Model training method, advertisement putting processing method, device and program product

Abstract

The invention provides a model training method and device, an advertisement putting processing method and device, a computer program product and electronic equipment, and relates to the technical field of computers. The model training method comprises the steps of performing anonymization processing on an original data set to obtain an anonymous data set, and determining an initial feature set according to the anonymous data set; evaluating the importance of the features in the initial feature set to obtain an importance evaluation result; random noise is added to the importance evaluation result, and a feature subset is determined from the initial feature set according to the importance evaluation result added with the random noise; and training the advertisement putting model based on the feature subset to obtain a trained advertisement putting model. According to the invention, the accuracy of the advertisement putting model can be improved under the condition of ensuring the safety of user data.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a model training method, an advertisement delivery processing method, a model training device, an advertisement delivery processing device, a computer program product, and an electronic device. Background Art

[0002] With the development of Internet technology, people have been able to obtain information, life, entertainment and work through the Internet, which has become a part of their lives. In order to increase their visibility and promote their products, businesses often place advertisements on the Internet. Internet online advertising has become the main way to place advertisements besides television and newspapers. For promoters, they hope to recommend advertisements to target groups in a targeted manner and provide users with products that they are interested in. For users, they hope to receive targeted advertising recommendations so that they can directly obtain advertisements for products that they are interested in. Currently, many machine learning models for recommending advertisements have emerged.

[0003] However, currently, with the training of models and the application of models for advertising recommendations, there are problems of user data leakage, illegal acquisition or abuse of data.

[0004] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0005] The purpose of the present disclosure is to provide a model training method, an advertisement delivery processing method, a model training device, an advertisement delivery processing device, a computer program product and an electronic device, thereby improving the accuracy of the advertisement delivery model while at least ensuring the security of user data.

[0006] According to a first aspect of the present disclosure, a model training method is provided, comprising: anonymizing an original data set to obtain an anonymous data set, and determining an initial feature set based on the anonymous data set; evaluating the importance of features in the initial feature set to obtain an importance evaluation result; adding random noise to the importance evaluation result, and determining a feature subset from the initial feature set based on the importance evaluation result after adding random noise; training an advertising delivery model based on the feature subset to obtain a trained advertising delivery model.

[0007] In an exemplary embodiment of the present disclosure, an advertising delivery model is trained based on a feature subset to obtain a trained advertising delivery model, including: obtaining the gradient of a network parameter of the advertising delivery model based on the feature subset; adding noise to the gradient of the network parameter, and updating the parameters of the advertising delivery model using the gradient of the network parameter after adding noise.

[0008] In an exemplary embodiment of the present disclosure, before adding noise to the gradient of the network parameter, the method further includes: clipping the gradient of the network parameter based on a preset gradient clipping threshold.

[0009] In an exemplary embodiment of the present disclosure, an original data set is anonymized to obtain an anonymous data set, including: obtaining a preset privacy budget and noise distribution, and determining a first noise scale according to the privacy budget; adding random noise to data in the original data set based on the noise distribution and the first noise scale to obtain interference data; and determining the anonymous data set based on the interference data.

[0010] In an exemplary embodiment of the present disclosure, random noise is added to an importance evaluation result, and a feature subset is determined from an initial feature set based on the importance evaluation result after adding the random noise, including: adding random noise to the importance evaluation result based on a preset privacy budget to obtain the importance evaluation result after adding the random noise; comparing the importance evaluation result after adding the random noise with a preset evaluation threshold, and determining a feature subset from the features of the initial feature set based on the comparison result.

[0011] In an exemplary embodiment of the present disclosure, an evaluation is performed on the importance of features in an initial feature set based on an importance evaluation model; and random noise is added to the importance evaluation result based on a preset privacy budget, including: obtaining sensitivity information of a query function corresponding to the importance evaluation model; determining a second noise scale based on the preset privacy budget and the sensitivity information; and adding random noise to the importance evaluation result based on the second noise scale.

[0012] In an exemplary embodiment of the present disclosure, before anonymizing the original data set to obtain the anonymous data set, the method also includes: performing data cleaning on the initial data set; standardizing the initial data set after data cleaning to obtain standardized data; and performing data enhancement on the standardized data to obtain the original data set.

[0013] According to a second aspect of the present disclosure, there is provided an advertisement delivery processing method, comprising: obtaining advertisement information to be processed; inputting the advertisement information to be processed into a trained advertisement delivery model to obtain a processing result of the advertisement information to be processed; and performing advertisement delivery processing based on the processing result. The trained advertisement delivery model is obtained by training the advertisement delivery model based on any of the above methods.

[0014] In an exemplary embodiment of the present disclosure, advertisement delivery processing is performed based on the processing result, including: obtaining target data corresponding to the processing result, the target data at least including advertisement delivery data and an original data set for training an advertisement delivery model; performing compliance checking on the original data set and the advertisement delivery data, so as to perform advertisement delivery processing on the processing result according to the compliance checking result.

[0015] In an exemplary embodiment of the present disclosure, the method further includes: generating a compliance detection file according to the compliance detection result, wherein the compliance detection file at least includes potential risk information and location information corresponding to the potential risk information.

[0016] According to a third aspect of the present disclosure, a model training device is provided, including: an anonymization processing module, used to anonymize the original data set to obtain an anonymous data set, and determine an initial feature set based on the anonymous data set; a feature evaluation module, used to evaluate the importance of features in the initial feature set to obtain an importance evaluation result; a feature selection module, used to add random noise to the importance evaluation result, and determine a feature subset from the initial feature set according to the importance evaluation result after adding random noise; a model training module, used to train an advertising delivery model based on the feature subset to obtain a trained advertising delivery model.

