Click rate model generation method, click rate determination method, and related devices

Abstract

The disclosure provides a click rate model generation method, a click rate determination method and related equipment, and relates to the technical field of computers, to solve the problem that the best feature combination cannot be confirmed. The method comprises the following steps: obtaining a plurality of browsing sample data; training a first estimation model based on the plurality of browsing sample data to obtain a second estimation model; calling an input control layer in the first estimation model to perform binary processing on initial first weight parameters in the first estimation model to obtain an input control coefficient vector; adjusting a plurality of attribute features corresponding to each piece of browsing sample data, and inputting target attribute features determined after the adjustment into an input layer in the first estimation model; determining a target element that meets a preset condition from elements in corresponding parameters in the second estimation model; and retraining the second estimation model based on a plurality of attribute features corresponding to the target element to obtain a trained click rate model.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of computer, and particularly relates to a click rate model generation method, a click rate determination method and related equipment. BACKGROUND

[0002] At present, an estimation model for estimating a click rate of a media resource usually needs to filter features in browsing sample data, and a click rate model composed of the filtered features can be used for actual click rate prediction.

[0003] In the related art, feature filtering in the estimation model is realized based on an area under curve (AUC) evaluation method. Specifically, the importance of each feature corresponding to the browsing sample data is calculated, and the importance of all features is sorted. According to the sorting result, part of the features are manually discarded, and a model is re-established based on the remaining features. This filtering method is manually selected based on the performance of each feature, and it cannot confirm whether the remaining features are the best feature combination, and it cannot confirm that the estimation model obtained based on the remaining features is the best estimation model. SUMMARY

[0004] The present disclosure provides a click rate model generation method, a click rate determination method and related equipment to at least solve the problem that the best feature combination cannot be confirmed in the prior art. The technical solutions of the present disclosure are as follows:

[0005] According to a first aspect of the present disclosure, a click rate model generation method is provided, which includes: an electronic device obtains a plurality of browsing sample data, each piece of browsing sample data in the plurality of browsing sample data including a plurality of attribute features of a browsing object and a click rate of the browsing object; a first estimation model is trained based on the plurality of browsing sample data to obtain a second estimation model; wherein, in the training process, an input control layer in the first estimation model is called to perform binary processing on an initial first weight parameter in the first estimation model to obtain an input control coefficient vector; and the input control layer is called to adjust the plurality of attribute features corresponding to each piece of browsing sample data based on each element in the input control coefficient vector, and input the target attribute features determined by the adjustment into an input layer in the first estimation model; the plurality of attribute features, the elements in the first weight parameter, the elements in the input control coefficient vector, and the elements in the corresponding parameters in the second estimation model are in one-to-one correspondence; a target element satisfying a preset condition is determined from the elements in the corresponding parameters in the second estimation model; the plurality of attribute features corresponding to the target element and the second estimation model are trained again based on the target element to obtain a trained click rate model.

[0006] Optionally, the input control layer is specifically configured to input the first weight parameter into the input control function to obtain an input control coefficient vector.

[0007] Optionally, the first estimation model is trained based on the plurality of browsing sample data to obtain a second estimation model, including: training the first estimation model based on the plurality of browsing sample data and a loss function to obtain the second estimation model; wherein the loss function is used to adjust parameters in the first estimation model so that the estimation result of the first estimation model meets a preset requirement, and the first weight parameter tends to be binary in the process of adjusting the parameters in the first estimation model.

[0008] Optionally, the loss function includes a cross-entropy loss function and a target loss function, and the target loss function is related to the first weight parameter.

[0009] Optionally, the target element satisfying the preset condition is determined from the elements in the corresponding parameters in the second estimation model, including: determining an element greater than or equal to a preset threshold from the elements in the corresponding parameters in the second estimation model as the target element satisfying the preset condition.

[0010] Optionally, the second estimation model is retrained based on the plurality of attribute features corresponding to the target element to obtain a trained click rate model, including: constructing a third estimation model based on the plurality of attribute features corresponding to the target element and the second estimation model; retraining the third estimation model to obtain the trained click rate model.

