Click rate estimation model training method and device, click rate estimation method and device

Abstract

The disclosure provides a click rate estimation model training method and device. A specific embodiment of the method includes: obtaining a preset sample set, the sample set including at least one sample; obtaining a pre-constructed click rate estimation network, the click rate estimation network including a vector module, a group division module, and a main network module, wherein the vector module is configured to convert each sample in the sample set into a feature vector; the group division module is configured to generate dynamic group parameters for each feature vector; and the main network module is configured to predict a click rate estimation value based on the group parameters and the feature vector; performing the following training steps: selecting a sample from the sample set; inputting the sample into the click rate estimation network to obtain the click rate estimation value; calculating a loss value of the click rate estimation network; and in response to the click rate estimation network satisfying a training completion condition, taking the click rate estimation network as a click rate estimation model. The embodiment improves the accuracy of click rate estimation.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, specifically to the field of artificial intelligence technology, and in particular to click-through rate prediction model training methods and apparatus, click-through rate prediction methods and apparatus, electronic devices, and computer-readable storage media. Background Technology

[0002] In traditional click-through rate (CTR) prediction models, all user samples typically share a set of model parameters. To improve parameter diversity, existing methods usually divide the samples into multiple groups, allowing each group to learn its own unique parameters.

[0003] Existing multi-group learning faces two problems. On the one hand, group division needs to be done manually, and this manual division method cannot traverse all possible groups, thus making it difficult to achieve optimal results. On the other hand, after group division, each sample can only belong to one group, which ignores the diversity of the samples themselves and limits the expression of user interests. Summary of the Invention

[0004] Embodiments of this disclosure provide a click-through rate (CTR) prediction model training method and apparatus, a click-through rate prediction method and apparatus, an electronic device, and a computer-readable storage medium.

[0005] In a first aspect, embodiments of this disclosure provide a method for training a click-through rate (CTR) prediction model. The method includes: acquiring a pre-set sample set, the sample set including at least one sample, the sample including: user features, features of the manipulated object, and contextual features between the user and the manipulated object; acquiring a pre-constructed CTR prediction network, the CTR prediction network including: a vector module, a group partitioning module, and a main network module, wherein the vector module is used to convert each sample in the sample set into a feature vector; the group partitioning module is used to generate dynamic group parameters for each feature vector, the group parameters being used to calculate the interest of the group to which the sample belongs; the main network module predicts a CTR prediction value based on the group parameters and feature vectors; and performing the following training steps: selecting a sample from the sample set; inputting the sample into the CTR prediction network to obtain the CTR prediction value output by the CTR prediction network; calculating the loss value of the CTR prediction network; and, in response to the CTR prediction network meeting the training completion conditions, using the CTR prediction network as a CTR prediction model.

[0006] In some embodiments, the aforementioned group segmentation module includes: a group information selection submodule, a group representation learning submodule, and a group parameter generation submodule; the group information selection submodule is used to assign importance weights to the user attribute feature vector and the user behavior feature vector in the feature vector to obtain a new feature vector; the group representation learning submodule is used to segment the new feature vector into group vectors and dynamically update the group probability of each user belonging to each group vector, where the group probability is used to characterize the probability of a user belonging to each group; the group parameter generation submodule generates independent group parameters for each user based on the group probability and the group parameters preset for each group vector; calculating the loss value of the click-through rate prediction network includes: constructing the cross-entropy loss function of the click-through rate prediction network; and calculating the loss value of the click-through rate estimation network based on the cross-entropy loss function.

[0007] In some embodiments, the above-mentioned group segmentation module further includes: a group splicing submodule and a multilayer perceptron; the group splicing submodule aggregates the group vectors of all users based on the group probability and the group vectors of each user, and splices the feature vector of the manipulated object and the group vectors of all users in the feature vector to obtain the spliced ​​vector; the multilayer perceptron obtains the group output of each group on the sample based on the spliced ​​vector, and the group output is used to characterize the interest probability of each group on the manipulated object; calculating the loss value of the click-through rate prediction network includes: constructing a first loss function for the vector module and the main network module, and a second loss function for the vector module and the group segmentation module respectively; calculating the main network loss value based on the first loss function; calculating the group segmentation loss value based on the second loss function; and calculating the loss value of the click-through rate prediction network based on the main network loss value and the group segmentation loss value.

[0008] In some embodiments, the main network module includes a fully connected layer and an output layer; the output layer is used to obtain the parameters of the fully connected layer during the training of the click-through rate prediction network, add the parameters of the fully connected layer to the group parameters, and obtain the click-through rate prediction value based on the added parameters.

[0009] In some embodiments, the main network module further includes: a behavior encoding layer, a deep network layer, and a main network splicing submodule; the behavior encoding layer is used to encode the user behavior feature vector in the feature vector to obtain the behavior encoding vector; the deep network layer is used to calculate the user attribute feature vector, the operated object feature vector, and the context feature vector in the feature vector to obtain the calculation vector; the main network splicing submodule is used to splice the behavior encoding vector and the calculation vector, and input the splicing result into the fully connected layer.

[0010] Secondly, embodiments of this disclosure provide a click-through rate (CTR) prediction method. The method includes: acquiring user data, object data, and context data between the user and the object on a webpage or application during a historical time period; obtaining user features, object features, and context features based on the user data, object data, and context data; and inputting the user features, object features, and context features into a CTR prediction model generated by the CTR prediction model training method of any embodiment of the first aspect to obtain a CTR prediction value output by the CTR prediction model.

[0011] Thirdly, embodiments of this disclosure provide a click-through rate (CTR) prediction model training apparatus, comprising: a sample acquisition unit configured to acquire a preset sample set, the sample set including at least one sample, the sample including: user features, features of the operated object, and context features; a network acquisition unit configured to acquire a pre-constructed CTR prediction network, the CTR prediction network including: a vector module, a group partitioning module, and a main network module, wherein the vector module is used to convert each sample in the sample set into a feature vector; the group partitioning module is used to generate dynamic group parameters for each feature vector, the group parameters being used to calculate the interest of the group to which the sample belongs; the main network module predicts the CTR prediction value based on the group parameters and feature vectors; a selection unit configured to select a sample from the sample set; an input unit configured to input the sample into the CTR prediction network to obtain the CTR prediction value output by the CTR prediction network; a calculation unit configured to calculate the loss value of the CTR prediction network; and an acquisition unit configured to, in response to the CTR prediction network meeting the training completion condition, use the CTR prediction network as a CTR prediction model.

