CN116992292A

CN116992292A - Click rate estimation model training method and device and click rate estimation method and device

Info

Publication number: CN116992292A
Application number: CN202311121149.0A
Authority: CN
Inventors: 马璇; 代彬丁; 张泽华
Original assignee: Beijing Jingdong Century Trading Co Ltd; Beijing Wodong Tianjun Information Technology Co Ltd
Current assignee: Beijing Jingdong Century Trading Co Ltd; Beijing Wodong Tianjun Information Technology Co Ltd
Priority date: 2023-09-01
Filing date: 2023-09-01
Publication date: 2023-11-03

Abstract

The disclosure provides a click rate estimation model training method and device. One embodiment of the method comprises the following steps: acquiring a preset sample set, wherein the sample set at least comprises one sample; obtaining a pre-constructed click rate estimation network, wherein the click rate estimation network comprises: the system comprises a vector module, a group dividing module and a main network module, wherein the vector module is used for converting each sample in a sample set into a characteristic vector; the group dividing module is used for generating dynamic group parameters for each feature vector; the main network module predicts and obtains a click rate predicted value based on the group parameters and the feature vectors; the following training steps are performed: selecting a sample from the sample set; inputting the sample into a click rate estimation network to obtain a click rate estimated value; calculating a click rate estimated loss value of the network; and responding to the click rate estimation network meeting the training completion condition, and taking the click rate estimation network as a click rate estimation model. According to the embodiment, the accuracy of click rate estimation is improved.

Description

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

Technical Field

The disclosure relates to the technical field of computers, in particular to the technical field of artificial intelligence, and especially relates to a click rate estimation model training method and device, a click rate estimation method and device, electronic equipment and a computer readable storage medium.

Background

In conventional click-through rate estimation models, samples of all users typically share a set of model parameters. To improve parameter diversity, existing methods generally divide a sample into multiple populations, allowing each population to learn its own parameters;

existing multi-population learning faces two problems. On the one hand, the division of the groups needs to be completed manually, and the manual division mode cannot traverse all possible groups, so that the optimal effect is difficult to achieve; on the other hand, after the group division is completed, each sample can only belong to one group, and the mode ignores the diversity of the samples, and limits the expression of the user interest.

Disclosure of Invention

The embodiment of the disclosure provides a click rate estimation model training method and device, a click rate estimation method and device, electronic equipment and a computer readable storage medium.

In a first aspect, an embodiment of the present disclosure provides a click rate estimation model training method, including: obtaining a preset sample set, wherein the sample set at least comprises one sample, and the sample comprises: user features, operated object features, and contextual features between the user and operated object; obtaining a pre-constructed click rate estimation network, wherein the click rate estimation network comprises: the system comprises a vector module, a group dividing module and a main network module, wherein the vector module is used for converting each sample in a sample set into a characteristic vector; the group dividing module is used for generating dynamic group parameters for each feature vector, and the group parameters are used for calculating the interests of the group to which the sample belongs; the main network module predicts and obtains a click rate predicted value based on the group parameters and the feature vectors; the following training steps are performed: selecting a sample from the sample set; inputting the sample into a click rate estimation network to obtain a click rate estimated value output by the click rate estimation network; calculating a click rate estimated loss value of the network; and responding to the click rate estimation network meeting the training completion condition, and taking the click rate estimation network as a click rate estimation model.

In some embodiments, the above population partitioning module comprises: a group information selection sub-module, a group representation learning sub-module and a group parameter generation sub-module; the group information selection submodule is used for carrying out importance weight assignment on the user attribute feature vector and the user behavior feature vector in the feature vectors to obtain new feature vectors; the group representation learning submodule is used for carrying out group vector division on the new feature vectors and dynamically updating the group probability of each user belonging to each group vector, wherein the group probability is used for representing the probability of each user belonging to each group; the group parameter generation submodule generates independent group parameters for each user based on the group probability and group parameters preset for each group vector; calculating a click rate estimation network loss value includes: constructing a cross entropy loss function of a click rate estimation network; and calculating a loss value of the click rate estimation network based on the cross entropy loss function.

In some embodiments, the above population division module further comprises: group splicing sub-module and multi-layer perceptron; the group splicing sub-module is used for acquiring group vectors of all users by aggregation based on the group probability and the group vectors of all users, and acquiring spliced vectors by the operated object feature vectors and the group vectors of all users in the spliced feature vectors; the multi-layer perceptron obtains group output of each group to the sample based on the spliced vector, wherein the group output is used for representing interest probability of each group to the operated object; calculating a click rate estimation network loss value includes: respectively constructing a first loss function of the vector module and the main network module, and a second loss function of the vector module and the group dividing module; calculating a main network loss value based on the first loss function; calculating a group loss value based on the second loss function; and calculating the loss value of the click rate estimation network based on the loss value of the main network and the group loss value.

In some embodiments, the master network module includes: a full connection layer and an output layer; the output layer is used for acquiring parameters of the full-connection layer when the click rate pre-estimated network is trained, adding the parameters of the full-connection layer with the group parameters, and obtaining a click rate pre-estimated value based on the added parameters.

In some embodiments, the master network module further includes: a behavior coding layer, a depth network layer and a main network splicing sub-module; the behavior coding layer is used for coding the user behavior feature vector in the feature vectors to obtain a behavior coding vector; the depth network layer is used for calculating the user attribute feature vector, the operated object feature vector and the context feature vector in the feature vectors to obtain calculation vectors; the main network splicing sub-module is used for splicing the behavior coding vector and the calculation vector and inputting the splicing result into the full-connection layer.

In a second aspect, an embodiment of the present disclosure provides a click rate estimation method, including: acquiring user data of an operated object, operated object data and context data between a user and the operated object on a webpage or application to be detected operated by the user in a historical time period; obtaining user characteristics, operated object characteristics and context characteristics based on the user data, the operated object data and the context data; inputting the user characteristics, the operated object characteristics and the context characteristics into the click rate estimation model generated by the click rate estimation model training method of any embodiment of the first aspect to obtain a click rate estimated value output by the click rate estimation model.

In a third aspect, an embodiment of the present disclosure provides a click rate estimation model training apparatus, including: a sample acquisition unit configured to acquire a preset sample set including at least one sample including: user features, operated object features, and contextual features; the network acquisition unit is configured to acquire a pre-constructed click rate estimation network, and the click rate estimation network comprises: the system comprises a vector module, a group dividing module and a main network module, wherein the vector module is used for converting each sample in a sample set into a characteristic vector; the group dividing module is used for generating dynamic group parameters for each feature vector, and the group parameters are used for calculating the interests of the group to which the sample belongs; the main network module predicts and obtains a click rate predicted value based on the group parameters and the feature vectors; a selecting unit configured to select a sample from a sample set; the input unit is configured to input the sample into a click rate estimation network to obtain a click rate estimated value output by the click rate estimation network; a calculation unit configured to calculate a loss value of the click rate estimation network; the obtaining unit is configured to respond to the click rate estimation network to meet the training completion condition, and takes the click rate estimation network as a click rate estimation model.