[0011] Optionally, the third estimation model is constructed based on the plurality of attribute features corresponding to the target element and the second estimation model, including: obtaining an adjustment coefficient vector based on the elements in the corresponding parameters in the second estimation model; the adjustment coefficient vector is used to filter the plurality of attribute features of the browsing object; replacing the corresponding parameters in the second estimation model with a target parameter to obtain a replaced second estimation model; the target parameter is composed of the target element; adding an adjustment layer before the input control layer of the replaced second estimation model to obtain the third estimation model, and the adjustment layer is used to filter the plurality of attribute features of the browsing object using the adjustment coefficient vector.

[0012] According to a second aspect of the present disclosure, a click rate determination method is provided, including: an electronic device obtaining browsing data; the browsing data including a plurality of attribute features of a browsing object; processing the browsing data using a click rate model to obtain a click rate corresponding to the browsing object, the click rate model being obtained according to the click rate model generation method of any one of the first aspect.

[0013] According to a third aspect of the present disclosure, a click rate model generation apparatus is provided, which comprises an acquisition module and a processing module. The acquisition module is configured to acquire a plurality of browsing sample data, each of the plurality of browsing sample data comprising a plurality of attribute features of a browsing object and a click rate of the browsing object; the processing module is configured to train a first estimation model based on the plurality of browsing sample data to obtain a second estimation model; wherein, in the training process, an input control layer in the first estimation model is invoked to perform binary processing on an initial first weight parameter in the first estimation model to obtain an input control coefficient vector; and the input control layer is invoked to adjust the plurality of attribute features corresponding to each browsing sample data based on elements in the input control coefficient vector, and input target attribute features determined by the adjustment into an input layer in the first estimation model; the plurality of attribute features, elements in the first weight parameter, elements in the input control coefficient vector, and elements in corresponding parameters in the second estimation model correspond one-to-one; the processing module is further configured to determine target elements satisfying a preset condition from the elements in the corresponding parameters in the second estimation model; and the processing module is further configured to retrain the second estimation model based on the plurality of attribute features corresponding to the target elements to obtain a trained click rate model.

[0014] Optionally, the input control layer is specifically configured to input the first weight parameter into an input control function to obtain the input control coefficient vector.

[0015] Optionally, the processing module is further configured to train the first estimation model based on the plurality of browsing sample data and a loss function to obtain the second estimation model; wherein, the loss function is used to adjust parameters in the first estimation model to make the estimation result of the first estimation model satisfy a preset requirement, and make the first weight parameter tend to binary in the process of adjusting the parameters in the first estimation model.

[0016] Optionally, the loss function comprises a cross-entropy loss function and a target loss function, and the target loss function is related to the first weight parameter.

[0017] Optionally, the processing module is further configured to determine, from the elements in the corresponding parameters in the second estimation model, elements greater than or equal to a preset threshold as the target elements satisfying the preset condition.

[0018] Optionally, the processing module is further configured to construct a third estimation model based on the plurality of attribute features corresponding to the target elements and the second estimation model; and the processing module is further configured to retrain the third estimation model to obtain the trained click rate model.

[0019] Optionally, the processing module is further configured to obtain an adjustment coefficient vector based on the elements in the corresponding parameters in the second estimation model; the adjustment coefficient vector is used to filter the multiple attribute features of the browsing object; the processing module is further configured to replace the corresponding parameters in the second estimation model with target parameters to obtain a replaced second estimation model; the target parameters are composed of target elements; and the processing module is further configured to add an adjustment layer before the input control layer of the replaced second estimation model to obtain a third estimation model, the adjustment layer being used to filter the multiple attribute features of the browsing object by using the adjustment coefficient vector.

[0020] According to a fourth aspect of the present disclosure, a click rate determination apparatus is provided, which comprises an acquisition module and a processing module. The acquisition module is configured to acquire browsing data; the browsing data comprises multiple attribute features of a browsing object; and the processing module is further configured to process the browsing data by using a click rate model to obtain a corresponding click rate of the browsing object, the click rate model being obtained according to the click rate model generation method of any one of the first aspect.