[0017] According to a fourth aspect of the present disclosure, an advertising delivery processing device is provided, comprising: an information acquisition module, used to acquire advertising information to be processed; a first processing module, used to input the advertising information to be processed into a trained advertising delivery model to obtain a processing result of the advertising information to be processed; and a second processing module, used to perform advertising delivery processing based on the processing result. The trained advertising delivery model is obtained by training the advertising delivery model based on any of the above methods.

[0018] According to one aspect of the present disclosure, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, any one of the above methods is implemented.

[0019] According to one aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform any one of the above methods by executing the executable instructions.

[0020] In the technical solution provided by the embodiments of the present disclosure, on the one hand, the original data set is anonymized to obtain an anonymous data set, and the initial feature set is determined based on the anonymous data set, which can avoid the presence of data information that leaks user privacy in the initial feature set. On the other hand, the importance of the features in the initial feature set is evaluated to obtain an importance evaluation result, random noise is added to the importance evaluation result, and a feature subset is determined from the initial feature set based on the importance evaluation result after adding random noise. The key features for model training can be extracted to train the advertising delivery model, improve the accuracy of model training, and add random noise to increase data privacy security, further avoiding privacy leakage in the process of screening data. In summary, the exemplary embodiments of the present disclosure can improve the accuracy of the advertising delivery model while ensuring the security of user data.

[0021] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The drawings herein are incorporated into the specification and constitute a part of the specification, showing embodiments consistent with the present disclosure, and together with the specification, are used to explain the principles of the present disclosure. Obviously, the drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0023] Figure 1 A schematic diagram showing the stages involved in the training scheme of the advertisement delivery model according to an embodiment of the present disclosure is shown;

[0024] Figure 2 A flowchart schematically illustrates a model training method of an exemplary embodiment of the present disclosure;

[0025] Figure 3 A flowchart schematically illustrates an implementation method of a training advertisement delivery model according to an exemplary embodiment of the present disclosure;

[0026] Figure 4 A flowchart of a training advertisement delivery model in an exemplary embodiment of the present disclosure is schematically shown;

[0027] Figure 5 A flowchart schematically illustrates a training process of a training batch in an exemplary embodiment of the present disclosure;

[0028] Figure 6 A flowchart schematically illustrates an implementation method of obtaining an anonymous data set in an exemplary embodiment of the present disclosure;

[0029] Figure 7 A flowchart schematically illustrates a method for implementing a screening feature according to an exemplary embodiment of the present disclosure;

[0030] Figure 8 A flowchart schematically illustrates an advertisement delivery processing method in an exemplary embodiment of the present disclosure;

[0031] Fig. 9 A schematic diagram schematically shows the composition of a model training device in an exemplary embodiment of the present disclosure;

[0032] Fig.10 A schematic diagram schematically shows the composition of an advertisement delivery processing device in an exemplary embodiment of the present disclosure;

[0033] Fig.11A schematic diagram of an electronic device to which the embodiments of the present disclosure can be applied is shown. DETAILED DESCRIPTION

[0034] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and to fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0035] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0036] Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making.

[0037] Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics and other technologies. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0038] Machine Learning (ML) is a multi-disciplinary subject that involves probability theory, statistics, approximation theory, convex analysis, and algorithmic complexity theory. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and teaching learning.

[0039] The technical solution provided by the exemplary embodiment of the present disclosure involves machine learning technology of artificial intelligence, which uses machine learning technology to anonymize the original data set to obtain an anonymous data set, and determines the initial feature set based on the anonymous data set; evaluates the importance of the features in the initial feature set to obtain an importance evaluation result; adds random noise to the importance evaluation result, and determines a feature subset from the initial feature set based on the importance evaluation result after adding random noise; trains the advertising delivery model based on the feature subset to obtain a trained advertising delivery model. The model training method of the exemplary embodiment of the present disclosure and the advertising delivery model obtained by training can be applied to the advertising delivery scenario, and can improve the accuracy of advertising delivery while ensuring the security of user data.

[0040] Figure 1 A schematic diagram of the stages involved in the technical solution of the exemplary embodiment of the present disclosure is shown, such as Figure 1 As shown, the technical solution of the exemplary embodiment of the present disclosure includes a model training stage and a model application stage.

[0041] In an exemplary embodiment of the present disclosure, the provided model training method can be executed by a terminal device. During the training stage of the model, the terminal device can anonymize the original data set to obtain an anonymous data set, and determine an initial feature set based on the anonymous data set; evaluate the importance of the features in the initial feature set to obtain an importance evaluation result; add random noise to the importance evaluation result, and determine a feature subset from the initial feature set based on the importance evaluation result after adding random noise; train the advertising delivery model based on the feature subset to obtain a trained advertising delivery model, and use the trained advertising delivery model to execute the advertising delivery processing process.

[0042] In this way, when all steps in the model training method and the advertisement delivery processing method provided by the exemplary embodiment of the present disclosure can be executed by the terminal device, all steps can be executed by the processor of the terminal device.

[0043] The terminal device can be an intelligent device with data processing capabilities, such as monitoring equipment, smart phones, computers, tablet computers, vehicle-mounted devices, wearable devices and other intelligent devices. The terminal device can also be called a mobile terminal, terminal, mobile device, etc. The present disclosure does not limit the type of terminal device.

[0044] Furthermore, the technical solution provided by the exemplary embodiment of the present disclosure may also be executed by a server. Correspondingly, in this mode of execution by the server, the server may start executing the steps in the technical solution of the exemplary embodiment of the present disclosure in response to a trigger instruction, wherein the trigger instruction may be sent by a terminal device used by a user, or may be triggered locally by the server in response to some automation events. The server may be a background system that provides related services in the exemplary embodiment of the present disclosure, and may include one electronic device with computing functions such as a portable computer, a desktop computer, a smart phone, or a cluster formed by multiple electronic devices.

