Training method of advertisement putting analysis model and advertisement putting analysis method

Abstract

The invention provides a training method of an advertisement putting analysis model and an advertisement putting analysis method, and relates to the technical field of information, and the method comprises the steps: obtaining sample data which comprises target sample behavior data of a plurality of users, the target sample behavior data of each user comprises a user behavior sequence, context information and an actual conversion label of each user under the target advertisement contact. Predicting a user behavior sequence and context information in the target sample behavior data of each user by adopting the initial advertisement putting analysis model to obtain a prediction conversion tag corresponding to the target sample behavior data; and adjusting parameters of the initial advertisement putting analysis model according to the actual conversion label and the predicted conversion label to obtain an advertisement putting analysis model. According to the method, the accuracy of user conversion prediction of the advertisement putting analysis model and the optimization efficiency of advertisement putting can be remarkably improved.

Description

Technical Field

[0001] The present application relates to the field of information technology, and in particular to a training method for an advertising delivery analysis model and an advertising delivery analysis method. Background Art

[0002] In recent years, the development of deep learning has brought new opportunities for advertising strategies. In particular, the emergence of the self-attention mechanism enables the model to better capture long-distance dependencies in the data. The self-attention mechanism automatically assigns different weights to each part of the input data by calculating the correlation between different positions, thereby focusing on key information. This mechanism has achieved remarkable results in natural language processing, image recognition and other fields, providing technical support for the innovation of advertising strategies.

[0003] At present, existing advertising strategies mainly rely on traditional machine learning models, such as logistic regression and decision trees, which analyze user attribution based on user historical behavior, interest preferences, and demographic information.

[0004] However, most existing advertising attribution models use rules or simple machine learning models, which cannot capture complex user behavior paths and cross-modal data associations. In addition, due to the lack of multi-level correlation analysis methods, user advertising interaction behaviors lead to inaccurate attribution and limited optimization effects. Summary of the invention

[0005] The purpose of this application is to provide a training method for an advertising delivery analysis model and an advertising delivery analysis method to address the deficiencies in the above-mentioned prior art, which can significantly improve the accuracy of user conversion prediction of the advertising delivery analysis model and the optimization efficiency of advertising delivery.

[0006] To achieve the above purpose, the technical solution adopted in the embodiment of the present application is as follows:

[0007] In a first aspect, an embodiment of the present application provides a method for training an advertising delivery analysis model, the method comprising:

[0008] Acquire a sample data set, wherein the sample data set includes: target sample behavior data of multiple users, wherein the target sample behavior data of each user includes: a user behavior sequence, context information, and an actual conversion label of each user at a target advertising contact point; the actual conversion label is used to indicate whether each user completes an advertising conversion after executing the user behavior sequence for the target advertising contact point under the context information;

[0009] Using the initial advertising delivery analysis model to predict the user behavior sequence and context information in the target sample behavior data of each user, and obtaining a predicted conversion label corresponding to the target sample behavior data;

[0010] According to the actual conversion label and the predicted conversion label, the initial advertising delivery analysis model is adjusted to obtain an advertising delivery analysis model.

[0011] Optionally, obtaining a sample data set includes:

[0012] Acquire multiple groups of original behavior data of each user, each group of original behavior data includes: historical behavior data of each user at the target advertising touch point, basic user information, advertising feature data, and historical conversion events;

[0013] Performing feature processing on the multiple groups of original behavior data to obtain multiple groups of initial sample behavior data for each user;

[0014] Generating target sample behavior data for each user according to the multiple groups of initial sample behavior data for each user;

[0015] The sample data set is constructed according to the target sample behavior data of the multiple users.

[0016] Optionally, the performing feature processing on the multiple groups of original behavior data to obtain the multiple groups of initial sample behavior data of each user includes:

[0017] Sort each behavior in the historical behavior data in each set of original behavior data in chronological order to obtain a user behavior sequence corresponding to each set of original behavior data;

[0018] Performing feature transformation processing on the user basic information and the advertisement features in each set of original behavior data to obtain context information corresponding to each set of original behavior data;

[0019] According to the conversion events in each set of original behavior data, an actual conversion label corresponding to each set of original behavior data is obtained;

[0020] A set of initial sample behavior data for each user is generated according to the user behavior sequence, the context information, and the actual conversion label.

[0021] Optionally, generating target sample behavior data for each user according to the multiple groups of initial sample behavior data for each user includes:

[0022] Calculating a behavior weight of each user according to the multiple groups of initial sample behavior data of each user;

[0023] According to the behavior weight of each user, a corresponding number of initial sample behavior data are determined from the multiple groups of initial sample behavior data of each user as the target sample behavior data of each user.

[0024] Optionally, calculating the behavior weight of each user according to the multiple groups of initial sample behavior data of each user includes:

[0025] Calculating a behavior activity index of each user according to the user behavior sequences in the multiple groups of initial sample behavior data of each user;

[0026] Calculating a historical conversion index for each user according to actual conversion labels in the multiple groups of initial sample behavior data for each user;

[0027] The behavior weight of each user is calculated according to the behavior activity index and the historical conversion index.

[0028] Optionally, the method further comprises:

[0029] Randomly deleting behaviors in a user behavior sequence in the target sample behavior data of each user or modifying the behavior sequence to obtain new sample behavior data of each user;

[0030] The new sample behavior data of each user is added to the sample data set.

[0031] Optionally, before using the initial advertising delivery analysis model to predict the user behavior sequence and context information in the target sample behavior data of each user to obtain the predicted conversion label corresponding to each group of sample user behavior data, the method further includes:

[0032] Using a graph neural network, structurally processing the context information in the target sample behavior data of each user to obtain a structured context feature of each user;

[0033] Performing feature fusion on the structured context features and user behavior sequence of each user to obtain fused behavior features of each user;

[0034] The initial advertising delivery analysis model is used to predict the user behavior sequence and context information in the target sample behavior data of each user to obtain the predicted conversion label corresponding to the target sample behavior data, including:

[0035] The initial advertisement delivery analysis model is used to predict the integrated behavior characteristics of each user to obtain a predicted conversion label corresponding to the target sample behavior data.

