user's meeting. Related technologies usually rely on manual adjustment of prompt modes, such as switching to silent mode during rest or meeting, and it is difficult to determine whether the scenario in which the user receives advertising messages is convenient. If the user forgets to manually adjust the prompt mode, the advertising message may disturb the user.

[0003] The information disclosed in the background technology section of this application is only intended to deepen the understanding of the general background technology of this application, and should not be regarded as an admission or any form of suggestion that the information constitutes the prior art already known to those skilled in the art. Summary of the Invention

[0004] The present invention provides a method and system for optimizing mobile programmatic advertising traffic, which can solve the technical problems that related technologies have difficulty in determining whether the scenario in which users receive advertising messages is convenient and that advertising messages may disturb users.

[0005] According to a first aspect of the present invention, a method for optimizing mobile programmatic advertising traffic is provided, comprising:

[0006] Determine whether there is an advertisement message pushed by the application at the current moment;

[0007] If the advertisement message exists, obtaining usage status information, current location information, and current time information of the mobile device;

[0008] Obtaining user reading behavior information for multiple historical advertising messages within a first preset time period before the current moment;

[0009] Determining a push evaluation index for the advertisement message based on the reading behavior information;

[0010] determining a push suppression coefficient according to the usage status information, the current location information, and the current time;

[0011] determining whether to push the advertisement message according to the push evaluation index and the push suppression coefficient;

[0012] If it is determined that the advertisement message is not to be pushed, temporarily storing the advertisement message of the application;

[0013] A traffic optimization scheme for the application is determined according to the number of advertisement messages pushed by the application temporarily stored at the current moment and within a second preset time period before the current moment, and a push suppression coefficient corresponding to the temporarily stored advertisement messages.

[0014] According to the present invention, determining the push evaluation index of the advertising message based on the reading behavior information includes:

[0015] Obtaining historical pop-up texts of historical advertising messages and the pop-up texts of the advertising messages;

[0016] Acquiring historical content information of historical advertisement messages and content information of the advertisement messages;

[0017] Obtain historical type information of historical advertisement messages and type information of advertisement messages;

[0018] Obtaining a first input vector of a node corresponding to a historical advertising message based on historical pop-up text, historical content information, and historical type information;

[0019] Obtaining a second input vector of a node corresponding to the advertisement information according to the pop-up window text, the content information, and the type information;

[0020] Processing the first input vector and the second input vector using a trained graph neural network model to obtain a first output vector for a node corresponding to a historical advertisement message and a second output vector for a node corresponding to the advertisement information;

[0021] filtering target historical advertisement messages from the historical advertisement messages according to the first output vector and the second output vector;

[0022] Determine the push evaluation indicators of the advertising message based on the reading behavior information of the target historical advertising message.

[0023] According to the present invention, filtering target historical advertisement messages from historical advertisement messages according to the first output vector and the second output vector includes:

[0024] Determining the similarity between each first output vector and the second output vector;

[0025] determining, as a target output vector, a first output vector having a similarity greater than a first similarity threshold and less than or equal to a second similarity threshold, wherein the second similarity threshold is greater than the first similarity threshold;

[0026] determining the historical advertising message corresponding to the target output vector as the target historical advertising message;

[0027] If there is a first output vector whose similarity is greater than the second similarity threshold, the push evaluation index is set to 0.

[0028] According to the present invention, the reading behavior information includes whether the user clicks and reads the historical advertising message, and the behavior data when reading the historical advertising message;

[0029] Based on the target's historical advertising message reading behavior information, determine the push evaluation indicators of the advertising message, including:

[0030] According to the formula , Determine the push evaluation indicators for advertising messages ,in, The behavior weight for users who did not click on the target historical advertising message and cleared the target historical advertising message, The identification result of whether the user clears the target i-th historical advertising message without reading the target i-th historical advertising message. The historical content information behavior weight of the user reading the target historical advertising message, The reading progress of the historical content information of the i-th target historical advertising message for the user, The weight of the user's interactive operation in the target historical advertising message, is the recognition result of whether the user performs an interactive operation in the i-th target historical advertising message, is the first output vector of the node corresponding to the i-th target historical advertising message, is the second output vector of the node corresponding to the advertising information, for and similarity, n is the number of target historical advertising messages, i≤n, and i and n are positive integers.

[0031] According to the present invention, the training method of the graph neural network model includes:

[0032] Selecting a plurality of training advertisement information of a plurality of training type information from a training advertisement information library;

[0033] Obtaining a plurality of training input vectors of training advertisement information;

[0034] The training input vector is processed through the graph neural network model to obtain the training output vector of each training advertisement information;

[0035] The training output vector is input into the fully connected layer and the activation layer to obtain the predicted reading behavior information of the training advertisement information;

[0036] Determine the loss function of the graph neural network model based on the predicted reading behavior information and the labeled information of the training advertisement information;

[0037] According to the loss function, the graph neural network model is trained to obtain a trained graph neural network model.

[0038] According to the present invention, the loss function of the graph neural network model is determined based on the predicted reading behavior information and the labeled information of the training advertisement information, including:

[0039] According to the formula , Determine the loss function LOSS of the graph neural network model, where is the probability that the j-th training advertisement information is not read and is cleared according to the labeled information, is the probability that the j-th training advertisement information is not read and is cleared according to the predicted reading behavior information, is the probability that the j-th training advertisement information is not processed within the preset time after being pushed, determined based on the labeled information. is the probability that the j-th training advertisement information is not processed within the preset time after being pushed, determined based on the predicted reading behavior information. is the probability that the jth training advertisement information is clicked and read but no interaction is performed based on the labeled information, is the probability that the jth training advertisement information is clicked and read but no interaction is performed, determined based on the predicted reading behavior information. is the probability of the user interacting with the jth training advertisement information determined based on the labeled information, is the probability of the user interacting with the jth training advertisement information determined based on the predicted reading behavior information, m is the number of training advertisement information, j≤m, and both j and m are positive integers.

