Model training method, method for processing promoted content, and related devices

Abstract

The present disclosure provides a model training method, comprising: obtaining a sample feature vector, the sample feature vector being used to characterize the attributes of a user's behavior toward promoted content; obtaining at least one auxiliary feature vector based on the sample feature vector and at least one auxiliary task network, the at least one auxiliary task network being used to assist in training a main task network, the objective function of the at least one auxiliary task network being different from the objective function of the main task network; fusing the sample feature vector and the at least one auxiliary feature vector to obtain a first fused vector; inputting the first fused vector into the main task network to obtain an output of the main task network, determining a first loss value based on the output of the main task network, the label value of the first fused vector, and the objective function of the main task network; and updating the weights of the main task network based on the first loss value until the main task network satisfies a first training stop condition. This method can achieve optimization for a single core objective and meet business needs.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to a model training method, a promotional content processing method, an apparatus, a device, a computer-readable storage medium, and a computer program product. Background Art

[0002] The conversion rate (CVR) is the probability that a user will be willing to consume, or even actually consume, the promoted item (e.g., digital products, tourist attractions, literary works, etc.) after receiving promotional content (e.g., an advertisement). Generally, the probability of a user converting after receiving promotional content can be predicted by training an artificial intelligence model.

[0003] Currently, in the process of training AI models, multiple objectives are used as supervision to incorporate more prior knowledge and simultaneously optimize the performance of the AI model. These multiple objectives can include the user's viewing time of promoted content, the presentation type of promoted content clicked by the user, and the information about the promoted object in the promoted content clicked by the user.

[0004] However, for promotional content, artificial intelligence models only need to optimize the core goal of predicting conversion rate. How to achieve the optimization of a single core goal in a multi-objective scenario is a technical problem that the industry urgently needs to solve. Summary of the Invention

[0005] The purpose of the present disclosure is to provide a model training method, a promotional content processing method, an apparatus, a device, a computer-readable storage medium, and a computer program product, which can achieve optimization for a single core goal and meet business needs.

[0006] In a first aspect, the present disclosure provides a model training method, the method comprising:

[0007] Obtaining a sample feature vector, where the sample feature vector is used to characterize attributes of a user's behavior toward the promoted content;

[0008] Obtaining at least one auxiliary feature vector based on the sample feature vector and at least one auxiliary task network, wherein the at least one auxiliary task network is used to assist in training the main task network, and the objective function of the at least one auxiliary task network is different from the objective function of the main task network;

[0009] fusing the sample feature vector and the at least one auxiliary feature vector to obtain a first fused vector;

[0010] Inputting the first fused vector into the main task network to obtain an output of the main task network, and determining a first loss value according to the output of the main task network, the label value of the first fused vector, and the objective function of the main task network;

[0011] The weight of the main task network is updated according to the first loss value until the main task network meets a first training stop condition.

[0012] In a second aspect, the present disclosure provides a method for processing promotional content, comprising:

[0013] Obtaining attributes of user behavior towards promoted content;

[0014] Obtaining a conversion rate of the user converted by the promoted content based on the attributes of the user's behavior toward the promoted content and a conversion rate model; wherein the conversion rate model includes a primary task network and at least one auxiliary task network; the primary task network is trained based on a first fusion vector, the first fusion vector is obtained by fusing a sample feature vector and at least one auxiliary feature vector, and the at least one auxiliary feature vector is obtained based on the sample feature vector and at least one auxiliary task network;

[0015] The promotional content is processed according to a conversion rate of the user being converted by the promotional content.

[0016] In a third aspect, the present disclosure provides a model training device, comprising:

[0017] A sample acquisition unit, configured to acquire a sample feature vector, wherein the sample feature vector is used to characterize an attribute of a user's behavior toward the promotional content;

[0018] a vector fusion unit, configured to obtain at least one auxiliary feature vector based on the sample feature vector and at least one auxiliary task network, wherein the at least one auxiliary task network is used to assist in training the main task network, and the objective function of the at least one auxiliary task network is different from the objective function of the main task network; and fuse the sample feature vector and the at least one auxiliary feature vector to obtain a first fused vector;

[0019] A model training unit is configured to input the first fused vector into the main task network to obtain an output of the main task network, determine a first loss value based on the output of the main task network, the label value of the first fused vector, and the objective function of the main task network; and update the weight of the main task network based on the first loss value until the main task network meets a first training stop condition.

[0020] In a fourth aspect, the present disclosure provides a promotional content processing device, comprising:

[0021] An acquisition unit, used to acquire attributes of a user's behavior on the promoted content;

[0022] a prediction unit, configured to obtain a conversion rate of the user being converted by the promoted content based on the attributes of the user's behavior toward the promoted content and a conversion rate model; wherein the conversion rate model includes a primary task network and at least one auxiliary task network; the primary task network is trained based on a first fusion vector, the first fusion vector is obtained by fusing a sample feature vector and at least one auxiliary feature vector, and the at least one auxiliary feature vector is obtained based on the sample feature vector and at least one auxiliary task network;

[0023] The processing unit is configured to process the promotional content according to a conversion rate of the user being converted by the promotional content.

[0024] In a fifth aspect, the present disclosure provides a computer-readable medium having a computer program stored thereon, which, when executed by a processing device, implements the steps of the method described in any one of the first aspect or the second aspect of the present disclosure.

[0025] In a sixth aspect, the present disclosure provides an electronic device, including:

[0026] a storage device having a computer program stored thereon;

[0027] A processing device is used to execute the computer program in the storage device to implement the steps of the method described in any one of the first aspect or the second aspect of the present disclosure.

[0028] In a seventh aspect, the present disclosure provides a computer program product comprising instructions, which, when executed on a device, enables the device to execute the method described in any one of the implementations of the first or second aspect.