In some embodiments, the above population partitioning module comprises: a group information selection sub-module, a group representation learning sub-module and a group parameter generation sub-module; the group information selection submodule is used for carrying out importance weight assignment on the user attribute feature vector and the user behavior feature vector in the feature vectors to obtain new feature vectors; the group representation learning submodule is used for carrying out group vector division on the new feature vectors and dynamically updating the group probability of each user belonging to each group vector, wherein the group probability is used for representing the probability of each user belonging to each group; the group parameter generation submodule generates independent group parameters for each user based on the group probability and group parameters preset for each group vector; the above-mentioned calculation unit is further configured to: constructing a cross entropy loss function of a click rate estimation network; and calculating a loss value of the click rate estimation network based on the cross entropy loss function.

In some embodiments, the above population division module further comprises: group splicing sub-module and multi-layer perceptron; the group splicing sub-module is used for acquiring group vectors of all users by aggregation based on the group probability and the group vectors of all users, and acquiring spliced vectors by the operated object feature vectors and the group vectors of all users in the spliced feature vectors; the multi-layer perceptron obtains group output of each group to the sample based on the spliced vector, wherein the group output is used for representing interest probability of each group to the operated object; the above-mentioned calculation unit is further configured to: respectively constructing a first loss function of the vector module and the main network module, and a second loss function of the vector module and the group dividing module; calculating a main network loss value based on the first loss function; calculating a group loss value based on the second loss function; and calculating the loss value of the click rate estimation network based on the loss value of the main network and the group loss value.

In a fourth aspect, an embodiment of the present disclosure provides a click rate estimating apparatus, including: a data acquisition unit configured to acquire user data, operated object data, and context data of an operated object on a web page or application to be detected operated by a user in a history period; a processing unit configured to obtain a user feature, an operated object feature, and a context feature based on the user data, the operated object data, and the context data; the prediction unit is configured to input the user characteristics, the operated object characteristics and the context characteristics into the click rate prediction model generated by the click rate prediction model training device in any embodiment of the third aspect, so as to obtain a click rate predicted value output by the click rate prediction model.

In a fifth aspect, embodiments of the present disclosure provide an electronic device comprising: one or more processors; a storage device having one or more programs stored thereon; the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method as described in any embodiment of the first or second aspects.

In a sixth aspect, embodiments of the present disclosure provide a computer readable medium having stored thereon a computer program which when executed by a processor implements a method as described in any of the embodiments of the first or second aspects.

The click rate estimation model training method and device provided by the embodiment of the disclosure first obtain a preset sample set, where the sample set includes at least one sample, and the sample includes: user features, operated object features, and contextual features between the user and operated object; secondly, a pre-constructed click rate estimation network is obtained, wherein the click rate estimation network comprises: the system comprises a vector module, a group dividing module and a main network module, wherein the vector module is used for converting each sample in a sample set into a characteristic vector; the group dividing module is used for generating dynamic group parameters for each feature vector, and the group parameters are used for calculating the interests of the group to which the sample belongs; the main network module predicts and obtains a click rate predicted value based on the group parameters and the feature vectors; again, selecting a sample from the sample set; inputting the sample into a click rate estimation network to obtain a click rate estimated value output by the click rate estimation network; calculating a loss value of the click rate prediction network again; and finally, responding to the click rate estimation network to meet the training completion condition, and taking the click rate estimation network as a click rate estimation model. Therefore, the constructed group division module can dynamically divide the sample into a plurality of different groups in the training process of the click rate prediction model, so that the reliability of group division is ensured, the interest expression of a user is ensured, and the robustness and the accuracy of the click prediction model are improved.

Drawings

Other features, objects and advantages of the present disclosure will become more apparent upon reading of the detailed description of non-limiting embodiments, made with reference to the following drawings:

FIG. 1 is an exemplary system architecture diagram in which an embodiment of the present disclosure may be applied;

FIG. 2 is a flow chart of one embodiment of a click rate estimation model training method according to the present disclosure;

FIG. 3 is a schematic diagram of a structure of a click rate estimation network of the present disclosure;

FIG. 4 is a flow chart of one embodiment of a click rate estimation method according to the present disclosure;

FIG. 5 is a schematic diagram of one embodiment of a click rate estimation model training apparatus according to the present disclosure;

FIG. 6 is a schematic diagram illustrating a configuration of one embodiment of a click rate estimation device according to the present disclosure;

fig. 7 is a schematic structural diagram of an electronic device suitable for use in implementing embodiments of the present disclosure.

Detailed Description

The present disclosure is described in further detail below with reference to the drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the invention and are not limiting of the invention. It should be noted that, for convenience of description, only the portions related to the present invention are shown in the drawings.

It should be noted that, without conflict, the embodiments of the present disclosure and features of the embodiments may be combined with each other. The present disclosure will be described in detail below with reference to the accompanying drawings in conjunction with embodiments.

FIG. 1 illustrates an exemplary system architecture 100 in which the click rate estimation model training method or click rate estimation method of the present disclosure may be applied.

As shown in fig. 1, the system architecture 100 may include terminals 101, 102, a network 103, a database server 104, and a server 105. The network 103 serves as a medium for providing a communication link between the terminals 101, 102, the database server 104 and the server 105. The network 103 may include various connection types, such as wired, wireless communication links, or fiber optic cables, among others.

The user 110 may interact with the server 105 via the network 103 using the terminals 101, 102 to receive or send messages or the like. The terminals 101, 102 may have various client applications installed thereon, such as model training class applications, image recognition applications, shopping class applications, payment class applications, web browsers, instant messaging tools, and the like.

The terminals 101 and 102 may be hardware or software. When the terminals 101, 102 are hardware, they may be various electronic devices with display screens, including but not limited to smartphones, tablet computers, electronic book readers, MP3 players (Moving Picture Experts Group Audio Layer III, dynamic video experts compression standard audio layer 3), laptop and desktop computers, and the like. When the terminals 101, 102 are software, they can be installed in the above-listed electronic devices. Which may be implemented as multiple software or software modules (e.g., to provide distributed services), or as a single software or software module. The present invention is not particularly limited herein.