[0045] In addition, the technical solution of the exemplary embodiment of the present disclosure can also be executed collaboratively by the terminal device and the server. In this way of collaborative execution by the electronic device and the server, some steps in the technical solution provided by the exemplary embodiment of the present disclosure are executed by the terminal device, while other steps are executed by the server. For example, the technical solution provided by the exemplary embodiment of the present disclosure can be executed by the server in the model training phase, and the trained advertising delivery model is sent to the terminal device, and the terminal device executes the model application phase, that is, the trained advertising delivery model is used to perform advertising delivery processing, and the advertisements are targeted and recommended to the target group.

[0046] It should be noted that in this mode of collaborative execution by the terminal device and the server, the steps respectively executed by the terminal device and the server can be dynamically adjusted according to actual conditions, and the exemplary embodiments of the present disclosure do not impose special restrictions on this.

[0047] In the field of advertising through the Internet, promoters hope to recommend advertisements to the target group in a targeted manner. Accurate advertising is the key. In the current related fields, there are methods of selecting features that have a greater impact on advertising through performance statistics or machine learning theory, and relying on the knowledge of domain experts when necessary. There are also methods of discovering potential features through traditional machine learning algorithms, and providing reference opinions on advertising by predicting user behavior and advertising click-through rate. There are also methods of recommending advertisements to users based on their historical behavior and preferences through collaborative filtering, content recommendation and other technologies.

[0048] However, there are currently problems with user data leakage, illegal acquisition or abuse of data during the model training phase and when the model is applied for advertising recommendation. For example, when processing user data using traditional feature selection methods and machine learning algorithms, the user's original information, such as browsing history, purchase behavior, etc., will be directly exposed, increasing the risk of leaking user privacy. Correspondingly, during the advertising process, user data will be shared between advertising platforms and advertisers to optimize the advertising effect, and data sharing can also cause security issues such as data leakage. Therefore, there is an urgent need for a solution to improve the accuracy of advertising delivery models while ensuring data security.

[0049] Exemplarily, the advertising delivery model of the exemplary embodiment of the present disclosure may be constructed using lightweight neural networks, such as CNN (Convolutional Neural Network), RNN (Recurrent Neural Networks), DNN (Deep Neural Networks), etc. Compared with lightweight neural networks, an advertising delivery model with a more complex network structure may also be adopted, such as one constructed using a heavyweight neural network, and any heavyweight neural network that can be used for advertising delivery processing. Of course, the advertising delivery model of the exemplary embodiment of the present disclosure may also be a large model at the billion level, which refers to a deep learning model trained using a large amount of sample data. The exemplary embodiment of the present disclosure takes the training of a deep learning model as an advertising delivery model as an example.

[0050] The exemplary embodiment of the present disclosure first provides a model training method. Figure 2 As shown, the model training method of the embodiment of the present disclosure may include the following steps S210 and S240:

[0051] In step S210, the original data set is anonymized to obtain an anonymous data set, and an initial feature set is determined based on the anonymous data set.

[0052] In the exemplary embodiments of the present disclosure, the original data set is a data set used for model training, which can be obtained from various channels or platforms, including but not limited to user behavior data, social data, and geographic location information, etc., without limitation. Anonymization processing refers to processing the data in the original data set by masking, generalization, shuffling, etc., deleting or hiding user sensitive information of the data in the original data set. The anonymous data after anonymization processing is difficult for attackers to de-anonymize. At the same time, anonymization processing will retain the useful information of the data in the original data set to the maximum extent.

[0053] Among them, determining the initial feature set based on the anonymous data set can be to extract the required features from the anonymous data set according to the model training requirements to construct the initial feature set, or the anonymous data set can be directly subjected to feature processing such as feature encoding to obtain the initial feature set. The exemplary embodiments of the present disclosure do not impose any special restrictions on the implementation method of constructing the initial feature set.

[0054] It should be noted that in the technical solution disclosed herein, the collection / collection, updating, analysis, use, transmission, storage and other aspects of user personal information involved are in compliance with the provisions of relevant laws and regulations, are used for legal and reasonable purposes, are not shared, leaked or sold outside of these legal uses, and are subject to supervision and management by national regulatory authorities. Necessary measures should be taken for user personal information to selectively block the use or access to personal information data to prevent illegal access to such personal information data, ensure that persons who have access to personal information data comply with the provisions of relevant laws and regulations, and ensure the security of user personal information. In addition, once these user personal information data are no longer needed, the risks should be minimized by limiting or even prohibiting data collection and / or deleting data.

[0055] In step S220, the importance of the features in the initial feature set is evaluated to obtain an importance evaluation result.

[0056] In an exemplary embodiment of the present disclosure, importance assessment is an evaluation process of the degree of influence of features in the initial feature set on the effect of advertising delivery, and the importance of features in the initial feature set can be evaluated based on an importance assessment model. The importance assessment model can be a pre-trained machine learning model, such as a linear model, a regularization model, and a tree model, etc. The exemplary embodiment of the present disclosure does not impose any special restrictions on the type of importance assessment model. The importance assessment result can be an importance score, and the importance score is used to highly characterize the degree of influence of the feature on the effect of advertising delivery.

[0057] In step S230, random noise is added to the importance evaluation result, and a feature subset is determined from the initial feature set according to the importance evaluation result after adding the random noise.

[0058] In an exemplary embodiment of the present disclosure, adding random noise to the importance evaluation result is to further ensure the security of the data while retaining the useful information of the importance evaluation result to the maximum extent. Among them, a preset noise distribution can be used to add random noise to the importance evaluation result, such as Laplace distribution, normal distribution, uniform distribution, Gaussian distribution, etc., without limitation. For example, random noise can be obtained by sampling from a Laplace distribution and added to the importance evaluation result.

[0059] Based on the importance evaluation results after adding random noise, a feature subset is determined from the initial feature set to ensure data security in the feature screening process, screen out features that have a significant impact on the advertising effect, provide high-quality feature input for training the advertising model, and improve the model training accuracy.

