Advertisement information pushing method and device, electronic equipment and storage medium

Abstract

The invention provides an advertisement information pushing method and apparatus, an electronic device and a storage medium. The method comprises the steps of obtaining a multi-dimensional object feature of a target object; inputting the multi-dimensional object features into the trained coding network to obtain a target coding vector output by the trained coding network; inputting the target coding vector and the first price reduction variable into the trained prediction network to obtain a first target probability output by the trained prediction network; inputting the target coding vector and the second price reduction variable into the trained prediction network to obtain a second target probability output by the trained prediction network; and based on the first target probability and the second target probability, determining whether to push the target price reduction advertisement information to the target object. According to the method, price reduction sensitive users can be intelligently screened to perform price reduction advertisement pushing on the users, and the price reduction marketing efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of information push technology, and in particular to an advertising information push method, device, electronic device, and storage medium. Background Technology

[0002] In financial marketing scenarios, increasing user credit adoption is key to business growth. One related technology involves using price reduction strategies to increase user adoption (i.e., "credit adoption") of credit products. For example, precise adjustments to interest rates or fees can attract more users to use credit services. Intuitively, the larger the price reduction, the more attractive it is to users and the higher the credit adoption rate. However, price reduction strategies also incur significant costs, and it's impossible to lower prices to the lowest level for all users indefinitely. Therefore, how to identify potential price-sensitive users and push price reductions to them is a crucial problem that needs to be solved to maximize the efficiency of price reduction marketing. Summary of the Invention

[0003] This application provides a method, apparatus, electronic device, and storage medium for pushing advertising information, which can intelligently filter users sensitive to price reductions and push price reduction advertisements to them, thereby improving the efficiency of price reduction marketing. The technical solution is as follows: According to one aspect of this application, a method for pushing advertising information is provided, the method comprising: Obtain multidimensional object characteristics of the target object; The multidimensional object features are input into the trained encoding network to obtain the target encoding vector output by the trained encoding network. The target encoding vector and the first price reduction variable are input into the trained prediction network to obtain the first target probability output by the trained prediction network. The first target probability indicates the prediction confidence rate of the target object under the first price reduction variable. The target encoding vector and the second price reduction variable are input into the trained prediction network to obtain the second target probability output by the trained prediction network. The second target probability indicates the prediction confidence rate of the target object under the second price reduction variable. Based on the first target probability and the second target probability, it is determined whether to push target price reduction advertisement information to the target object.

[0004] According to another aspect of this application, an advertising information push device is provided, the device comprising: The first acquisition module is used to acquire the multidimensional object features of the target object; The first encoding module is used to input the multidimensional object features into the trained encoding network to obtain the target encoding vector output by the trained encoding network. The first prediction module is used to input the target encoding vector and the first price reduction variable into the trained prediction network to obtain the first target probability output by the trained prediction network. The first target probability indicates the prediction confidence rate of the target object under the first price reduction variable. The second prediction module is used to input the target encoding vector and the second price reduction variable into the trained prediction network to obtain the second target probability output by the trained prediction network. The second target probability indicates the prediction confidence rate of the target object under the second price reduction variable. The information push module is used to determine whether to push target price reduction advertisement information to the target object based on the first target probability and the second target probability.

[0005] According to one aspect of this application, an electronic device is provided, comprising: a processor and a memory storing a program, the program including instructions that, when executed by the processor, cause the processor to perform the advertising information push method as described above.

[0006] According to another aspect of this application, a non-transitory computer-readable storage medium is provided that stores computer instructions for causing the computer to perform the advertising information push method described above.

[0007] According to another aspect of this application, a computer program product is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the aforementioned advertising information push method.