[0036] Optionally, the initial advertisement delivery analysis model includes: an input layer, a coding layer and an output layer;

[0037] The initial advertising delivery analysis model is used to predict the user behavior sequence and context information in the target sample behavior data of each user to obtain the predicted conversion label corresponding to the target sample behavior data, including:

[0038] Using the input layer, feature conversion is performed on the user behavior sequence and context information in the target sample behavior data of each user to obtain a user behavior feature vector and a context feature vector corresponding to the target sample behavior data;

[0039] Using the coding layer, based on the user behavior feature vector and the context feature vector, obtaining the conversion attribution probability of the target sample behavior data corresponding to the target advertising touch point;

[0040] The output layer is used to determine the conversion attribution probability distribution of the target advertising touchpoint according to the conversion attribution probability under the target advertising touchpoint, and the predicted conversion label corresponding to the target sample behavior data is determined according to the conversion attribution probability distribution.

[0041] Optionally, the encoding layer includes: a contact-level attention module, a session-level attention module, a user-level attention module, and a probability calculation module; the encoding layer is used to obtain the predicted conversion label corresponding to the target sample behavior data and the conversion attribution probability under the target advertising contact based on the user behavior feature vector and the context feature vector, including:

[0042] The contact-level attention module is used to calculate the contact-level weight of each target advertising contact according to the user behavior feature vector and the context feature vector of all users at each target advertising contact;

[0043] The session-level attention module is used to calculate the session-level weights of all the target advertising contacts according to the user behavior feature vectors and the context feature vectors of all users at all the target advertising contacts;

[0044] The user-level attention module is used to calculate the user-level weight of each user according to the user behavior feature vector and the context feature vector of each user at all the target advertisement touch points;

[0045] The probability calculation module is used to calculate the conversion attribution probability of each user at each target advertising touch point based on the touch point level weight of the target advertising touch point, the session level weight of the target advertising touch point and the user level weight of each user.

[0046] In a second aspect, another embodiment of the present application provides an advertising delivery analysis method, the method comprising: obtaining current behavior data of a current user at a target advertising touch point, the current behavior data comprising: a current behavior sequence and current context information;

[0047] An advertising delivery analysis model is used to process the current behavior sequence and the current context information to obtain the conversion prediction result and conversion attribution probability of the current user under the target advertising touch point, wherein the advertising delivery analysis model is the advertising delivery analysis model trained by the method described in the first aspect above.

[0048] In a third aspect, another embodiment of the present application provides a training device for an advertising delivery analysis model, the device comprising:

[0049] An acquisition module is used to acquire a sample data set, wherein the sample data set includes: target sample behavior data of multiple users, wherein the target sample behavior data of each user includes: a user behavior sequence, context information, and an actual conversion label of each user at a target advertising contact point; the actual conversion label is used to indicate whether each user completes an advertising conversion after executing the user behavior sequence for the target advertising contact point under the context information;

[0050] A prediction module, used to predict the user behavior sequence and context information in the target sample behavior data of each user using the initial advertising delivery analysis model to obtain a predicted conversion label corresponding to the target sample behavior data;

[0051] A determination module is used to adjust the parameters of the initial advertising delivery analysis model according to the actual conversion label and the predicted conversion label to obtain the advertising delivery analysis model.

[0052] In the fourth aspect, another embodiment of the present application provides a computer device, comprising: a processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the computer device is running, the processor communicates with the storage medium through the bus, and the processor executes the machine-readable instructions to perform the steps of any method described in the first aspect above.

[0053] In a fifth aspect, another embodiment of the present application provides a storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any method described in the first aspect are executed.

[0054] The beneficial effects of this application are:

[0055] The present application provides a training method for an advertising delivery analysis model and an advertising delivery analysis method, by acquiring sample data, wherein the sample data set includes: target sample behavior data of multiple users, and the target sample behavior data of each user includes: the user behavior sequence, context information and actual conversion label of each user under the target advertising touch point. The initial advertising delivery analysis model is used to predict the user behavior sequence and context information in the target sample behavior data of each user, and the predicted conversion label corresponding to the target sample behavior data is obtained; according to the actual conversion label and the predicted conversion label, the initial advertising delivery analysis model is adjusted to obtain the advertising delivery analysis model. The present application utilizes the user behavior sequence, context information and actual conversion label of each user under the target advertising touch point, realizes attribution analysis of the user behavior sequence and context information under the target advertising touch point through the initial advertising delivery analysis model, and determines the advertising delivery analysis model based on the comparison of the attribution analysis results with the actual conversion label, which can significantly improve the accuracy of the user conversion prediction of the advertising delivery analysis model and the optimization efficiency of advertising delivery. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0057] Figure 1 A training method for an advertising delivery analysis model provided in an embodiment of the present application;

[0058] Figure 2 A schematic diagram of a process for obtaining a sample data set in a training method for an advertising delivery analysis model provided in an embodiment of the present application;

[0059] Figure 3 A schematic diagram of a process for determining initial sample behavior data in a training method for an advertising delivery analysis model provided in an embodiment of the present application;

[0060] Figure 4 A schematic diagram of a process for determining target sample behavior data in a training method for an advertising delivery analysis model provided in an embodiment of the present application;

[0061] Figure 5 A schematic diagram of a process for determining behavior weights in a training method for an advertising delivery analysis model provided in an embodiment of the present application;

[0062] Figure 6A schematic diagram of a process for updating a sample data set in a training method for an advertising delivery analysis model provided in an embodiment of the present application;

[0063] Figure 7 A schematic diagram of a process for determining a predicted conversion label in a training method for an advertising delivery analysis model provided in an embodiment of the present application;

[0064] Figure 8 A schematic diagram of a process for determining a predicted conversion label in another method for training an advertisement delivery information model provided in an embodiment of the present application;

[0065] Fig. 9 A schematic diagram of a process for determining conversion attribution probability in a training method for an advertising delivery analysis model provided in an embodiment of the present application;

[0066] Fig.10 A schematic diagram of the process of advertising analysis and amplification provided in an embodiment of the present application;

[0067] Fig.11 A schematic diagram of the structure of a training device for an advertising delivery analysis model provided in an embodiment of the present application;

[0068] Fig.12 A schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0069] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the drawings in the present application only serve the purpose of explanation and description and are not used to limit the scope of protection of the present application. In addition, it should be understood that the schematic drawings are not drawn in real proportion. The flowchart used in this application shows the operations implemented according to some embodiments of the present application. It should be understood that the operations of the flowchart can be implemented out of order, and the steps without logical context can be reversed in order or implemented simultaneously. In addition, those skilled in the art can add one or more other operations to the flowchart under the guidance of the content of the present application, or remove one or more operations from the flowchart.

[0070] In addition, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0071] It should be noted that the term "comprising" will be used in the embodiments of the present application to indicate the existence of the features declared thereafter, but does not exclude the addition of other features.