[0040] According to the present invention, determining a push suppression coefficient based on the usage status information, the current location information, and the current time includes:

[0041] determining, based on the usage status information, a first identification result of whether the mobile device is in a state that is inconvenient for directly reading the advertisement message;

[0042] determining, based on the current location information, a second identification result of whether the mobile device is in a location where it is inconvenient to directly read the advertisement message;

[0043] a third recognition result of determining whether the current moment is within a preset time period that is inconvenient for reading the advertisement message;

[0044] A push suppression coefficient is determined according to the first recognition result, the second recognition result, and the third recognition result.

[0045] According to the present invention, determining a push suppression coefficient based on the first recognition result, the second recognition result, and the third recognition result includes:

[0046] According to the formula , Determine the push suppression coefficient S, where is the kth recognition result, when k=1, is the first recognition result, when k=2, is the second recognition result, k=3, is the third recognition result, For The corresponding suppression weight, k is a positive integer, and k≤3.

[0047] According to the present invention, a traffic optimization scheme for the application is determined based on the number of advertisement messages pushed by the application temporarily stored at the current moment and within a second preset time period before the current moment, and a push suppression coefficient corresponding to the temporarily stored advertisement messages, including:

[0048] Obtaining an average value of historical push suppression coefficients each time the application generates an advertisement message within a second preset time period;

[0049] If the relative gap between the push suppression coefficient and the average value of the historical push suppression coefficient is less than or equal to the preset ratio threshold, and the number of temporarily stored advertising messages pushed by the application is greater than or equal to the preset number threshold, the advertising push traffic of the application is suspended.

[0050] According to a second aspect of the present invention, a mobile programmatic advertising traffic optimization system is provided, comprising:

[0051] A judgment module, used to judge whether there is an advertisement message pushed by the application at the current moment;

[0052] an acquisition module, configured to acquire usage status information, current location information, and current time information of the mobile device if the advertisement message exists;

[0053] A reading behavior information module is used to obtain user reading behavior information for multiple historical advertising messages within a first preset time period before the current moment;

[0054] A push evaluation index module, configured to determine a push evaluation index for the advertisement message based on the reading behavior information;

[0055] a push suppression coefficient module, configured to determine a push suppression coefficient based on the usage status information, the current location information, and the current time;

[0056] a determination module, configured to determine whether to push the advertisement message based on the push evaluation index and the push suppression coefficient;

[0057] A temporary storage module, configured to temporarily store the advertisement message of the application program when it is determined that the advertisement message is not to be pushed;

[0058] The optimization module is used to determine a traffic optimization plan for the application based on the number of advertisement messages pushed by the application temporarily stored at the current moment and within a second preset time period before the current moment, and a push suppression coefficient corresponding to the temporarily stored advertisement messages.

[0059] By adopting the above technical solution, the present invention can achieve the following technical effects:

[0060] According to the present invention, a push evaluation index for a current advertisement message can be determined based on a user's reading behavior information for historical advertisement messages. This is combined with the user's current location information and the current time to determine a push suppression coefficient. This automatically determines the necessity of pushing the current advertisement message. If push is not necessary, the advertisement message is temporarily stored rather than pushed immediately. This reduces the likelihood of disturbing the user when it is inconvenient for them to receive advertisement messages, eliminates the need for manual intervention, and improves user convenience. When determining the push evaluation index, target advertisement messages that are similar but different from the current advertisement message can be screened from multiple historical advertisement messages. This not only provides a data basis for determining the user's interest in the advertisement message, but also eliminates interference from duplicate advertisement messages. If duplicate historical advertisement messages exist, this prevents repeated interruptions from the current advertisement message, improving user convenience. Furthermore, a score for each target advertisement message can be calculated based on the user's various reading behavior information for the target advertisement message, thereby determining the user's interest in each target advertisement message. This allows for a comprehensive determination of the user's interest in the current advertisement message and the necessity of pushing the advertisement message, improving the accuracy of the push evaluation index. When training a graph neural network model, a loss function can be obtained by predicting reading behavior information and comparing it with the user's actual processing of the training advertising information. This error can be reduced during the training process to improve the accuracy of the predicted reading behavior information, thereby improving the accuracy of the training output vector extracted by the graph neural network model. In the loss function, the scores of the user's various reading behavior information are made consistent with the weights of the loss function, thereby improving the targeted training, thereby improving training efficiency and the accuracy of the graph neural network model. When determining the push suppression coefficient, a comprehensive analysis of the mobile device's status, location, and current time can be performed to determine whether the push advertising message is likely to disturb the user. This can then be used to obtain a push suppression coefficient. In the event that the advertising message is likely to disturb the user, a push suppression coefficient less than 1 can be generated to reduce the possibility of disturbing the user.

[0061] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and not limiting of the present invention. Other features and aspects of the present invention will become more apparent from the following detailed description of exemplary embodiments with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can derive other embodiments based on these drawings without inventive efforts.

[0063] Figure 1 A schematic diagram exemplarily illustrates a flow chart of a method for optimizing mobile programmatic advertising traffic according to an embodiment of the present invention;

[0064] Figure 2 A block diagram of a mobile programmatic advertising traffic optimization system according to an embodiment of the present invention is exemplarily shown. DETAILED DESCRIPTION

[0065] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0066] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0067] Figure 1 A flow chart illustrating a method for optimizing mobile programmatic advertising traffic according to an embodiment of the present invention is provided. The method includes:

[0068] Step S1, determining whether there is an advertisement message pushed by the application at the current moment;

[0069] Step S2: if the advertisement message exists, obtaining the usage status information, current location information and current time information of the mobile device;

[0070] Step S3, obtaining user reading behavior information for multiple historical advertising messages within a first preset time period before the current moment;

[0071] Step S4, determining a push evaluation index for the advertisement message based on the reading behavior information;

[0072] Step S5, determining a push suppression coefficient based on the usage status information, the current location information, and the current time;

[0073] Step S6, determining whether to push the advertisement message based on the push evaluation index and the push suppression coefficient;

[0074] Step S7: If it is determined that the advertisement message is not to be pushed, temporarily storing the advertisement message of the application;

[0075] Step S8: determining a traffic optimization solution for the application program according to the number of advertisement messages pushed by the application program temporarily stored at the current moment and within a second preset time period before the current moment, and the push suppression coefficient corresponding to the temporarily stored advertisement messages.