Database server 104 may be a database server that provides various services. For example, a database server may have stored therein a sample set. The sample set contains a large number of samples, each of which is different, and the samples may include user features, manipulated object features, and contextual features between the user and the manipulated object. The user 110 may also select samples from the set of samples stored by the database server 104 via the terminals 101, 102.

The server 105 may also be a server providing various services, such as a background server providing support for various applications displayed on the terminals 101, 102. The background server may train the click rate estimation model by using samples in the sample set sent by the terminals 101 and 102, and may send the click rate estimation model obtained by training to the terminals 101 and 102. In this way, the user may apply the generated click rate prediction model to predict the click rate prediction value.

The database server 104 and the server 105 may be hardware or software. When they are hardware, they may be implemented as a distributed server cluster composed of a plurality of servers, or as a single server. When they are software, they may be implemented as a plurality of software or software modules (e.g., to provide distributed services), or as a single software or software module. The present invention is not particularly limited herein.

It should be noted that, the click rate estimation model training method or the click rate estimation method provided by the embodiments of the present disclosure is generally executed by the server 105. Accordingly, a click rate estimation model training device or a click rate estimation device is also typically provided in the server 105.

It should be noted that the database server 104 may not be provided in the system architecture 100 in cases where the server 105 may implement the relevant functions of the database server 104.

It should be understood that the number of terminals, networks, database servers, and servers in fig. 1 are merely illustrative. There may be any number of terminals, networks, database servers, and servers, as desired for implementation.

Aiming at the problems that manual division is needed for population division in the prior art and each sample can only belong to one population after the population division, the disclosure provides a click rate estimation model training method, which learns the population to which the sample belongs and dynamically divides the sample into a plurality of populations with various interests, so that the information characterization of the sample is richer and more stable, the accuracy of click rate estimation is improved, as shown in fig. 2, a flow 200 of one embodiment of the click rate estimation model training method according to the disclosure is shown, and the click rate estimation model training method comprises the following steps:

Step 201, a preset sample set is obtained.

In this embodiment, the sample set is a sample set collected in advance for training the click rate estimation model, where samples in the sample set may be obtained by collecting sample data from the internet and performing feature extraction (for example, by a feature extractor) on the sample data, where the sample set includes at least one sample, and each sample includes a user feature, an object feature to be operated, and a context feature between the user and the object to be operated.

In this embodiment, the feature of the operated object may be a feature of the operated object in the same website or application specified content, where the operated object includes: commodity or advertisement, the characteristics of commodity include: the color of the commodity, the type of the commodity, the place of origin of the commodity, etc. The advertisement features include: advertisement content, advertisement exposure time, etc.

In this embodiment, the user features are features of interested persons in the website or the application-specific content, and the user features include: the attribute features of the user and the behavior features of the user, wherein the user features refer to user attribute class features such as age, sex and the like. The user behavior feature refers to a user behavior sequence feature, i.e., a sequence of goods that the user has clicked on in the past period of time.

In this embodiment, the context feature between the user and the operated object is a feature related to the user feature and the operated object feature, for example, the context feature includes: whether the user clicks the similar items, whether the user clicks the same type of advertisements, and the like.

In this embodiment, the execution subject of the click rate estimation model training method (e.g., the server 105 shown in fig. 1) may acquire the sample set in various ways. For example, the executing entity may obtain the sample set stored therein from a database server (e.g., database server 104 shown in fig. 1) through a wired connection or a wireless connection. As another example, a user may collect a sample through a terminal (e.g., terminals 101, 102 shown in fig. 1). In this way, the executing body may receive samples collected by the terminal and store the samples locally, thereby generating a sample set.

Step 202, obtaining a pre-constructed click rate estimation network.

In this embodiment, the click rate estimation network includes: the system comprises a vector module, a group dividing module and a main network module, wherein the vector module is used for converting each sample in a sample set into a characteristic vector; the group dividing module is used for generating dynamic group parameters for each feature vector, and the group parameters are used for calculating the interests of the group to which the sample belongs; the main network module predicts and obtains a click rate predicted value based on the group parameters and the feature vectors.

In this embodiment, the vector module is a module for converting a sample into a vector, where the user feature, the operated object feature and the context feature in the sample belong to an ID class feature and are sparse, and the vector module can convert the user feature, the operated object feature and the context feature in the sample into feature vectors with dense multidimensional features.

In this embodiment, the Click Rate predicted value is an average predicted Click Rate obtained by predicting the feature, the average predicted Click Rate is an average of predicted values of Click-Through-Rate (CTR), and the Click-Through Rate is a ratio of the number of times that specified content is clicked and exposed on a website or an application, and is generally an important indicator for measuring recommendation efficiency in a recommendation system. In general, the click through rate may be the actual number of clicks of the operated object on the web page or application (also the number of target pages reached) divided by the amount of display of the operated object on the web page or application (Show content).

In this embodiment, compared with a traditional click rate estimation model, the click rate estimation network of the present disclosure has a group division module for providing group parameters, the main network module needs to consider the group parameters in real time when predicting the click rate estimation value, the group parameters are representations of specific distributions of samples belonging to different samples, and the main network module can more clearly understand the group in which each sample is located by adding the group parameters.

In this embodiment, the population parameter is a parameter value of a population to which a sample (such as a user or a commodity or an advertisement in the sample) belongs, the population parameter is represented as a network parameter in network calculation, and can be changed along with click rate prediction network iterative training, after the sample is attributed to different populations, the interest of the population to which the sample belongs can be calculated through the population parameter, for example, the overall division module is equivalent to y '=wx' +b, where W is the population parameter, x 'is an input sample of the population division module, y' is an output value of the population division module, and b is a super parameter. After the sample is acquired, the group dividing module divides the feature vector into preset groups based on the similarity between the feature vector of the sample and the vector of the preset groups to obtain the probability that the sample belongs to the groups, namely the group probability; based on the interested aspects of the sample, setting corresponding group parameters for each group, and based on the probability that the sample belongs to the group and each group parameter, obtaining the group parameters of the group dividing module.

In this embodiment, after the group parameter is obtained, the main network module combines (e.g., adds points and multiplies points) the group parameter with the parameter of the main network module to obtain a new parameter of the main network module, and when the click rate prediction network trains, the new parameter of the main network module changes along with the change until the click rate prediction network reaches the training completion condition.

In this embodiment, the group parameters of the group division module may change with the parameters of the vector module and the main network module. When training the click rate prediction network, parameters of a group parameter, a vector module and a main network module are required to be adjusted through a loss function of the click rate prediction network, convergence of the click rate prediction network is determined, and the training completion condition is achieved.