[0029] It can be seen from the above technical solutions that the present disclosure has the following advantages:

[0030] The present disclosure provides a model training method, which first obtains a sample feature vector, which is used to characterize the attributes of user behavior toward promoted content; then obtains at least one auxiliary feature vector based on the sample feature vector and at least one auxiliary task network; wherein the objective function of the at least one auxiliary task network is different from the objective function of the main task network; then, the sample feature vector and the at least one auxiliary feature vector are fused to obtain a first fused vector, which is then input into the main task network to obtain the output of the main task network; and a first loss value is determined based on the output of the main task network, the label value of the first fused vector, and the objective function of the main task network. In this way, after the at least one auxiliary feature vector obtained by the at least one auxiliary task network is fused with the sample feature vector to obtain the first fused vector, based on the label value of the first fused vector, the output of the main task network, and the objective function of the main task network, the at least one auxiliary task network can be biasedly optimized to assist in training the main task network and updating the weights of the main task network. When the model to be trained meets the training stop condition, a biased model can be trained to optimize a single core objective in a multi-objective scenario, thereby meeting business needs.

[0031] Other features and advantages of the present disclosure will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical methods of the embodiments of the present disclosure, the following briefly introduces the drawings required for use in the embodiments.

[0033] Figure 1 A system architecture diagram of a model training system provided in an embodiment of the present disclosure;

[0034] Figure 2 A flowchart of a model training method provided in an embodiment of the present disclosure;

[0035] Figure 3 A schematic diagram of a multi-objective optimization provided in an embodiment of the present disclosure;

[0036] Figure 4 A flowchart of another model training method provided in an embodiment of the present disclosure;

[0037] Figure 5 A schematic diagram of another multi-objective optimization provided in an embodiment of the present disclosure;

[0038] Figure 6 A flowchart of a method for processing promotional content provided by an embodiment of the present disclosure;

[0039] Figure 7 A schematic diagram of a model training device provided in an embodiment of the present disclosure;

[0040] Figure 8 A schematic diagram of a promotional content processing device provided by an embodiment of the present disclosure;

[0041] Figure 9 A schematic structural diagram of an electronic device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0042] The terms "first" and "second" in the embodiments of the present disclosure are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of technical features indicated. Therefore, the terms "first" and "second" may explicitly or implicitly include one or more of the features.

[0043] First, some technical terms involved in the embodiments of the present disclosure are introduced.

[0044] Promotional content refers to content used to promote the promoted content, such as advertisements. Generally, AI models can be pre-trained to predict the conversion rate of users who are converted to promoted content. This conversion rate can then be used to process the promoted content. For example, when the conversion rate is high, the number of times the promoted content is pushed to the user can be increased, while when the conversion rate is low, the number of times the promoted content is pushed to the user can be reduced. This can reduce the ineffective delivery of promotional content and minimize the waste of resources (such as computing and storage resources).

[0045] Currently, when training AI models, multiple objectives are often used as supervisors to incorporate more prior knowledge, simultaneously optimizing the performance of these objectives. However, for promotional content, AI models only need to optimize the core objective of predicting conversion rates. Optimizing a single core objective in a multi-objective scenario is a pressing technical challenge for the industry.

[0046] In view of this, an embodiment of the present disclosure provides a model training method, which includes obtaining a sample feature vector, which is used to characterize the attributes of the user's behavior on the promoted content; obtaining at least one auxiliary feature vector based on the sample feature vector and at least one auxiliary task network; at least one auxiliary task network is used to assist in training the main task network, and the objective function of at least one auxiliary task network is different from the objective function of the main task network; then fusing the sample feature vector and the at least one auxiliary feature vector to obtain a first fusion vector, inputting the first fusion vector into the main task network to obtain the output of the main task network, and determining a first loss value based on the output of the main task network, the label value of the first fusion vector, and the objective function of the main task network; then updating the weight of the main task network according to the first loss value until the main task network meets the first training stop condition.

[0047] In this method, after fusing the auxiliary feature vector obtained by the auxiliary task network with the sample feature vector to obtain a first fused vector, the auxiliary task network is then biasedly optimized based on the label value of the first fused vector, the output of the main task network, and the objective function of the main task network to assist in training the main task network and update its weights. When the model to be trained meets the training stopping conditions, a biased model can be trained to optimize a single core objective in multi-objective scenarios, meeting business needs.

[0048] In some possible implementations, the above-mentioned model training method can be performed by an electronic device. The electronic device can be a device with data processing capabilities. The electronic device can be a terminal or a server. Terminals include but are not limited to smartphones, tablets, and laptops. The server can be a cloud server, such as a central server in a central cloud computing cluster, or an edge server in an edge cloud computing cluster. Of course, the server can also be a server in a local data center. A local data center refers to a data center directly controlled by a user. The model training method provided in the embodiments of the present disclosure can be collaboratively performed by a terminal and a server.

[0049] In order to make the technical solution of the present disclosure clearer and easier to understand, the architecture of the model training system provided by the embodiment of the present disclosure is introduced below with reference to the accompanying drawings.

[0050] See also Figure 1 The system architecture diagram of the model training system 100 shown in FIG. 1 includes a terminal 110 and a server 120. The terminal 110 and the server 120 can communicate through a network.

[0051] Among them, the terminal 110 is used to present a human-computer interaction interface. Developers can select samples based on the human-computer interaction interface. The terminal 110 can send the sample selected by the user to the server 120. The server 120 vectorizes the sample to obtain a sample feature vector. In this way, the server 120 can obtain the sample feature vector. Then, the server 120 obtains at least one auxiliary feature vector based on the sample feature vector and at least one auxiliary task network, fuses the sample feature vector with the at least one auxiliary feature vector to obtain a first fused vector, inputs the first fused vector into the main task network, obtains the output of the main task network, and determines a first loss value based on the output of the main task network, the label value of the first fused vector, and the objective function of the main task network. Finally, the server 120 updates the weights of the main task network and the weights of at least one auxiliary task network based on the first loss value until the model to be trained meets the training stop condition.