[0072] In order to clearly describe the training method of the advertising delivery analysis model provided in the embodiment of the present application, the method provided in the embodiment of the present application is described below with reference to a plurality of drawings. Figure 1 A training method for an advertising delivery analysis model provided in an embodiment of the present application is as follows: Figure 1 As shown, the method includes:

[0073] Step 101: Obtain a sample data set.

[0074] Among them, the sample data set includes: target sample behavior data of multiple users. The target sample behavior data of each user includes: the user behavior sequence, context information and actual conversion label of each user under the target advertising contact; the actual conversion label is used to indicate whether each user completes the advertising conversion after executing the user behavior sequence for the target advertising contact under the context information. The advertising contact can be a social media, a search engine or a recommendation platform, etc., and the target advertising contact can be one or more of the advertising contacts, which is not limited in the embodiment of the present application. User behavior can be click, browse, search and other behaviors, which is not limited in the embodiment of the present application. The user behavior of each user under the target advertising contact can be one or more, which is not required in the embodiment of the present application, and the number of user behavior sequences is the same as the number of user behaviors. The actual conversion label is used to indicate whether each user completes the advertising conversion after executing the user behavior sequence for the target advertising contact under the context information. When the advertising conversion is completed, the actual conversion label is 1, and when the advertising conversion is not completed, the actual conversion label is 0.

[0075] Optionally, the sample data set is determined to be obtained based on the original behavior data of the advertisement delivery, wherein the original behavior data includes original behavior data of multiple users.

[0076] Optionally, the sample data set is randomly divided into a training set, a validation set, and a test set according to a ratio, and the specific ratio may be 70% for the training set, 15% for the validation set, and 15% for the test set. The training set is used to train the initial advertising delivery analysis model. The validation set is used to evaluate the performance of the advertising delivery analysis model, for example, indicators such as accuracy and area under the curve. The test set is used to evaluate the final performance of the advertising delivery analysis model.

[0077] Step 102: Use the initial advertising delivery analysis model to predict the user behavior sequence and context information in the target sample behavior data of each user to obtain a predicted conversion label corresponding to the target sample behavior data.

[0078] Among them, the initial advertising delivery analysis model is a Transformer model based on the self-attention mechanism, which includes a multimodal attention mechanism, a hierarchical attention mechanism, and dynamic adjustment of attribution weights. The multimodal attention mechanism is used to enable the initial advertising delivery analysis model to better understand the association between multiple types of information such as text and images. The hierarchical attention mechanism is used to apply the attention mechanism at different levels of abstraction, such as calculating the attention weights at the user level, session level, and touch level, so as to capture information of different granularities. The dynamic adjustment of the attribution weight ensures that the initial advertising delivery analysis model automatically calculates the attribution weight of each data source through the self-attention layer, and can dynamically adjust as the data input changes, ensuring the real-time and accuracy of the attribution results. The predicted conversion label corresponding to the target sample behavior data is the prediction result of the initial advertising delivery analysis model on the user behavior sequence and context information in the target sample behavior data, and the predicted conversion label is 1 or 0.

[0079] Optionally, the user behavior sequence and context information in the target sample behavior data of each user is used as the input of the initial advertising delivery analysis model, and the input data is predicted by the initial advertising delivery analysis model to obtain a predicted conversion label corresponding to the target sample behavior data.

[0080] Step 103: According to the actual conversion label and the predicted conversion label, the initial advertising delivery analysis model is adjusted to obtain the advertising delivery analysis model.

[0081] Optionally, the loss values ​​of the actual conversion label and the predicted conversion label are determined, and the parameters of the initial water meter reading recognition model are adjusted according to the loss values ​​of the actual conversion label and the predicted conversion label to obtain the target advertising delivery analysis model. The parameter adjustment includes calculating the gradient through the back propagation algorithm according to the loss value and updating the model parameters. The model parameters can be weight parameters, bias parameters, etc. The embodiment of the present application does not limit this.

[0082] Optionally, after updating the model parameters, you can adjust hyperparameters such as learning rate and batch size based on the performance on the validation set. When the model performs best on the validation set, save the model parameters.

[0083] The present application provides a training method for an advertising delivery analysis model, by acquiring sample data, wherein the sample data set includes: target sample behavior data of multiple users, and the target sample behavior data of each user includes: the user behavior sequence, context information and actual conversion label of each user under the target advertising touch point. The initial advertising delivery analysis model is used to predict the user behavior sequence and context information in the target sample behavior data of each user, and the predicted conversion label corresponding to the target sample behavior data is obtained; according to the actual conversion label and the predicted conversion label, the initial advertising delivery analysis model is adjusted to obtain the advertising delivery analysis model. The present application utilizes the user behavior sequence, context information and actual conversion label of each user under the target advertising touch point, and realizes attribution analysis of the user behavior sequence and context information under the target advertising touch point through the initial advertising delivery analysis model, and determines the advertising delivery analysis model based on the comparison of the attribution analysis results with the actual conversion label, which can significantly improve the accuracy of the user conversion prediction of the advertising delivery analysis model and the optimization efficiency of advertising delivery.

[0084] Based on the above embodiments, the present application also provides a process for obtaining a sample data set in a training method of an advertising delivery analysis model. Figure 2 A schematic diagram of a flow chart of obtaining a sample data set in a training method of an advertising delivery analysis model provided in an embodiment of the present application, such as Figure 2 As shown, in the above step 101, obtaining a sample data set includes:

[0085] Step 201: Obtain multiple groups of original behavior data of each user.

[0086] Among them, each set of original behavior data includes: each user's historical behavior data under the target advertising touchpoint, user basic information, advertising feature data and historical conversion events. Historical behavior data refers to behavioral data such as clicks, searches, and browsing. By setting the corresponding embedding tool, the user's behavior data is determined by the embedding tool in each user's interactive element code. User basic information includes the user's geographic location, device type, age, gender and other information. User basic information users segment user groups. Advertising features include advertising time, delivery time, advertising position, etc. Advertising position features can include information such as page location and advertising shape. Historical conversion events are used to record whether the user finally completed the conversion, such as purchase, registration, etc. Different conversion events are defined according to different business scenarios.

[0087] Step 202: Perform feature processing on multiple groups of original behavior data to obtain multiple groups of initial sample behavior data for each user.

[0088] The feature processing is to convert the original behavior data into an original behavior sequence, and the multiple groups of initial sample behavior data include multiple original behavior sequences.

[0089] Optionally, feature processing is performed on multiple groups of original behavior data to obtain multiple groups of original behavior sequences, and multiple groups of initial sample behavior data of each user are formed through the original behavior sequences.

[0090] Step 203: Generate target sample behavior data for each user based on the multiple groups of initial sample behavior data for each user.