[0012] In some embodiments, the aforementioned group segmentation module includes: a group information selection submodule, a group representation learning submodule, and a group parameter generation submodule; the group information selection submodule is used to assign importance weights to the user attribute feature vector and the user behavior feature vector in the feature vector to obtain a new feature vector; the group representation learning submodule is used to segment the new feature vector into group vectors and dynamically update the group probability of each user belonging to each group vector, where the group probability is used to characterize the probability of a user belonging to each group; the group parameter generation submodule generates independent group parameters for each user based on the group probability and the group parameters preset for each group vector; the aforementioned calculation unit is further configured to: construct the cross-entropy loss function of the click-through rate estimation network; and calculate the loss value of the click-through rate estimation network based on the cross-entropy loss function.

[0013] In some embodiments, the aforementioned group segmentation module further includes: a group splicing submodule and a multilayer perceptron; the group splicing submodule aggregates the group vectors of all users based on the group probability and the group vectors of each user, and splices the feature vector of the manipulated object and the group vectors of all users in the feature vector to obtain a spliced ​​vector; the multilayer perceptron obtains the group output of each group on the sample based on the spliced ​​vector, and the group output is used to characterize the interest probability of each group on the manipulated object; the aforementioned computing unit is further configured to: construct a first loss function for the vector module and the main network module, and a second loss function for the vector module and the group segmentation module respectively; calculate the main network loss value based on the first loss function; calculate the group segmentation loss value based on the second loss function; and calculate the loss value of the click-through rate prediction network based on the main network loss value and the group segmentation loss value.

[0014] In some embodiments, the main network module includes a fully connected layer and an output layer; the output layer is used to obtain the parameters of the fully connected layer during the training of the click-through rate prediction network, add the parameters of the fully connected layer to the group parameters, and obtain the click-through rate prediction value based on the added parameters.

[0015] In some embodiments, the main network module further includes: a behavior encoding layer, a deep network layer, and a main network splicing submodule; the behavior encoding layer is used to encode the user behavior feature vector in the feature vector to obtain the behavior encoding vector; the deep network layer is used to calculate the user attribute feature vector, the operated object feature vector, and the context feature vector in the feature vector to obtain the calculation vector; the main network splicing submodule is used to splice the behavior encoding vector and the calculation vector, and input the splicing result into the fully connected layer.

[0016] Fourthly, embodiments of this disclosure provide a click-through rate (CTR) prediction apparatus, comprising: a data acquisition unit configured to acquire user data, object data, and context data of a user interacting with an object on a webpage or application during a historical time period; a processing unit configured to obtain user features, object features, and context features based on the user data, object data, and context data; and a prediction unit configured to input the user features, object features, and context features into a CTR prediction model generated by a CTR prediction model training apparatus according to any embodiment of the third aspect, to obtain a CTR prediction value output by the CTR prediction model.

[0017] Fifthly, embodiments of this disclosure provide an electronic device comprising: one or more processors; a storage device having one or more programs stored thereon; and, when the one or more programs are executed by the one or more processors, causing the one or more processors to implement the method described in any embodiment of the first or second aspect.

[0018] In a sixth aspect, embodiments of the present disclosure provide a computer-readable medium having a computer program stored thereon that, when executed by a processor, implements the methods described in any of the embodiments of the first or second aspect.

[0019] The click-through rate (CTR) prediction model training method and apparatus provided in the embodiments of this disclosure firstly acquire a preset sample set, which includes at least one sample, and the sample includes: user features, features of the manipulated object, and contextual features between the user and the manipulated object; secondly, acquire a pre-constructed CTR prediction network, which includes: a vector module, a group partitioning module, and a main network module, wherein the vector module is used to convert each sample in the sample set into a feature vector; the group partitioning module is used to generate dynamic group parameters for each feature vector, and the group parameters are used to calculate the interest of the group to which the sample belongs; the main network module predicts the CTR prediction value based on the group parameters and feature vectors; thirdly, select a sample from the sample set; then, input the sample into the CTR prediction network to obtain the CTR prediction value output by the CTR prediction network; next, calculate the loss value of the CTR prediction network; finally, in response to the CTR prediction network meeting the training completion condition, use the CTR prediction network as the CTR prediction model. Therefore, the constructed group segmentation module can dynamically divide samples into multiple different groups during the training process of the click-through rate prediction model, ensuring the reliability of group segmentation, guaranteeing the expression of user interests, and improving the robustness and accuracy of the click prediction model. Attached Figure Description

[0020] Other features, objects, and advantages of this disclosure will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0021] Figure 1 This is an exemplary system architecture diagram to which one embodiment of this disclosure can be applied;

[0022] Figure 2 This is a flowchart of an embodiment of the click-through rate prediction model training method according to the present disclosure;

[0023] Figure 3 This is a schematic diagram of the structure of this publicly available click-through rate prediction network;

[0024] Figure 4 This is a flowchart of an embodiment of the click-through rate prediction method according to this disclosure;

[0025] Figure 5 This is a schematic diagram of the structure of an embodiment of a click-through rate prediction model training apparatus according to the present disclosure;

[0026] Figure 6This is a schematic diagram of a structure of an embodiment of the click-through rate prediction device according to the present disclosure;

[0027] Figure 7 This is a schematic diagram of the structure of an electronic device suitable for implementing embodiments of the present disclosure. Detailed Implementation

[0028] The present disclosure will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0029] It should be noted that, unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other. This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0030] Figure 1 An exemplary system architecture 100 is shown that can be applied to the click-through rate prediction model training method or click-through rate prediction method of this disclosure.

[0031] like Figure 1 As shown, system architecture 100 may include terminals 101 and 102, network 103, database server 104, and server 105. Network 103 serves as the medium for providing communication links between terminals 101 and 102, database server 104, and server 105. Network 103 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0032] User 110 can use terminals 101 and 102 to interact with server 105 via network 103 to receive or send messages, etc. Various client applications can be installed on terminals 101 and 102, such as model training applications, image recognition applications, shopping applications, payment applications, web browsers, and instant messaging tools.

[0033] The terminals 101 and 102 here can be either hardware or software. When terminals 101 and 102 are hardware, they can be various electronic devices with displays, including but not limited to smartphones, tablets, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), laptops, and desktop computers. When terminals 101 and 102 are software, they can be installed in the electronic devices listed above. They can be implemented as multiple software programs or software modules (e.g., to provide distributed services) or as a single software program or software module. No specific limitations are set here.

[0034] Database server 104 can be a database server that provides various services. For example, the database server may store a sample set. The sample set contains a large number of samples, each of which is different. The samples may include user characteristics, characteristics of the object being operated on, and contextual characteristics between the user and the object being operated on. User 110 can also select samples from the sample set stored in database server 104 through terminals 101 and 102.