[0052] In some embodiments, the server 120 may also deploy the trained model to the terminal 110. Based on this, the terminal 110 may obtain the attributes of the user's behavior toward the promoted content, and then, based on the attributes of the user's behavior toward the promoted content and the trained model (e.g., a conversion rate model), obtain the conversion rate of the user being converted by the promoted content. The terminal may then process the promoted content based on the conversion rate.

[0053] In some examples, when the conversion rate of users being converted by promoted content exceeds a preset threshold, terminal 110 may increase the number of times the promoted content is pushed to the user and / or the duration of the promotional content presentation; when the conversion rate of users being converted by promoted content is less than or equal to the preset threshold, terminal 110 may reduce the number of times the promoted content is pushed to the user and / or the duration of the promotional content presentation. This can reduce ineffective promotional content delivery and reduce resource waste.

[0054] In order to make the technical solution of the present disclosure clearer and easier to understand, the following is an example of the collaborative execution of the model training method by the terminal and the server. Figure 2 , which is a flow chart of a model training method provided by an embodiment of the present disclosure, the method comprising:

[0055] S201: The terminal 110 determines the sample selected by the developer.

[0056] In some examples, the terminal 110 may present a human-computer interaction interface, and the developer may select a desired sample based on the human-computer interaction interface. For example, the terminal 110 may present candidate samples on the human-computer interaction interface, and the developer may select a sample by triggering a click operation on the candidate sample. After the terminal 110 detects the click operation, it determines the selected sample.

[0057] S202 : The terminal 110 sends the sample selected by the developer to the server 120 .

[0058] Next, terminal 110 can send the sample selected by the developer to server 120, so that server 120 can vectorize the sample to obtain a sample feature vector. The sample feature vector is used to characterize the attributes of the user's behavior towards the promoted content. In some examples, the sample feature vector can represent the length of time the user views the promoted content, whether the user clicks on the promoted content, the presentation type of the promoted content clicked by the user (e.g., video type, image type, etc.), whether the user is converted by the promoted content, and so on.

[0059] It should be noted that the above is merely an exemplary introduction to the information that can be represented by the sample feature vector, and the embodiments of the present disclosure are not limited thereto.

[0060] S203: The server 120 obtains a sample feature vector.

[0061] After the server 120 vectorizes the sample, it can obtain the sample feature vector so as to subsequently train the model to be trained.

[0062] S204: The server 120 obtains at least one auxiliary feature vector according to the sample feature vector and at least one auxiliary task network.

[0063] In the multi-objective optimization scenario, such as Figure 3 As shown, this figure is a schematic diagram of a multi-objective optimization provided by an embodiment of the present disclosure. Each auxiliary task network in at least one auxiliary task network includes a feature extraction subnetwork and an output subnetwork. The feature extraction subnetwork is used to extract features to obtain auxiliary feature vectors, and the output subnetwork is used to obtain the output of the auxiliary task network based on the auxiliary feature vectors.

[0064] Server 120 can input the sample feature vector into at least one auxiliary task network, such as feature extraction subnetwork 211, feature extraction subnetwork 212, ..., and feature extraction subnetwork 21N, to obtain an auxiliary feature vector output by each feature extraction subnetwork. Server 120 then inputs the auxiliary feature vector output by the feature extraction subnetwork into the output subnetwork to obtain the output of the auxiliary task network.

[0065] Next, the server 120 determines a second loss value based on the output of the auxiliary task network, the label value of the sample feature vector, and the objective function of the auxiliary task network, and updates the weights of the auxiliary task network based on the second loss value until the auxiliary task network meets the second training stop condition. For example, the server updates the weights of the feature extraction subnetwork 211 based on the output of the auxiliary task network obtained by the output subnetwork until the auxiliary task network meets the second training stop condition.

[0066] In some embodiments, taking the feature extraction subnetwork 211 as an example, after the feature extraction subnetwork 211 outputs the auxiliary feature vector, the auxiliary feature vector can also be input into the bias network 311 corresponding to the feature extraction subnetwork 211. The bias network 311 is used to further process the auxiliary feature vector, such as adjusting the coefficient of the auxiliary feature vector in the subsequent fusion process, etc., which will be introduced in detail later.

[0067] In the disclosed embodiment, the main task network 400 can be used to obtain the conversion rate of users converted by promoted content. The auxiliary task network corresponding to the feature extraction sub-network 211 can be used to obtain the user's viewing time of the promoted content. The auxiliary task network corresponding to the feature extraction sub-network 212 can be used to obtain the presentation type of the promoted content clicked by the user. The auxiliary task network corresponding to the feature extraction sub-network 21N can be used to obtain information about the promoted object in the promotional content clicked by the user. Different auxiliary task networks have different goals and are separated from each other.

[0068] It should be noted that the embodiments of the present disclosure do not specifically limit the types of the above-mentioned main task network, auxiliary task network, output sub-network, and bias network. For example, they can be deep neural networks, convolutional neural networks, etc. Those skilled in the art can select the network type with characteristics according to actual needs.

[0069] S205: The server 120 fuses the sample feature vector and at least one auxiliary feature vector to obtain a first fused vector.

[0070] After obtaining at least one auxiliary feature vector, the server 120 may fuse the sample feature vector with the at least one auxiliary feature vector to obtain a first fused vector. Figure 3 Taking the bias network 311 shown as an example, the bias network 311 can be used to provide coefficients of auxiliary feature vectors. The server 120 can fuse the sample feature vector and at least one auxiliary feature vector according to the coefficient of at least one auxiliary feature vector to obtain a first fusion vector. When the coefficient of the auxiliary feature vector is large, it indicates that the auxiliary feature vector has a greater impact on the output of the main task network (such as the core target); when the coefficient of the auxiliary feature vector is large, it indicates that the auxiliary feature vector has a smaller impact on the output of the main task network. In this way, the coefficient of the auxiliary feature vector can be adjusted by updating the weight of the bias network 311, thereby achieving optimization for a single core target in a multi-target scenario.