[0091] Optionally, the target sample behavior data of each user is determined based on the initial sample behavior data of each user. The target sample behavior data of each user may be multiple groups of initial sample behavior data; or multiple groups of initial sample behavior data of each user may be randomly deleted to generate the target sample behavior data of each user; or multiple groups of initial sample behavior data of each user may be randomly sorted to generate the target sample behavior data of each user, etc., and the embodiments of the present application do not limit this.

[0092] Step 204: construct a sample data set based on target sample behavior data of multiple users.

[0093] In the embodiment of the present application, multiple sets of original behavior data of each user are obtained, and feature processing is performed on the multiple sets of original behavior data to obtain multiple sets of initial sample behavior data of each user, generate target sample behavior data of each user, and construct a sample data set. The multiple sets of data of the present application can provide a more comprehensive behavior pattern, make the sample data set more representative, help to build a model with stronger generalization ability, and can extract the most influential information on the prediction target, reduce the influence of noise and irrelevant variables, and thus improve the prediction accuracy of the model.

[0094] Based on the above embodiments, the present application also provides a process for determining initial sample behavior data in a training method for an advertising delivery analysis model. Figure 3 A schematic diagram of a process for determining initial sample behavior data in a training method for an advertising delivery analysis model provided in an embodiment of the present application, such as Figure 3 As shown, in the above step 202, multiple groups of original behavior data are feature processed to obtain multiple groups of initial sample behavior data of each user, including:

[0095] Step 301: Sort the behaviors in the historical behavior data in each set of original behavior data in chronological order to obtain a user behavior sequence corresponding to each set of original behavior data.

[0096] Optionally, each original behavior is sorted according to the time from each behavior in the historical behavior data in each set of original behavior data to the present, so as to obtain a user behavior sequence corresponding to each set of original behavior data. The closer the time from the present to each behavior in the user behavior data, the higher the sorting, and the farther the time from the present to each behavior in the user behavior data, the lower the sorting.

[0097] Optionally, the time weight of each behavior in the historical behavior data in each set of original behavior data is determined according to a time weight model, and each behavior in the historical behavior data in each set of original behavior data is sorted according to the time weight to obtain a user behavior sequence corresponding to each set of original behavior data. The time weight model can be formula (1).

[0098] T w =e -a×t (1)

[0099] Among them, T w is the time weight, α is the time decay coefficient, and t represents the time interval since the last behavior.

[0100] Step 302: Perform feature transformation processing on the user basic information and advertisement features in each set of original behavior data to obtain context information corresponding to each set of original behavior data.

[0101] Optionally, the user basic information and advertising features in each set of original behavior data are subjected to feature transformation processing to be converted into corresponding user basic information sequences and advertising feature sequences, and each set of user basic information sequences and advertising feature sequences are used as context information corresponding to each set of original behavior data.

[0102] Step 303: Obtain the actual conversion label corresponding to each set of original behavior data according to the conversion events in each set of original behavior data.

[0103] Optionally, when a conversion event in a set of original behavior data is a successful conversion, the actual conversion label corresponding to the set of original behavior data is 1; when a conversion event in a set of original behavior data is a failed conversion, the actual conversion label corresponding to the set of original behavior data is 0.

[0104] Step 304: Generate a set of initial sample behavior data for each user based on the user behavior sequence, context information, and actual conversion labels.

[0105] Each set of initial sample behavior data includes: user behavior sequence, context information and actual conversion label.

[0106] In the embodiment of the present application, each behavior in the historical behavior data in each set of original behavior data is sorted in chronological order to obtain the user behavior sequence corresponding to each set of original behavior data, and the user basic information and advertising features in each set of original behavior data are subjected to feature transformation processing to obtain the context information corresponding to each set of original behavior data, and the actual conversion label corresponding to each set of original behavior data is obtained according to the conversion events in each set of original behavior data, and a set of initial sample behavior data for each user is generated. The accuracy of the generated initial sample behavior data is ensured, so that the initial sample behavior data is structured data, which is convenient for model processing and learning, thereby improving the accuracy of model prediction.

[0107] Based on the above embodiments, the present application also provides a process for determining target sample behavior data in a training method for an advertising delivery analysis model. Figure 4 A flow chart of determining target sample behavior data in a training method for an advertising delivery analysis model provided in an embodiment of the present application, such as Figure 4 As shown, in the above step 203, target sample behavior data of each user is generated according to multiple groups of initial sample behavior data of each user, including:

[0108] Step 401: Calculate the behavior weight of each user based on multiple groups of initial sample behavior data of each user.

[0109] The user's behavior weight is used to determine the proportion of multiple groups of initial sample behavior data of each user in the sample data set.

[0110] Optionally, the behavior weight of each user is calculated based on the user's click frequency, search frequency, browsing time or any other behavior-related data in the user behavior sequence of multiple groups of initial sample behavior data of each user, as well as the user's actual conversion label.

[0111] Step 402: According to the behavior weight of each user, a corresponding number of initial sample behavior data are determined from multiple groups of initial sample behavior data of each user as target sample behavior data of each user.

[0112] For example, when the behavior weight of a user is 0.2, the target sample behavior data of the user in the sample data set accounts for 0.2. The target sample behavior data of the user is determined based on the target sample behavior data of all users in the sample data set and the behavior weight of the user. A corresponding number of initial sample behavior data are determined from the multiple groups of initial sample behavior data of the user as the target sample behavior data of each user.

[0113] In the embodiment of the present application, the behavior weight of each user is calculated based on multiple sets of initial sample behavior data of each user, and the initial sample behavior data corresponding to the user's behavior weight is determined from the multiple sets of initial sample behavior data of each user as the target sample behavior data of each user. The present application selects more representative sample behavior data as the target sample, which helps the model learn more common and stable user behavior patterns and enhances the generalization ability of the model.

[0114] Based on the above embodiments, the present application also provides a process for determining behavior weights in a training method for an advertising delivery analysis model. Figure 5 A flowchart of determining behavior weights in a training method for an advertising delivery analysis model provided in an embodiment of the present application is shown in FIG. Figure 5 As shown, in the above step 401, the behavior weight of each user is calculated according to the multiple groups of initial sample behavior data of each user, including:

[0115] Step 501: Calculate the behavior activity index of each user according to the user behavior sequences in multiple groups of initial sample behavior data of each user.

[0116] The activity index of each user behavior may be determined based on one behavior in the user behavior sequence in the multiple groups of initial sample behavior data of each user, or may be determined based on multiple behaviors. For example, it may be determined based on one or more of the user's click frequency, search frequency, and browsing time.