[0060] In step S240, the advertisement delivery model is trained based on the feature subset to obtain a trained advertisement delivery model.

[0061] In an exemplary embodiment of the present disclosure, after obtaining a feature subset, the feature subset is used to train an advertisement delivery model to obtain a trained advertisement delivery model. The advertisement delivery model is used to predict response information to advertisements, and the response information is used to guide the delivery of advertisements. The advertisement delivery model may be a deep learning model, such as DNN, CNN, or RNN, etc., and there is no special limitation on this.

[0062] In the technical solution provided by the embodiments of the present disclosure, on the one hand, the original data set is anonymized to obtain an anonymous data set, and the initial feature set is determined based on the anonymous data set, which can avoid the presence of data information that leaks user privacy in the initial feature set. On the other hand, the importance of the features in the initial feature set is evaluated to obtain an importance evaluation result, random noise is added to the importance evaluation result, and a feature subset is determined from the initial feature set based on the importance evaluation result after adding random noise. The key features for model training can be extracted to train the advertising delivery model, improve the accuracy of model training, and add random noise to increase data privacy security, further avoiding privacy leakage. In summary, the exemplary embodiments of the present disclosure can improve the accuracy of the advertising delivery model while ensuring the security of user data.

[0063] In an exemplary embodiment, a method for implementing a training advertisement delivery model is provided. Figure 3 As shown, training the advertisement delivery model based on the feature subset to obtain the trained advertisement delivery model may include step S310 and step S320:

[0064] Step S310: Obtaining the gradient of the network parameters of the advertisement delivery model based on the feature subset.

[0065] Before training the advertising delivery model, the input layer, hidden layer and output layer of the advertising delivery model can be determined so that the dimension of the input layer matches the number of features of the feature subset, and the activation function, loss function and optimizer of the advertising delivery model are set. The exemplary embodiment of the present disclosure does not impose any special restrictions on the above setting information of the advertising delivery model. For example, the loss function can be a cross entropy loss, etc. Among them, the feature subset can also be divided into a training set, a validation set and a test set, and the specific division method is also not limited.

[0066] Among them, the gradient of the network parameters of the advertising delivery model refers to the derivative of the loss function with respect to the model, which is used to reflect the rate of change of the loss function in the parameter space, that is, to reflect the adjustment of the network parameters and the change of the loss function, thereby guiding the direction and size of the network parameter update and making the advertising delivery model perform better on the training set.

[0067] In an exemplary embodiment of the present disclosure, a training set may be used to train an advertisement delivery model, prediction values ​​may be calculated through forward propagation, and model parameters may be updated through a back propagation algorithm to minimize a loss function.

[0068] Step S320: adding noise to the gradient of the network parameters, and using the gradient of the network parameters after adding the noise to update the parameters of the advertisement delivery model.

[0069] Noise can be added to the gradient of network parameters or a preset noise distribution can be used to add random noise to the importance evaluation result to ensure data security during training. The noise distribution can be selected according to actual needs and there is no restriction on this.

[0070] In each batch of training, random noise is added to the gradient of the network parameters, so as to update the parameters of the advertising delivery model using the gradient of the network parameters after adding the noise.

[0071] In some optional embodiments, before adding noise to the gradient of the network parameter, the method may further include: clipping the gradient of the network parameter based on a preset gradient clipping threshold.

[0072] Among them, gradient clipping is to limit the gradient size by clipping the gradient to avoid the gradient being too large or too small, which ensures the stability of model training and helps to control the amount of noise added to the gradient.

[0073] As an example, Figure 4 A flowchart of a training advertising delivery model is shown below. Figure 4 The training process of the advertising delivery model is described. The input of the model training may include but is not limited to: training data set D, privacy budget ε, advertising delivery model (such as deep learning model) M, learning rate η, gradient clipping threshold C, and noise standard deviation σ.

[0074] Step S410: Initialize model parameters θ.

[0075] Step S420: In each training batch, the gradient of the network parameters of the advertising delivery model is obtained, and the gradient of the network parameters is clipped based on a preset gradient clipping threshold, and noise is added to the clipped gradient of the network parameters to update the parameters of the advertising delivery model using the gradient of the network parameters after adding noise.

[0076] Specifically, Figure 5 As shown, for each training batch B∈D, the following steps can be included:

[0077] Step S510: Calculate the gradient of the network parameters corresponding to the training batch. For example, the gradient of the network parameters is calculated by formula (1):

[0078] g=▽ θ L(B;θ) (1)

[0079] Among them, g is the gradient of network parameters, L is the loss function, ▽ θ represents the gradient with respect to θ.

[0080] Step S520: Clipping the gradient of the network parameters.

[0081] The gradient of the network parameters can be clipped by the following formula (2):

[0082] g'=g / max(1,||g|| / C) (2)

[0083] Among them, g' is the gradient of the network parameters after clipping, || || is the norm, and the absolute value can be taken here.

[0084] Step S530: Add noise to the gradient of the pruned network parameters.

[0085] Noise can be added using the following formula (3):

[0086]

[0087] in, is the gradient of the network parameters after adding noise, and N is a normal distribution.

[0088] Step S540: update the parameters of the advertisement delivery model using the gradient of the network parameters after adding noise.

[0089] The parameters can be updated using the following formula (4):

[0090]

[0091] The training process of one training batch is completed through step S510 to step S540, and steps S510 to step S540 are repeatedly executed until the model training stop condition is met, for example, a preset number of training rounds is reached or the loss function converges, and there is no restriction on this.

[0092] It should be noted that the formulas and noise distribution information involved in the above training process are only examples and can be flexibly adjusted according to the specific training process.

[0093] The exemplary embodiments of the present disclosure can ensure the data security of the training process by adding random noise to the gradient of the network parameters during the model training process, and can increase the stability of the model training by clipping the gradient of the network parameters to avoid the gradient being too large or too small. Since the training data set comes from a screened feature subset, the training accuracy of the model can also be improved.