[0071] It should be noted that the embodiments of the present disclosure do not specifically limit the manner of fusing multiple features, and those skilled in the art may select a specific manner to fuse multiple features according to actual needs.

[0072] S206: The server 120 inputs the first fusion vector into the main task network to obtain the output of the main task network, and determines a first loss value based on the output of the main task network, the label value of the first fusion vector, and the objective function of the main task network.

[0073] Continuing with the above example, the main task network can be used to obtain the conversion rate of users converted by the promoted content. After the server 120 inputs the first fusion vector into the main task network, the output of the main task network can be the first conversion rate. Correspondingly, the label value of the first fusion vector can be the second conversion rate.

[0074] The server 120 can then determine the first loss value based on the output of the main task network, the label value of the first fusion vector, and the objective function of the main task network. For example, the server 120 can calculate the difference between the first conversion rate and the second conversion rate, and determine the first loss value based on the difference between the first conversion rate and the second conversion rate and the objective function of the main task network. In other embodiments, the server 120 can also use other methods to determine the first loss value. The objective function of the main task network can be a loss function or a regularized function based on the loss function; similarly, the objective function of the auxiliary task network can also be a loss function or a regularized function based on the loss function.

[0075] S207: The server 120 updates the weight of the main task network according to the first loss value until the main task network meets the first training stop condition.

[0076] After determining the first loss value, the server 120 may update the weight of the main task network based on the first loss value, and adjust the coefficient of at least one auxiliary feature vector.

[0077] For example, when the auxiliary task network corresponding to the feature extraction subnetwork 211 predicts the user's viewing time of the promoted content, and has a greater impact on the main task network 400's prediction of the user's conversion rate of the promoted content, the bias network 311 corresponding to the feature extraction subnetwork 211 will give more consideration to the auxiliary feature vector output by the feature extraction subnetwork 211, such as increasing the coefficient of the auxiliary feature vector; when the auxiliary task network corresponding to the feature extraction subnetwork 212 predicts the presentation type of the promoted content clicked by the user, and has a smaller impact on the main task network 400's prediction of the user's conversion rate of the promoted content, the bias network 312 corresponding to the feature extraction subnetwork 212 will give less consideration to the auxiliary feature vector output by the feature extraction subnetwork 212, such as reducing the coefficient of the auxiliary feature vector. In this way, this method can make the model to be trained more inclined to optimize the core goal (such as the conversion rate of the user converted by the promoted content) in a multi-objective scenario, so that after the model to be trained meets the first training stop condition, an asymmetric biased model is obtained to meet business needs.

[0078] It should be noted that the embodiments of the present disclosure do not specifically limit the training stop conditions (such as the first training stop condition or the second training stop condition described above). Those skilled in the art may set the training stop conditions according to actual needs. For example, the training stop condition may be when the model prediction accuracy is greater than a preset accuracy rate.

[0079] In some embodiments, the present disclosure provides a method for inputting the sample feature vector into at least one auxiliary task network. Figure 4 As shown, the method includes:

[0080] S401: The server 120 inputs the sample feature vector into the sharing network to obtain the shared component of the sample feature vector.

[0081] like Figure 5 As shown in FIG, this figure is a schematic diagram of another multi-objective optimization provided by an embodiment of the present disclosure. Server 120 can first input the sample feature vector into the shared network to obtain a shared component. In some examples, when the objectives of feature extraction subnetwork 211 and feature extraction subnetwork 212 are highly correlated, the shared component can be used to improve the learning efficiency of the auxiliary task network corresponding to feature extraction subnetwork 211 and the auxiliary task network corresponding to feature extraction subnetwork 212.

[0082] S402: The server 120 inputs the sample feature vector into at least one separation network to obtain an independent component of at least one feature extractor.

[0083] In some examples, when the target correlation of different feature extraction sub-networks is weak, if the shared component is still used, the learning efficiency of the auxiliary task networks corresponding to the different feature extraction sub-networks will be reduced, causing negative transfer in the different auxiliary task networks and reducing the prediction accuracy of the different auxiliary task networks (for example, the accuracy of the output of the output sub-network).

[0084] To this end, the server 120 can input the sample feature vector into at least one separation network, such as the separation network 21, the separation network 22...the separation network 2N (not shown in the figure), so as to achieve grouping of auxiliary task networks for different targets. Among them, at least one auxiliary task network includes at least one group, and each group corresponds to a separation network. For example, the separation network 21 can correspond to the auxiliary task network corresponding to the feature extraction subnetwork 211, the auxiliary task network corresponding to the feature extraction subnetwork 212...the auxiliary task network corresponding to the feature extraction subnetwork 21N, and the separation network 22 can correspond to the auxiliary task network corresponding to the feature extraction subnetwork 221, the auxiliary task network corresponding to the feature extraction subnetwork 222...the auxiliary task network corresponding to the feature extraction subnetwork 22N. As a result, there is a strong correlation between the auxiliary task networks in each group, thereby improving the learning efficiency of the auxiliary task network in each group and reducing the negative transfer phenomenon between different auxiliary task networks.

[0085] When grouping different auxiliary task networks based on the separation network, the grouping can be done manually or based on preset rules. In some examples, the separation network 21 can be a landing page group, and the separation network 22 can be a video group; accordingly, the auxiliary task network corresponding to the feature extraction subnetwork 211 can be used to predict the length of time a user stays on the landing page, the auxiliary task network corresponding to the feature extraction subnetwork 212 can be used to predict whether the user clicks on the landing page, the auxiliary task network corresponding to the feature extraction subnetwork 221 can be used to predict whether the user clicks on the prompt information of the promotional content in the video, and the auxiliary task network corresponding to the feature extraction subnetwork 222 can be used to predict whether the user closes the prompt information of the promotional content in the video.