[0117] Step 502: Calculate the historical conversion index of each user according to the actual conversion labels in the multiple groups of initial sample behavior data of each user.

[0118] Optionally, the number of successful conversions and failed conversions of each user is determined based on the actual conversion labels in the multiple groups of initial sample behavior data of each user, and the conversion success rate of each user is determined, and the historical conversion index of each user is determined based on the conversion success rate of each user. For example, the conversion success rate of each user can be used as the historical conversion index of each user.

[0119] Step 503: Calculate the behavior weight of each user based on the behavior activity index and the historical conversion index.

[0120] Optionally, the behavior weight of each user is calculated according to the behavior activity index and historical conversion index of each user by using a preset formula (2).

[0121] W i =α·A i +β·H i (2)

[0122] Among them, W iis the behavior weight of user i, A i is the behavioral activity index of user i, H i is the historical conversion index of user i, α and β are preset adjustment parameters used to balance the impact of different features on the weight.

[0123] In an embodiment of the present application, the behavioral activity index of each user is calculated based on the user behavior sequence in the multiple groups of initial sample behavior data of each user, the historical conversion index of each user is calculated based on the actual conversion label in the multiple groups of initial sample behavior data of each user, and the behavioral weight of each user is calculated. The present application calculates the behavioral weight to realize the personalized weight allocation for different users or different behaviors, so that the model pays more attention to the behaviors that have a greater and more active impact on the conversion, and can improve the model's ability to predict the user's conversion behavior.

[0124] Based on the above embodiments, the present application also provides a process for updating a sample data set in a training method for an advertising delivery analysis model. Figure 6 A schematic diagram of a flow chart of updating a sample data set in a training method of an advertising delivery analysis model provided in an embodiment of the present application, such as Figure 6 As shown, based on the above steps 201 to 204, the method further includes:

[0125] Step 601: Randomly delete behaviors in a user behavior sequence in target sample behavior data of each user or modify the behavior sequence to obtain new sample behavior data of each user.

[0126] For example, the behaviors in the user behavior sequence in a target sample behavior data of a certain user include A, B, C and D, in the order of ABCD. B and C are deleted, and the behaviors in the user behavior sequence in a target sample behavior data of the user include A and D, and new sample behavior data of the user is obtained.

[0127] For example, the behaviors in the user behavior sequence in a target sample behavior data of a certain user include A, B, C and D, and the order is ABCD. The order of B and C is swapped, and the behaviors in the user behavior sequence in a target sample behavior data of the user include A, C, B and D, and new sample behavior data of the user is obtained.

[0128] Step 602: Add new sample behavior data of each user to the sample data set.

[0129] In an embodiment of the present application, the behaviors in the user behavior sequence in the target sample behavior data of each user are randomly deleted or the behavior order is modified to obtain new sample behavior data for each user and add them to the sample data set to train the initial advertising delivery analysis model, which can enhance the robustness and generalization ability of the model.

[0130] Based on the above embodiments, the present application also provides a process for determining predicted conversion labels in a training method of an advertising delivery analysis model. Figure 7 A schematic diagram of a process for determining a predicted conversion label in a training method for an advertising delivery analysis model provided in an embodiment of the present application, such as Figure 7 As shown, in the above step 102, before using the initial advertising delivery analysis model to predict the user behavior sequence and context information in the target sample behavior data of each user to obtain the predicted conversion label corresponding to each group of sample user behavior data, the method also includes:

[0131] Step 701: Use a graph neural network to perform structured processing on the context information in the target sample behavior data of each user to obtain the structured context features of each user.

[0132] Among them, Graph Neural Networks (GNN) is a deep learning model for processing graph structured data.

[0133] For example, when the context information in the target sample behavior data of each user is "background, mobile phone, 25, male", the graph neural network is used to process the context information, and the obtained structured context features are "[1,0,0,0],[1,0,0],[0,1,0,0],[1,0]".

[0134] Step 702: perform feature fusion on the structured context features and user behavior sequence of each user to obtain a fused behavior feature of each user.

[0135] Among them, the fusion method can be a direct splicing or weighted summation method, which is not limited in the embodiment of the present application.

[0136] For example, when a user's structured context feature is “[1,0,0,0],[1,0,0],[0,1,0,0],[1,0]” and the user's behavior sequence is “[1,0,0,0],[0,1,0,0],[0,0,1,0],[0,0,0,1]”, the structured context feature and the user behavior sequence are fused. When the fusion method is concatenation, the obtained fused behavior feature is “[1,0,0,0,1,0,0,0],[1,0,0,0,1,0,0],[0,1,0,0,0,0,1,0],[1,0,0,0,0,0,0,1]”.

[0137] As in step 102 above, the initial advertising delivery analysis model is used to predict the user behavior sequence and context information in the target sample behavior data of each user, and the predicted conversion label corresponding to the target sample behavior data is obtained, including:

[0138] Step 703: Use the initial advertising delivery analysis model to predict the integrated behavior characteristics of each user to obtain a predicted conversion label corresponding to the target sample behavior data.

[0139] Optionally, the fused behavior characteristics of each user are used as input to the initial advertising delivery analysis model, and the initial advertising delivery analysis model is used to predict the fused behavior characteristics of each user to obtain a predicted conversion label corresponding to the target sample behavior data.

[0140] In the embodiment of the present application, the structured context features and user behavior sequences of each user are fused to obtain the fused behavior features of each user, and the fused behavior features of each user are predicted using the initial advertising delivery analysis model to obtain the predicted conversion label corresponding to the target sample behavior data. The present application can enhance the model's understanding of user context information, realize the fusion modeling of complex network features, and improve the overall understanding ability of the model.

[0141] Based on the above embodiment, the initial advertising delivery analysis model includes: an input layer, an encoding layer and an output layer; the present application also provides a process for determining a predicted conversion label in another training method of an advertising delivery information model, Figure 8 A schematic diagram of a process for determining a predicted conversion label in another method for training an advertisement delivery information model provided in an embodiment of the present application, such as Figure 8 As shown, in the above step 102, the initial advertising delivery analysis model is used to predict the user behavior sequence and context information in the target sample behavior data of each user, and the predicted conversion label corresponding to the target sample behavior data is obtained, including:

[0142] Step 801: Using the input layer, perform feature conversion on the user behavior sequence and context information in the target sample behavior data of each user to obtain a user behavior feature vector and a context feature vector corresponding to the target sample behavior data.

[0143] The input layer is used to perform feature transformation on the user behavior sequence and context information in the target sample behavior data of each user to obtain the corresponding feature vector.