[0021] According to a fifth aspect of the present disclosure, an electronic device is provided, which comprises a processor and a memory for storing instructions executable by the processor; and the processor is configured to execute the instructions to implement the optional click rate model generation method of any one of the first aspect and the click rate determination method of the second aspect.

[0022] According to a sixth aspect of the present disclosure, a computer readable storage medium is provided, which stores instructions; when the instructions in the computer readable storage medium are executed by a processor of an electronic device, the electronic device can execute the optional click rate model generation method of any one of the first aspect and the click rate determination method of the second aspect.

[0023] According to a seventh aspect of the present disclosure, a computer program product is provided, which comprises instructions; when the instructions in the computer program product are executed by a processor of an electronic device, the optional click rate model generation method of any one of the first aspect and the click rate determination method of the second aspect are implemented.

[0024] The technical solutions provided by the embodiments of the present disclosure at least bring the following beneficial effects:

[0025] In the above scheme, the first estimation model is trained based on the browsing sample data to obtain the second estimation model by obtaining a plurality of browsing sample data. The target element is filtered out from the elements in the corresponding parameters in the second estimation model, and the target click rate estimation model is trained based on the plurality of attribute features corresponding to the target element and the second estimation model. Compared with the prior art which needs to manually retain and discard the attribute features of the browsing sample data, the importance of each attribute feature is determined as a whole by training according to the present disclosure. Not only does it not need to be manually set, but also the feature combination determined based on the overall performance is more objective, avoiding the problem that the features selected by humans are not the optimal feature combination. At the same time, the click rate estimation result of the click rate model obtained based on the features filtered according to the present disclosure will be more accurate.

[0026] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure. BRIEF DESCRIPTION OF DRAWINGS

[0027] The accompanying drawings incorporated in the specification and forming a part of it, illustrate embodiments consistent with the present disclosure and together with the description, serve to explain the principles of the present disclosure, and do not limit the present disclosure.

[0028] Figure 1 is one of flowcharts of a click rate model generation method according to an exemplary embodiment;

[0029] Figure 2 is a structural schematic diagram of a first estimation model according to an exemplary embodiment;

[0030] Figure 3 is another one of flowcharts of a click rate model generation method according to an exemplary embodiment;

[0031] Figure 4 is a third one of flowcharts of a click rate model generation method according to an exemplary embodiment;

[0032] Figure 5 is a fourth one of flowcharts of a click rate model generation method according to an exemplary embodiment;

[0033] Figure 6 is one of flowcharts of a click rate determination method according to an exemplary embodiment;

[0034] Figure 7 is a structural block diagram of a click rate model generation device according to an exemplary embodiment;

[0035] Figure 8 is a structural block diagram of a click rate determination device according to an exemplary embodiment;

[0036] Figure 9 is a structural schematic diagram of an electronic device according to an exemplary embodiment. DETAILED DESCRIPTION

[0037] In order for those skilled in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings.

[0038] It should be noted that the terms "first", "second", and the like in the specification and claims of the present disclosure and the above-described drawings are used to distinguish similar objects, and do not necessarily have to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The data involved in the present disclosure can be data authorized by the user or sufficiently authorized by all parties. The implementation described in the following exemplary embodiments does not represent all implementations consistent with the present disclosure. Rather, they are only examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0039] It should also be understood that the term "comprising" indicates the presence of described features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, and / or components.

[0040] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, analyzed data, etc.) involved in the present disclosure are all information and data authorized by the user or sufficiently authorized by all parties.

[0041] At present, a media resource system is usually composed of multiple parts such as targeting, recall, sorting, and display. Among them, targeting is mainly to filter all media resources that meet the conditions for the user according to the user's own attributes (such as gender, age, etc.); recall is mainly to preliminarily filter all media resources filtered by targeting, greatly reducing the number of media resources; sorting is mainly to further filter N media resources according to the media resources obtained by recall, and to predict the click value of the N media resources.

[0042] The click value of the media resource is determined according to a click rate, a conversion rate and a conversion bid. Since the conversion bid is fixed, the conversion rate is determined based on the accuracy of the click rate recommendation. To improve the click value of the media resource, the click rate of the media resource needs to be improved. To improve the click rate of the media resource, the user needs to be recommended the media resource that the user is interested in. Therefore, before pushing the media resource to the user, the click rate of the media resource is usually estimated by using an estimation model, and the media resource with a high estimated click rate is pushed.