[0008] The beneficial effects of the technical solutions provided in this application include at least the following: By inputting the multidimensional features of the target object into the encoding network, a target encoding vector is obtained. The target encoding vector, along with the first and second price reduction variables, are then input into the prediction network to predict the acceptance rate. Based on the predicted first and second target probabilities, it is determined whether to push the target price reduction advertisement information to the target object. By using the target encoding vector output by the shared encoding network, the acceptance probability is predicted under whether the price reduction is accepted. Compared with related technologies that use two models for probability prediction, this avoids subsequent prediction errors caused by encoding vector extraction errors, further improving the prediction accuracy of price reduction gain. Attached Figure Description

[0009] Further details, features, and advantages of this application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which: Figure 1A flowchart of an advertising information push method according to an exemplary embodiment of this application is shown; Figure 2 This is a flowchart of the training methods for the encoding and prediction networks; Figure 3 This is a model training architecture diagram provided in an exemplary embodiment of this application; Figure 4 This is a schematic diagram of the structure of an advertising information push device provided in an embodiment of this application; Figure 5 A structural block diagram of an exemplary electronic device that can be used to implement embodiments of this application is shown. Detailed Implementation

[0010] Embodiments of this application will now be described in more detail with reference to the accompanying drawings. While some embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this application. It should be understood that the drawings and embodiments of this application are for illustrative purposes only and are not intended to limit the scope of protection of this application.

[0011] It should be understood that the steps described in the method embodiments of this application may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this application is not limited in this respect.

[0012] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the following description. It should be noted that the concepts of "first", "second", etc., mentioned in this application are only used to distinguish different devices, modules, or units, and are not intended to limit the order of functions performed by these devices, modules, or units or their interdependencies. It should be noted that the modifications "a" and "a plurality" mentioned in this application are illustrative and not restrictive, and those skilled in the art should understand that unless explicitly indicated in the context, they should be understood as "one or more". The names of messages or information exchanged between multiple devices in the embodiments of this application are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0013] The present invention will now be described with reference to the accompanying drawings. The technical solutions provided by the embodiments of the present invention will be explained in detail through specific examples and application scenarios.

[0014] Taking the price reduction of financial lending products as an example, it is necessary to screen users who can maximize the credit usage rate for the price reduction. Users' reactions to the price reduction can be divided into the following four categories: (1) Price reduction sensitive group: they will use credit if the price is reduced, and will not use credit if the price is not reduced; (2) Natural conversion group: they will use credit regardless of whether the price is reduced; (3) Indifferent group: they will not use credit regardless of whether the price is reduced; (4) Reactive group: they will not use credit if the price is reduced.

[0015] To maximize the marketing efficiency of price reductions, accurately identifying the first category of price-reduction-sensitive individuals is crucial. Traditional machine learning algorithms can only simply correlate user characteristics with trust levels, failing to predict the increase in trust levels resulting from price reductions compared to when prices are not reduced. This prevents maximizing user trust levels, as many individuals who opt for price reductions already possess high trust levels, rather than those who can maximize the increase in trust levels. Therefore, modeling the gain in trust levels resulting from price reductions is essential for accurately identifying price-reduction-sensitive individuals and improving the efficiency of price reduction marketing.

[0016] Please refer to Figure 1 The diagram illustrates a flowchart of an advertising information delivery method according to an exemplary embodiment of this application. The method is described using an electronic device as an example. Figure 1 As shown, the method includes: Step 101: Obtain the multidimensional object features of the target object; Step 102: Input the multidimensional object features into the trained encoding network to obtain the target encoding vector output by the trained encoding network; Step 103: Input the target encoding vector and the first price reduction variable into the trained prediction network to obtain the first target probability output by the trained prediction network. The first target probability indicates the prediction confidence rate of the target object under the first price reduction variable. Step 104: Input the target encoding vector and the second price reduction variable into the trained prediction network to obtain the second target probability output by the trained prediction network. The second target probability indicates the prediction confidence rate of the target object under the second price reduction variable. Step 105: Based on the first target probability and the second target probability, determine whether to push the target price reduction advertisement information to the target object.

[0017] Taking the adjustment of credit product pricing in a peer-to-peer lending platform as an example, the purpose of this solution is to increase new users' willingness to use credit by lowering prices, thereby increasing user usage and ultimately increasing the total number of users. To predict the increase in user credit usage caused by price reductions, this application constructs a credit usage prediction model. This model includes an encoding network and a prediction network. The encoding network encodes the multi-dimensional features of the input to obtain an encoding vector. The prediction network predicts the user's credit usage in both cases: receiving a price reduction and not receiving a price reduction. By comparing the two credit usage rates, it is determined whether to push a price reduction strategy to the user.