[0035] Server 105 can also be a server providing various services, such as a backend server supporting various applications displayed on terminals 101 and 102. The backend server can use samples from the sample set sent by terminals 101 and 102 to train a click-through rate (CTR) prediction model, and can send the trained CTR prediction model back to terminals 101 and 102. In this way, users can apply the generated CTR prediction model to predict the estimated CTR value.

[0036] The database server 104 and server 105 here can be either hardware or software. When they are hardware, they can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. When they are software, they can be implemented as multiple software programs or software modules (e.g., used to provide distributed services), or as a single software program or software module. No specific limitations are made here.

[0037] It should be noted that the click-through rate (CTR) prediction model training method or CTR prediction method provided in the embodiments of this disclosure are generally executed by the server 105. Accordingly, the CTR prediction model training device or CTR prediction device is also generally located in the server 105.

[0038] It should be noted that if server 105 can perform the relevant functions of database server 104, database server 104 may not be set up in system architecture 100.

[0039] It should be understood that Figure 1 The number of terminals, networks, database servers, and servers shown is merely illustrative. Depending on implementation needs, any number of terminals, networks, database servers, and servers can be included.

[0040] To address the problems of existing technologies requiring manual group segmentation and resulting in each sample belonging to only one group, this disclosure provides a click-through rate (CTR) prediction model training method. This method learns the group to which a sample belongs and dynamically assigns samples to multiple groups with diverse interests, enriching and stabilizing the information representation of the samples and improving the accuracy of CTR prediction. Figure 2The diagram illustrates a flow 200 of an embodiment of a click-through rate (CTR) prediction model training method according to the present disclosure, which includes the following steps:

[0041] Step 201: Obtain the preset sample set.

[0042] In this embodiment, the sample set is a set of samples collected in advance for training the click-through rate prediction model. The samples in the sample set can be obtained by collecting sample data from the Internet and extracting features from the sample data (e.g., by using a feature extractor). The sample set includes at least one sample, and each sample includes user features, features of the manipulated object, and contextual features between the user and the manipulated object.

[0043] In this embodiment, the characteristics of the manipulated object can be those of the manipulated object within the specified content of the same website or application. The manipulated object includes: goods or advertisements. The characteristics of goods include: the color of the goods, the type of the goods, the place of origin of the goods, etc. The characteristics of advertisements include: the content of the advertisement, the exposure time of the advertisement, etc.

[0044] In this embodiment, user characteristics refer to the characteristics of individuals who are interested in the objects being operated on within the specified content of the aforementioned website or application over a certain period of time. User characteristics include: user attribute characteristics and user behavioral characteristics. User attributes refer to user attribute features, such as age and gender. User behavioral characteristics refer to user behavior sequence characteristics, that is, the sequence of products clicked by the user over a past period of time.

[0045] In this embodiment, the contextual features between the user and the operated object are features related to the user's features and the operated object's features. For example, contextual features include: whether the user has clicked on similar products, whether the user has clicked on the same type of advertisement, etc.

[0046] In this embodiment, the execution entity of the click-through rate prediction model training method (e.g., Figure 1 The server 105 shown can obtain the sample set in various ways. For example, the executing entity can obtain it from a database server (e.g., via a wired or wireless connection). Figure 1 The sample set stored in the database server 104 shown can be retrieved. For example, a user can access the sample set via a terminal (e.g., ...). Figure 1 The terminals 101 and 102 shown are used to collect samples. In this way, the executing entity can receive the samples collected by the terminals and store these samples locally, thereby generating a sample set.

[0047] Step 202: Obtain the pre-built click-through rate prediction network.

[0048] In this embodiment, the click-through rate prediction network includes a vector module, a group segmentation module, and a main network module. The vector module is used to convert each sample in the sample set into a feature vector. The group segmentation module is used to generate dynamic group parameters for each feature vector. The group parameters are used to calculate the interest of the group to which the sample belongs. The main network module predicts the click-through rate based on the group parameters and the feature vectors.

[0049] In this embodiment, the vector module is a module that converts samples into vectors. The user features, manipulated object features, and context features in the samples belong to the ID class and are relatively sparse. The vector module can convert the user features, manipulated object features, and context features in the samples into multi-dimensional dense feature vectors.

[0050] In this embodiment, the click-through rate (CTR) estimate is the average estimated CTR obtained by estimating the features. The average estimated CTR is the average of the estimated CTR values. CTR refers to the ratio of the number of times specified content on a website or application is clicked to the number of times it is displayed. CTR is usually an important indicator for measuring recommendation efficiency in recommendation systems. Generally, CTR can be calculated by dividing the actual number of clicks (i.e., the number of times the target page is reached) of the manipulated object on the webpage or application by the number of times the manipulated object is displayed on the webpage or application.

[0051] In this embodiment, compared with the traditional click-through rate (CTR) prediction model, the CTR prediction network of this disclosure has a group segmentation module that provides group parameters. When the main network module predicts the CTR prediction value, it needs to consider the group parameters in real time. The group parameters are a representation of the specific distribution of samples belonging to different samples. By adding the group parameters, the main network module can more clearly understand the group to which each sample belongs.

[0052] In this embodiment, the group parameter is the parameter value of the group to which the sample (such as a user, product, or advertisement in the sample) belongs. This group parameter is represented as a network parameter in network computation and can change as the click-through rate prediction network iterates and trains. After classifying the sample into different groups, the interest of the group to which the sample belongs can be calculated through the group parameter. For example, the group partitioning module can be equivalent to y' = Wx' + b, where W is the group parameter, x' is the input sample of the group partitioning module, y' is the output value of the group partitioning module, and b is a hyperparameter. After acquiring the sample, the group partitioning module classifies the feature vector into the preset group based on the similarity between the feature vector of the sample and the vector of the preset group, obtaining the probability that the sample belongs to the group, i.e., the group probability. Based on the aspects of interest of the sample, corresponding group parameters are set for each group. Based on the probability that the sample belongs to the group and the parameters of each group, the group parameters of the group partitioning module are obtained.

[0053] In this embodiment, after obtaining the group parameters, the main network module combines the group parameters with the parameters of the main network module (e.g., dot addition, dot multiplication) to obtain new parameters of the main network module. During the training of the click-through rate prediction network, the new parameters of the main network module change until the click-through rate prediction network reaches the training completion condition.

[0054] In this embodiment, the group parameters of the group partitioning module can change with the parameters of the vector module and the main network module. When training the click-through rate (CTR) prediction network, the parameters of the group, vector, and main network modules need to be adjusted using the loss function of the CTR prediction network to ensure convergence and meet the training completion conditions.