[0076] According to the mobile programmatic advertising traffic optimization method of an embodiment of the present invention, the push evaluation index of the current advertising message can be determined through the user's reading behavior information of historical advertising messages, and the push suppression coefficient can be determined in combination with the current location information and the current time, so as to automatically judge the necessity of pushing the current advertising message. When the advertising message does not need to be pushed, the advertising message is temporarily stored instead of being pushed immediately, which reduces the probability of disturbing the user when it is inconvenient for the user to receive the advertising message, and does not require manual intervention, thereby improving the convenience of use.

[0077] According to one embodiment of the present invention, in step S1, it can be determined whether each application pushes an advertising message. If no application pushes an advertising message, the waiting process continues. If an application pushes an advertising message, a prompt for the advertising message is temporarily not issued. For example, a sound prompt or a vibration prompt is temporarily not performed. Instead, the necessity of pushing the advertising message and whether it is convenient for the user to receive the advertising message are first determined. If the necessity of pushing the advertising message is high, or it is convenient for the user to receive the advertising message, a prompt is issued and the advertising message is displayed in the message bar. Otherwise, no prompt is temporarily issued and the advertising message is temporarily stored. When it is convenient for the user to receive the advertising message, a prompt is issued and the advertising message is displayed in the message bar.

[0078] According to one embodiment of the present invention, in step S2, if an application is currently pushing an advertising message, the aforementioned process is performed to obtain the mobile device's usage status information, current location information, and current time information. Furthermore, in step S3, information is obtained regarding the user's reading behavior for multiple historical advertising messages within a first preset time period (e.g., one week, three days, or one day) prior to the current time. This information includes, for example, whether the user directly deleted the historical advertising message without reading it after seeing its title, whether the user did not process the historical advertising message, whether the user read the historical advertising message, and whether the user performed any interactive actions (e.g., leaving a message, adding a favorite, or liking) on ​​the historical advertising message.

[0079] According to one embodiment of the present invention, in step S4, the push evaluation index of the advertising information pushed at the current moment can be determined based on the user's reading behavior information of historical advertising messages, so as to determine whether the advertising information belongs to the content that the user is interested in, and thus determine the necessity of pushing the current advertising information.

[0080] According to one embodiment of the present invention, the push evaluation index of the advertising message is determined based on the reading behavior information, including: obtaining the historical pop-up text of the historical advertising message and the pop-up text of the advertising message; obtaining the historical content information of the historical advertising message and the content information of the advertising message; obtaining the historical type information of the historical advertising message and the type information of the advertising message; obtaining the first input vector of the node corresponding to the historical advertising message based on the historical pop-up text, the historical content information and the historical type information; obtaining the second input vector of the node corresponding to the advertising information based on the pop-up text, the content information and the type information; processing the first input vector and the second input vector through a trained graph neural network model to obtain the first output vector of the node corresponding to the historical advertising message and the second output vector of the node corresponding to the advertising information; screening the target historical advertising message from the historical advertising messages based on the first output vector and the second output vector; and determining the push evaluation index of the advertising message based on the reading behavior information of the target historical advertising message.

[0081] According to one embodiment of the present invention, target historical advertising messages similar to the current advertising messages can be screened out from numerous historical advertising messages, and combined with the user's reading behavior information for the target historical advertising messages, it can be judged whether the user is interested in the current advertising message, thereby determining the necessity of pushing the current advertising information.

[0082] According to one embodiment of the present invention, a graph structure can be constructed and processed using a graph neural network model to determine the similarity between the current advertising message and historical advertising messages, thereby filtering the historical advertising messages and obtaining a target historical advertising message. In the process of constructing the graph structure, each historical advertising message can be used as a node in the graph structure. Through processing using the graph neural network model, the association between each node can be determined, that is, the association between the current advertising message and historical advertising messages can be determined, and then the similarity between the current advertising message and historical advertising messages can be determined, so as to filter the target historical advertising message from multiple historical advertising messages.

[0083] According to one embodiment of the present invention, the input information of each node can be determined first, and the pop-up text of the advertising message (i.e., the text displayed in the message bar, usually the title of the advertising message or part of the text in the title), the content information of the advertising message, and the type information of the advertising message (for example, whether the advertising message is news or a shopping advertisement, if it is news, whether the news is social news or financial news, and if it is a shopping advertisement, what category of goods the shopping advertisement corresponds to) can be obtained. The pop-up text and content information can be processed by a natural language processing model to obtain a semantic vector of the pop-up text and a semantic vector of the content information. The type information can also be represented as a vector. For example, the text representation of the type information (for example, text representation of type information such as shopping advertisements and sneakers) can be processed by a natural language processing model to obtain a semantic vector. By splicing the above multiple semantic vectors, the second input vector of the node corresponding to the advertising information can be obtained. Similarly, the first input vector of the historical advertising information can be obtained in the same way, which will not be repeated here.