[0144] Optionally, an input layer is used to perform feature conversion on the user behavior sequence and context information in the target sample behavior data of each user, where the user behavior sequence and context information are the user behavior sequence and context information for the target advertising touch point, and the user behavior sequence and context information in the converted target sample behavior data are normalized to obtain a corresponding feature vector. The input layer determines the corresponding position code according to the time sequence of the user behavior sequence in the target sample behavior data, and sorts the feature vectors according to the position code.

[0145] Step 802: Using the coding layer, based on the user behavior feature vector and the context feature vector, obtain the conversion attribution probability of the target sample behavior data corresponding to the target advertising touch point.

[0146] Optionally, the long-term and short-term relationships in the user behavior feature vector are captured according to the multi-head self-attention mechanism of the encoding layer, and the feature vector output by the self-attention mechanism is linearly transformed through a feedforward neural network to obtain the conversion attribution probability under the target advertising touch point corresponding to the target sample behavior data.

[0147] Step 803: using the output layer, according to the conversion attribution probability under the target advertising touch point, determine the conversion attribution probability distribution of the target advertising touch point, and determine the predicted conversion label corresponding to the target sample behavior data according to the conversion attribution probability distribution.

[0148] Optionally, an output layer is used to determine the conversion attribution probability distribution of the target advertising touchpoints according to a preset formula (3) based on the conversion attribution probability under the target advertising touchpoints.

[0149]

[0150] Among them, γ c is the conversion attribution probability of touchpoint C, ∑ j γ j Contribution is the sum of the attribution probabilities of all touch points. c is the conversion attribution probability distribution of touchpoint C.

[0151] Optionally, the conversion attribution probability distribution of the target advertising contact is determined by setting a preset conversion attribution probability distribution threshold of the target advertising contact. When the conversion attribution probability distribution of the target advertising contact is greater than the conversion attribution probability distribution threshold of the target advertising contact, the predicted conversion label corresponding to the target sample behavior data is determined to be 1; when the conversion attribution probability distribution of the target advertising contact is less than or equal to the conversion attribution probability distribution threshold of the target advertising contact, the predicted conversion label corresponding to the target sample behavior data is determined to be 0. The target advertising contact can be multiple or 1, which is determined according to actual needs. The specific judgment process can start when the conversion attribution probability distribution of several target advertising contacts is greater than the conversion attribution probability distribution threshold of the target advertising contact, or the conversion attribution probability distribution of a certain target advertising contact is greater than the conversion attribution probability distribution threshold of the target advertising contact, or the conversion attribution probability distribution of all target advertising contacts is greater than the conversion attribution probability distribution threshold of the target advertising contact, then the predicted conversion label corresponding to the target sample behavior data is determined to be 1. The specific implementation method is determined according to actual needs, and the embodiments of the present application do not limit this.

[0152] On the basis of the above embodiment, the encoding layer includes: a contact-level attention module, a session-level attention module, a user-level attention module and a probability calculation module; the embodiment of the present application also provides a process for determining conversion attribution probability in a training method of an advertising delivery analysis model, Fig. 9 A schematic diagram of a process for determining conversion attribution probability in a training method for an advertising delivery analysis model provided in an embodiment of the present application, such as Fig. 9 As shown, in the above step 802, the coding layer is used to obtain the predicted conversion label corresponding to the target sample behavior data and the conversion attribution probability under the target advertising touch point based on the user behavior feature vector and the context feature vector, including:

[0153] Step 901: Use a contact-level attention module to calculate the contact-level weight of each target advertising contact based on the user behavior feature vectors and context feature vectors of all users at each target advertising contact.

[0154] Among them, the touch-level weight is used to reflect the importance of the target advertising touch in the user behavior sequence.

[0155] Specifically, the contact-level weight of each target advertising contact can be determined according to formula (4) in the contact-level attention module.

[0156]

[0157] Among them, Q c and K c is the query and key representation of contact C, which is determined according to the user behavior feature vector and context feature vector under contact C. kis the dimension of the key vector, a c is the contact level weight of contact C.

[0158] Step 902: Use a session-level attention module to calculate the session-level weights of all target advertising touch points based on the user behavior feature vectors and context feature vectors of all users at all target advertising touch points.

[0159] Optionally, the session-level weights of all target advertising touch points are determined according to the user behavior feature vectors and context feature vectors of all users at all target advertising touch points through formula (5).

[0160] a s =softmax(W s H s )(5)

[0161] Among them, a s is the session-level weight of all target ad touch points, H s is the session feature of all target advertising contacts, W s is the session-level projection matrix.

[0162] Step 903: Use a user-level attention module to calculate the user-level weight of each user based on the user behavior feature vector and context feature vector of each user at all target advertising touch points.

[0163] Optionally, the user-level weight of each user is calculated according to the user behavior feature vector and context feature vector of each user at all target advertising touch points using formula (6).

[0164] a u =softmax(W u H u )(6)

[0165] Among them, a u is the user-level weight, H u is the user behavior feature vector, W u is the user-level projection matrix for the user.

[0166] Step 904: Use a probability calculation module to calculate the conversion attribution probability of each user at each target advertising touchpoint based on the touchpoint-level weight of the target advertising touchpoint, the session-level weight of the target advertising touchpoint, and the user-level weight of each user.

[0167] Optionally, the conversion attribution probability of each user at each target advertising touchpoint is calculated according to the touchpoint-level weight of the target advertising touchpoint, the session-level weight of the target advertising touchpoint, and the user-level weight of each user through formula (7).

[0168] γc =W i ·a c ·a s ·a u

[0169] Where: W i =α·A i +β·H i is the user behavior weight, a c is the contact level weight of contact C, a s The session-level weight of all target ad touchpoints, a u is the user-level weight, γ c is the conversion attribution probability of touch point c.

[0170] In an embodiment of the present application, the conversion attribution probability of the target touch is determined by the touch-level weight, the session-level weight of all target advertising touches, and the user-level weight. The present application takes into account multiple levels of the advertising conversion process, and can more comprehensively understand user conversion behavior. Through detailed weight allocation, the specific contribution of each touch to user conversion can be more accurately measured.

[0171] Another embodiment of the present application further provides an advertisement delivery analysis method. Fig.10 A schematic diagram of the process of advertising analysis and amplification provided in the embodiment of the present application, such as Fig.10 As shown, the method includes:

[0172] Step 1001: Obtain the current behavior data of the current user at the target advertising touch point.

[0173] The current behavior data includes: the current behavior sequence and the current context information. The context information includes the user basic information sequence and the advertisement feature sequence.