[0018] In one possible implementation, to predict a user's credit usage rate, the data input into the credit usage rate prediction model consists of multidimensional object features (also known as multidimensional covariate features) of the target object (user). These multidimensional object features can include: the user's internet finance characteristics, credit characteristics, historical behavioral data, and basic personal characteristics such as age, education level, and length of employment. Naturally, the more comprehensive and high-quality the features, the better the user profile can be depicted, resulting in more accurate predictions.

[0019] When determining whether to push a price reduction strategy to a target audience, multi-dimensional object features of the target audience can be obtained. First, the multi-dimensional object features are input into a trained encoding network for feature encoding, resulting in a target encoding vector output by the trained encoding network. In the confidence prediction stage, the target encoding vector and the label indicating whether the price reduction is accepted are input into a trained prediction network to obtain the confidence rate of the target audience under the two conditions of accepting the price reduction. That is, the target encoding vector and the first price reduction variable are input into the trained prediction network to obtain the first target probability output by the trained prediction network, and the target encoding vector and the second price reduction variable are input into the trained prediction network to obtain the second target probability output by the trained prediction network. Then, based on the first target probability and the second target probability, it is determined whether to push the target price reduction advertisement information to the target audience.

[0020] The first price reduction variable can be receiving a price reduction (X=1), and the first target probability indicates the predicted credit rate of the target object under the first price reduction variable, that is, the first target probability is the credit rate of the target object under the condition of receiving a price reduction; the second price reduction variable can be not receiving a price reduction (X=0), and the second target probability indicates the predicted credit rate of the target object under the second price reduction variable, that is, the second target probability is the credit rate of the target object under the condition of not receiving a price reduction.

[0021] By using the target encoding vectors output by the shared encoding network to predict the probability of accepting the price reduction, compared to related technologies that use two models for probability prediction, the error in encoding vector extraction can be avoided, which leads to subsequent prediction errors and further improves the prediction accuracy of the price reduction gain.

[0022] Considering that during the training of the prediction network, there is often confusion bias between the experimental and control groups in the observation experiment, i.e., the covariate distributions of the control and experimental groups are too different, which may affect the accuracy of subsequent confidence probability predictions, a discriminant network is introduced during the training phase of both the encoding and prediction networks to reduce confusion bias and enable the encoding network to encode a more balanced encoding vector. This discriminant network is used to train the encoding network adversarially.

[0023] In one possible implementation, the flowchart of the training method for the encoding network and the prediction network can be as follows: Figure 2 As shown, the method includes: Step 201: Obtain training data for the experimental group and the control group; Step 202: Based on the training data of the experimental group and the training data of the control group, train the encoding network and the discriminant network; Step 203: After the encoding network is trained, a prediction network is trained based on the training data of the experimental group, the experimental data of the control group, and the trained encoding network.

[0024] The training data for the experimental group consisted of the multidimensional object features of each sample object under the condition of price reduction, while the training data for the control group consisted of the multidimensional object features of each sample object under the condition of not receiving price reduction.

[0025] Observational experiments often suffer from confusion bias between the experimental and control groups, meaning that the covariate distributions of the control and experimental groups differ significantly. Employing adversarial learning to balance the vector representations of the experimental and control group data generated by the encoding network can better reduce this confusion bias. In the traditional GAN ​​(Generative Adversarial Network) framework, noise is mapped to generate images through a generator, and a discriminator is trained to distinguish whether the input is a real image generated by the generator or a synthetic image. A GAN trains a reliable discriminator to differentiate between fake and real images, and then uses this discriminator to train the generator, which generates images attempting to fool the discriminator.

[0026] In one possible implementation, the encoding network and discriminator network are first trained adversarially using training data from both the experimental and control groups. This makes it impossible for the discriminator to distinguish between the encoded vectors generated by the encoder network based on the experimental group training data and the vector representations corresponding to the control group training data. This achieves the goal of generating more balanced encoded vectors in the encoding network, resolving the confusion and bias problem between the experimental and control group data. After the encoding network training is completed, the prediction network is then trained using the experimental group training data, the control group experimental data, and the trained encoding network.