In this embodiment, the main Network module may use an NN (Neural Network) Network or a Network result after the NN Network is transformed, and analyze, by using the main Network module, group parameters, user characteristics, operated object characteristics and context characteristics to predict a click rate estimated value of clicking the operated object by the user.

At step 203, samples are selected from the sample set.

In this embodiment, the executing body may select a sample from the sample set obtained in step 201, and execute the training steps from step 204 to step 206. The selection manner and the selection number of the samples are not limited in the present application. For example, at least one sample may be randomly selected.

And 204, inputting the sample into a click rate estimation network to obtain a click rate estimated value output by the click rate estimation network.

In this embodiment, the samples include user features, operated object features and context features, the samples selected from the sample set are input into the click rate estimation network, the click rate estimation network analyzes the user features, the operated object features and the context features, determines click rate pre-estimated values corresponding to the current user features, the operated object features and the context features, and can measure the recommendation efficiency of each operated object in the recommendation system through the click rate pre-estimated values, thereby providing an effective means for recommending and displaying the operated objects in the recommendation system of the web pages and the applications.

In this embodiment, the click rate estimation network is used to characterize a correspondence between a sample and a click rate estimated value, where the sample includes a user feature and an operated object feature, and the click rate estimated value is an estimated value of a click rate of the user on the operated object under the current user feature and the operated object feature.

Step 205 calculates the click rate estimation network loss value.

In this embodiment, in the training step of the click rate estimation network, multiple iterative training is generally required to be performed to converge the overall loss of the click rate estimation network, each iterative training sequentially performs steps 203 to 205, after the loss value of step 205 is calculated, the parameters of the click rate estimation network are adjusted based on the loss value (the loss value reflects the error), and then the next iterative training is performed, and steps 203 to 205 are continuously performed until the training completion condition is satisfied.

In this embodiment, in order to effectively adjust parameters of the click rate estimation network, a loss function needs to be set for the click rate estimation network, and the loss function can calculate errors between a forward calculation result and a true value of each iteration of the click rate estimation network, so that the next training is guided to be performed in a correct direction through the errors.

It should be noted that, in the training step of the current iteration number, if the click rate estimation network already meets the training completion condition, the executing body will not input a sample into the click rate estimation network, and the click rate estimation network is the click rate estimation model after training is completed.

And step 206, responding to the click rate estimation network to meet the training completion condition, and taking the click rate estimation network as a click rate estimation model.

In this embodiment, the training completion condition includes at least one of: the training iteration times of the click rate estimation network reach a preset iteration threshold; and when the change rate of the model parameters of the click rate prediction network (including the parameters of the group parameter, the vector module and the main network module) is smaller than a preset threshold, determining that the click rate prediction network meets the training completion condition. For example, training iterations reach 5 thousand times. The change rate of the model parameters of the click rate estimation network is less than 0.05. In this embodiment, setting the training completion condition can accelerate the model convergence speed.

In step 206, when the model parameter changes less than the predetermined threshold, it is determined that the loss value of the click rate estimation network converges, and the click rate estimation model training is completed.

In some optional implementations of the disclosure, in response to the click rate estimation network not meeting the training completion condition, the model parameters of the click rate estimation network are adjusted so that the loss function of the click rate estimation network converges, and the steps 203-205 are continuously performed.

In this embodiment, the adjusting the model parameters of the click rate estimation network in response to the click rate estimation network not meeting the training completion condition so that the loss function convergence of the click rate estimation network includes: maintaining the group parameters unchanged, adjusting the parameters of the vector module and the main network module, and repeatedly executing the steps 203-205 until the loss value of the click rate estimated network reaches a first loss threshold; and adjusting the group parameters to gradually reduce the loss value of the click rate estimation network until the loss value is stabilized to a second loss threshold value, wherein the second loss threshold value is smaller than the first loss threshold value.

According to the click rate estimation model training method provided by the embodiment of the disclosure, firstly, a preset sample set is obtained, the sample set at least comprises one sample, and the sample comprises: user features, operated object features, and contextual features between the user and operated object; secondly, a pre-constructed click rate estimation network is obtained, wherein the click rate estimation network comprises: the system comprises a vector module, a group dividing module and a main network module, wherein the vector module is used for converting each sample in a sample set into a characteristic vector; the group dividing module is used for generating dynamic group parameters for each feature vector, and the group parameters are used for calculating the interests of the group to which the sample belongs; the main network module predicts and obtains a click rate predicted value based on the group parameters and the feature vectors; again, selecting a sample from the sample set; inputting the sample into a click rate estimation network to obtain a click rate estimated value output by the click rate estimation network; calculating a loss value of the click rate prediction network again; and finally, responding to the click rate estimation network to meet the training completion condition, and taking the click rate estimation network as a click rate estimation model. Therefore, the constructed group division module can dynamically divide the sample into a plurality of groups with different interests in the training process of the click rate prediction model, so that the reliability of group and sample division is improved, the interest expression of a user is improved, and the robustness and the accuracy of the click prediction model are improved.

In some optional implementations of the disclosure, the group partitioning module includes: a group information selection sub-module, a group representation learning sub-module and a group parameter generation sub-module; the group information selection submodule is used for carrying out importance weight assignment on the user attribute feature vector and the user behavior feature vector in the feature vectors to obtain new feature vectors; the group representation learning submodule is used for carrying out group vector division on the new feature vectors and dynamically updating the group probability of each user belonging to each group vector, wherein the group probability is used for representing the probability of each user belonging to each group; the group parameter generation submodule generates independent group parameters for each user based on the group probability and group parameters preset for each group vector; calculating a click rate estimation network loss value includes: constructing a cross entropy loss function of a click rate estimation network; and calculating a loss value of the click rate estimation network based on the cross entropy loss function.

In this alternative implementation, the execution subject may set corresponding population parameters for each population based on characteristics of the population vector, e.g., interests of each sample in the population. And the group probability can be used for determining the group parameters of the users.

In this optional implementation manner, in the group information selection sub-module, after the user behavior vector and the user feature vector are spliced, the user behavior vector and the user feature vector are input into a SENET module (a Squeeze-and-Excitation Networks module, a compression and excitation module, and a module for selecting the importance of the feature) to perform the importance selection of the feature, and the weight of the important feature is amplified (i.e. a corresponding weight value is set) to obtain a new feature vector. In the group representation learning sub-module, each group corresponds to a unique vector representation, namely a group vector; and then updating the group vector by the new feature vector through read operation and write operation in the neuropsychiatric machine. The reading operation is used for carrying out similarity calculation between the new feature vector and the group vector to obtain a group probability for representing the selected group; the write operation updates the population vector based on information in the new feature vector. In the group parameter generation submodule, each group corresponds to a set of independent group parameters, and each sample can dynamically combine the group parameters according to the group probability to obtain the group parameters specific to the sample.