[0094] In an exemplary embodiment, a method for obtaining an anonymous data set is provided. Figure 6 As shown, performing anonymization processing on the original data set to obtain an anonymous data set may include steps S610 to S630:

[0095] Step S610: Obtain a preset privacy budget and noise distribution, and determine a first noise scale according to the privacy budget.

[0096] The privacy budget can determine the size of the noise. The smaller the privacy budget, the greater the amount of noise added, the stronger the privacy protection, and the accuracy of the model may be reduced. Correspondingly, the larger the privacy budget, the smaller the amount of noise added, and the higher the accuracy of the model, but the weaker the privacy protection. An appropriate privacy budget can be selected according to the needs of differential privacy. When selecting a privacy budget, the sensitivity of the data, the accuracy requirements of the model, and the legal and compliance requirements need to be considered. The exemplary embodiments of the present disclosure can determine the privacy budget according to actual conditions.

[0097] The selection method of noise distribution can refer to the selection method of adding random noise mentioned above, and there is no special restriction on this. For example, the noise distribution can be selected as Laplace distribution. The first noise scale refers to the amount of noise added, which is controlled by the privacy budget. After determining the privacy budget, the first noise scale is obtained accordingly.

[0098] Step S620: adding random noise to the data in the original data set based on the noise distribution and the first noise scale to obtain interference data.

[0099] After determining the noise distribution and the first noise scale, random noise can be added to the original data set according to the random noise obtained from the noise distribution to obtain interference data. It should be understood that by selecting a suitable noise distribution and the first noise scale, it is necessary to ensure that adding or subtracting a piece of data in the original data set does not have a significant impact on the query results, ensuring data security while avoiding affecting the accuracy of the original data set for the training model.

[0100] As an example, for the original data set, random noise is added to each numerical data (x) in the original data set using the following formula:

[0101] C=(Lap(△f / ε)) (5)

[0102] Wherein, △f is the sensitivity of the query function (f), which is used to indicate the maximum variation range of the query result when a data record in the original data set changes. Lap is the Laplace distribution, and C is the random noise. The interference data after adding noise is: x'=x+Lap(△f / ε).

[0103] Step S630: Determine an anonymous data set according to the interference data.

[0104] An anonymous data set is determined based on each noise-processed interference data.

[0105] The exemplary embodiments of the present disclosure anonymize the original data set so that the anonymous data set will not leak private data, thereby providing more secure sample data for the entire model training process.

[0106] In an exemplary embodiment, a method for implementing a screening feature is also provided. Figure 7 As shown, adding random noise to the importance evaluation result, and determining a feature subset from the initial feature set according to the importance evaluation result after adding the random noise may include step S710 and step S720:

[0107] Step S710: adding random noise to the importance assessment result based on a preset privacy budget to obtain the importance assessment result after adding the random noise.

[0108] The importance evaluation result of a feature can reflect the contribution of the feature to the prediction of the target variable, that is, the impact of the feature on the advertising effect can be determined by evaluating the importance of the feature. As mentioned above, the privacy budget affects the scale of adding random noise, so random noise can be added to the importance evaluation result according to the privacy budget to obtain the importance evaluation result after adding random noise.

[0109] In some optional embodiments, adding random noise to the importance assessment result based on a preset privacy budget may include:

[0110] First, the sensitivity information of the query function corresponding to the importance assessment model is obtained. Secondly, the second noise scale is determined according to the preset privacy budget and sensitivity information. Finally, random noise is added to the importance assessment result based on the second noise scale.

[0111] The importance assessment model corresponds to the sensitivity information of the query function, which is used to indicate the maximum range of change of the query result when the importance assessment result changes. The preset privacy budget can be divided by the sensitivity information to determine the second noise scale. Of course, it can also be determined by other calculation methods, and there is no special limitation on this.

[0112] It should be noted that the noise distribution of the random noise added here can also be selected according to needs. The specific method of adding noise can refer to the content of adding random noise in the above scenario, which will not be repeated here.

[0113] Step S720: Compare the importance evaluation result after adding random noise with a preset evaluation threshold, and determine a feature subset from the features of the initial feature set according to the comparison result.

[0114] The preset evaluation threshold is a threshold used to filter the importance evaluation results. After obtaining the importance evaluation result after adding noise, by comparing the evaluation result with the preset evaluation threshold, the features corresponding to the evaluation results greater than the preset evaluation threshold can be obtained to determine the feature subset.

[0115] As an example, the process of obtaining a feature subset is described by taking the initial feature set as F and the privacy budget as ε.

[0116] First, the initial feature subset S is an empty set.

[0117] Secondly, for each feature in the initial feature set, the importance score S_f of the feature is calculated by the importance evaluation model, and the Laplace noise scale (a second noise scale) is calculated based on the privacy budget ε and the sensitivity Δf of the query function. Then, Laplace noise is added to the importance score S_f according to the Laplace noise to obtain the noisy importance score S_f'.

[0118] Finally, all the importance scores S_f' after adding noise are sorted from large to small, and the top k features with the highest scores are selected from the sorting results and added to the feature subset S. The k value can be flexibly set according to the actual scenario and requirements.

[0119] In some optional embodiments, the screened feature subset can be used to train the model and verify the trained advertising delivery model, so as to verify the effect of the advertising delivery model on advertising delivery, that is, to verify the performance of the advertising delivery model. Then, according to the verification results, the privacy budget involved in the above steps can be adjusted or the feature selection process can be re-executed to achieve a balance between ensuring model performance and protecting privacy, thereby improving the accuracy of subsequent advertising delivery processing based on the advertising delivery model.

[0120] When screening features, the exemplary embodiment of the present disclosure can ensure the data security of the feature screening process while effectively screening feature subsets that have a significant impact on advertising delivery effects by calculating the importance scores of the features and adding random noise to the importance evaluation.