[0027] In an exemplary example, step 202 may also include steps 202A to 202D.

[0028] Step 202A: Input the training data of the experimental group into the encoding network to obtain the first encoding vector output by the encoding network; Step 202B: After embedding the training data of the control group, concatenate it with the noise vector to obtain the sample control vector; Step 202C: Input the first encoding vector and the sample control vector into the discriminant network for differentiation, and obtain the sample discrimination probability output by the discriminant network. The sample discrimination probability is used to indicate the similarity between the first encoding vector and the sample control vector. Step 202D: Train the encoding network and the discriminant network based on the sample discrimination probability adversarial training.

[0029] In each round of adversarial training, the experimental group training data is input into the encoding network to obtain the first encoded vector output by the encoding network. At the same time, the control group training data is embedded into a vector representation, and a noise vector is generated based on a uniform distribution. The vector representation and the noise vector are concatenated to obtain the sample control vector. Then, the first encoded vector and the sample control vector are input into the discriminant network, which distinguishes between the two and obtains the sample discrimination probability output by the discriminant network. This sample discrimination probability is used to indicate the similarity between the first encoded vector and the sample control vector, that is, to distinguish whether the two are similar vectors. Then, the encoding network and the discriminant network are trained adversarially according to the sample discrimination probability until the discriminant network can no longer distinguish between the first encoded vector and the sample control vector, indicating that the encoding network has generated a balanced encoded vector.

[0030] The training objective of adversarial training is to maximize the sample discrimination probability. A maximum discrimination probability indicates that the discriminator network cannot distinguish between the first encoding vector and the sample control vector, considering them similar. Optionally, a threshold can be set. When the sample discrimination probability exceeds this threshold, the encoding and discriminator networks are considered to have completed training, resulting in a fully trained encoding network. This fully trained encoding network can generate more balanced encoding vectors for the experimental and control data of the same sample object, reducing confusion bias between the two, thereby reducing training bias in the subsequent prediction network and improving its prediction performance.

[0031] Optionally, a noise vector is introduced when generating the sample comparison vector, which mainly plays an important role in enhancing the model's learning ability and ensuring training stability.

[0032] Considering that the encoding network loses information when encoding based on the multidimensional object features of the input, and in order to avoid losing too much important information for subsequent confidence prediction, an information preservation network can be introduced before the encoding network is adversarially trained based on the discriminant network to train the encoding network and minimize the loss of encoded information.

[0033] In an exemplary example, the following steps (steps one through four) may also be included before step 201: Step 1: Obtain the features of the sample object, the sample embedding representation corresponding to the features of the sample object, and the sample pseudo-features; Step 2: Input the features of the sample object into the encoding network to obtain the fourth encoding vector output by the encoding network; Step 3: Input the fourth encoding vector, sample embedding representation, and sample pseudo-features into the information preservation network to obtain the sample scalar value output by the information preservation network. The sample scalar value is composed of the mutual information expectation between the fourth encoding vector and the sample embedding representation, and the mutual information expectation between the fourth encoding vector and the sample pseudo-features. Step 4: Train the encoding network and the information retention network based on the sample scalar values.

[0034] The encoding network and the information-preserving network are trained and optimized simultaneously. The encoding network loses information when encoding high-dimensional input features. Therefore, the purpose of using the information-preserving network here is to minimize the information loss between the output encoded vector and the input high-dimensional covariates, thereby maximizing the retention of predictive information for causal effect estimation in the encoded vector.

[0035] In one possible implementation, sample embedding representations and sample pseudo-features are first constructed based on the characteristics of the sample objects. The sample embedding representations are obtained by embedding the characteristics of the sample objects, and the sample pseudo-features are created by shuffling the characteristics of the sample objects. These sample object characteristics belong to both the experimental group data and the training group data, and are multi-dimensional object characteristics of the sample objects, such as internet finance characteristics, credit characteristics, historical behavioral data, and basic personal characteristics, such as age, education level, work experience, etc.