[0055] In this embodiment, the main network module can use an NN (Neural Network) network or a network result transformed from an NN network. The main network module analyzes group parameters, user characteristics, characteristics of the manipulated object, and contextual characteristics to predict the estimated click-through rate of the user clicking on the manipulated object.

[0056] Step 203: Select samples from the sample set.

[0057] In this embodiment, the executing entity can select samples from the sample set obtained in step 201 and execute the training steps from 204 to 206. The method and number of samples selected are not limited in this application. For example, at least one sample can be randomly selected.

[0058] Step 204: Input the sample into the click-through rate prediction network to obtain the click-through rate prediction value output by the click-through rate prediction network.

[0059] In this embodiment, the samples include user features, manipulated object features, and context features. Samples selected from the sample set are input into the click-through rate (CTR) prediction network. The CTR prediction network analyzes the user features, manipulated object features, and context features to determine the CTR prediction value corresponding to the current user features, manipulated object features, and context features. The CTR prediction value can be used to measure the recommendation efficiency of each manipulated object in the recommendation system, thereby providing an effective means for recommending and displaying manipulated objects in the recommendation system of web pages and applications.

[0060] In this embodiment, the click-through rate (CTR) prediction network is used to characterize the correspondence between samples and CTR prediction values. The samples include user features and object features, and the CTR prediction value is the estimated CTR of the user on the object under the current user features and object features.

[0061] Step 205: Calculate the loss value of the click-through rate prediction network.

[0062] In this embodiment, multiple iterations of training are generally required in the training steps of the click-through rate (CTR) prediction network to bring the overall loss of the CTR prediction network to converge. Each iteration of training executes steps 203-205 in sequence. After calculating the loss value in step 205, the parameters of the CTR estimation network are adjusted based on the loss value (which reflects the magnitude of the error), and then the next iteration of training is performed. Steps 203-205 are executed again until the training completion condition is met.

[0063] In this embodiment, in order to effectively adjust the parameters of the click-through rate (CTR) prediction network, a loss function needs to be set for the CTR prediction network. The loss function can calculate the error between the forward calculation result and the true value of each iteration of the CTR prediction network, thereby guiding the next step of training in the right direction through the error.

[0064] It should be noted that in the training step of the current iteration, if the click-through rate estimation network has met the training completion condition, the execution subject will no longer input samples into the click-through rate estimation network, and the click-through rate estimation network is the click-through rate estimation model that has been trained.

[0065] Step 206: In response to the click-through rate prediction network meeting the training completion conditions, the click-through rate prediction network is used as the click-through rate prediction model.

[0066] In this embodiment, the training completion condition includes at least one of the following: the number of training iterations of the click-through rate (CTR) prediction network reaches a predetermined iteration threshold; the rate of change of the model parameters of the CTR prediction network (including the parameters of the population parameters, vector modules, and main network modules) is less than a predetermined threshold. When the rate of change of the model parameters of the CTR prediction network is less than the predetermined threshold, the CTR prediction network is determined to meet the training completion condition. For example, the training iterations reach 5,000. The rate of change of the model parameters of the CTR prediction network is less than 0.05. In this embodiment, setting the training completion condition can accelerate the model convergence speed.

[0067] In step 206, when the model parameters do not change significantly and are less than a predetermined threshold, the loss value of the click-through rate prediction network is determined to have converged, and the click-through rate prediction model training is completed.

[0068] In some alternative implementations of this disclosure, in response to the click-through rate prediction network not meeting the training completion conditions, the model parameters of the click-through rate prediction network are adjusted so that the loss function of the click-through rate prediction network converges, and steps 203-205 above are continued.

[0069] In this embodiment, the above-mentioned response to the click-through rate (CTR) prediction network not meeting the training completion condition, adjusting the model parameters of the CTR prediction network to make the loss function of the CTR prediction network converge includes: keeping the group parameters unchanged, adjusting the parameters of the vector module and the main network module, and repeatedly executing steps 203-205 until the loss value of the CTR prediction network reaches the first loss threshold; adjusting the group parameters so that the loss value of the CTR prediction network gradually decreases until it stabilizes at the second loss threshold, wherein the second loss threshold is less than the first loss threshold.

[0070] The click-through rate (CTR) prediction model training method provided in the embodiments of this disclosure firstly obtains a preset sample set, which includes at least one sample. The sample includes user features, features of the manipulated object, and contextual features between the user and the manipulated object. Secondly, it obtains a pre-constructed CTR prediction network, which includes a vector module, a group partitioning module, and a main network module. The vector module converts each sample in the sample set into a feature vector. The group partitioning module generates dynamic group parameters for each feature vector, which are used to calculate the interest of the group to which the sample belongs. The main network module predicts the CTR prediction value based on the group parameters and feature vectors. Thirdly, it selects a sample from the sample set. The sample is then input into the CTR prediction network to obtain the CTR prediction value output by the network. The loss value of the CTR prediction network is calculated. Finally, in response to the CTR prediction network meeting the training completion conditions, the CTR prediction network is used as the CTR prediction model. Therefore, the constructed group segmentation module can dynamically divide samples into multiple groups with different interests during the training process of the click-through rate prediction model, thereby improving the reliability of group and sample segmentation, enhancing the expression of user interests, and improving the robustness and accuracy of the click prediction model.

[0071] In some optional implementations of this disclosure, the aforementioned group partitioning module includes: a group information selection submodule, a group representation learning submodule, and a group parameter generation submodule; the group information selection submodule is used to assign importance weights to the user attribute feature vector and user behavior feature vector in the feature vector to obtain new feature vectors; the group representation learning submodule is used to partition the new feature vectors into group vectors and dynamically update the group probability of each user belonging to each group vector, where the group probability is used to characterize the probability of a user belonging to each group; the group parameter generation submodule generates independent group parameters for each user based on the group probability and the preset group parameters for each group vector; calculating the loss value of the click-through rate prediction network includes: constructing the cross-entropy loss function of the click-through rate prediction network; and calculating the loss value of the click-through rate estimation network based on the cross-entropy loss function.

[0072] In this optional implementation, the executing entity can set corresponding group parameters for each group based on the characteristics of the group vector, such as the interests of each sample in the group. The group parameters for each user can then be determined using the group probability.