[0121] In an exemplary embodiment, a method for implementing data preprocessing is also provided, which may include:

[0122] First, the initial data set is cleaned, then the initial data set after data cleaning is standardized to obtain standardized data, and finally the standardized data is enhanced to obtain the original data set.

[0123] Specifically, data cleaning can be the use of big data processing and other technologies to identify and remove duplicate, erroneous or invalid information in the data to improve the accuracy of the data set. Standardization of the initial data set refers to normalization, discretization, feature encoding, data transformation, principal component analysis and other operations on the data in the initial data set to obtain standardized data to provide a standardized data basis for subsequent model training.

[0124] Among them, for the positive and negative samples in the initial data set, the imbalance of positive and negative samples can be avoided and the diversity of data can be increased based on data enhancement processing. For example, data enhancement can be performed through upsampling or downsampling. The exemplary embodiments of the present disclosure do not impose specific restrictions on the above-mentioned data preprocessing methods.

[0125] Furthermore, the exemplary embodiments of the present disclosure also provide an advertisement delivery processing method, such as Figure 8 As shown, the advertisement delivery process may include:

[0126] Step S810: Obtain advertisement information to be processed.

[0127] Step S820: Input the advertisement information to be processed into the trained advertisement delivery model to obtain the processing result of the advertisement information to be processed. The trained advertisement delivery model is obtained by training the advertisement delivery model based on the model training method of any one of the above exemplary embodiments. Step S830: Perform advertisement delivery processing based on the processing result.

[0128] The pending advertisement information is the relevant information of the pending advertisement. The pending advertisement information is processed by the advertisement delivery model, and the processing result is used to reflect the user's response information to the pending advertisement, so as to make a decision on whether to deliver the pending advertisement based on the response information.

[0129] It should be noted that the training process of the advertisement delivery model can be referred to the above exemplary embodiments and will not be described in detail here.

[0130] In an exemplary embodiment, performing advertisement delivery processing based on the processing result may include:

[0131] Obtaining target data corresponding to the processing result, where the target data at least includes advertising delivery data and an original data set used to train an advertising delivery model;

[0132] Compliance checks are performed on the original data set and the advertisement delivery data, so as to process the processed results for advertisement delivery according to the compliance check results.

[0133] Among them, the target data corresponding to the processing result refers to the advertising data of the advertisements to be placed corresponding to the processing result, and the original data set used to train the advertising delivery model. That is, in order to further avoid data leakage in the process of training the model and applying the model for advertising delivery, the data of these two stages can also be obtained for compliance testing.

[0134] Compliance detection can be implemented through the compliance detection module of the advertising system. Compliance detection may include at least one of the following: for example, according to the attributes of the data involved in the above two stages, determine the data source, purpose, storage location and sharing status (such as whether it can be used for model training) and other information, and determine whether the data is correctly applied to the corresponding scenario based on this information. For another example, the privacy policy statement and other documents involved in the advertising system can be extracted, and the commitment information on the collection, use, storage and sharing of user data can be obtained through text recognition or extraction technology, and the commitment information can be compared with the data processing practices involved in the above two stages (such as whether it is used in appropriate scenarios) to determine whether there is a risk of data leakage based on the comparison results. For another example, the data involved in the data processing process in the advertising system can be checked based on technical means such as static code analysis and dynamic monitoring to determine whether there is a risk of data leakage.

[0135] By performing compliance checks on the target data of the processing results, we can ensure the compliance of data use and avoid data leakage throughout the entire process of model training and advertising processing, thereby improving the security of the entire advertising solution and enhancing the user experience.

[0136] Based on the aforementioned exemplary embodiments, a compliance detection file may also be generated according to the compliance detection result, and the compliance detection file includes at least potential risk information and location information corresponding to the potential risk information.

[0137] Among them, a compliance inspection report can be generated, and the issues involved in the inspection process and their potential risks can be listed in detail to facilitate relevant personnel to locate and solve problems.

[0138] In the technical solution provided by the embodiments of the present disclosure, on the one hand, the original data set is anonymized to obtain an anonymous data set, and the initial feature set is determined based on the anonymous data set, which can avoid the presence of data information that leaks user privacy in the initial feature set. On the other hand, the importance of the features in the initial feature set is evaluated to obtain an importance evaluation result, random noise is added to the importance evaluation result, and a feature subset is determined from the initial feature set based on the importance evaluation result after adding random noise, which can extract key features for model training to train the advertising delivery model and improve the accuracy of model training. The addition of random noise increases data privacy security and further avoids privacy leakage. On the other hand, data preprocessing can avoid the problem of unbalanced training samples and further improve the accuracy of model training. In summary, the exemplary embodiments of the present disclosure can improve the accuracy of the advertising delivery model while ensuring the security of user data.

[0139] It should be noted that the above figures are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present disclosure, and are not intended to be limiting. It is easy to understand that the processes shown in the above figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0140] For further reference, Fig. 9 As shown, in an exemplary embodiment of the present disclosure, a model training device 900 is provided, comprising an anonymity processing module 910, a feature evaluation module 920, a feature selection module 930 and a model training module 940. Among them:

[0141] The anonymity processing module 910 is used to anonymize the original data set to obtain an anonymous data set, and determine the initial feature set based on the anonymous data set; the feature evaluation module 920 is used to evaluate the importance of the features in the initial feature set to obtain an importance evaluation result; the feature selection module 930 is used to add random noise to the importance evaluation result, and determine a feature subset from the initial feature set based on the importance evaluation result after adding random noise; the model training module 940 is used to train the advertising delivery model based on the feature subset to obtain a trained advertising delivery model.

[0142] In an exemplary embodiment of the present disclosure, the model training module 940 is configured to perform: obtaining the gradient of the network parameters of the advertising delivery model based on the feature subset; adding noise to the gradient of the network parameters, and updating the parameters of the advertising delivery model using the gradient of the network parameters after adding noise.