The step of dynamically combining the group parameters to obtain the group parameters specific to the sample comprises the following steps: and carrying out weighted summation on the group probability and the group parameters of each group to obtain the group parameters specific to the sample. Optionally, the dynamically combining the group parameters to obtain the group parameters specific to the sample further includes: and carrying out dot multiplication on the population probability and the population parameters of each population to obtain the population parameters specific to the sample.

In this alternative implementation, the cross entropy loss function of the click rate estimation network is a function for calculating a difference between a predicted value and a true value of the click rate estimation network, and in this disclosure, the cross entropy loss function of the click rate estimation network is a loss function based on a master network module output and a sample label, where the loss function is related to a true value of the predicted output and the sample of the click rate estimation network, that is, to a predicted value (a value between 0 and 1) of the click rate estimation network and a sample label (0/1).

The group dividing module provided by the alternative implementation mode carries out importance weight assignment on the user attribute feature vector and the user behavior feature vector, can divide users in the sample into a plurality of different groups, provides a reliable implementation mode for realizing group parameters of the users of the sample, and improves the reliability of click rate estimation model generation.

Optionally, the above group dividing module further includes: the system comprises an object information selection sub-module, a first representation learning sub-module and a first parameter generation sub-module; the object information selection submodule is used for carrying out importance weight assignment on the feature vector of the operated object in the feature vector to obtain a new feature vector; the first representation learning submodule is used for carrying out group vector division on the new feature vectors and dynamically updating the group probability that each operated object belongs to each group vector, wherein the group probability is used for representing the probability that each operated object belongs to each group; the first parameter generation submodule generates independent group parameters for each operated object based on the group probability and group parameters preset for each group vector; calculating a click rate estimation network loss value includes: constructing a cross entropy loss function of a click rate estimation network; and calculating a loss value of the click rate estimation network based on the cross entropy loss function.

In some optional implementations of the disclosure, the above group partitioning module further includes: group splicing sub-module and multi-layer perceptron; the group splicing sub-module is used for acquiring group vectors of all users by aggregation based on the group probability and the group vectors of all users, and acquiring spliced vectors by the operated object feature vectors and the group vectors of all users in the spliced feature vectors; the multi-layer perceptron obtains group output of each group to the sample based on the spliced vector, wherein the group output is used for representing interest probability of each group to the operated object; calculating a click rate estimation network loss value includes: respectively constructing a first loss function of the vector module and the main network module, and a second loss function of the vector module and the group dividing module; calculating a main network loss value based on the first loss function; calculating a group loss value based on the second loss function; and calculating the loss value of the click rate estimation network based on the loss value of the main network and the group loss value.

In this alternative implementation, the first loss function and the second loss function are functions for calculating a difference between a predicted value and a true value of the click rate prediction network, and since the present disclosure sets the output of the main network module and the output of the group division module, the first loss function is a loss function based on the output of the main network module and the sample tag, and the second loss function is a loss function based on the output of the group division module and the sample tag.

In this alternative implementation, as shown in fig. 3, an initial value may be set for the group vector in advance by constructing a vector matrix, and the similarity between the feature vector of the user and the group vector is calculated through a read operation, so as to determine the group probability; the group vector is updated (the existing group vector is deleted or the group vector is added) through the reading operation, so that the purpose of dynamically changing the group vector is achieved.

As shown in fig. 3, dynamic population selection and dynamic population parameter generation may be implemented by a population partitioning module, where dynamic population selection: dynamic group selection is carried out through the attribute characteristics of the sample, so that group probability is obtained instead of single group division. Dynamic population parameter generation: and dynamically combining the group parameters according to the group selection result, and generating specific group parameters for each sample.

In fig. 3, the group probability and the group vector of each group are multiplied by each other to obtain a group vector of a sample, the group vector of the sample and the feature vector of an operated object (such as commodity or advertisement) in the feature vector are spliced and then input into a single multi-layer perceptron to obtain a group output y2, and the group output y2 is used for representing the interest of the group in the sample. Taking the vector module and the main network module as a first part, and inputting a sample x into the first part to obtain a click rate predicted value y1; taking the vector module and the group dividing module as a second part, and inputting a sample x into the second part to obtain a group output y2; and respectively calculating the losses of the first part and the second part, and then carrying out weighted summation to obtain the final loss, wherein the final loss is used for adjusting the model parameters of the click rate estimation network. When the click rate is estimated to be used for network training, whether a user clicks or not is adopted as a label, a positive sample is clicked, and a negative sample is exposed but not clicked.

The group division module and the loss value calculation mode provided by the alternative implementation mode respectively construct a first loss function for the vector module and the main network module, and construct a second loss function for the vector module and the group division module, so that the study of group vectors can be effectively supervised while the click rate prediction representation is monitored, and the training reliability of the click rate prediction model is improved.

In some optional implementations of the disclosure, the master network module includes: a full connection layer and an output layer; the output layer is used for acquiring parameters of the full-connection layer when the click rate pre-estimated network is trained, adding the parameters of the full-connection layer with the group parameters, and obtaining a click rate pre-estimated value based on the added parameters.

In this alternative 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) into a sample mark space, that is, integrate the feature vector into a value, and classify the value.

In the alternative implementation mode, the output layer refers to parameters of the full-connection layer and population parameters in the click rate estimation network training process, population attributes of samples are added in the process of processing the values processed by the full-connection layer, and accuracy of click rate estimation is improved.

In some optional implementations of the disclosure, the master network module further includes: a behavior coding layer, a depth network layer and a main network splicing sub-module; the behavior coding layer is used for coding the user behavior feature vector in the feature vectors to obtain a behavior coding vector; the depth network layer is used for calculating the user attribute feature vector, the operated object feature vector and the context feature vector in the feature vectors to obtain calculation vectors; the main network splicing sub-module is used for splicing the behavior coding vector and the calculation vector and inputting the splicing result into the full-connection layer.

In this alternative implementation, the user behavior feature is a sequence feature that the user has collected in the past from the browsing history of a particular network or application, and has sequence characteristics such as a sequence feature composed of sku (Stock Keeping Unit, stock level unit), id, etc. The user behavior feature vector is a vector obtained by mapping and converting the user behavior feature into a set of dense vectors and then averaging the set. The behavior coding layer may map the user behavior feature vector through a matrix to obtain a dense vector.