[0174] Step 1002: Use the advertising delivery analysis model to process the current behavior sequence and current context information to obtain the conversion prediction result and conversion attribution probability of the current user under the target advertising touch point.

[0175] Among them, the advertising delivery analysis model is the advertising delivery analysis model trained by the above method.

[0176] Optionally, the current behavior sequence and current context information are used as inputs of an advertising delivery analysis model, and the advertising delivery analysis model is used to process the current behavior sequence and current context information to obtain the conversion attribution probability of the current user at the target advertising touchpoint, determine the conversion attribution probability distribution based on the conversion attribution probability, and determine the predicted conversion label based on the conversion attribution probability distribution.

[0177] Based on the same inventive concept, the embodiment of the present application also provides a training device for an advertising delivery analysis model corresponding to the training method for the advertising delivery analysis model. Since the principle of solving the problem by the device in the embodiment of the present application is similar to the training method for the advertising delivery analysis model in the embodiment of the present application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0178] Fig.11 A structural diagram of a training device for an advertising delivery analysis model provided in an embodiment of the present application, the device comprising:

[0179] The acquisition module 1101 is used to acquire a sample data set, where the sample data set includes: target sample behavior data of multiple users, where the target sample behavior data of each user includes: a user behavior sequence, context information, and an actual conversion label of each user under a target advertising contact point; the actual conversion label is used to indicate whether each user completes an advertising conversion after executing the user behavior sequence for the target advertising contact point under the context information;

[0180] Prediction module 1102, used to predict the user behavior sequence and context information in the target sample behavior data of each user using the initial advertising delivery analysis model, and obtain the predicted conversion label corresponding to the target sample behavior data;

[0181] The determination module 1103 is used to adjust the parameters of the initial advertising delivery analysis model according to the actual conversion label and the predicted conversion label to obtain the advertising delivery analysis model.

[0182] Optionally, the acquisition module 1101 is specifically used to: acquire multiple groups of original behavior data of each user, each group of original behavior data includes: historical behavior data of each user under the target advertising touch point, basic user information, advertising feature data and historical conversion events;

[0183] Perform feature processing on multiple groups of original behavior data to obtain multiple groups of initial sample behavior data for each user;

[0184] Generate target sample behavior data for each user based on multiple groups of initial sample behavior data for each user;

[0185] Construct a sample data set based on target sample behavior data of multiple users.

[0186] Optionally, the acquisition module 1101 is specifically used to: sort each behavior in the historical behavior data in each set of original behavior data in chronological order to obtain a user behavior sequence corresponding to each set of original behavior data;

[0187] Perform feature transformation processing on user basic information and advertisement features in each set of original behavior data to obtain context information corresponding to each set of original behavior data;

[0188] According to the conversion events in each set of original behavior data, the actual conversion labels corresponding to each set of original behavior data are obtained;

[0189] Based on the user behavior sequence, context information, and actual conversion labels, a set of initial sample behavior data for each user is generated.

[0190] Optionally, the acquisition module 1101 is specifically used to: calculate the behavior weight of each user according to multiple groups of initial sample behavior data of each user;

[0191] According to the behavior weight of each user, a corresponding number of initial sample behavior data are determined from multiple groups of initial sample behavior data of each user as target sample behavior data of each user.

[0192] Optionally, the acquisition module 1101 is specifically used to: calculate the behavior activity index of each user according to the user behavior sequences in the multiple groups of initial sample behavior data of each user;

[0193] Calculate each user's historical conversion indicators based on the actual conversion labels in multiple groups of initial sample behavior data of each user;

[0194] Calculate each user's behavior weight based on behavioral activity indicators and historical conversion indicators.

[0195] Optionally, the acquisition module 1101 is further used to: randomly delete behaviors in the user behavior sequence in the target sample behavior data of each user or modify the behavior sequence to obtain new sample behavior data of each user;

[0196] Add each user's new sample behavior data to the sample data set.

[0197] Optionally, the prediction module 1102 is further used to: use a graph neural network to perform structured processing on the context information in the target sample behavior data of each user to obtain a structured context feature of each user;

[0198] Perform feature fusion on each user's structured context features and user behavior sequence to obtain each user's fused behavior features;

[0199] Optionally, the prediction module 1102 is specifically used to: use the initial advertising delivery analysis model to predict the integrated behavior characteristics of each user, and obtain a predicted conversion label corresponding to the target sample behavior data.

[0200] Optionally, the initial advertisement delivery analysis model includes: an input layer, a coding layer, and an output layer; a prediction module 1102, specifically used for:

[0201] The input layer is used to perform feature conversion on the user behavior sequence and context information in the target sample behavior data of each user, and obtain the user behavior feature vector and context feature vector corresponding to the target sample behavior data;

[0202] The encoding layer is used to obtain the conversion attribution probability of the target sample behavior data corresponding to the target advertising touchpoint based on the user behavior feature vector and the context feature vector.

[0203] The output layer is used to determine the conversion attribution probability distribution of the target advertising touchpoints based on the conversion attribution probability under the target advertising touchpoints, and to determine the predicted conversion label corresponding to the target sample behavior data based on the conversion attribution probability distribution.

[0204] Optionally, the encoding layer includes: a contact-level attention module, a session-level attention module, a user-level attention module, and a probability calculation module; the prediction module 1102 is specifically used to: use the contact-level attention module to calculate the contact-level weight of each target advertising contact according to the user behavior feature vectors and context feature vectors of all users at each target advertising contact;

[0205] A session-level attention module is used to calculate the session-level weights of all target ad touchpoints based on the user behavior feature vectors and context feature vectors of all users at all target ad touchpoints.

[0206] A user-level attention module is used to calculate the user-level weight of each user based on the user behavior feature vector and context feature vector of each user at all target advertising touch points.

[0207] A probability calculation module is used to calculate the conversion attribution probability of each user at each target advertising touchpoint based on the touchpoint-level weight of the target advertising touchpoint, the session-level weight of the target advertising touchpoint, and the user-level weight of each user.

[0208] For descriptions of the processing flow of each module in the device and the interaction flow between each module, reference may be made to the relevant descriptions in the above method embodiment, which will not be described in detail here.

[0209] The present application also provides a computer device, Fig.12 A schematic diagram of a computer device structure provided in an embodiment of the present application is shown in FIG. Fig.12 As shown, the computer and device include: a processor 1201, a memory 1202, and optionally, a bus 1203. The memory 1202 stores machine-readable instructions executable by the processor 1201. When the computer device is running, the processor 1201 communicates with the memory 1202 via the bus 1203. When the machine-readable instructions are executed by the processor 1201, the steps of the training method of the above-mentioned advertising delivery analysis model are executed.