[0084] According to one embodiment of the present invention, a trained graph neural network model can be used to process a first input vector and a second input vector to obtain a first output vector for a node corresponding to a historical advertising message and a second output vector for a node corresponding to advertising information. In this example, the graph neural network model's attention mechanism can be used to process the input vectors (first input vector or second input vector) of each node to obtain an association weight between the two nodes. For example, a weight matrix can be multiplied by each of the two input vectors to obtain two processed vectors. These two processed vectors are then concatenated and input into a multilayer perceptron layer (e.g., a network layer consisting of multiple fully connected layers and activation layers) to obtain an association weight between the two input vectors. For a node's input vector, the association weight can be used to perform a weighted summation of the input vector and the other input vectors. This allows the node's output vector to incorporate features from more similar nodes and differences from more dissimilar nodes. The weighted summed vectors are then multiplied by a coefficient matrix to obtain the node's output vector. This processing can be performed on each node's input vector to obtain each node's output vector, namely, a first output vector and a second output vector. The above weight matrix, coefficient matrix and parameters in the multi-layer perception network layer can all be obtained through training.

[0085] According to one embodiment of the present invention, the first output vector and the second output vector are both vectors obtained by integrating the correlation relationship between historical advertising messages and advertising messages, and can describe more comprehensive features of each historical advertising message and advertising message based on the correlation relationship. Compared with the first input vector and the second input vector, in the first output vector and the second output vector, similar output vectors are closer in the feature space, that is, they incorporate more features of advertisements with higher correlation and more similarity, and dissimilar output vectors are farther in the feature space, that is, they incorporate more differences between advertisements with lower correlation and dissimilarity, making the features of each historical advertising message and advertising message more obvious, facilitating classification and screening, and helping to improve the accuracy of screening target historical advertising messages.

[0086] According to one embodiment of the present invention, target historical advertising messages are screened from historical advertising messages based on a first output vector and a second output vector, including: determining the similarity between each first output vector and a second output vector; determining a first output vector whose similarity is greater than a first similarity threshold and less than or equal to a second similarity threshold as a target output vector, wherein the second similarity threshold is greater than the first similarity threshold; determining the historical advertising message corresponding to the target output vector as the target historical advertising message; and setting the push evaluation index to 0 if there is a first output vector whose similarity is greater than the second similarity threshold.

[0087] According to one embodiment of the present invention, the cosine similarity of each first output vector and the second output vector can be calculated. If the cosine similarity between the first output vector and the second output vector is greater than a first similarity threshold (e.g., 0.6) and less than or equal to a second similarity threshold (e.g., 0.9), then there is a certain similarity between the historical advertising message corresponding to the first output vector and the advertising message at the current moment. The historical advertising message may be of the same type as the advertising message at the current moment, but the specific content may be different (e.g., the historical advertising message is a shopping advertisement for a top, and the advertising message at the current moment is a shopping advertisement for pants). The user's reading behavior information for historical advertising messages of the same type can be used to determine whether the user is likely to be interested in the advertising message at the current moment, thereby determining the necessity of pushing the advertising information at the current moment. Therefore, the first output vector with a similarity greater than the first similarity threshold and less than or equal to the second similarity threshold can be used as the target output vector. If there is a first output vector with a similarity greater than the second similarity threshold, then the type and specific content of the historical advertising message corresponding to the first output vector are highly similar to the type and specific content of the advertising message at the current moment (for example, the historical advertising message and the advertising message at the current moment are shopping advertisements for the same product pushed by different applications). In this case, since highly similar or even identical historical advertising messages have been pushed before, there is no need to push the advertising message at the current moment again, and the push evaluation index of the advertising message is directly set to 0.

[0088] According to one embodiment of the present invention, if there is no first output vector with a similarity greater than a second similarity threshold, it means that a highly similar or even completely identical historical advertising message has not been pushed before. Based on the reading behavior information of the target historical advertising messages screened out above, the necessity of pushing an advertising message with a certain similarity at the current moment is determined, that is, the push evaluation index of the advertising message is solved.

[0089] According to one embodiment of the present invention, the reading behavior information includes whether the user clicks and reads the historical advertising message, as well as the behavior data when reading the historical advertising message; according to the reading behavior information of the target historical advertising message, determining the push evaluation index of the advertising message includes: determining the push evaluation index of the advertising message according to formula (1) , (1), in, The behavior weight for users who did not click on the target historical advertising message and cleared the target historical advertising message, The identification result of whether the user clears the target i-th historical advertising message without reading the target i-th historical advertising message. The historical content information behavior weight of the user reading the target historical advertising message, The reading progress of the historical content information of the i-th target historical advertising message for the user, The weight of the user's interactive operation in the target historical advertising message, is the recognition result of whether the user performs an interactive operation in the i-th target historical advertising message, is the first output vector of the node corresponding to the i-th target historical advertising message, is the second output vector of the node corresponding to the advertising information, for and similarity, n is the number of target historical advertising messages, i≤n, and i and n are positive integers.

[0090] According to one embodiment of the present invention, in formula (1), 0.5 is the basic score of the target historical advertisement message. If the user does not perform any processing on the target historical advertisement message, the score of the target historical advertisement message is 0.5. The identification result of whether the user clears the i-th target historical advertising message without reading the i-th target historical advertising message. If the user does not read the target historical advertising message and directly clears the target historical advertising message, then ,otherwise , in the example, If the user does not read the target historical advertising message and directly clears the target historical advertising message, it means that the user is not interested in the title of the target historical advertising message. At this time, the score of the target historical advertising message is .

[0091] According to one embodiment of the present invention, if a user clicks on a target historical advertising message, the user can enter the content page of the target historical advertising message to read it. The user's interest in the target historical advertising message can be expressed according to the reading progress. The higher the reading progress, the more interested the user is in the content of the target historical advertising message. In this example, If the user clicks on the target historical advertising message and reads the specific content without performing other interactive operations, the score of the target advertising message is obtained based on the reading progress. In this example, the reading progress is 60%, and the score of the target advertising message is Furthermore, if the user performs an interactive operation, ,otherwise, , in the example, If the user not only reads the content of the target advertising message but also performs interactive operations such as leaving a message, the score of the target advertising message is .