In this alternative implementation, the deep network layer is a network structure composed of a plurality of fully connected layers, where the basic calculation formula of the fully connected layers is y=f (wx+b), f is an activation function, w is a weight parameter, and b is a bias parameter. The feature space mapping sample mark space calculated by the front layer (such as a vector module) can be used by the full connection layer, namely, the feature vector is subjected to specific value transformation.

In the alternative implementation manner, because the user behavior characteristics in the sample are greatly different from other samples, the user behavior characteristic vectors are respectively encoded through the behavior encoding layer, the other characteristic vectors in the sample are calculated through the depth network layer, and then the user behavior characteristic vectors and the other characteristic vectors are spliced to obtain a splicing result, so that the characteristic vectors can be subjected to different types of information distinguishing processing, and the accuracy of the information input by the main network module information is improved.

Referring to fig. 4, a flowchart 400 of an embodiment of a click rate estimation method provided by the present disclosure is shown, where the click rate estimation method may include the following steps:

step 401, obtaining user data of an operated object, operated object data and context data between a user and the operated object on a webpage or application to be tested operated by the user in a historical time period.

In this embodiment, the execution body on which the click rate estimation method operates may communicate with a terminal (such as terminals 101 and 102 in fig. 1) to obtain user data, operated object data, and context data between the user and the operated object sent by the terminal.

In this embodiment, the user data, the operated object data, and the context data between the user and the operated object may be related data when the user operates the operated object on the same web page or under the application, for example, the user data includes: user attribute data and user behavior data, wherein the user attribute data includes user age, gender, and the like. The user behavior data includes: the sequence of operated objects clicked by the user in the history period. The operated object data includes: operated object name, operated object type, operated object color, operated object place of origin, and the like. The context data between the user and the operated object is the time, the kind of operation, etc. at which the user operates the operated object.

Step 402, obtaining user features, operated object features and context features based on the user data, the operated object data and the context data.

In this embodiment, after obtaining the user data, the operated object data, and the context data, feature extraction is performed on the user data, the operated object data, and the context data by the feature extractor, so as to obtain the user feature, the operated object feature, and the context feature.

And step 403, inputting the user characteristics, the operated object characteristics and the contextual characteristics into a click rate estimation model generated by a click rate estimation model training method to obtain a click rate estimated value output by the click rate estimation model.

In this embodiment, the click rate estimation model is obtained by training based on the click rate estimation model training method of this embodiment. The click rate estimation model is generated by adopting the click rate estimation model training method of the embodiment, wherein the specific generation process of the click rate estimation model can be referred to the related description of the embodiment of fig. 2, and the description is omitted here.

It should be noted that, the click rate estimation method of the present embodiment may be used to test the click rate estimation model generated in each embodiment. And then the click rate estimation model can be continuously optimized according to the test result. The method may also be a practical application method of the click rate estimation model generated in the above embodiments. The click rate estimation model generated by the embodiments is adopted to identify the entity type in the test sentence, which is beneficial to improving the identification efficiency of the entity.

The click rate estimation task aims at modeling the interest of the user in candidate commodities and estimating the click probability of the user, is an important link in a recommendation system, and has wide application in an e-commerce platform. The click rate estimation model obtained through dynamic group parameter modeling can improve the click through rate estimation accuracy, and recommend commodities which are more in line with the interests of users to the users, so that the user satisfaction degree of a recommendation system is improved.

The click rate estimation method provided by the embodiment of the disclosure includes the steps of firstly, acquiring user data of an operated object, operated object data and context data between a user and the operated object on a webpage or an application to be detected operated by the user in a historical time period; secondly, obtaining user characteristics, operated object characteristics and context characteristics based on the user data, the operated object data and the context data; and finally, inputting the user characteristics, the operated object characteristics and the contextual characteristics into a click rate estimation model generated by a click rate estimation model training method to obtain a click rate estimated value output by the click rate estimation model. Therefore, the click rate estimated value can be accurately and rapidly obtained by estimating the user data, the operated object data and the context data between the user and the operated object through the click rate estimated model obtained through pre-training.

With further reference to fig. 5, as an implementation of the method shown in the foregoing figures, the present disclosure provides an embodiment of a click rate estimation model training apparatus, where the embodiment of the apparatus corresponds to the embodiment of the method shown in fig. 2, and the apparatus may be specifically applied to various electronic devices.

As shown in fig. 5, an embodiment of the present disclosure provides a click rate estimation model training apparatus 500, where the apparatus 500 includes: sample acquisition unit 501, network acquisition unit 502, selection unit 503, input unit 504, calculation unit 505, and acquisition unit 506. The sample acquiring unit 501 may be configured to acquire a preset sample set, where the sample set includes at least one sample, and the sample includes: user features, manipulated object features, and contextual features. The network obtaining unit 502 may be configured to obtain a pre-constructed click rate estimation network, where the click rate estimation network includes: the system comprises a vector module, a group dividing module and a main network module, wherein the vector module is used for converting each sample in a sample set into a characteristic vector; the group dividing module is used for generating dynamic group parameters for each feature vector, and the group parameters are used for calculating the interests of the group to which the sample belongs; the main network module predicts and obtains a click rate predicted value based on the group parameters and the feature vectors. The selection unit 503 may be configured to select a sample from a sample set. The input unit 504 may be configured to input the sample into a click rate estimation network, so as to obtain a click rate estimated value output by the click rate estimation network. The calculation unit 505 may be configured to calculate a loss value of the click rate estimation network. The obtaining unit 506 may be configured to use the click rate estimation network as the click rate estimation model in response to the click rate estimation network satisfying the training completion condition.

In the embodiment, in the click rate estimation model training apparatus 500, the specific processes of the sample acquiring unit 501, the network acquiring unit 502, the selecting unit 503, the input unit 504, the calculating unit 505, the obtaining unit 506 and the technical effects thereof may refer to the steps 201, 202, 203, 204, 205 and 206 in the corresponding embodiment of fig. 2.

In some embodiments, the above population partitioning module comprises: a group information selection sub-module, a group representation learning sub-module and a group parameter generation sub-module; the group information selection submodule is used for carrying out importance weight assignment on the user attribute feature vector and the user behavior feature vector in the feature vectors to obtain new feature vectors; the group representation learning submodule is used for carrying out group vector division on the new feature vectors and dynamically updating the group probability of each user belonging to each group vector, wherein the group probability is used for representing the probability of each user belonging to each group; the group parameter generation submodule generates independent group parameters for each user based on the group probability and group parameters preset for each group vector; the above-described calculation unit 505 is further configured to: constructing a cross entropy loss function of a click rate estimation network; and calculating a loss value of the click rate estimation network based on the cross entropy loss function.