[0143] In an exemplary embodiment of the present disclosure, the model training module 940 is further configured to perform: before adding noise to the gradient of the network parameter, clipping the gradient of the network parameter based on a preset gradient clipping threshold.

[0144] In an exemplary embodiment of the present disclosure, the anonymity processing module 910 is configured to perform: obtaining a preset privacy budget and noise distribution, and determining a first noise scale based on the privacy budget; adding random noise to the data in the original data set based on the noise distribution and the first noise scale to obtain interference data; and determining an anonymous data set based on the interference data.

[0145] In an exemplary embodiment of the present disclosure, the feature selection module 930 is configured to perform: adding random noise to the importance assessment result based on a preset privacy budget to obtain the importance assessment result after adding the random noise; comparing the importance assessment result after adding the random noise with a preset assessment threshold, and determining a feature subset from the features of the initial feature set based on the comparison result.

[0146] In an exemplary embodiment of the present disclosure, the importance of features in an initial feature set is evaluated based on an importance assessment model; the feature selection module 930 is configured to execute: obtaining sensitivity information of a query function corresponding to the importance assessment model; determining a second noise scale based on a preset privacy budget and sensitivity information; and adding random noise to the importance assessment result based on the second noise scale.

[0147] In an exemplary embodiment of the present disclosure, the model training device 900 also includes a data preprocessing module, which is configured to perform: before anonymizing the original data set to obtain an anonymous data set, performing data cleaning on the initial data set; standardizing the initial data set after data cleaning to obtain standardized data; and performing data enhancement processing on the standardized data to obtain the original data set.

[0148] Since the details of each functional module of the model training device of the exemplary embodiment of the present disclosure have been recorded in the exemplary embodiment of the above-mentioned model training method, they will not be repeated here.

[0149] For further reference, Fig.10 As shown, in an exemplary embodiment of the present disclosure, an advertisement delivery processing device 1000 is provided, comprising an information acquisition module 1010, a first processing module 1020 and a second processing module 1030. Among them:

[0150] The information acquisition module 1010 is used to acquire the advertisement information to be processed;

[0151] The first processing module 1020 is used to input the advertisement information to be processed into the trained advertisement delivery model to obtain the processing result of the advertisement information to be processed;

[0152] The second processing module 1030 is used to perform advertisement delivery processing based on the processing result.

[0153] The trained advertisement delivery model is obtained by training the advertisement delivery model based on any one of the methods described in the above exemplary embodiments.

[0154] In an exemplary embodiment of the present disclosure, the second processing module 1030 is configured to execute: obtaining target data corresponding to the processing result, the target data at least including advertising delivery data and an original data set for training an advertising delivery model; performing compliance checking on the original data set and the advertising delivery data, so as to perform advertising delivery processing on the processing result according to the compliance checking result.

[0155] In an exemplary embodiment of the present disclosure, the advertisement delivery processing device 1000 further includes a file generation module, which is configured to execute: generating a compliance detection file according to the compliance detection result, the compliance detection file at least including potential risk information and location information corresponding to the potential risk information.

[0156] Since the details of the various functional modules of the advertisement delivery processing device of the exemplary embodiment of the present disclosure have been described in the exemplary embodiment of the advertisement delivery processing method described above, they will not be repeated here.

[0157] It should be noted that although several modules or units of the model training device and the advertisement delivery processing device are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be concretized in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units for concretization.

[0158] The exemplary embodiments of the present disclosure also provide a computer program product, which includes a computer program, and when the computer program is executed by a processor, the above method is implemented.

[0159] In one embodiment, the computer program product may be a tangible product containing a computer program, such as a computer-readable storage medium storing a computer program. The readable storage medium may be a storage medium based on electrical, magnetic, optical, electromagnetic, infrared, or other signals, including but not limited to: random access memory (RAM), read-only memory (ROM), magnetic tape, floppy disk, flash memory (Flash), mechanical hard disk (HDD), solid-state drive (SSD), and the like. Exemplarily, the computer program product may be implemented as a non-volatile storage medium storing a computer program, such as a read-only memory, a NAND flash memory, and the like.

[0160] In one embodiment, the computer program product may be an intangible product including a computer program. Exemplarily, the computer program product may be implemented as a virtual digital product, such as a digital file storing an executable file, an installation package, etc. of the computer program.

[0161] The code of the computer program can be written in one or more programming languages. Programming languages ​​such as C language, Java, C++, etc. The program code can be executed entirely on the user computing device, or partially on the user computing device, or as a separate software package, or partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, such as a local area network (LAN), a wide area network (WAN), etc., or can be connected to an external computing device (e.g., an Internet connection provided by an operator).

[0162] The computer program may be carried or transmitted via electrical, magnetic, optical, electromagnetic, infrared, or other signals. The electronic device may convert the signal carrying the computer program into a digital signal, thereby running the computer program. When the computer program is run on an electronic device, its code is used to enable the electronic device to execute (more specifically, the processor of the electronic device may execute) the method steps of various exemplary embodiments of the present disclosure, such as the above-mentioned method.

[0163] In addition, in an exemplary embodiment of the present disclosure, an electronic device capable of implementing the above method is also provided. It will be appreciated by those skilled in the art that various aspects of the present disclosure may be implemented as a system, method, or program product. Therefore, various aspects of the present disclosure may be specifically implemented in the following forms, namely: a complete hardware embodiment, a complete software embodiment (including firmware, microcode, etc.), or an embodiment combining hardware and software aspects, which may be collectively referred to herein as a "circuit", "module", or "system".