[0086] It should be noted that the embodiment of the present disclosure does not specifically limit the execution order of S401 and S402. S401 and S402 can be executed simultaneously or sequentially. In other examples, S402 can be executed first and then S401.

[0087] S403: The server 120 fuses the shared component with each of the at least one independent component to obtain a second fusion vector corresponding to each separation network.

[0088] In some examples, server 120 may fuse the shared component with the independent component obtained from separation network 21 to obtain a second fused vector corresponding to separation network 21, and fuse the shared component with the independent component obtained from separation network 22 to obtain a second fused vector corresponding to separation network 22.

[0089] S404: The server 120 inputs the second fusion vector corresponding to each separation network into the auxiliary task network corresponding to each separation network to obtain an auxiliary feature vector.

[0090] Continuing with the above example, server 120 can input the second fused vector corresponding to separation network 21 into feature extraction sub-networks 211, 212, ..., 21N, and input the second fused vector corresponding to separation network 22 into feature extraction sub-networks 221, 222, ..., 22N. In this way, this method can train auxiliary task networks in different groups using different second fused vectors, reducing negative transfer between auxiliary task networks in different groups and improving the learning efficiency of the auxiliary task networks.

[0091] In some embodiments, the server 120 may further adjust the weights of the shared network 500 and the weights of at least one separate network. Specifically, the server 120 may determine at least one second loss value based on the label value of at least one second fusion vector output by the at least one output sub-network and the objective function of the at least one auxiliary task network; and update the weights of the shared network 500, the at least one auxiliary task network, and the at least one separate network based on the at least one second loss value.

[0092] For example, the goal of the feature extraction subnetwork 211 may be whether the user clicks upon landing, the output of the feature extraction subnetwork 211 may be that the user did not click, and the label value of the second fusion vector may be that the user clicked. The server 120 may obtain the loss value of the auxiliary task network corresponding to the feature extraction subnetwork 211 based on the output of the output subnetwork corresponding to the feature extraction subnetwork 211 and the label value of the second fusion vector, and adjust the weight of the auxiliary task network corresponding to the feature extraction subnetwork 211, the weight of the separation network 21, and the weight of the shared network 500 based on the loss value. In actual scenarios, there will be multiple auxiliary task networks and multiple separation networks, and the principles are similar, so they will not be repeated here.

[0093] Based on the above description, the embodiment of the present disclosure provides a model training method, which models the core target (such as conversion rate) separately from the non-core target. The update of the main task network corresponding to the core target is fully responsible for the improvement of the core target, thereby enhancing the model representation capability. Without affecting the learning of the main task network corresponding to the core target, the auxiliary feature vector and the sample feature vector are fused and provided to the main task network corresponding to the core target. In this way, the auxiliary task network corresponding to the non-core target will tend to optimize the core target during learning, making the entire model an asymmetric and biased multi-target model.

[0094] Furthermore, by separating the networks, we grouped the auxiliary task networks corresponding to different non-core objectives, constructing multiple groups of auxiliary task networks and a shared network. The input to the auxiliary task network is a fused vector formed by fusing the independent components of the sample feature vectors output by the separation network with the shared components of the sample feature vectors output by the shared network. This balances the positive and negative transfer of the auxiliary task networks during learning, improving their learning ability and efficiency.

[0095] The present disclosure also provides a method for processing promotional content, which can be executed by the server 120 or by the terminal 110. When the method is executed by the terminal 110, a conversion rate model can be deployed on the terminal 110. The conversion rate model can be trained by the server 120.

[0096] For ease of understanding, the following describes a method for processing promotional content provided by an embodiment of the present disclosure from the perspective of the terminal 110 . Figure 6 A flowchart of a method for processing promotional content provided in an embodiment of the present disclosure, the method comprising:

[0097] S601: The terminal 110 obtains attributes of a user's behavior on promotional content.

[0098] It should be noted that the terminal 110 needs to obtain the user's authorization in advance. Only after obtaining the user's authorization to use the corresponding data (such as the attributes of the user's behavior on the promotional content mentioned above) can the terminal 110 obtain data such as the attributes of the user's behavior on the promotional content.

[0099] S602: The terminal 110 obtains the conversion rate of users converted by the promoted content according to the attributes of the user's behavior on the promoted content and the conversion rate model.

[0100] The conversion rate model includes a main task network and at least one auxiliary task network; the main task network is trained based on the first fusion vector, the first fusion vector is obtained by fusing the sample feature vector and at least one auxiliary feature vector, and the at least one auxiliary feature vector is obtained based on the sample feature vector and at least one auxiliary task network. The training process of the conversion rate model is similar to the training process of the model to be trained in the above example, and will not be repeated here.

[0101] S603: The terminal 110 processes the promotional content according to the conversion rate of users converted by the promotional content.

[0102] In some examples, when the conversion rate of users being converted by promoted content exceeds a preset threshold, terminal 110 may increase the number of times promoted content is pushed to the user and / or the duration of the promotional content presentation; when the conversion rate of users being converted by promoted content is less than or equal to the preset threshold, terminal 110 may reduce the number of times promoted content is pushed to the user and / or the duration of the promotional content presentation. This can reduce ineffective promotional content delivery and reduce resource waste.

[0103] Figure 7 is a schematic diagram of a model training device according to an exemplary disclosed embodiment. Figure 7 As shown, the model training device 700 includes:

[0104] The sample acquisition unit 701 is used to acquire a sample feature vector, wherein the sample feature vector is used to characterize the attributes of the user's behavior toward the promotional content;

[0105] a vector fusion unit 702 configured to obtain at least one auxiliary feature vector based on the sample feature vector and at least one auxiliary task network, wherein the at least one auxiliary task network is used to assist in training the main task network, and wherein the objective function of the at least one auxiliary task network is different from the objective function of the main task network; and to fuse the sample feature vector and the at least one auxiliary feature vector to obtain a first fused vector;

[0106] The model training unit 703 is used to input the first fusion vector into the main task network to obtain the output of the main task network, determine a first loss value based on the output of the main task network, the label value of the first fusion vector, and the objective function of the main task network; and update the weight of the main task network based on the first loss value until the main task network meets the first training stop condition.