[0043] After predicting the click rate, the prediction effect of the estimation model is evaluated by using an AUC evaluation method. In the case where the prediction effect does not meet the requirements, the features in the estimation model need to be screened, and a final target estimation model is created based on the screened features.

[0044] Before screening the features in the estimation model, the importance of each feature needs to be calculated. Then, the features are selected for retention and discard according to the importance. Finally, the target estimation model is constructed based on the retained features. In general, the retained features are manually set features, which cannot guarantee the best effect. In addition, since the multiple attribute features retained are determined by analyzing the importance separately, they cannot be guaranteed to be optimal as a whole.

[0045] Based on the above problems, the embodiment of the present disclosure provides a click rate model generation method. The method includes obtaining a plurality of browsing sample data, training a first estimation model based on the plurality of browsing sample data to obtain a second estimation model, screening a target element from elements in corresponding parameters in the second estimation model, and training the second estimation model again based on the target element and the second estimation model to obtain a click rate model.

[0046] The click rate model generation method provided by the embodiment of the present disclosure is exemplarily described as follows.

[0047] The click rate model generation method provided by the present disclosure can be applied to an electronic device.

[0048] In some embodiments, the electronic device can be a server, a terminal or other electronic device for click rate estimation, and the present disclosure does not limit the electronic device.

[0049] The server can be a single server or a server cluster composed of multiple servers. In some embodiments, the server cluster can also be a distributed cluster. The present disclosure does not limit the specific implementation of the server.

[0050] The terminal can be a mobile phone, a tablet computer, a desktop computer, a laptop computer, a handheld computer, a notebook computer, an ultra-mobile personal computer (UMPC), a netbook, and a cellular phone, a personal digital assistant (PDA), an augmented reality (AR) \ virtual reality (VR) device, or the like, which can install and use a content community application (such as a short video application), and the specific form of the electronic device is not specially limited in the present disclosure. It can interact with the user through one or more ways such as a keyboard, a touchpad, a touch screen, a remote controller, voice interaction, or a handwriting device.

[0051] The technical solutions in the embodiments of the present disclosure will be described clearly and completely below in combination with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, rather than all the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the scope of protection of the present disclosure.

[0052] As shown in Figure 1 When the click rate model generation method is applied to an electronic device, the click rate model generation method can include:

[0053] Step 11, the electronic device acquires a plurality of browsing sample data.

[0054] Each piece of browsing sample data in the plurality of browsing sample data includes a plurality of attribute features of a browsing object and a click rate of the browsing object.

[0055] In some embodiments, the electronic device acquires a plurality of browsing sample data from a log record, and the plurality of browsing sample data includes a browsing object that is clicked by a user and a click rate of the browsing object. According to a plurality of preset feature categories, a plurality of attribute features of each piece of browsing sample data are determined.

[0056] A feature category contains all features after the feature category is exhausted; the attribute feature of the browsing sample data refers to the value corresponding to the browsing sample data for a certain feature category. The click rate of the browsing object specifically refers to the proportion of the number of clicks of a certain browsing object to the total number of clicks of all browsing objects within a period of time. The click rate of the browsing object can reflect the interest of the user to some extent, and the user can be recommended a browsing object that the user is more interested in based on the click rate in the future. Exemplarily, the browsing object can be an advertisement, a commodity, a video, an article, or the like.

[0057] Step 12, the electronic device trains the first estimation model based on the plurality of browsing sample data to obtain a second estimation model.

[0058] In the training process, the input control layer in the first estimation model is called to binarize the initial first weight parameter in the first estimation model to obtain an input control coefficient vector; and the input control layer is called to adjust the plurality of attribute features corresponding to each piece of browsing sample data based on each element in the input control coefficient vector, and input the target attribute features determined by the adjustment into the input layer in the first estimation model; the plurality of attribute features, the elements in the first weight parameter, the elements in the input control coefficient vector, and the elements in the corresponding parameters in the second estimation model are in one-to-one correspondence.