[0073] In this optional implementation, in the group information selection submodule, the user behavior vector and user feature vector are concatenated and input into the SENET module (Squeeze-and-Excitation Networks, a compression and excitation module used for feature importance selection) to select the importance of features and amplify the weights of important features (i.e., set corresponding weight values), resulting in a new feature vector. In the group representation learning submodule, each group corresponds to a unique vector representation, i.e., a group vector; then, the new feature vector updates the group vector through read and write operations in the neural Turing machine. The read operation is used to calculate the similarity between the new feature vector and the group vector, obtaining a group probability to represent the selected group; the write operation updates the group vector based on the information in the new feature vector. In the group parameter generation submodule, each group corresponds to an independent set of group parameters, and each sample can dynamically combine the group parameters according to its group probability to obtain sample-specific group parameters.

[0074] The above-mentioned dynamic combination of population parameters to obtain sample-specific population parameters includes: weighted summation of the population probability and the population parameters of each group to obtain sample-specific population parameters. Optionally, the above-mentioned dynamic combination of population parameters to obtain sample-specific population parameters also includes: dot product of the population probability and the population parameters of each group to obtain sample-specific population parameters.

[0075] In this optional implementation, the cross-entropy loss function of the click-through rate prediction network is a function used to calculate the difference between the predicted value and the true value of the click-through rate prediction network. In this disclosure, the cross-entropy loss function of the click-through rate prediction network is a loss function based on the output of the main network module and the sample label. This loss function is related to the predicted output of the click-through rate prediction network and the true value of the sample, that is, it is related to the predicted value of the click-through rate prediction network (a value between 0 and 1) and the sample label (0 / 1).

[0076] The optional implementation provides a group segmentation module that assigns importance weights to user attribute feature vectors and user behavior feature vectors. This allows for the segmentation of users in the sample into multiple different groups, providing a reliable implementation method for realizing the group parameters of the sample users and improving the reliability of the click-through rate prediction model.

[0077] Optionally, the aforementioned group segmentation module further includes: an object information selection submodule, a first representation learning submodule, and a first parameter generation submodule; the object information selection submodule is used to assign importance weights to the feature vectors of the operated objects in the feature vectors to obtain new feature vectors; the first representation learning submodule is used to perform group vector segmentation on the new feature vectors and dynamically update the group probability of each operated object belonging to each group vector, which is used to characterize the probability of the operated object belonging to each group; the first parameter generation submodule generates independent group parameters for each operated object based on the group probability and the preset group parameters for each group vector; calculating the loss value of the click-through rate prediction network includes: constructing the cross-entropy loss function of the click-through rate prediction network; and calculating the loss value of the click-through rate estimation network based on the cross-entropy loss function.

[0078] In some optional implementations of this disclosure, the aforementioned group segmentation module further includes: a group splicing submodule and a multilayer perceptron; the group splicing submodule aggregates the group vectors of all users based on the group probability and the group vectors of each user, and splices the feature vector of the manipulated object and the group vectors of all users in the feature vector to obtain the spliced ​​vector; the multilayer perceptron obtains the group output of each group on the sample based on the spliced ​​vector, and the group output is used to characterize the interest probability of each group on the manipulated object; calculating the loss value of the click-through rate prediction network includes: constructing a first loss function for the vector module and the main network module, and a second loss function for the vector module and the group segmentation module respectively; calculating the main network loss value based on the first loss function; calculating the group segmentation loss value based on the second loss function; and calculating the loss value of the click-through rate prediction network based on the main network loss value and the group segmentation loss value.

[0079] In this optional implementation, the first loss function and the second loss function are functions used to calculate the difference between the predicted value and the true value of the click-through rate prediction network. Since this disclosure sets the output of the main network module and the output of the group segmentation module, the first loss function is a loss function based on the output of the main network module and the sample labels, and the second loss function is a loss function based on the output of the group segmentation module and the sample labels.

[0080] In this optional implementation method, such as Figure 3 By constructing a vector matrix, initial values ​​can be set for the group vectors in advance. The similarity between the user's feature vector and the group vector can be calculated through read operations to determine the group probability. The group vectors can be updated (deleting existing group vectors or adding group vectors) through read operations to achieve the purpose of making the group vectors change dynamically.

[0081] like Figure 3As shown, the population partitioning module enables dynamic population selection and dynamic population parameter generation. Dynamic population selection involves dynamically selecting populations based on sample attribute features to obtain population probabilities rather than simple population partitions. Dynamic population parameter generation dynamically combines population parameters based on the results of population selection, generating unique population parameters for each sample.

[0082] exist Figure 3 In the process, the group probability is multiplied by the group vector of each group to obtain the group vector of the sample. The group vector of the sample is concatenated with the feature vector of the object being operated on (such as a product or advertisement) in the feature vector and then input into a separate multilayer perceptron to obtain the group output y2. The group output y2 is used to represent the group's interest in the sample. The vector module and the main network module are used as the first part. The sample x is input into the first part to obtain the click-through rate prediction value y1. The vector module and the group segmentation module are used as the second part. The sample x is input into the second part to obtain the group output y2. The losses of the first part and the second part are calculated separately and then weighted and summed to obtain the final loss, which is used to adjust the model parameters of the click-through rate prediction network. In the training of the click-through rate prediction network, whether the user clicks or not is used as the label. Clicked samples are positive samples, and exposed samples without clicks are negative samples.

[0083] The optional implementation provides a group partitioning module and a loss value calculation method that construct a first loss function for the vector module and the main network module, and a second loss function for the vector module and the group partitioning module. This can effectively supervise the learning of group vectors while supervising the click-through rate prediction representation, thereby improving the reliability of the click-through rate prediction model training.

[0084] In some optional implementations of this disclosure, the main network module includes a fully connected layer and an output layer; the output layer is used to obtain the parameters of the fully connected layer during the training of the click-through rate prediction network, add the parameters of the fully connected layer to the group parameters, and obtain the click-through rate prediction value based on the added parameters.

[0085] In this optional implementation, the main function of the fully connected layer is to map the feature space calculated by the previous layer (such as the vector module) to the sample label space, that is, to integrate the feature vectors into a single value and classify that value.

[0086] In this optional implementation, the output layer, during the training of the click-through rate prediction network, not only refers to the parameters of the fully connected layer, but also to the group parameters. The group attributes of the samples are added during the processing of the values ​​after the fully connected layer, which improves the accuracy of the click-through rate prediction.

[0087] In some optional implementations of this disclosure, the main network module further includes: a behavior encoding layer, a deep network layer, and a main network splicing submodule; the behavior encoding layer is used to encode the user behavior feature vector in the feature vector to obtain the behavior encoding vector; the deep network layer is used to calculate the user attribute feature vector, the operated object feature vector, and the context feature vector in the feature vector to obtain the calculation vector; the main network splicing submodule is used to splice the behavior encoding vector and the calculation vector, and input the splicing result into the fully connected layer.