[0036] In each round of training, the sample object features are input into the encoding network to obtain the fourth encoding vector output by the encoding network. Then, the fourth encoding vector, the sample embedding representation, and the sample pseudo-features are input into the information retention network. The information retention network compares the mutual information difference between the fourth encoding vector and the sample embedding representation, as well as the mutual information difference between the fourth encoding vector and the sample pseudo-features, to obtain the sample scalar value output by the information retention network. This sample scalar value is composed of the expected mutual information between the fourth encoding vector and the sample embedding representation, and the expected mutual information between the fourth encoding vector and the sample pseudo-features. Then, the encoding network and the information retention network are trained based on the sample scalar value.

[0037] The training objectives of the encoding network and the information retention network are: to maximize the expected mutual information between the fourth encoded vector and the sample embedded representation, and to minimize the expected mutual information between the fourth encoded vector and the sample pseudo-features. In other words, the smaller the difference between the fourth encoded vector and the sample embedded representation, the larger the difference between the fourth encoded vector and the sample pseudo-features. The expected mutual information of these two networks constitutes the mutual information estimation loss of the encoding network and the information retention network. The encoding network and the information retention network are then trained based on this mutual information estimation loss, and the mutual information estimation loss is minimized through iterative training.

[0038] It should be noted that the encoding network is trained twice. First, it is trained with the information preservation network. After minimizing the mutual information estimation loss, the encoding network is then trained adversarially based on the discriminant network. After the two-stage training is completed, the trained encoding network is obtained.

[0039] After the encoding network is trained, the prediction network can be trained based on the training data from the experimental group and the control group. In an exemplary example, step 203 may also include steps 203A to 203F.

[0040] Step 203A: Input the training data of the experimental group into the trained encoding network to obtain the second encoding vector output by the trained encoding network; Step 203B: Input the second encoding network and the first price reduction variable into the prediction network to obtain the first prediction probability output by the prediction network; Step 203C: Train the prediction network based on the difference between the first predicted probability and the first label value; Step 203D: Input the training data of the control group into the trained encoding network to obtain the third encoding vector output by the trained encoding network; Step 203E: Input the third encoding vector and the second price reduction variable into the prediction network to obtain the second prediction probability output by the prediction network; Step 203D: Train the prediction network based on the difference between the second predicted probability and the second label value.

[0041] In one possible implementation, the experimental group training data is input into the trained encoding network to obtain a second encoding vector output by the trained encoding network. Considering that the experimental group training data is training data under the condition of receiving a price reduction, the second encoding vector and the first price reduction variable (receiving a price reduction) are input into the prediction network to obtain a first prediction probability output by the prediction network. Then, the prediction network is trained based on the difference between the first prediction probability and the first label value (whether to use credit). Similarly, the control group training data is also input into the trained encoding network to obtain a third encoding vector output by the trained encoding network. Considering that the control group training data is training data under the condition of not receiving a price reduction, the third encoding vector and the second price reduction variable (not receiving a price reduction) are input into the prediction network to obtain a second prediction probability output by the prediction network. Then, the prediction network is trained based on the difference between the second prediction probability and the second label value (whether to use credit). After multiple rounds of iterative training, until the prediction loss of the prediction network is less than a preset value, the prediction network training is determined to be complete, and the trained prediction network can be obtained.

[0042] Please refer to Figure 3 This is a model training architecture diagram provided in an exemplary embodiment of this application. For example... Figure 3 As shown, the model training includes three stages: (1) Input the training data into the encoder (encoding network) to obtain the encoding vector, and input the positive and negative samples corresponding to the encoding network and the training data into the information retention network to calculate the mutual information loss to train the encoder and the information retention network; (2) Input the training data (experimental group training data) into the encoder to obtain the encoding vector, and input the noise data and the control group training data and the encoding vector into the discriminator network (discriminator network) to calculate the adversarial loss to train the encoder and the discriminator network; (3) Input the training data into the encoder again to obtain the encoding vector, input it into the prediction network to perform confidence prediction, and calculate the prediction loss to train the prediction network.

[0043] It should be noted that although an information retention network and a discriminator network were introduced during the model training phase, they were only used to improve the encoding performance of the encoding network. Therefore, during the model application phase, only the trained encoding network and prediction network were used to predict the price reduction gain.