[0059] In some embodiments, the electronic device trains the first estimation model based on the plurality of browsing sample data to obtain a second estimation model. In the training process, the electronic device calls the input control layer in the first estimation model to binarize the first weight parameter in the first estimation model to obtain an input control coefficient vector corresponding to the first weight parameter. Then, each element in the input control coefficient vector is multiplied by the plurality of attribute features corresponding to each piece of browsing sample data to obtain target attribute features, and finally the target attribute features are input into the input layer for training to obtain the second estimation model. The first weight parameter includes a plurality of elements, and the number of elements in the first weight parameter is consistent with the number of the plurality of attribute features of the browsing object in each piece of browsing sample data. Each element in the first weight parameter can be understood as the weight of the plurality of attribute features of the browsing object in each piece of browsing sample data in the plurality of browsing sample data.

[0060] For example, the first estimation model is trained based on the plurality of browsing sample data to obtain a second estimation model. In combination with the above Figure 2 It can be seen that the first estimation model includes an input control layer, an input layer, a hidden layer, and an output layer. The input control layer in the first estimation model binarizes the first weight parameter to obtain an input control coefficient vector. Part of the elements in the input control coefficient vector approaches 0, and part of the elements approaches 1.

[0061] Each piece of browsing sample data includes a plurality of features (feature 1, feature 2, feature 3, feature 4,..., feature n) of one browsing object (1 PV), Xi = [v1, v2, v3,..., vk]. Xi is all features of the i-th piece of browsing sample data, and vk is the k-th feature in the i-th piece of browsing sample data. k is the feature length.

[0062] Then each element in the input control coefficient vector is multiplied by each attribute feature of each browsing sample data to obtain a target attribute feature. The target attribute feature is input into the input layer, the input layer inputs the target attribute feature into the hidden layer, the hidden layer performs activation processing on the target attribute feature of the input layer, and the activated result is input into the output layer.

[0063] The activation processing is realized by an activation function. The activation function satisfies the expression: H = relu(WH) H X new ). In the expression, H is the result obtained after the activation processing of the hidden layer, relu is the activation function, WH is the matrix corresponding to the activation function, X new is the target attribute feature.

[0064] The output layer normalizes the result of the activation processing of the hidden layer to obtain the click rate of the browsing object in the plurality of browsing sample data.

[0065] The normalization processing satisfies the following expression: P = softmax(WpH). In the expression, P is the click rate obtained after the normalization processing of the output layer, softmax is the normalization, and Wp is the matrix corresponding to the normalization.

[0066] Optionally, the input control layer is specifically configured to input the first weight parameter into an input control function to obtain the input control coefficient vector.

[0067] In some embodiments, the first weight parameter is binarized to obtain the input control coefficient vector. Specifically, the first weight parameter is input into the input control function, and the input control coefficient vector is determined based on the calculation of the input control function. The first weight parameter includes n elements, and the n elements are in the form of an n-dimensional vector, and one dimension represents one element. For example, the input control function can be a gating function, and the input control function satisfies the following expression:

[0068]

[0069] In the expression, w is the first weight parameter, the value range of each element in w is (0, 1), and ε is a constant greater than 0, for example, ε = 10 -5 .

[0070] The technical scheme provided by the above embodiments can at least bring the following beneficial effects: in order to distinguish the influence of each feature on the final result as much as possible, the input control function is used to make the first weight parameter tend to two extremes, so as to further quickly determine the influence of each attribute feature in the browsing sample data on the final click rate estimation.

[0071] In combination with the aboveFigure 1 As shown in Figure 3 The step 12 of training the first estimation model based on the plurality of browsing sample data to obtain the second estimation model includes the following steps.

[0072] In step 121, the first estimation model is trained based on the plurality of browsing sample data and a loss function to obtain the second estimation model.

[0073] The loss function is used to adjust the parameters in the first estimation model so that the estimation result of the first estimation model meets the preset requirement, and the first weight parameter tends to be binary in the process of adjusting the parameters in the first estimation model.