[0088] In this optional implementation, user behavior features are collected from the user's past browsing history on a specific network or application, and possess sequential characteristics such as sequence features composed of SKU (Stock Keeping Unit) and ID. The user behavior feature vector is a set of dense vectors obtained by mapping user behavior features, and then averaging the set. The behavior encoding layer can obtain a dense vector from the user behavior feature vector through a matrix mapping.

[0089] In this optional implementation, the deep network layer is a network structure composed of multiple fully connected layers. The basic calculation formula for a fully connected layer is y = f(wx + b), where f is the activation function, w is the weight parameter, and b is the bias parameter. The fully connected layer can map the feature space calculated by the previous layer (such as the vector module) to the sample label space, that is, to transform the feature vectors into specific values.

[0090] In this optional implementation, since the user behavior features in the sample are significantly different from other features in the sample, the user behavior feature vector is encoded separately through a behavior encoding layer, and other feature vectors in the sample are calculated through a deep network layer. The two are then concatenated to obtain the concatenated result. This allows for the differentiation of different types of information in the feature vectors, improving the accuracy of the information input to the main network module.

[0091] Please see Figure 4 The diagram illustrates a flow 400 of an embodiment of the click-through rate (CTR) prediction method provided in this disclosure, which may include the following steps:

[0092] Step 401: Obtain user data, data of the manipulated object, and context data between the user and the manipulated object on the webpage or application under test during the historical time period.

[0093] In this embodiment, the execution entity running the click-through rate prediction method can communicate with a terminal (such as...). Figure 1 The user communicates with terminals 101 and 102 to obtain user data, data of the operated object, and context data between the user and the operated object sent by the terminal.

[0094] In this embodiment, user data, manipulated object data, and contextual data between the user and the manipulated object can be related data when the user operates on the manipulated object within the same webpage or application. For example, user data includes user attribute data and user behavior data. User attribute data includes user age, gender, etc. User behavior data includes the sequence of manipulated objects clicked by the user in historical periods. Manipulated object data includes the name, type, color, and origin of the manipulated object, etc. The contextual data between the user and the manipulated object includes the time and type of user operation on the manipulated object.

[0095] Step 402: Based on user data, manipulated object data, and context data, obtain user features, manipulated object features, and context features.

[0096] In this embodiment, after obtaining user data, manipulated object data, and context data, a feature extractor is used to extract features from the user data, manipulated object data, and context data to obtain user features, manipulated object features, and context features.

[0097] Step 403: Input the user features, the features of the object being operated on, and the context features into the click-through rate prediction model generated by the click-through rate prediction model training method to obtain the click-through rate prediction value output by the click-through rate prediction model.

[0098] In this embodiment, the click-through rate (CTR) prediction model is trained based on the CTR prediction model training method of this embodiment. The CTR prediction model is generated using the CTR prediction model training method described above. The specific generation process of the CTR prediction model can be found in [link to relevant documentation]. Figure 2 The relevant descriptions of the embodiments will not be repeated here.

[0099] It should be noted that the click-through rate (CTR) prediction method in this embodiment can be used to test the CTR prediction models generated in the above embodiments. Furthermore, the CTR prediction model can be continuously optimized based on the test results. This method can also be a practical application of the CTR prediction models generated in the above embodiments. Using the CTR prediction models generated in the above embodiments to identify entity types in test sentences helps improve entity recognition efficiency.

[0100] Click-through rate (CTR) prediction aims to model user interest in candidate products and estimate the probability of a user clicking on them. It's a crucial part of recommendation systems and is widely used in e-commerce platforms. CTR prediction models derived through dynamic group parameter modeling can improve the accuracy of click-through rate prediction, recommending products that better match user interests and ultimately enhancing user satisfaction with the recommendation system.

[0101] The click-through rate (CTR) prediction method provided in this disclosure first acquires user data, object data, and contextual data between the user and the object on a webpage or application under test during a historical time period. Second, based on the user data, object data, and contextual data, user features, object features, and contextual features are obtained. Finally, the user features, object features, and contextual features are input into a CTR prediction model generated by a CTR prediction model training method to obtain the CTR prediction value output by the model. Therefore, by using a pre-trained CTR prediction model to predict user data, object data, and contextual data between the user and the object, an accurate and rapid CTR prediction value can be obtained.

[0102] Further reference Figure 5 As an implementation of the methods shown in the above figures, this disclosure provides an embodiment of a click-through rate prediction model training device, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0103] like Figure 5 As shown, an embodiment of this disclosure provides a click-through rate (CTR) prediction model training device 500, which includes: a sample acquisition unit 501, a network acquisition unit 502, a selection unit 503, an input unit 504, a calculation unit 505, and an acquisition unit 506. The sample acquisition unit 501 can be configured to acquire a preset sample set, which includes at least one sample. The sample includes user features, features of the manipulated object, and context features. The network acquisition unit 502 can be configured to acquire a pre-constructed CTR prediction network, which includes a vector module, a group partitioning module, and a main network module. The vector module converts each sample in the sample set into a feature vector; the group partitioning module generates dynamic group parameters for each feature vector, which are used to calculate the interest of the group to which the sample belongs; and the main network module predicts the CTR prediction value based on the group parameters and feature vectors. The selection unit 503 can be configured to select samples from the sample set. The input unit 504 can be configured to input the sample into the click-through rate (CTR) prediction network to obtain the CTR prediction value output by the CTR prediction network. The calculation unit 505 can be configured to calculate the loss value of the CTR prediction network. The output unit 506 can be configured to use the CTR prediction network as the CTR prediction model in response to the CTR prediction network meeting the training completion conditions.

[0104] In this embodiment, the specific processing of the sample acquisition unit 501, network acquisition unit 502, selection unit 503, input unit 504, calculation unit 505, and acquisition unit 506 in the click-through rate prediction model training device 500, and the resulting technical effects, can be found in the following references: Figure 2 The corresponding steps in the embodiment are 201, 202, 203, 204, 205, and 206.

[0105] In some embodiments, the aforementioned group segmentation module includes: a group information selection submodule, a group representation learning submodule, and a group parameter generation submodule; the group information selection submodule is used to assign importance weights to the user attribute feature vector and the user behavior feature vector in the feature vector to obtain a new feature vector; the group representation learning submodule is used to perform group vector segmentation on the new feature vector and dynamically update the group probability of each user belonging to each group vector, where the group probability is used to characterize the probability of a user belonging to each group; the group parameter generation submodule generates independent group parameters for each user based on the group probability and the group parameters preset for each group vector; the aforementioned calculation unit 505 is further configured to: construct the cross-entropy loss function of the click-through rate estimation network; and calculate the loss value of the click-through rate estimation network based on the cross-entropy loss function.