[0044] In step 105, when determining whether to push the target price reduction advertisement information to the target audience based on the first target probability and the second target probability, specifically, the decision to push the price reduction advertisement is made by comparing whether there is a difference between the two probabilities and whether it represents a gain. For example, if the first target probability is greater than the second target probability, it means that the user's trust rate is higher when the price reduction is received than when the price reduction is not received, and the price reduction leads to an increase in the user's trust rate. In this case, the target audience is determined to be a price reduction-sensitive user, and the target price reduction advertisement information can be pushed to the target audience. Conversely, if the first target probability is less than or equal to the second target probability, it means that the user's trust rate is less than or equal to the user's trust rate when the price reduction is received, and the price reduction does not lead to an increase in the user's trust rate. In this case, the target audience is determined to be not a price reduction-sensitive user, and the push of the target price reduction advertisement information to the target audience can be stopped.

[0045] Optionally, if the probability of the first target is greater than the probability of the second target, a probability threshold can be set. If the difference between the probability of the first target and the probability of the second target is greater than the probability threshold, then the target price reduction advertisement information is pushed to the target object.

[0046] In summary, this application provides an advertising information push method: by inputting the multi-dimensional features of the target object into an encoding network to obtain a target encoding vector, and then inputting the target encoding vector and the first and second price reduction variables into a prediction network to predict the confidence rate, and then determining whether to push the target price reduction advertising information to the target object based on the predicted first and second target probabilities; by using the target encoding vector output by the shared encoding network to predict the confidence rate under whether the price reduction is accepted, compared with the related technology that uses two models for probability prediction, it can avoid the subsequent prediction error caused by the encoding vector extraction error, and further improve the prediction accuracy of the price reduction gain.

[0047] Please refer to Figure 4 This is a schematic diagram of the structure of an advertising information push device provided in an embodiment of this application. For example, as shown below... Figure 4 As shown, the device 400 includes: The first acquisition module 401 is used to acquire the multidimensional object features of the target object; The first encoding module 402 is used to input the multidimensional object features into the trained encoding network to obtain the target encoding vector output by the trained encoding network. The first prediction module 403 is used to input the target encoding vector and the first price reduction variable into the trained prediction network to obtain the first target probability output by the trained prediction network. The first target probability indicates the prediction confidence rate of the target object under the first price reduction variable. The second prediction module 404 is used to input the target encoding vector and the second price reduction variable into the trained prediction network to obtain the second target probability output by the trained prediction network. The second target probability indicates the prediction confidence rate of the target object under the second price reduction variable. The information push module 405 is used to determine whether to push target price reduction advertisement information to the target object based on the first target probability and the second target probability.

[0048] Optionally, the device further includes: The second acquisition module is used to acquire training data from the experimental group and training data from the control group. The first training module is used to train the encoding network and the discriminant network based on the training data of the experimental group and the training data of the control group. The second training module is used to train a prediction network based on the training data of the experimental group, the experimental data of the control group, and the trained encoding network after the encoding network has been trained.

[0049] Optionally, the first training module is further configured to: The training data of the experimental group is input into the encoding network to obtain the first encoding vector output by the encoding network. After embedding the training data of the control group, it is concatenated with the noise vector to obtain the sample control vector; The first encoding vector and the sample comparison vector are input into the discriminant network for differentiation, and the sample discrimination probability output by the discriminant network is obtained. The sample discrimination probability is used to indicate the similarity between the first encoding vector and the sample comparison vector. The encoding network and the discriminant network are trained adversarially based on the sample discrimination probability.

[0050] Optionally, the second training module is further configured to: The training data of the experimental group is input into the trained encoding network to obtain the second encoding vector output by the trained encoding network; The second encoding network and the first price reduction variable are input into the prediction network to obtain the first prediction probability output by the prediction network. The prediction network is trained based on the difference between the first predicted probability and the first label value; The training data of the control group is input into the trained encoding network to obtain the third encoding vector output by the trained encoding network; The third encoding vector and the second price reduction variable are input into the prediction network to obtain the second prediction probability output by the prediction network. The prediction network is trained based on the difference between the second predicted probability and the second label value.