[0092] According to one embodiment of the present invention, The similarity between the first output vector of the node corresponding to the i-th target historical advertisement message and the second output vector of the node corresponding to the advertisement information can be used as a weight for the score of the i-th target historical advertisement message, indicating the likelihood that the advertisement message will receive the same score or the likelihood that the user will be interested in the advertisement message. The scores of multiple target historical advertisement messages can be weighted and summed using the similarity corresponding to each target advertisement message to obtain an advertisement message push evaluation index, which can be used to indicate the user's level of interest in the advertisement message at the current moment and the necessity of pushing the advertisement message.

[0093] In this way, target advertising messages that are similar but different from the current advertising message can be filtered from multiple historical advertising messages. This not only provides a data basis for determining the user's interest in advertising messages, but also eliminates interference from duplicate advertising messages. If the same historical advertising messages exist, the current advertising message can be prevented from interrupting the user again, improving user convenience. Furthermore, based on the user's various reading behavior information for the target advertising message, a score can be calculated for each target advertising message to determine the user's interest in each target advertising message. This can then comprehensively determine the user's interest in the current advertising message and the necessity of pushing the advertising message, improving the accuracy of the push evaluation index.

[0094] According to one embodiment of the present invention, the above graph neural network model can be trained before use, and the training method of the graph neural network model includes: selecting multiple training advertising information based on multiple training type information from the training advertising information library; obtaining training input vectors of multiple training advertising information; processing the training input vectors through the graph neural network model to obtain a training output vector for each training advertising information; inputting the training output vectors into the fully connected layer and the activation layer to obtain predicted reading behavior information of the training advertising information; determining the loss function of the graph neural network model based on the predicted reading behavior information and the annotation information of the training advertising information; and training the graph neural network model based on the loss function to obtain a trained graph neural network model.

[0095] According to one embodiment of the present invention, the training advertising information library is a database composed of historical advertising messages received by the same user within a longer historical period of time (for example, within a month). These historical advertising messages can be used as training advertising information, and whether the user is interested in these training advertising messages can be determined based on the user's reading behavior information of the training advertising information.

[0096] According to one embodiment of the present invention, the training input vector of the training advertising information is obtained in a similar manner to the first input vector and the second input vector, and the training output vector of each training advertising information is obtained in a similar manner to the first output vector and the second output vector, which will not be repeated here.

[0097] According to one embodiment of the present invention, each training output vector can be processed through a fully connected layer and an activation layer to obtain predicted reading behavior information of the training advertising information, for example, whether the user clicked and read the training advertising information, whether the user performed an interactive operation, whether the user directly cleared the training advertising information, etc. The predicted reading behavior information can be information in the form of probability, for example, the probability that the user directly cleared the training advertising information, the probability that the user clicked and read the training advertising information but did not interact, and the probability that the user performed an interactive behavior can be obtained. Based on the predicted reading behavior information and the actual processing of the training advertising information by the user, the error between the two can be determined, and the error can be reduced during the training process to improve the accuracy of the predicted reading behavior information, thereby improving the accuracy of the training output vectors extracted by the graph neural network model. After training, the graph neural network model can be enabled to obtain accurate first output vectors and second output vectors.

[0098] According to one embodiment of the present invention, the loss function of the graph neural network model is determined based on the predicted reading behavior information and the annotation information of the training advertisement information, including: determining the loss function LOSS of the graph neural network model according to formula (2), (2), in, is the probability that the j-th training advertisement information is not read and is cleared according to the labeled information, is the probability that the j-th training advertisement information is not read and is cleared according to the predicted reading behavior information, is the probability that the j-th training advertisement information is not processed within the preset time after being pushed, determined based on the labeled information. is the probability that the j-th training advertisement information is not processed within the preset time after being pushed, determined based on the predicted reading behavior information. is the probability that the jth training advertisement information is clicked and read but no interaction is performed based on the labeled information, is the probability that the jth training advertisement information is clicked and read but no interaction is performed, determined based on the predicted reading behavior information. is the probability of the user interacting with the jth training advertisement information determined based on the labeled information, is the probability of the user interacting with the jth training advertisement information determined based on the predicted reading behavior information, m is the number of training advertisement information, j≤m, and both j and m are positive integers.

[0099] According to an embodiment of the present invention, the annotation information of the training advertisement information is the user's actual reading behavior information of the training advertisement information. When , it means that the user's actual reading behavior information of the training advertisement information is directly cleared. This means that the user did not directly clear the training advertisement information but performed other operations. represents the cross entropy loss function composed of the probability that the j-th training advertisement information is not read and cleared according to the predicted reading behavior information and the annotation information. The weight of the cross entropy loss function is That is, a smaller weight is set for the cross entropy loss function when the user directly clears the training advertising information. This weight is consistent with the score of the training advertising information when the user directly clears the training advertising information, thereby improving the targetedness of the training.

[0100] According to one embodiment of the present invention, When , it means that the user's actual reading behavior information of the training advertisement information is not processed. Indicates that the user has actually performed other processing on the training advertising information. The cross entropy loss function is the probability that the j-th training advertising information is not processed within the preset time after being pushed based on the predicted reading behavior information and the labeled information. Its weight is 0.5, which is still consistent with the score of the training advertising information when the user does not process the training advertising information, which can improve the targeted training.

[0101] According to one embodiment of the present invention, When , it means that the user's actual reading behavior information for the training ad information is click and read but no interaction is performed. The information indicating the user's actual reading behavior of the training advertisement information is other. is the cross entropy loss function between the probability of the jth training advertisement being clicked and read but not interacted with according to the labeled information and the labeled information, and its weight is , which is consistent with the score of users who have completely read the training advertising information without interacting with it, and can improve the targetedness of training.

[0102] According to one embodiment of the present invention, When , it means that the user's actual reading behavior information of the training advertisement information is reading and interacting. The information indicating the user's actual reading behavior of the training advertisement information is other. is the cross entropy loss function between the probability of the user interacting with the jth training advertisement information determined based on the annotation information and the annotation information, and its weight is , which is consistent with the score that the user has completely read and interacted with the training advertising information, which can improve the targetedness of training.