[0106] In some embodiments, the aforementioned group segmentation module further includes: a group splicing submodule and a multilayer perceptron; the group splicing submodule aggregates the group vectors of all users based on the group probability and the group vectors of each user, and splices the feature vector of the manipulated object and the group vectors of all users in the feature vector to obtain a spliced ​​vector; the multilayer perceptron obtains the group output of each group on the sample based on the spliced ​​vector, and the group output is used to characterize the interest probability of each group on the manipulated object; the aforementioned computing unit 505 is further configured to: construct a first loss function for the vector module and the main network module, and a second loss function for the vector module and the group segmentation module respectively; calculate the main network loss value based on the first loss function; calculate the group segmentation loss value based on the second loss function; and calculate the loss value of the click-through rate prediction network based on the main network loss value and the group segmentation loss value.

[0107] In some embodiments, the main network module includes a fully connected layer and an output layer; the output layer is used to obtain the parameters of the fully connected layer during the training of the click-through rate prediction network, add the parameters of the fully connected layer to the group parameters, and obtain the click-through rate prediction value based on the added parameters.

[0108] In some embodiments, the main network module further includes: a behavior encoding layer, a deep network layer, and a main network splicing submodule; the behavior encoding layer is used to encode the user behavior feature vector in the feature vector to obtain the behavior encoding vector; the deep network layer is used to calculate the user attribute feature vector, the operated object feature vector, and the context feature vector in the feature vector to obtain the calculation vector; the main network splicing submodule is used to splice the behavior encoding vector and the calculation vector, and input the splicing result into the fully connected layer.

[0109] The click-through rate (CTR) prediction model training apparatus provided in the embodiments of this disclosure firstly includes a sample acquisition unit 501 acquiring a preset sample set, which includes at least one sample. The sample includes user features, features of the manipulated object, and contextual features between the user and the manipulated object. Secondly, a network acquisition unit 502 acquires a pre-constructed CTR prediction network, which includes a vector module, a group partitioning module, and a main network module. The vector module is used to convert each sample in the sample set into a feature vector. The group partitioning module is used to generate dynamic group parameters for each feature vector, and the group parameters are used to calculate the interest of the group to which the sample belongs. The main network module predicts the CTR prediction value based on the group parameters and feature vectors. Thirdly, a selection unit 503 selects a sample from the sample set. Fourthly, an input unit 504 inputs the sample into the CTR prediction network to obtain the CTR prediction value output by the CTR prediction network. Fifthly, a calculation unit 505 calculates the loss value of the CTR prediction network. Finally, an acquisition unit 506, in response to the CTR prediction network meeting the training completion conditions, uses the CTR prediction network as the CTR prediction model. Therefore, the constructed group segmentation module can dynamically divide samples into multiple groups with different interests during the training process of the click-through rate prediction model, thereby improving the reliability of group and sample segmentation, enhancing the expression of user interests, and improving the robustness and accuracy of the click prediction model.

[0110] Further reference Figure 6 As an implementation of the methods shown in the above figures, this disclosure provides an embodiment of a click-through rate prediction device, which is similar to... Figure 4 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0111] like Figure 6As shown, an embodiment of this disclosure provides a click-through rate (CTR) prediction device 600, which includes a data acquisition unit 601, a processing unit 602, and a prediction unit 603. The data acquisition unit 601 can be configured to acquire user data, object data, and context data of user interactions with a webpage or application under test during historical time periods. The processing unit 602 can be configured to obtain user features, object features, and context features based on the user data, object data, and context data. The prediction unit 603 can be configured to input the user features, object features, and context features into a CTR prediction model generated by a CTR prediction model training device to obtain a CTR prediction value output by the CTR prediction model.

[0112] In this embodiment, the click-through rate prediction model is trained using a click-through rate prediction model training device.

[0113] In this embodiment, the specific processing of the data acquisition unit 601, the processing unit 602, and the prediction unit 603, and the resulting technical effects, can be found by referring to [the relevant documentation]. Figure 4 The corresponding steps are 401, 402, and 403 in the embodiment.

[0114] The following is for reference. Figure 7 It shows a schematic diagram of the structure of an electronic device 700 suitable for implementing embodiments of the present disclosure.

[0115] like Figure 7 As shown, the electronic device 700 may include a processing unit (e.g., a central processing unit, a graphics processor, etc.) 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage device 708 into a random access memory (RAM) 703. The RAM 703 also stores various programs and data required for the operation of the electronic device 700. The processing unit 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0116] Typically, the following devices can be connected to I / O interface 705: input devices 706 including, for example, touchscreens, touchpads, keyboards, mice, etc.; output devices 707 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 708 including, for example, magnetic tapes, hard disks, etc.; and communication devices 709. Communication device 709 allows electronic device 700 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 7An electronic device 700 with various devices is shown; however, 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 alternatively. Figure 7 Each box shown can represent a device or multiple devices as needed.

[0117] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 709, or installed from storage device 708, or installed from ROM 702. When the computer program is executed by processing device 701, it performs the functions defined in the methods of embodiments of this disclosure.

[0118] It should be noted that the computer-readable medium in the embodiments of this disclosure may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In the embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In the embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (Radio Frequency), etc., or any suitable combination thereof.

[0119] The aforementioned computer-readable medium may be included in the aforementioned server; or it may exist independently and not assembled into the server. The aforementioned computer-readable medium carries one or more programs that, when executed by the server, cause the server to: acquire a pre-set sample set, the sample set including at least one sample, the sample including: user features, features of the manipulated object, and contextual features between the user and the manipulated object; acquire a pre-constructed click-through rate (CTR) prediction network, the CTR prediction network including: a vector module, a group partitioning module, and a main network module, wherein the vector module is used to convert each sample in the sample set into a feature vector; the group partitioning module is used to generate dynamic group parameters for each feature vector, the group parameters representing the group to which the sample corresponding to the feature vector belongs; the main network module predicts the CTR prediction value based on the group parameters and feature vectors; and perform the following training steps: select a sample from the sample set; input the sample into the CTR prediction network to obtain the CTR prediction value output by the CTR prediction network; calculate the loss value of the CTR prediction network; and, in response to the CTR prediction network meeting the training completion condition, use the CTR prediction network as the CTR prediction model.