In some embodiments, the above population division module further comprises: group splicing sub-module and multi-layer perceptron; the group splicing sub-module is used for acquiring group vectors of all users by aggregation based on the group probability and the group vectors of all users, and acquiring spliced vectors by the operated object feature vectors and the group vectors of all users in the spliced feature vectors; the multi-layer perceptron obtains group output of each group to the sample based on the spliced vector, wherein the group output is used for representing interest probability of each group to the operated object; the above-described calculation unit 505 is further configured to: respectively constructing a first loss function of the vector module and the main network module, and a second loss function of the vector module and the group dividing module; calculating a main network loss value based on the first loss function; calculating a group loss value based on the second loss function; and calculating the loss value of the click rate estimation network based on the loss value of the main network and the group loss value.

In the click rate estimation model training device provided in the embodiments of the present disclosure, first, a sample obtaining unit 501 obtains a preset sample set, where the sample set includes at least one sample, and the sample includes: user features, operated object features, and contextual features between the user and operated object; next, the network obtaining unit 502 obtains a click rate estimation network constructed in advance, where the click rate estimation network includes: the system comprises a vector module, a group dividing module and a main network module, wherein the vector module is used for converting each sample in a sample set into a characteristic vector; the group dividing module is used for generating dynamic group parameters for each feature vector, and the group parameters are used for calculating the interests of the group to which the sample belongs; the main network module predicts and obtains a click rate predicted value based on the group parameters and the feature vectors; again, the selection unit 503 selects a sample from the sample set; from time to time, the input unit 504 inputs the sample into a click rate estimation network to obtain a click rate estimated value output by the click rate estimation network; again, the calculation unit 505 calculates a loss value of the click rate estimation network; finally, the obtaining unit 506 responds to the click rate estimation network to meet the training completion condition, and takes the click rate estimation network as a click rate estimation model. Therefore, the constructed group division module can dynamically divide the sample into a plurality of groups with different interests in the training process of the click rate prediction model, so that the reliability of group and sample division is improved, the interest expression of a user is improved, and the robustness and the accuracy of the click prediction model are improved.

With further reference to fig. 6, as an implementation of the method shown in the foregoing figures, the present disclosure provides an embodiment of a click rate estimating apparatus, where the embodiment of the apparatus corresponds to the embodiment of the method shown in fig. 4, and the apparatus may be specifically applied to various electronic devices.

As shown in fig. 6, an embodiment of the present disclosure provides a click rate estimating apparatus 600, the apparatus 600 including: a data acquisition unit 601, a processing unit 602, and a prediction unit 603. The data obtaining unit 601 may be configured to obtain user data, operated object data, and context data of an operated object on a web page or an application to be tested operated by a user in a historical period. The processing unit 602 may be configured to obtain the user feature, the operated object feature, and the context feature based on the user data, the operated object data, and the context data. The prediction unit 603 may be configured to input the user feature, the operated object feature, and the context feature into the click rate prediction model generated by the click rate prediction model training device, to obtain the click rate predicted value output by the click rate prediction model.

In this embodiment, the click rate estimation model is obtained by training using a click rate estimation model training device.

In this embodiment, the specific processes of the data acquisition unit 601, the processing unit 602, the prediction unit 603 and the technical effects thereof may refer to step 401, step 402, and step 403 in the corresponding embodiment of fig. 4, respectively.

Referring now to fig. 7, a schematic diagram of an electronic device 700 suitable for use in implementing embodiments of the present disclosure is shown.

As shown in fig. 7, the electronic device 700 may include a processing means (e.g., a central processor, a graphics processor, etc.) 701, which may 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 means 708 into a Random Access Memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the electronic device 700 are also stored. The processing device 701, the ROM 702, and the RAM 703 are connected to each other through a bus 704. An input/output (I/O) interface 705 is also connected to bus 704.

In general, the following devices may be connected to the I/O interface 705: input devices 706 including, for example, a touch screen, touchpad, keyboard, mouse, etc.; an output device 707 including, for example, a liquid crystal display (LCD, liquid Crystal Display), a speaker, a vibrator, and the like; storage 708 including, for example, magnetic tape, hard disk, etc.; and a communication device 709. The communication means 709 may allow the electronic device 700 to communicate wirelessly or by wire with other devices to exchange data. While fig. 7 shows an electronic device 700 having various means, it is to be understood that not all of the illustrated means are required to be implemented or provided. More or fewer devices may be implemented or provided instead. Each block shown in fig. 7 may represent one device or a plurality of devices as needed.

In particular, according to embodiments of the present disclosure, the processes described above with reference to flowcharts may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program embodied on a computer readable medium, the computer program comprising program code for performing the method shown in the flowcharts. In such an embodiment, the computer program may be downloaded and installed from a network via communication device 709, or installed from storage 708, or installed from ROM 702. The above-described functions defined in the methods of the embodiments of the present disclosure are performed when the computer program is executed by the processing device 701.

It should be noted that the computer readable medium of the embodiments of the present disclosure may be a computer readable signal medium or a computer readable storage medium, or any combination of the two. The computer readable storage medium can be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples of the computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer diskette, 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 disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In an embodiment of the present disclosure, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. Whereas in embodiments of the present disclosure, the computer-readable signal medium may comprise a data signal propagated in baseband or as part of a carrier wave, with computer-readable program code embodied therein. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination of the foregoing. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to: electrical wires, fiber optic cables, RF (Radio Frequency), and the like, or any suitable combination thereof.

The computer readable medium may be contained in the server; or may exist alone without being assembled into the server. The computer readable medium carries one or more programs which, when executed by the server, cause the server to: obtaining a preset sample set, wherein the sample set at least comprises one sample, and the sample comprises: user features, operated object features, and contextual features between the user and operated object; obtaining a pre-constructed click rate estimation network, wherein the click rate estimation network comprises: the system comprises a vector module, a group dividing module and a main network module, wherein the vector module is used for converting each sample in a sample set into a characteristic vector; the group dividing module is used for generating dynamic group parameters for each feature vector, and the group parameters are used for representing the group to which the sample corresponding to the feature vector belongs; the main network module predicts and obtains a click rate predicted value based on the group parameters and the feature vectors; the following training steps are performed: selecting a sample from the sample set; inputting the sample into a click rate estimation network to obtain a click rate estimated value output by the click rate estimation network; calculating a click rate estimated loss value of the network; and responding to the click rate estimation network meeting the training completion condition, and taking the click rate estimation network as a click rate estimation model.