[0210] An embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the training method of the above-mentioned advertising delivery analysis model are executed.

[0211] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, the specific working process of the system and device described above can refer to the corresponding process in the method embodiment, and will not be repeated in this application. In the several embodiments provided in this application, it should be understood that the disclosed system, device and method can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation. For example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0212] In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or part of the technical solution that contributes to the prior art or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, RandomAccess Memory), disk or optical disk and other media that can store program code.

[0213] The above are only specific implementation methods of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be covered by the protection scope of the present application.

Claims

1. A training method for an advertising delivery analysis model, characterized in that: The method comprises: Acquire a sample data set, wherein the sample data set includes: target sample behavior data of multiple users, wherein the target sample behavior data of each user includes: a user behavior sequence, context information, and an actual conversion label of each user at a target advertising contact point; the actual conversion label is used to indicate whether each user completes an advertising conversion after executing the user behavior sequence at the target advertising contact point under the context information; Using the initial advertising delivery analysis model to predict the user behavior sequence and context information in the target sample behavior data of each user, and obtain a predicted conversion label corresponding to the target sample behavior data; According to the actual conversion label and the predicted conversion label, the initial advertising delivery analysis model is adjusted to obtain an advertising delivery analysis model.

2. The method according to claim 1, characterized in that The obtaining of the sample data set comprises: Acquire multiple groups of original behavior data of each user, each group of original behavior data includes: historical behavior data of each user at the target advertising touch point, basic user information, advertising feature data, and historical conversion events; Performing feature processing on the multiple groups of original behavior data to obtain multiple groups of initial sample behavior data for each user; Generating target sample behavior data for each user according to the multiple groups of initial sample behavior data for each user; The sample data set is constructed according to the target sample behavior data of the multiple users.

3. The method according to claim 2, characterized in that The performing feature processing on the multiple groups of original behavior data to obtain multiple groups of initial sample behavior data of each user includes: Sort each behavior in the historical behavior data in each set of original behavior data in chronological order to obtain a user behavior sequence corresponding to each set of original behavior data; Performing feature transformation processing on the user basic information and the advertisement features in each set of original behavior data to obtain context information corresponding to each set of original behavior data; According to the conversion events in each set of original behavior data, an actual conversion label corresponding to each set of original behavior data is obtained; A set of initial sample behavior data for each user is generated according to the user behavior sequence, the context information, and the actual conversion label.

4. The method according to claim 2, characterized in that: The step of generating target sample behavior data for each user according to the multiple groups of initial sample behavior data for each user comprises: Calculating a behavior weight of each user according to the multiple groups of initial sample behavior data of each user; According to the behavior weight of each user, a corresponding number of initial sample behavior data are determined from the multiple groups of initial sample behavior data of each user as the target sample behavior data of each user.

5. The method according to claim 4, characterized in that The calculating the behavior weight of each user according to the multiple groups of initial sample behavior data of each user includes: Calculating a behavior activity index of each user according to the user behavior sequences in the multiple groups of initial sample behavior data of each user; Calculating a historical conversion index for each user according to actual conversion labels in the multiple groups of initial sample behavior data for each user; The behavior weight of each user is calculated according to the behavior activity index and the historical conversion index.

6. The method according to claim 2, characterized in that The method further comprises: Randomly deleting behaviors in a user behavior sequence in the target sample behavior data of each user or modifying the behavior sequence to obtain new sample behavior data of each user; The new sample behavior data of each user is added to the sample data set.

7. The method according to claim 1, characterized in that Before using the initial advertising delivery analysis model to predict the user behavior sequence and context information in the target sample behavior data of each user to obtain the predicted conversion label corresponding to each group of sample user behavior data, the method further includes: Using a graph neural network, structurally processing the context information in the target sample behavior data of each user to obtain a structured context feature of each user; Performing feature fusion on the structured context features and user behavior sequence of each user to obtain fused behavior features of each user; The initial advertising delivery analysis model is used to predict the user behavior sequence and context information in the target sample behavior data of each user to obtain the predicted conversion label corresponding to the target sample behavior data, including: The initial advertisement delivery analysis model is used to predict the integrated behavior characteristics of each user to obtain a predicted conversion label corresponding to the target sample behavior data.

8. The method according to claim 1, characterized in that The initial advertisement delivery analysis model includes: an input layer, a coding layer and an output layer; The initial advertising delivery analysis model is used to predict the user behavior sequence and context information in the target sample behavior data of each user to obtain the predicted conversion label corresponding to the target sample behavior data, including: Using the input layer, feature conversion is performed on the user behavior sequence and context information in the target sample behavior data of each user to obtain a user behavior feature vector and a context feature vector corresponding to the target sample behavior data; Using the coding layer, based on the user behavior feature vector and the context feature vector, obtaining the conversion attribution probability of the target sample behavior data corresponding to the target advertising touch point; The output layer is used to determine the conversion attribution probability distribution of the target advertising touchpoint according to the conversion attribution probability under the target advertising touchpoint, and the predicted conversion label corresponding to the target sample behavior data is determined according to the conversion attribution probability distribution.

9. The method according to claim 8, characterized in that The encoding layer includes: a contact-level attention module, a session-level attention module, a user-level attention module, and a probability calculation module; the encoding layer is used to obtain the predicted conversion label corresponding to the target sample behavior data and the conversion attribution probability under the target advertising contact based on the user behavior feature vector and the context feature vector, including: The contact-level attention module is used to calculate the contact-level weight of each target advertising contact according to the user behavior feature vector and the context feature vector of all users at each target advertising contact; The session-level attention module is used to calculate the session-level weights of all the target advertising contacts according to the user behavior feature vectors and the context feature vectors of all users at all the target advertising contacts; The user-level attention module is used to calculate the user-level weight of each user according to the user behavior feature vector and the context feature vector of each user at all the target advertisement touch points; The probability calculation module is used to calculate the conversion attribution probability of each user at each target advertising touch point based on the touch point level weight of the target advertising touch point, the session level weight of the target advertising touch point and the user level weight of each user.

10. An advertising delivery analysis method, characterized in that: The method comprises: Acquire current behavior data of the current user at the target advertisement contact point, wherein the current behavior data includes: current behavior sequence and current context information; An advertising delivery analysis model is used to process the current behavior sequence and the current context information to obtain the conversion prediction result and conversion attribution probability of the current user under the target advertising touch point, wherein the advertising delivery analysis model is the advertising delivery analysis model trained by any of the methods described in claims 1-9 above.

Images