[0120] Computer program code for performing the operations of embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—and conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0121] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0122] The units described in the embodiments of this disclosure can be implemented in software or hardware. The described units can also be housed in a processor; for example, it can be described as: a processor including a sample acquisition unit, a network acquisition unit, a selection unit, an input unit, a calculation unit, and an acquisition unit. The names of these units do not necessarily limit the unit itself; for example, a sample acquisition unit can also be described as a unit "configured to acquire a preset sample set, the sample set including at least one sample, the sample including: user features, characteristics of the object being operated on, and context features".

[0123] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for training a click-through rate (CTR) prediction model, the method comprising: Obtain a preset sample set, the sample set including at least one sample, the sample including: user features, features of the operated object, and context features between the user and the operated object; A pre-constructed click-through rate (CTR) prediction network is obtained. This network includes a vector module, a group segmentation module, and a main network module. The vector module transforms each sample in the sample set into a feature vector. The group segmentation module generates dynamic group parameters for each feature vector, which are used to calculate the interest of the group to which a sample belongs. The main network module predicts the CTR based on the group parameters and the feature vectors. The group parameters are combined with the parameters of the main network module to obtain new parameters for the main network module, and these new parameters change as the click-through rate prediction network is trained. The group segmentation module includes: a group information selection submodule, a group representation learning submodule, and a group parameter generation submodule. The group information selection submodule assigns importance weights to the user attribute feature vector and user behavior feature vector in the feature vector to obtain new feature vectors. The group representation learning submodule performs group vector segmentation on the new feature vectors and dynamically updates the group probability of each user belonging to each group vector. The group probability is used to represent the probability that a user belongs to each group. The group parameter generation submodule generates independent group parameters for each user based on the group probability and the preset group parameters for each group vector. The following training steps are performed: a sample is selected from the sample set; the sample is input into the click-through rate (CTR) prediction network to obtain the CTR prediction value output by the CTR prediction network; the loss value of the CTR prediction network is calculated; in response to the CTR prediction network meeting the training completion condition, the CTR prediction network is used as the CTR prediction model.

2. The method according to claim 1, wherein, The calculation of the loss value of the click-through rate prediction network includes: constructing the cross-entropy loss function of the click-through rate prediction network; and calculating the loss value of the click-through rate prediction network based on the cross-entropy loss function.

3. The method according to claim 2, wherein, The group segmentation module further includes a group splicing submodule and a multilayer perceptron; the group splicing submodule aggregates the group vectors of all users based on the group probability and the group vectors of each user, and splices the feature vector of the manipulated object in the feature vector and the group vectors of all users to obtain a spliced ​​vector; the multilayer perceptron obtains the group output of each group on the sample based on the spliced ​​vector, and the group output is used to characterize the interest probability of each group on the manipulated object; The calculation of the loss value of the click-through rate prediction network includes: constructing a first loss function for the vector module and the main network module, and a second loss function for the vector module and the group segmentation module; calculating the main network loss value based on the first loss function; calculating the group segmentation loss value based on the second loss function; and calculating the loss value of the click-through rate prediction network based on the main network loss value and the group segmentation loss value.

4. The method according to claim 1, wherein, The main network module includes a fully connected layer and an output layer. The output layer is used to obtain the parameters of the fully connected layer during the training of the click-through rate prediction network, add the parameters of the fully connected layer to the group parameters, and obtain the click-through rate prediction value based on the added parameters.

5. The method according to claim 4, wherein, The main network module further includes: a behavior encoding layer, a deep network layer, and a main network splicing submodule; the behavior encoding layer is used to encode the user behavior feature vector in the feature vector to obtain a behavior encoding vector; the deep network layer is used to calculate the user attribute feature vector, the manipulated object feature vector, and the context feature vector in the feature vector to obtain a calculation vector; the main network splicing submodule is used to splice the behavior encoding vector and the calculation vector, and input the splicing result into the fully connected layer.

6. A click-through rate (CTR) prediction method, the method comprising: Acquire user data, data of the manipulated object, and contextual data between the user and the manipulated object on the webpage or application under test during a historical time period. Based on the user data, the manipulated object data, and the context data, user features, manipulated object features, and context features are obtained. The user features, the features of the object being operated on, and the context features are input into the click-through rate prediction model generated by the click-through rate prediction model training method according to any one of claims 1-5, and the click-through rate prediction value output by the click-through rate prediction model is obtained.

7. A click-through rate prediction model training device, the device comprising: The sample acquisition unit is configured to acquire a preset sample set, the sample set including at least one sample, the sample including: user features, features of the object being operated on, and context features; The network acquisition unit is configured to acquire a pre-constructed click-through rate (CTR) prediction network, which includes a vector module, a group segmentation module, and a main network module. The vector module transforms each sample in the sample set into a feature vector. The group segmentation module generates dynamic group parameters for each feature vector, which are used to calculate the interest of the group to which the sample belongs. The main network module predicts the CTR based on the group parameters and the feature vectors. The group parameters are combined with the parameters of the main network module to obtain new parameters for the main network module, and these new parameters change as the click-through rate prediction network is trained. The group segmentation module includes: a group information selection submodule, a group representation learning submodule, and a group parameter generation submodule. The group information selection submodule assigns importance weights to the user attribute feature vector and user behavior feature vector in the feature vector to obtain new feature vectors. The group representation learning submodule performs group vector segmentation on the new feature vectors and dynamically updates the group probability of each user belonging to each group vector. The group probability is used to represent the probability that a user belongs to each group. The group parameter generation submodule generates independent group parameters for each user based on the group probability and the preset group parameters for each group vector. The selection unit is configured to select a sample from the sample set; The input unit is configured to input the sample into the click-through rate prediction network to obtain the click-through rate prediction value output by the click-through rate prediction network; The computing unit is configured to calculate the loss value of the click-through rate prediction network; The obtained unit is configured to use the click-through rate prediction network as a click-through rate prediction model in response to the click-through rate prediction network meeting the training completion condition.

8. A click-through rate (CTR) prediction device, the device comprising: The data acquisition unit is configured to acquire user data, object data, and context data of the user's actions on the webpage or application under test during a historical time period. The processing unit is configured to obtain user features, manipulated object features, and context features based on the user data, manipulated object data, and context data. The prediction unit is configured to input the user features, the features of the manipulated object, and the context features into the click-through rate prediction model generated by the click-through rate prediction model training device according to claim 7, and obtain the click-through rate prediction value output by the click-through rate prediction model.

9. An electronic device, comprising: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-6.

10. A computer-readable medium having a computer program stored thereon, wherein, When the program is executed by the processor, it implements the method as described in any one of claims 1-6.

Images