[0103] According to one embodiment of the present invention, the aforementioned weights can be used to perform a weighted summation of the multiple cross-entropy loss functions to obtain a cross-entropy loss function corresponding to each training advertisement message. The cross-entropy loss functions for each training advertisement message are then averaged to obtain the loss function of the graph neural network model. Furthermore, the graph neural network model can be trained by backpropagating the loss function and updating the parameters of the graph neural network model using gradient descent. After multiple training cycles, a trained graph neural network model is obtained.

[0104] In this way, the loss function can be obtained by predicting reading behavior information and comparing it with the actual processing of the training advertising information by the user. The error can be reduced during the training process to improve the accuracy of the predicted reading behavior information, thereby improving the accuracy of the training output vector extracted by the graph neural network model. In the loss function, the scores of the user's various reading behavior information are made consistent with the weights of the loss function, thereby improving the targetedness of the training, thereby improving the training efficiency and the accuracy of the graph neural network model.

[0105] According to one embodiment of the present invention, in step S5, the push evaluation index for the current advertising message is obtained, which can be used to describe whether the user is interested in the current advertising message and the necessity of pushing the advertising message. Furthermore, whether the user is convenient for receiving advertising messages can be determined by using a push suppression coefficient to describe whether the user is convenient for receiving advertising messages.

[0106] According to one embodiment of the present invention, a push suppression coefficient is determined based on the usage status information, the current location information and the current time, including: a first identification result of determining whether the mobile device is in a state inconvenient for directly reading advertising messages based on the usage status information; a second identification result of determining whether the mobile device is in a place inconvenient for directly reading advertising messages based on the current location information; a third identification result of determining whether the current moment is within a preset time period inconvenient for reading advertising messages; and a push suppression coefficient is determined based on the first identification result, the second identification result, the third identification result and the average moving speed.

[0107] According to one embodiment of the present invention, states where it is inconvenient to directly read advertising messages may include states where the user is using other functions of the mobile device. For example, a user is using a navigation program while driving. In this case, it is inconvenient for the user to directly read advertising messages, and push advertising messages may affect the user's use of other functions. Locations where it is inconvenient to directly read advertising messages may include places such as conference rooms. Pushed advertising messages may disturb users during meetings. Preset time periods when it is inconvenient to read advertising messages may include the user's nighttime rest period or the user's lunch break period. Pushed advertising messages may disturb users during rest periods.

[0108] According to one embodiment of the present invention, determining a push suppression coefficient according to the first recognition result, the second recognition result, and the third recognition result includes: determining the push suppression coefficient S according to formula (3), (3), in, is the kth recognition result, when k=1, is the first recognition result, when k=2, is the second recognition result, k=3, is the third recognition result, For The corresponding suppression weight, k is a positive integer, and k≤3.

[0109] According to one embodiment of the present invention, When k=1, it means that the kth recognition result is 1. When k=1, it means that the mobile device is in a state where it is not convenient to read the advertisement message directly. When k=2, it means that the mobile device is in a place where it is not convenient to read the advertisement message directly. When k=3, it means that the current moment is in a preset time period where it is not convenient to read the advertisement message. Therefore, in any When , the push suppression coefficient will be less than 1, indicating that it is inconvenient for users to receive advertising messages. In other words, advertising messages may cause disturbances. When k=1, it means that the mobile device is not in a state where it is inconvenient to read the advertisement message directly. When k=2, it means that the mobile device is not in a place where it is inconvenient to read the advertisement message directly. When k=3, it means that the current moment is not in a preset time period where it is inconvenient to read the advertisement message. If the three All are 0, indicating that push advertising messages will not disturb users. In the example, It is a manually set value that indicates the degree to which the user does not want to be disturbed, and , three Can be consistent or inconsistent. In the example, three The values ​​of are all equal to 0.5, and the present invention does not impose any limitation on this.

[0110] In this way, a comprehensive analysis can be performed on the status, location and current time of the mobile device to determine whether the push advertising message may cause disturbance to the user, so as to obtain the push suppression coefficient. When the advertising message may cause disturbance to the user, a push suppression coefficient less than 1 can be generated to reduce the possibility of disturbing the user.

[0111] According to one embodiment of the present invention, in step S6, the push evaluation index and the push suppression coefficient may be multiplied together to obtain a push coefficient. If the push coefficient is above a preset coefficient threshold (e.g., 0.5), the advertisement message may be pushed; otherwise, the advertisement message may not be pushed to reduce the possibility of disturbing the user. For example, if both the push evaluation index and the push suppression coefficient are low, it may indicate that the user is not interested in the advertisement message and that the advertisement message may be disturbing to the user. In this case, the advertisement message may not be pushed and no prompt may be generated.

[0112] According to one embodiment of the present invention, in step S7, if the advertising message is not pushed, the advertising message can be temporarily stored and can be pushed again when the user is convenient to receive the advertising message, for example, when the push suppression coefficient changes (for example, the user leaves the meeting room).

[0113] According to one embodiment of the present invention, in step S8, the application's traffic for pushing advertising messages may also be optimized. Based on the number of advertisement messages pushed by the application temporarily stored during a second preset time period before and at the current moment, and the push suppression coefficient corresponding to the temporarily stored advertisement messages, a traffic optimization scheme for the application is determined, including: obtaining an average value of the historical push suppression coefficients each time the application generates an advertisement message during the second preset time period; if the relative difference between the push suppression coefficient and the average value of the historical push suppression coefficients is less than or equal to a preset ratio threshold, and the number of temporarily stored advertisement messages pushed by the application is greater than or equal to a preset number threshold, pausing the application's advertisement push traffic.