[0107] Optionally, each auxiliary task network in the at least one auxiliary task network includes a feature extraction subnetwork and an output subnetwork, the feature extraction subnetwork is used to perform feature extraction to obtain the auxiliary feature vector, and the output subnetwork is used to obtain the output of the auxiliary task network according to the auxiliary feature vector; the model training unit 703 is also used to determine a second loss value based on the output of the auxiliary task network and the label value of the sample feature vector and the objective function of the auxiliary task network; and update the weight of the auxiliary task network according to the second loss value until the auxiliary task network meets the second training stop condition

[0108] Optionally, each auxiliary task network in the at least one auxiliary task network is connected to a shared network and a separation network corresponding to the auxiliary task network, and the input of each auxiliary task network is a second fusion vector determined by the shared component and the independent component of the sample feature vector, the shared component is obtained by inputting the sample feature vector into the shared network, and the independent component is obtained by inputting the sample feature vector into the separation network.

[0109] Optionally, the at least one auxiliary task network includes at least one group, and each group corresponds to a separate network.

[0110] Optionally, the vector fusion unit 702 is specifically used to fuse the sample feature vector and the at least one auxiliary feature vector according to the coefficient of the at least one auxiliary feature vector to obtain a first fusion vector; the model training unit 703 is also used to adjust the coefficient of the at least one auxiliary feature vector according to the first loss value.

[0111] Figure 8 is a schematic diagram of a device for processing promotional content according to an exemplary disclosed embodiment. Figure 8 As shown, the promotion content processing device 800 includes:

[0112] The acquisition unit 801 is used to acquire attributes of the user's behavior on the promotion content;

[0113] Prediction unit 802 is configured to obtain a conversion rate of the user being converted by the promoted content based on the attributes of the user's behavior toward the promoted content and a conversion rate model; wherein the conversion rate model includes a primary task network and at least one auxiliary task network; the primary task network is trained based on a first fusion vector, the first fusion vector is obtained by fusing a sample feature vector and at least one auxiliary feature vector, and the at least one auxiliary feature vector is obtained based on the sample feature vector and at least one auxiliary task network;

[0114] The processing unit 803 is configured to process the promotion content according to the conversion rate of the user being converted by the promotion content.

[0115] Optionally, the processing unit 803 is specifically configured to increase the number of times the promotional content is pushed to the user and / or the presentation time of the promotional content when the conversion rate of the user converted by the promotional content is greater than a preset threshold; and reduce the number of times the promotional content is pushed to the user and / or the presentation time of the promotional content when the conversion rate of the user converted by the promotional content is less than or equal to the preset threshold.

[0116] The functions of the above modules have been described in detail in the method steps of the previous embodiment and will not be repeated here.

[0117] Reference below Figure 9 , which shows a schematic structural diagram of an electronic device 1000 suitable for implementing an embodiment of the present disclosure. The electronic device may be used to implement the following Figure 7 The function corresponding to the model training device 700 shown, or Figure 8 The functions corresponding to the promotional content processing device 800 are shown. Figure 9 The electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present disclosure.

[0118] like Figure 9 As shown, the electronic device 900 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage device 908 into a random access memory (RAM) 903. Various programs and data required for the operation of the electronic device 900 are also stored in the RAM 903. The processing device 901, the ROM 902, and the RAM 903 are connected to each other via a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0119] Typically, the following devices may be connected to the I / O interface 905: an input device 906 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 907 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 908 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 909. The communication device 909 may allow the electronic device 900 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 9 The electronic device 900 is shown with various devices, but it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed instead.

[0120] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device 909, or installed from the storage device 908, or installed from the ROM 902. When the computer program is executed by the processing device 901, the above-mentioned functions defined in the method of the embodiment of the present disclosure are performed.

[0121] It should be noted that the computer-readable medium mentioned above in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, device, or component. In the present disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to wires, optical cables, RF (radio frequency), etc., or any suitable combination thereof.

[0122] In some embodiments, the client and server can communicate using any currently known or future developed network protocol, such as HTTP (HyperText Transfer Protocol), and can be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0123] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0124] The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device is caused to: obtain a sample feature vector, wherein the sample feature vector is used to characterize the attributes of the user's behavior on the promoted content; obtain at least one auxiliary feature vector based on the sample feature vector and at least one auxiliary task network, wherein the at least one auxiliary task network is used to assist in training the main task network, and the objective function of the at least one auxiliary task network is different from the objective function of the main task network; fuse the sample feature vector and the at least one auxiliary feature vector to obtain a first fusion vector; input the first fusion vector into the main task network to obtain the output of the main task network, and determine a first loss value based on the output of the main task network, the label value of the first fusion vector, and the objective function of the main task network; update the weight of the main task network according to the first loss value until the main task network meets the first training stop condition; or,

[0125] Obtaining attributes of user behavior towards promoted content;

[0126] Obtaining a conversion rate of the user converted by the promoted content based on the attributes of the user's behavior toward the promoted content and a conversion rate model; wherein the conversion rate model includes a primary task network and at least one auxiliary task network; the primary task network is trained based on a first fusion vector, the first fusion vector is obtained by fusing a sample feature vector and at least one auxiliary feature vector, and the at least one auxiliary feature vector is obtained based on the sample feature vector and at least one auxiliary task network;

[0127] The promotional content is processed according to a conversion rate of the user being converted by the promotional content.