[0074] In some embodiments, the electronic device inputs the plurality of browsing sample data into the first estimation model, and the first estimation model outputs the estimated click rate of the browsing object in the plurality of browsing sample data. After obtaining the estimated click rate, the difference between the estimated click rate and the click rate is calculated, which is the loss value. The first estimation model is iteratively trained according to the loss value until the estimation result of the first estimation model meets the preset requirement, i.e., the loss value between the estimated click rate output by the first estimation model and the click rate is less than or equal to the preset threshold, thereby determining the second estimation model. Wherein, according to the loss value, the first estimation model is iteratively trained, which specifically includes iteratively training each element in the first weight parameter in the first estimation model according to the loss value, so that each element in the first weight parameter tends to be binary.

[0075] The technical solutions provided by the above embodiments can at least bring the following beneficial effects: training the first estimation model by the loss function can make the training result of the first estimation model quickly reach the expected effect, and further the loss function can also promote the binary of the first weight parameter to obtain the corresponding parameter in the second estimation model, which is convenient for subsequent screening of target elements based on the elements in the corresponding parameter in the second estimation model.

[0076] Further, the loss function includes a cross-entropy loss function and a target loss function.

[0077] The target loss function is related to the first weight parameter.

[0078] Specifically, the loss function for training the first estimation model includes a cross-entropy loss function and a target loss function, and the target loss function is related to the first weight parameter, specifically for promoting the binary of the first weight parameter. For example, the loss function satisfies the following expression:

[0079] L = L 交叉熵 + L 目标 = L 交叉熵 + a|w|

[0080] wherein w is a first weight parameter, || is an absolute value, and a is a hyperparameter (a parameter whose value is set before the learning process begins).

[0081] Step 13. The electronic device determines a target element that meets a preset condition from elements in the corresponding parameter of the second estimation model.

[0082] In some embodiments, after obtaining the corresponding parameter of the second estimation model, the electronic device screens a target element that meets the requirement according to the specific performance of each element in the corresponding parameter of the second estimation model. The screened target element is an element that has greater relevance to the estimated click rate result.

[0083] In combination with the above Figure 1 As Figure 3 illustrated, step 13 determines a target element that meets a preset condition from the second random vector, including:

[0084] Step 131. The electronic device determines an element that is greater than or equal to a preset threshold value from elements in the corresponding parameter of the second estimation model as a target element that meets a preset condition.

[0085] In some embodiments, after obtaining the specific value of each element in the corresponding parameter of the second estimation model, when it is determined that the value of an element is greater than or equal to a preset threshold value, the element is determined to be one of the target elements.

[0086] For example, the elements in the corresponding parameter of the second estimation model are basically either tending to 0.05 or tending to 0.99. Therefore, the target element is an element tending to 0.99.

[0087] The technical solutions provided by the above embodiments can at least bring the following beneficial effects: target elements are screened based on elements in the corresponding parameter of the second estimation model, so that the screened target elements are directly related to the click rate.

[0088] Step 14. The electronic device performs retraining based on the second estimation model and a plurality of attribute features corresponding to the target element, to obtain a trained click rate model.

[0089] In some embodiments, after the electronic device screens the target element from the corresponding parameter of the second estimation model, it is known from the foregoing that the elements in the corresponding parameter of the second estimation model are in one-to-one correspondence with the plurality of attribute features in each browsing sample data. Therefore, the attribute features corresponding to the target element can be determined, and finally the second estimation model is retrained based on the attribute features corresponding to the target element to create a click rate model. Since the click rate model is constructed based on the second estimation model, the network structures of the click rate model and the second estimation model are the same.

[0090] The technical solutions provided by the above embodiments can bring at least the following beneficial effects: As can be seen from steps 11-14, the disclosure obtains a plurality of browsing sample data, trains a first estimation model based on the browsing sample data, and obtains a second estimation model. In the second estimation model, target elements are selected from elements in corresponding parameters, and a target click rate estimation model is trained based on a plurality of attribute features corresponding to the target elements and the second estimation model. Compared with the prior art, which needs to manually retain and discard attribute features of browsing sample data, the disclosure determines the importance of each attribute feature as a whole by training, which not only does not need to be manually set, but also determines the feature combination based on the overall performance, which is more objective, avoiding the problem that the features selected by humans are not the optimal feature combination. At the same time, the click rate estimation result of the click rate model obtained based on the features selected by the disclosure will be more accurate.