[0164] Refer to the following Fig.111100 according to this embodiment of the present disclosure is described. Fig.11 The electronic device 1100 shown is only an example and should not limit the functions and scope of use of the embodiments of the present disclosure. Fig.11 As shown, the electronic device 1100 is in the form of a general computing device. The components of the electronic device 1100 may include, but are not limited to: the at least one processing unit 1110, the at least one storage unit 1120, a bus 1130 connecting different system components (including the storage unit 1120 and the processing unit 1110), and a display unit 1140.

[0165] The storage unit stores program codes, which can be executed by the processing unit 1110, so that the processing unit 1110 executes the steps according to various exemplary embodiments of the present disclosure described in the above “Exemplary Method” section of this specification.

[0166] The storage unit 1120 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 1121 and / or a cache storage unit 1122, and may further include a read-only storage unit (ROM) 1123. The storage unit 1120 may also include a program / utility 1124 having a set (at least one) of program modules 1125, such program modules 1125 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination thereof may include the implementation of a network environment.

[0167] Bus 1130 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0168] The electronic device 1100 may also communicate with one or more external devices 1200 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 1100, and / or communicate with any device that enables the electronic device 1100 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 1150. Furthermore, the electronic device 1100 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 1160. As shown, the network adapter 1160 communicates with other modules of the electronic device 1100 via a bus 1130. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 1100, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0169] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the embodiment of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the embodiment of the present disclosure.

[0170] In addition, the above-mentioned figures are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present disclosure, and are not intended to be limiting. It is easy to understand that the processes shown in the above-mentioned figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0171] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure. The description and examples are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the claims.

Claims

1. A model training method, characterized in that: include: Anonymizing the original data set to obtain an anonymous data set, and determining an initial feature set based on the anonymous data set; Evaluate the importance of features in the initial feature set to obtain an importance evaluation result; Adding random noise to the importance evaluation result, and determining a feature subset from the initial feature set according to the importance evaluation result after adding the random noise; The advertisement delivery model is trained based on the feature subset to obtain a trained advertisement delivery model.

2. The method according to claim 1, characterized in that The step of training the advertisement delivery model based on the feature subset to obtain a trained advertisement delivery model includes: Obtaining a gradient of a network parameter of the advertisement delivery model based on the feature subset; Noise is added to the gradient of the network parameter, and the parameters of the advertisement delivery model are updated using the gradient of the network parameter after the noise is added.

3. The method according to claim 2, characterized in that Before adding noise to the gradient of the network parameters, the method further comprises: The gradient of the network parameter is clipped based on a preset gradient clipping threshold.

4. The method according to claim 1, characterized in that The anonymization process of the original data set to obtain an anonymous data set includes: Obtaining a preset privacy budget and noise distribution, and determining a first noise scale according to the privacy budget; Adding random noise to the data in the original data set based on the noise distribution and the first noise scale to obtain interference data; The anonymous data set is determined based on the interference data.

5. The method according to claim 4, characterized in that The adding random noise to the importance evaluation result, and determining a feature subset from the initial feature set according to the importance evaluation result after adding the random noise, comprises: Adding random noise to the importance assessment result based on the preset privacy budget to obtain the importance assessment result after adding the random noise; The importance evaluation result after adding random noise is compared with a preset evaluation threshold, and the feature subset is determined from the features of the initial feature set according to the comparison result.

6. The method according to claim 5, characterized in that An evaluation of the importance of features in the initial feature set based on an importance evaluation model; The adding random noise to the importance evaluation result based on the preset privacy budget includes: Obtaining sensitivity information of the query function corresponding to the importance assessment model; Determining a second noise scale according to the preset privacy budget and the sensitivity information; Random noise is added to the significance evaluation result based on the second noise scale.

7. The method according to any one of claims 1 to 6, characterized in that Before anonymizing the original data set to obtain the anonymous data set, the method further includes: Perform data cleaning on the initial data set; The initial data set after data cleaning is standardized to obtain standardized data; The standardized data is subjected to data enhancement processing to obtain the original data set.

8. An advertisement delivery processing method, characterized in that: include: Get the advertisement information to be processed; Inputting the advertisement information to be processed into the trained advertisement delivery model to obtain a processing result of the advertisement information to be processed; Performing advertisement delivery processing based on the processing result; The trained advertising delivery model is obtained by training the advertising delivery model based on the method described in any one of claims 1 to 7.

9. The method according to claim 8, characterized in that The advertisement delivery processing based on the processing result includes: Acquire target data corresponding to the processing result, wherein the target data at least includes advertisement delivery data and an original data set used to train the advertisement delivery model; Compliance detection is performed on the original data set and the advertisement delivery data, so as to perform advertisement delivery processing on the processing result according to the compliance detection result.

10. The method according to claim 9, characterized in that The method further comprises: A compliance detection file is generated according to the compliance detection result, and the compliance detection file at least includes potential risk information and location information corresponding to the potential risk information.

11. A model training device, characterized in that: include: Anonymization processing module, used to perform anonymization processing on the original data set to obtain an anonymous data set, and determine an initial feature set based on the anonymous data set; A feature evaluation module, used to evaluate the importance of features in the initial feature set to obtain an importance evaluation result; A feature selection module, used for adding random noise to the importance evaluation result, and determining a feature subset from the initial feature set according to the importance evaluation result after adding the random noise; The model training module is used to train the advertisement delivery model based on the feature subset to obtain a trained advertisement delivery model.

12. An advertisement delivery processing device, characterized in that: include: An information acquisition module, used to acquire advertisement information to be processed; A first processing module, used for inputting the advertisement information to be processed into the trained advertisement delivery model to obtain a processing result of the advertisement information to be processed; A second processing module, used for performing advertisement delivery processing based on the processing result; The trained advertising delivery model is obtained by training the advertising delivery model based on the method described in any one of claims 1 to 7.

13. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 10 is implemented.

14. An electronic device, characterized in that: include: processor; as well as A memory, configured to store executable instructions of the processor; The processor is configured to perform the method of any one of claims 1 to 10 by executing the executable instructions.

Images