[0128] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages, or a combination thereof, including, but not limited to, object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on the remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0129] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the module, program segment, or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of the boxes in the block diagram and / or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0130] The modules described in the embodiments of the present disclosure may be implemented in software or hardware. In some cases, the name of a module does not limit the module itself. For example, the first acquisition module may also be described as a "module for acquiring at least two Internet Protocol addresses."

[0131] The functions described above herein may be performed, at least in part, by one or more hardware logic components. For example, and without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip (SOCs), complex programmable logic devices (CPLDs), and the like.

[0132] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0133] According to one or more embodiments of the present disclosure, Example 1 provides a model training method, including: obtaining a sample feature vector, the sample feature vector being used to characterize an attribute of a user's behavior toward promoted content;

[0134] Obtaining at least one auxiliary feature vector based on the sample feature vector and at least one auxiliary task network, wherein the at least one auxiliary task network is used to assist in training the main task network, and the objective function of the at least one auxiliary task network is different from the objective function of the main task network;

[0135] fusing the sample feature vector and the at least one auxiliary feature vector to obtain a first fused vector;

[0136] Inputting the first fused vector into the main task network to obtain an output of the main task network, and determining a first loss value according to the output of the main task network, the label value of the first fused vector, and the objective function of the main task network;

[0137] The weight of the main task network is updated according to the first loss value until the main task network meets a first training stop condition.

[0138] According to one or more embodiments of the present disclosure, Example 2 provides the method of Example 1, wherein each auxiliary task network in the at least one auxiliary task network includes a feature extraction subnetwork and an output subnetwork, wherein the feature extraction subnetwork is used to perform feature extraction to obtain the auxiliary feature vector, and the output subnetwork is used to obtain the output of the auxiliary task network according to the auxiliary feature vector;

[0139] The method further comprises:

[0140] Determining a second loss value according to the output of the auxiliary task network, the label value of the sample feature vector, and the objective function of the auxiliary task network;

[0141] The weight of the auxiliary task network is updated according to the second loss value until the auxiliary task network meets a second training stop condition.

[0142] According to one or more embodiments of the present disclosure, Example 3 provides the method of Example 1, wherein each auxiliary task network in the at least one auxiliary task network is connected to a shared network and a separation network corresponding to the auxiliary task network, and the input of each auxiliary task network is a second fusion vector determined by the shared component and the independent component of the sample feature vector, the shared component is obtained by inputting the sample feature vector into the shared network, and the independent component is obtained by inputting the sample feature vector into the separation network.

[0143] According to one or more embodiments of the present disclosure, Example 4 provides the method of Example 3, wherein the at least one auxiliary task network includes at least one group, each group corresponding to a separate network.

[0144] According to one or more embodiments of the present disclosure, Example 5 provides the method of Example 1, wherein fusing the sample feature vector and the at least one auxiliary feature vector to obtain a first fused vector includes:

[0145] fusing the sample feature vector and the at least one auxiliary feature vector according to a coefficient of the at least one auxiliary feature vector to obtain a first fused vector;

[0146] The method further comprises:

[0147] According to the first loss value, coefficients of the at least one auxiliary feature vector are adjusted.

[0148] According to one or more embodiments of the present disclosure, Example 6 provides the methods of Examples 1 to 5, wherein the main task network is used to obtain the conversion rate of the user converted by the promoted content; and the at least one auxiliary task network is used to obtain the user's viewing time of the promoted content, the presentation type of the promoted content clicked by the user, or the information of the promoted object in the promoted content clicked by the user.

[0149] According to one or more embodiments of the present disclosure, Example 7 provides a method for processing promotional content, including: obtaining attributes of a user's behavior on the promotional content;

[0150] Obtaining a conversion rate of the user converted by the promoted content based on the attributes of the user's behavior toward the promoted content and a conversion rate model; wherein the conversion rate model includes a primary task network and at least one auxiliary task network; the primary task network is trained based on a first fusion vector, the first fusion vector is obtained by fusing a sample feature vector and at least one auxiliary feature vector, and the at least one auxiliary feature vector is obtained based on the sample feature vector and at least one auxiliary task network;

[0151] The promotional content is processed according to a conversion rate of the user being converted by the promotional content.

[0152] According to one or more embodiments of the present disclosure, Example 8 provides the method of Example 7, wherein the promotional content is processed according to the conversion rate of the user to the promotional content, including:

[0153] When the conversion rate of the user converted by the promotional content is greater than a preset threshold, increasing the number of times the promotional content is pushed to the user and / or the presentation time of the promotional content;

[0154] When the conversion rate of the user converted by the promotional content is less than or equal to the preset threshold, the number of times the promotional content is pushed to the user and / or the presentation time of the promotional content are reduced.

[0155] The above description is merely a preferred embodiment of the present disclosure and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of disclosure involved in the present disclosure is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but also includes other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the above-mentioned disclosed concepts. For example, a technical solution formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in this disclosure.

[0156] In addition, although each operation is described in a specific order, this should not be understood as requiring these operations to be performed in the specific order shown or in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details have been included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Some features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination mode.

[0157] Although the subject matter has been described using language specific to structural features and / or methodological logical acts, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are merely example forms of implementing the claims. Regarding the apparatus in the above-described embodiments, the specific manner in which each module performs operations has been described in detail in the embodiments related to the method and will not be elaborated upon here.

Claims

1. A model training method, characterized in that: The method comprises: Obtaining a sample feature vector, where the sample feature vector is used to characterize attributes of a user's behavior toward the promoted content; Obtaining at least one auxiliary feature vector based on the sample feature vector and at least one auxiliary task network, wherein the at least one auxiliary task network is used to assist in training the main task network, and the objective function of the at least one auxiliary task network is different from the objective function of the main task network; fusing the sample feature vector and the at least one auxiliary feature vector to obtain a first fused vector; Inputting the first fused vector into the main task network to obtain an output of the main task network, and determining a first loss value according to the output of the main task network, the label value of the first fused vector, and the objective function of the main task network; Updating the weight of the main task network according to the first loss value until the main task network meets a first training stop condition; Wherein, each auxiliary task network in the at least one auxiliary task network is connected to a shared network and a separation network corresponding to the auxiliary task network, and the input of each auxiliary task network is a second fusion vector determined by the shared component and the independent component of the sample feature vector, the shared component is obtained by inputting the sample feature vector into the shared network, and the independent component is obtained by inputting the sample feature vector into the separation network.