[0114] According to one embodiment of the present invention, if the relative difference between the push suppression coefficient and the average value of the historical push suppression coefficients is less than or equal to a preset ratio threshold, this indicates that the application has repeatedly pushed advertisements at inappropriate times, i.e., the application has pushed advertisements when the user is not readily available to receive them. If the number of advertisements pushed by the application that is temporarily stored within a second time period (e.g., three days) is greater than or equal to a preset threshold, this indicates that the application has pushed a large number of advertisements that are of no interest to the user or that may be disruptive to the user. In this case, it can be determined that there is a problem with the application's processing of pushed advertisements, making it unsuitable for pushing advertisements to the user. The application's advertisement push traffic can be suspended, i.e., the application's use of pushing advertisements can be suspended to reduce disruption to the user.

[0115] According to an embodiment of the present invention, a method for optimizing mobile programmatic advertising traffic can determine a push evaluation index for a current ad message based on a user's reading behavior information for historical ad messages. Furthermore, a push suppression coefficient is determined based on the user's current location and time. This automatically determines the necessity of pushing the current ad message. If push is not necessary, the ad message is temporarily stored rather than pushed immediately. This reduces the likelihood of disturbing the user when it is inconvenient to receive ad messages and eliminates the need for manual intervention, thereby improving user convenience. When determining the push evaluation index, target ad messages that are similar but different from the current ad message can be screened from multiple historical ad messages. This not only provides a data basis for determining the user's interest in the ad message, but also eliminates interference from duplicate ad messages. If duplicate historical ad messages exist, this prevents further interruptions from the current ad message, thereby improving user convenience. Furthermore, a score for each target ad message can be calculated based on the user's various reading behavior information for the target ad message, thereby determining the user's interest in each target ad message. This can then be used to comprehensively determine the user's interest in the current ad message and the necessity of pushing the ad message, thereby improving the accuracy of the push evaluation index. When training a graph neural network model, a loss function can be obtained by predicting reading behavior information and comparing it with the user's actual processing of the training advertising information. This error can be reduced during the training process to improve the accuracy of the predicted reading behavior information, thereby improving the accuracy of the training output vector extracted by the graph neural network model. In the loss function, the scores of the user's various reading behavior information are made consistent with the weights of the loss function, thereby improving the targeted training, thereby improving training efficiency and the accuracy of the graph neural network model. When determining the push suppression coefficient, a comprehensive analysis of the mobile device's status, location, and current time can be performed to determine whether the push advertising message is likely to disturb the user. This can then be used to obtain a push suppression coefficient. In the event that the advertising message is likely to disturb the user, a push suppression coefficient less than 1 can be generated to reduce the possibility of disturbing the user.

[0116] Figure 2 A block diagram of a mobile programmatic advertising traffic optimization system according to an embodiment of the present invention is exemplarily shown, wherein the system includes:

[0117] A judgment module, used to judge whether there is an advertisement message pushed by the application at the current moment;

[0118] an acquisition module, configured to acquire usage status information, current location information, and current time information of the mobile device if the advertisement message exists;

[0119] A reading behavior information module is used to obtain user reading behavior information for multiple historical advertising messages within a first preset time period before the current moment;

[0120] A push evaluation index module, configured to determine a push evaluation index for the advertisement message based on the reading behavior information;

[0121] a push suppression coefficient module, configured to determine a push suppression coefficient based on the usage status information, the current location information, and the current time;

[0122] a determination module, configured to determine whether to push the advertisement message based on the push evaluation index and the push suppression coefficient;

[0123] A temporary storage module, configured to temporarily store the advertisement message of the application program when it is determined that the advertisement message is not to be pushed;

[0124] The optimization module is used to determine a traffic optimization plan for the application based on the number of advertisement messages pushed by the application temporarily stored at the current moment and within a second preset time period before the current moment, and a push suppression coefficient corresponding to the temporarily stored advertisement messages.

[0125] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0126] Those skilled in the art will appreciate that the embodiments of the present invention described above and shown in the accompanying drawings are intended to be illustrative only and are not intended to limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and illustrated in the embodiments. Any variations or modifications may be made to the embodiments of the present invention without departing from the principles described.

[0127] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing mobile programmatic advertising traffic, characterized in that: include: Determine whether there is an advertisement message pushed by the application at the current moment; If the advertisement message exists, obtaining usage status information, current location information, and current time information of the mobile device; Obtaining user reading behavior information for multiple historical advertising messages within a first preset time period before the current moment; Determining a push evaluation index for the advertisement message based on the reading behavior information; determining a push suppression coefficient according to the usage status information, the current location information, and the current time; determining whether to push the advertisement message according to the push evaluation index and the push suppression coefficient; If it is determined that the advertisement message is not to be pushed, temporarily storing the advertisement message of the application; A traffic optimization scheme for the application is determined according to the number of advertisement messages pushed by the application temporarily stored at the current moment and within a second preset time period before the current moment, and a push suppression coefficient corresponding to the temporarily stored advertisement messages.

2. The method for optimizing mobile programmatic advertising traffic according to claim 1, characterized in that: Determining a push evaluation index for the advertisement message based on the reading behavior information includes: Obtaining historical pop-up texts of historical advertising messages and the pop-up texts of the advertising messages; Acquiring historical content information of historical advertisement messages and content information of the advertisement messages; Obtain historical type information of historical advertisement messages and type information of advertisement messages; Obtaining a first input vector of a node corresponding to a historical advertising message based on historical pop-up text, historical content information, and historical type information; Obtaining a second input vector of a node corresponding to the advertisement information according to the pop-up window text, the content information, and the type information; Processing the first input vector and the second input vector using a trained graph neural network model to obtain a first output vector for a node corresponding to a historical advertisement message and a second output vector for a node corresponding to the advertisement information; filtering target historical advertisement messages from the historical advertisement messages according to the first output vector and the second output vector; Determine the push evaluation indicators of the advertising message based on the reading behavior information of the target historical advertising message.