Computer program code for carrying out operations of embodiments of the present disclosure may be written in one or more programming languages, including an object oriented programming language such as Java, smalltalk, C ++ and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computer (for example, through the Internet using an Internet service provider).

The flowcharts and block diagrams in the figures 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 the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

The units involved in the embodiments described in the present disclosure may be implemented by means of software, or may be implemented by means of hardware. The described units may also be provided in a processor, for example, described as: a processor comprises a sample acquisition unit, a network acquisition unit, a selection unit, an input unit, a calculation unit and an acquisition unit. Wherein the names of the units do not constitute a limitation of the unit itself in certain cases, for example, the sample acquisition unit may also be described as "configured to acquire a preset sample set including at least one sample including: user feature, operated object feature, and context feature.

The foregoing description is only of the preferred embodiments of the present disclosure and description of the principles of the technology being employed. It will be appreciated by those skilled in the art that the scope of the invention in the embodiments of the present disclosure is not limited to the specific combination of the above technical features, but encompasses other technical features formed by any combination of the above technical features or their equivalents without departing from the spirit of the invention. Such as the above-described features, are mutually substituted with (but not limited to) the features having similar functions disclosed in the embodiments of the present disclosure.

Claims

1. A click rate estimation model training method, the method comprising:

obtaining a preset sample set, wherein the sample set at least comprises one sample, and the sample comprises: user features, operated object features, and contextual features between the user and operated object;

obtaining a pre-constructed click rate estimation network, wherein the click rate estimation network comprises: the system comprises a vector module, a group dividing module and a main network module, wherein the vector module is used for converting each sample in a sample set into a characteristic vector; the group dividing module is used for generating dynamic group parameters for each feature vector, and the group parameters are used for calculating the interests of the group to which the sample belongs; the main network module predicts and obtains a click rate predicted value based on the group parameters and the feature vector;

the following training steps are performed: selecting a sample from the sample set; inputting the sample into the click rate estimation network to obtain a click rate estimated value output by the click rate estimation network; calculating the loss value of the click rate estimation network; and responding to the click rate estimation network meeting the training completion condition, and taking the click rate estimation network as a click rate estimation model.

2. The method of claim 1, wherein the population partitioning module comprises: a group information selection sub-module, a group representation learning sub-module and a group parameter generation sub-module; the group information selection submodule is used for carrying out importance weight assignment on the user attribute feature vector and the user behavior feature vector in the feature vectors to obtain new feature vectors; the group representation learning sub-module is used for carrying out group vector division on the new feature vectors and dynamically updating the group probability of each user belonging to each group vector, wherein the group probability is used for representing the probability of each user belonging to each group; the group parameter generation submodule generates independent group parameters for each user based on the group probability and group parameters preset for each group vector;

the calculating the loss value of the click rate estimation network comprises the following steps: constructing a cross entropy loss function of the click rate estimation network; and calculating a loss value of the click rate estimation network based on the cross entropy loss function.

3. The method of claim 2, wherein the population partitioning module further comprises: group splicing sub-module and multi-layer perceptron; the group stitching submodule is used for converging group vectors of all users based on the group probability and the group vectors of all users, stitching the feature vector of the operated object in the feature vector with the group vectors of all users to obtain a stitching vector; the multi-layer perceptron obtains group output of each group to the sample based on the splicing vector, wherein the group output is used for representing interest probability of each group to the operated object;

The calculating the loss value of the click rate estimation network comprises the following steps: respectively constructing a first loss function of the vector module and the main network module and a second loss function of the vector module and the group dividing module; calculating a main network loss value based on the first loss function; calculating a group loss value based on the second loss function; and calculating the loss value of the click rate estimation network based on the main network loss value and the group loss value.

4. The method of claim 1, wherein the master network module comprises: a full connection layer and an output layer; the output layer is used for acquiring the parameters of the full-connection layer when the click rate estimation network is trained, adding the parameters of the full-connection layer with the group parameters, and obtaining a click rate estimated value based on the added parameters.

5. The method of claim 4, wherein the master network module further comprises: a behavior coding layer, a depth network layer and a main network splicing sub-module; the behavior coding layer is used for coding the user behavior feature vectors in the feature vectors to obtain behavior coding vectors; the depth network layer is used for calculating a user attribute feature vector, an operated object feature vector and a context feature vector in the feature vectors to obtain calculation vectors; and the main network splicing sub-module is used for splicing the behavior coding vector and the calculation vector and inputting a splicing result into the full-connection layer.

6. A click rate estimation method, the method comprising:

acquiring user data of an operated object, operated object data and context data between a user and the operated object on a webpage or application to be detected operated by the user in a historical time period;

obtaining user characteristics, operated object characteristics and context characteristics based on the user data, the operated object data and the context data;

inputting the user features, the operated object features and the contextual features into a click rate estimation model generated by the click rate estimation model training method according to one of claims 1 to 5 to obtain a click rate estimated value output by the click rate estimation model.

7. A click rate estimation model training apparatus, the apparatus comprising:

a sample acquisition unit configured to acquire a preset sample set including at least one sample including: user features, operated object features, and contextual features;

a network acquisition unit configured to acquire a click rate estimation network constructed in advance, the click rate estimation network including: the system comprises a vector module, a group dividing module and a main network module, wherein the vector module is used for converting each sample in a sample set into a characteristic vector; the group dividing module is used for generating dynamic group parameters for each feature vector, and the group parameters are used for calculating the interests of the group to which the sample belongs; the main network module predicts and obtains a click rate predicted value based on the group parameters and the feature vector;

A selecting unit configured to select a sample from the sample set;

the input unit is configured to input the sample into the click rate estimation network to obtain a click rate estimated value output by the click rate estimation network;

a calculation unit configured to calculate a loss value of the click rate estimation network;

the obtaining unit is configured to respond to the click rate estimation network meeting training completion conditions, and take the click rate estimation network as a click rate estimation model.

8. A click rate estimation device, the device comprising:

a data acquisition unit configured to acquire user data, operated object data, and context data of an operated object on a web page or application to be detected operated by a user in a history period;

a processing unit configured to obtain a user feature, an operated object feature, and a context feature based on the user data, the operated object data, and the context data;

the prediction unit is configured to input the user feature, the operated object feature and the context feature into the click rate prediction model generated by the click rate prediction model training device according to claim 8, so as to obtain a click rate predicted value output by the click rate prediction model.

9. An electronic device, comprising:

one or more processors;

a storage device having one or more programs stored thereon;

when executed by the one or more processors, causes the one or more processors to implement the method of any of claims 1-6.

10. A computer readable medium having stored thereon a computer program, wherein the program when executed by a processor implements the method of any of claims 1-6.