2. The method according to claim 1, characterized in that Each auxiliary task network in the at least one auxiliary task network includes a feature extraction subnetwork and an output subnetwork, wherein the feature extraction subnetwork is used to perform feature extraction to obtain the auxiliary feature vector, and the output subnetwork is used to obtain the output of the auxiliary task network according to the auxiliary feature vector; The method further comprises: Determining a second loss value according to the output of the auxiliary task network, the label value of the sample feature vector, and the objective function of the auxiliary task network; The weight of the auxiliary task network is updated according to the second loss value until the auxiliary task network meets a second training stop condition.

3. The method according to claim 1, characterized in that The at least one auxiliary task network includes at least one group, and each group corresponds to a separate network.

4. The method according to claim 1, wherein The fusing the sample feature vector and the at least one auxiliary feature vector to obtain a first fused vector includes: fusing the sample feature vector and the at least one auxiliary feature vector according to a coefficient of the at least one auxiliary feature vector to obtain a first fused vector; The method further comprises: According to the first loss value, coefficients of the at least one auxiliary feature vector are adjusted.

5. The method according to any one of claims 1 to 4, characterized in that The main task network is used to obtain the conversion rate of the user converted by the promoted content; the at least one auxiliary task network is used to obtain the user's viewing time of the promoted content, the presentation type of the promoted content clicked by the user, or the promoted object information in the promoted content clicked by the user.

6. A method for processing promotional content, characterized in that: include: Obtaining attributes of user behavior towards promoted content; According to the attributes of the user's behavior on the promotional content and a conversion rate model, the conversion rate of the user converted by the promotional content is obtained; wherein, the conversion rate model includes a main task network and at least one auxiliary task network; the main task network is trained based on a first fusion vector, the first fusion vector is obtained by fusing a sample feature vector and at least one auxiliary feature vector, and the at least one auxiliary feature vector is obtained based on the sample feature vector and at least one auxiliary task network; wherein, each auxiliary task network in the at least one auxiliary task network is connected to a shared network and a separate network corresponding to the auxiliary task network, and the input of each auxiliary task network is a second fusion vector determined by a shared component and an independent component of the sample feature vector, the shared component is obtained by inputting the sample feature vector into the shared network, and the independent component is obtained by inputting the sample feature vector into the separate network; according to the conversion rate of the user converted by the promotional content, the promotional content is processed.

7. The method according to claim 6, characterized in that Processing the promotional content according to the conversion rate of the user to the promotional content includes: When the conversion rate of the user converted by the promotional content is greater than a preset threshold, increasing the number of times the promotional content is pushed to the user and / or the presentation time of the promotional content; When the conversion rate of the user converted by the promotional content is less than or equal to the preset threshold, the number of times the promotional content is pushed to the user and / or the presentation time of the promotional content are reduced.

8. A model training device, characterized in that: The device comprises: A sample acquisition unit, configured to acquire a sample feature vector, wherein the sample feature vector is used to characterize an attribute of a user's behavior toward the promotional content; a vector fusion unit, configured to obtain at least one auxiliary feature vector based on the sample feature vector and at least one auxiliary task network, wherein the at least one auxiliary task network is used to assist in training the main task network, and the objective function of the at least one auxiliary task network is different from the objective function of the main task network; and fuse the sample feature vector and the at least one auxiliary feature vector to obtain a first fused vector; a model training unit, configured to input the first fused vector into the main task network to obtain an output of the main task network, determine a first loss value based on the output of the main task network, the label value of the first fused vector, and the objective function of the main task network; and update the weight of the main task network based on the first loss value until the main task network meets a first training stop condition; Wherein, each auxiliary task network in the at least one auxiliary task network is connected to a shared network and a separation network corresponding to the auxiliary task network, and the input of each auxiliary task network is a second fusion vector determined by the shared component and the independent component of the sample feature vector, the shared component is obtained by inputting the sample feature vector into the shared network, and the independent component is obtained by inputting the sample feature vector into the separation network.

9. A promotional content processing device, characterized in that: include: An acquisition unit, used to acquire attributes of a user's behavior on the promoted content; A prediction unit is used to obtain the conversion rate of the user converted by the promotional content based on the attributes of the user's behavior on the promotional content and a conversion rate model; wherein the conversion rate model includes a main task network and at least one auxiliary task network; the main task network is trained based on a first fusion vector, the first fusion vector is obtained by fusing a sample feature vector and at least one auxiliary feature vector, and the at least one auxiliary feature vector is obtained based on the sample feature vector and at least one auxiliary task network; wherein each auxiliary task network in the at least one auxiliary task network is connected to a shared network and a separate network corresponding to the auxiliary task network, and the input of each auxiliary task network is a second fusion vector determined by a shared component and an independent component of the sample feature vector, the shared component is obtained by inputting the sample feature vector into the shared network, and the independent component is obtained by inputting the sample feature vector into the separate network. The processing unit is configured to process the promotional content according to a conversion rate of the user being converted by the promotional content.

10. An electronic device, characterized in that: include: a storage device having a computer program stored thereon; A processing device for executing the computer program in the storage device to implement the steps of the method according to any one of claims 1 to 6, or to implement the steps of the method according to claim 7 or 8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processing device, the program implements the steps of the method according to any one of claims 1 to 5, or the steps of the method according to claim 6 or 7.

12. A computer program product, characterized in that When the computer program product is run on a computer, the computer is enabled to execute the method according to any one of claims 1 to 5 , or the method according to claim 6 or 7 .

Images