3. The method for optimizing mobile programmatic advertising traffic according to claim 2, wherein: Filtering target historical advertisement messages from historical advertisement messages according to the first output vector and the second output vector includes: Determining the similarity between each first output vector and the second output vector; determining, as a target output vector, a first output vector having a similarity greater than a first similarity threshold and less than or equal to a second similarity threshold, wherein the second similarity threshold is greater than the first similarity threshold; determining the historical advertising message corresponding to the target output vector as the target historical advertising message; If there is a first output vector whose similarity is greater than the second similarity threshold, the push evaluation index is set to 0.

4. The method for optimizing mobile programmatic advertising traffic according to claim 2, wherein: The reading behavior information includes whether the user clicks and reads the historical advertising message, and the behavior data when reading the historical advertising message; Based on the target's historical advertising message reading behavior information, determine the push evaluation indicators of the advertising message, including: According to the formula ， Determine the push evaluation indicators for advertising messages ,in, The behavior weight for users who did not click on the target historical advertising message and cleared the target historical advertising message, The identification result of whether the user clears the target i-th historical advertising message without reading the target i-th historical advertising message. The historical content information behavior weight of the user reading the target historical advertising message, The reading progress of the historical content information of the i-th target historical advertising message for the user, The weight of the user's interactive operation in the target historical advertising message, is the recognition result of whether the user performs an interactive operation in the i-th target historical advertising message, is the first output vector of the node corresponding to the i-th target historical advertising message, is the second output vector of the node corresponding to the advertising information, for and similarity, n is the number of target historical advertising messages, i≤n, and i and n are positive integers.

5. The method for optimizing mobile programmatic advertising traffic according to claim 4, characterized in that: The training method of the graph neural network model includes: Selecting a plurality of training advertisement information of a plurality of training type information from a training advertisement information library; Obtaining a plurality of training input vectors of training advertisement information; The training input vector is processed through the graph neural network model to obtain the training output vector of each training advertisement information; The training output vector is input into the fully connected layer and the activation layer to obtain the predicted reading behavior information of the training advertisement information; Determine the loss function of the graph neural network model based on the predicted reading behavior information and the labeled information of the training advertisement information; According to the loss function, the graph neural network model is trained to obtain a trained graph neural network model.

6. The method for optimizing mobile programmatic advertising traffic according to claim 5, characterized in that: Based on the predicted reading behavior information and the labeled information of the training advertisement information, the loss function of the graph neural network model is determined, including: According to the formula ， Determine the loss function LOSS of the graph neural network model, where is the probability that the j-th training advertisement information is not read and is cleared according to the labeled information, is the probability that the j-th training advertisement information is not read and is cleared according to the predicted reading behavior information, is the probability that the j-th training advertisement information is not processed within the preset time after being pushed, determined based on the labeled information. is the probability that the j-th training advertisement information is not processed within the preset time after being pushed, determined based on the predicted reading behavior information. is the probability that the jth training advertisement information is clicked and read but no interaction is performed based on the labeled information, is the probability that the jth training advertisement information is clicked and read but no interaction is performed, determined based on the predicted reading behavior information. is the probability of the user interacting with the jth training advertisement information determined based on the labeled information, is the probability of the user interacting with the jth training advertisement information determined based on the predicted reading behavior information, m is the number of training advertisement information, j≤m, and both j and m are positive integers.

7. The method for optimizing mobile programmatic advertising traffic according to claim 1, wherein: Determining a push suppression coefficient according to the usage status information, the current location information, and the current time includes: determining, based on the usage status information, a first identification result of whether the mobile device is in a state that is inconvenient for directly reading the advertisement message; determining, based on the current location information, a second identification result of whether the mobile device is in a location where it is inconvenient to directly read the advertisement message; a third recognition result of determining whether the current moment is within a preset time period that is inconvenient for reading the advertisement message; A push suppression coefficient is determined according to the first recognition result, the second recognition result, and the third recognition result.

8. The method for optimizing mobile programmatic advertising traffic according to claim 7, characterized in that: Determining a push suppression coefficient according to the first recognition result, the second recognition result, and the third recognition result includes: According to the formula ， Determine the push suppression coefficient S, where is the kth recognition result, when k=1, is the first recognition result, when k=2, is the second recognition result, k=3, is the third recognition result, For The corresponding suppression weight, k is a positive integer, and k≤3.

9. The method for optimizing mobile programmatic advertising traffic according to claim 1, wherein: Determining a traffic optimization scheme for the application according to the number of advertisement messages pushed by the application temporarily stored at the current moment and within a second preset time period before the current moment, and a push suppression coefficient corresponding to the temporarily stored advertisement messages, including: Obtaining an average value of historical push suppression coefficients each time the application generates an advertisement message within a second preset time period; If the relative gap between the push suppression coefficient and the average value of the historical push suppression coefficient is less than or equal to the preset ratio threshold, and the number of temporarily stored advertising messages pushed by the application is greater than or equal to the preset number threshold, the advertising push traffic of the application is suspended.

10. A mobile programmatic advertising traffic optimization system, characterized in that: include: A judgment module, used to judge whether there is an advertisement message pushed by the application at the current moment; an acquisition module, configured to acquire usage status information, current location information, and current time information of the mobile device if the advertisement message exists; A reading behavior information module is used to obtain user reading behavior information for multiple historical advertising messages within a first preset time period before the current moment; A push evaluation index module, configured to determine a push evaluation index for the advertisement message based on the reading behavior information; a push suppression coefficient module, configured to determine a push suppression coefficient based on the usage status information, the current location information, and the current time; a determination module, configured to determine whether to push the advertisement message based on the push evaluation index and the push suppression coefficient; A temporary storage module, configured to temporarily store the advertisement message of the application program when it is determined that the advertisement message is not to be pushed; The optimization module is used to determine a traffic optimization plan for the application based on the number of advertisement messages pushed by the application temporarily stored at the current moment and within a second preset time period before the current moment, and a push suppression coefficient corresponding to the temporarily stored advertisement messages.

Images