[0051] Optionally, before acquiring the training data of the experimental group and the training data of the control group, the device further includes: The third acquisition module is used to acquire sample object features, sample embedding representations corresponding to the sample object features, and sample pseudo features; The second encoding module is used to input the features of the sample object into the encoding network to obtain the fourth encoding vector output by the encoding network; The information retention module is used to input the fourth encoding vector, the sample embedding representation, and the sample pseudo-feature into the information retention network to obtain the sample scalar value output by the information retention network. The sample scalar value is composed of the mutual information expectation between the fourth encoding vector and the sample embedding representation and the mutual information expectation between the fourth encoding vector and the sample pseudo-feature. The third training module is used to train the encoding network and the information retention network based on the sample scalar values.

[0052] Optionally, the information push module 405 is further configured to: If the first target probability is greater than the second target probability, it is determined to push the target price reduction advertisement information to the target object; If the first target probability is less than the second target probability, stop pushing the target price reduction advertisement information to the target object.

[0053] An exemplary embodiment of this application also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to cause the electronic device to perform a method according to an embodiment of this application.

[0054] An exemplary embodiment of this application also provides a non-transitory computer-readable storage medium storing a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform a method according to an embodiment of this application.

[0055] An exemplary embodiment of this application also provides a computer program product, including a computer program, wherein, when executed by a computer's processor, the computer program is used to cause the computer to perform a method according to an embodiment of this application.

[0056] refer to Figure 5The present invention describes a structural block diagram of an electronic device 500 that can serve as a server or client of this application, which is an example of a hardware device that can be applied to various aspects of this application. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the application described and / or claimed herein.

[0057] like Figure 5 As shown, the electronic device 500 includes a computing unit 501, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 502 or a computer program loaded from a storage unit 508 into a random access memory (RAM) 503. The RAM 503 may also store various programs and data required for the operation of the electronic device 500. The computing unit 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0058] Multiple components in electronic device 500 are connected to I / O interface 505, including: input unit 506, output unit 507, storage unit 508, and communication unit 509. Input unit 506 can be any type of device capable of inputting information to electronic device 500. Input unit 506 can receive input digital or character information and generate key signal inputs related to user settings and / or function control of electronic device. Output unit 507 can be any type of device capable of presenting information and may include, but is not limited to, a display, speaker, video / audio output terminal, vibrator, and / or printer. Storage unit 508 may include, but is not limited to, disk and optical disk. Communication unit 509 allows electronic device 500 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and may include, but is not limited to, modems, network cards, infrared communication devices, wireless communication transceivers, and / or chipsets, such as Bluetooth devices, WiFi devices, WiMax devices, cellular communication devices, and / or the like.

[0059] The computing unit 501 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 501 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 501 performs the various methods and processes described above. For example, in some embodiments, Figure 1 , Figure 2 The method shown can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as storage unit 508. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 500 via ROM 502 and / or communication unit 509. In some embodiments, computing unit 501 can be configured to execute by any other suitable means (e.g., by means of firmware). Figure 1 , Figure 2 The method shown.

[0060] The program code used to implement the methods of this application may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0061] In the context of this application, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0062] As used in this application, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or apparatus (e.g., disk, optical disk, memory, programmable logic device (PLD)) for providing machine instructions and / or data to a programmable processor, including machine-readable media that receive machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0063] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0064] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0065] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other.

Claims

1. A method for pushing advertising information, characterized in that, The method includes: Obtain multidimensional object characteristics of the target object; The multidimensional object features are input into the trained encoding network to obtain the target encoding vector output by the trained encoding network. The target encoding vector and the first price reduction variable are input into the trained prediction network to obtain the first target probability output by the trained prediction network. The first target probability indicates the prediction confidence rate of the target object under the first price reduction variable. The target encoding vector and the second price reduction variable are input into the trained prediction network to obtain the second target probability output by the trained prediction network. The second target probability indicates the prediction confidence rate of the target object under the second price reduction variable. Based on the first target probability and the second target probability, it is determined whether to push target price reduction advertisement information to the target object.

2. The method according to claim 1, characterized in that, The method further includes: Obtain training data for the experimental group and the control group; Based on the training data of the experimental group and the training data of the control group, the encoding network and the discriminant network are trained. After the encoding network is trained, a prediction network is trained based on the training data of the experimental group, the experimental data of the control group, and the trained encoding network.

3. The method according to claim 2, characterized in that, The training of the encoding network and the discriminant network based on the training data of the experimental group and the training data of the control group includes: The training data of the experimental group is input into the encoding network to obtain the first encoding vector output by the encoding network. After embedding the training data of the control group, it is concatenated with the noise vector to obtain the sample control vector; The first encoded vector and the sample comparison vector are input into the discriminant network for differentiation, and the sample discrimination probability output by the discriminant network is obtained. The sample discrimination probability is used to indicate the similarity between the first encoded vector and the sample comparison vector. The encoding network and the discriminant network are trained adversarially based on the sample discrimination probability.

4. The method according to claim 2, characterized in that, The training of the prediction network based on the training data of the experimental group, the experimental data of the control group, and the trained encoding network includes: The training data of the experimental group is input into the trained encoding network to obtain the second encoding vector output by the trained encoding network; The second encoding network and the first price reduction variable are input into the prediction network to obtain the first prediction probability output by the prediction network. The prediction network is trained based on the difference between the first predicted probability and the first label value; The training data of the control group is input into the trained encoding network to obtain the third encoding vector output by the trained encoding network; The third encoding vector and the second price reduction variable are input into the prediction network to obtain the second prediction probability output by the prediction network. The prediction network is trained based on the difference between the second predicted probability and the second label value.

5. The method according to claim 2, characterized in that, Before acquiring the training data of the experimental group and the training data of the control group, the method further includes: Obtain sample object features, sample embedding representations corresponding to the sample object features, and sample pseudo features; The features of the sample object are input into the encoding network to obtain the fourth encoding vector output by the encoding network; The fourth encoding vector, the sample embedding representation, and the sample pseudo-feature are input into the information preservation network to obtain the sample scalar value output by the information preservation network. The sample scalar value is composed of the mutual information expectation between the fourth encoding vector and the sample embedding representation and the mutual information expectation between the fourth encoding vector and the sample pseudo-feature. The encoding network and the information retention network are trained based on the sample scalar values.

6. The method according to any one of claims 1 to 5, characterized in that, The step of determining whether to push target price reduction advertisement information to the target object based on the first target probability and the second target probability includes: If the first target probability is greater than the second target probability, it is determined to push the target price reduction advertisement information to the target object; If the first target probability is less than or equal to the second target probability, stop pushing the target price reduction advertisement information to the target object.

7. An advertising information push device, characterized in that, The device includes: The first acquisition module is used to acquire the multidimensional object features of the target object; The first encoding module is used to input the multidimensional object features into the trained encoding network to obtain the target encoding vector output by the trained encoding network. The first prediction module is used to input the target encoding vector and the first price reduction variable into the trained prediction network to obtain the first target probability output by the trained prediction network. The first target probability indicates the prediction confidence rate of the target object under the first price reduction variable. The second prediction module is used to input the target encoding vector and the second price reduction variable into the trained prediction network to obtain the second target probability output by the trained prediction network. The second target probability indicates the prediction confidence rate of the target object under the second price reduction variable. The information push module is used to determine whether to push target price reduction advertisement information to the target object based on the first target probability and the second target probability.

8. The apparatus according to claim 7, characterized in that, The device further includes: The second acquisition module is used to acquire training data from the experimental group and training data from the control group. The first training module is used to train the encoding network and the discriminant network based on the training data of the experimental group and the training data of the control group. The second training module is used to train a prediction network based on the training data of the experimental group, the experimental data of the control group, and the trained encoding network after the encoding network has been trained.

9. An electronic device, comprising: processor; as well as Stored program memory, The program includes instructions that, when executed by the processor, cause the processor to perform the advertising information push method according to any one of claims 1-6.

10. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to execute the advertising information push method according to any one of claims 1-6.

Images