[0091] Optionally, in combination with the above Figure 1 As shown in Figure 4 The above step 14 re-trains based on the plurality of attribute features corresponding to the target elements and the second estimation model to obtain a trained click rate model, including:

[0092] Step 141, the electronic device constructs a third estimation model based on the plurality of attribute features corresponding to the target elements and the second estimation model.

[0093] In some embodiments, the electronic device determines a plurality of attribute features corresponding to the target elements according to the target elements, and re-trains based on the plurality of attribute features and the second estimation model to generate a third estimation model. The network structure of the third estimation model is the same as that of the second estimation model.

[0094] Optionally, in combination with the above Figure 1 As shown in Figure 5 The above step 141 constructs a third estimation model based on the plurality of attribute features corresponding to the target elements and the second estimation model, including:

[0095] Step 1411, the electronic device obtains an adjustment coefficient vector based on the elements in the corresponding parameters in the second estimation model.

[0096] The adjustment coefficient vector is used to filter the plurality of attribute features of the browsing object.

[0097] In some embodiments, since part of the plurality of attribute features of the browsing object cannot contribute to the click rate estimation result, the element in the corresponding parameter in the second estimation model can be used to determine whether the attribute feature corresponding to the element can contribute to the click rate estimation result. When the element in the corresponding parameter in the second estimation model tends to 1, it is considered that the attribute feature corresponding to the element can contribute to the click rate estimation result. When the element in the corresponding parameter in the second estimation model tends to 0, it is considered that the attribute feature corresponding to the element cannot contribute to the click rate estimation result. Therefore, the adjustment coefficient vector can be generated according to the value of the element in the corresponding parameter in the second estimation model, so as to screen the plurality of attribute features of the browsing object.

[0098] Step 1412, the electronic device replaces the corresponding parameter in the second estimation model with the target parameter to obtain a replaced second estimation model.

[0099] The target parameter is composed of target elements.

[0100] In some embodiments, the target elements are elements directly related to the click rate estimation result, so that the target elements in the corresponding parameters in the second estimation model need to be screened out, and the target parameter is composed of the target elements, and finally the corresponding parameters in the second estimation model are replaced by the target parameter. Therefore, the elements in the parameters of the second estimation model are all elements directly related to the click rate estimation.

[0101] Step 1413, the electronic device adds an adjustment layer before the input control layer of the replaced second estimation model to obtain a third estimation model.

[0102] The adjustment layer is used to screen the plurality of attribute features of the browsing object by using the adjustment coefficient vector.

[0103] In some embodiments, the electronic device adds an adjustment layer before the input control layer of the replaced second estimation model to generate a third estimation model. The third estimation model uses the adjustment coefficient vector in the adjustment layer to screen the plurality of attribute features of the browsing object, and directly multiplies the screened attribute features with the target parameter, and inputs the multiplication result into the click rate estimation result.

[0104] The technical solutions provided by the above embodiments can at least bring the following beneficial effects: by adding an adjustment layer directly related to the click rate estimation and replacing the parameters in the second estimation model, a third estimation model can be obtained.

[0105] Step 142, the electronic device re-trains the third estimation model to obtain a trained click rate model.

[0106] In some embodiments, the electronic device retrains the created third prediction model to obtain a more accurate click-through rate (CTR) model.

[0107] The technical solution provided by the above embodiments can bring at least the following beneficial effects: by training the third prediction model twice, a more accurate click-through rate model can be obtained.

[0108] like Figure 6 As shown, when the click-through rate (CTR) determination method is applied to electronic devices, the CTR determination method may include:

[0109] Step 61: The electronic device acquires browsing data.

[0110] The browsing data includes multiple attribute characteristics of the browsing object.

[0111] In some embodiments, the browsing data acquired by the electronic device may be browsing data that is potentially to be displayed on a third-party platform. The third-party platform may be a shopping platform, advertising platform, information platform, etc. The browsing data may be a data packet containing the viewed object and its attribute characteristics. For example, the viewed object may be media resources such as advertisements, products, videos, and articles. When the viewed object is an advertisement, multiple attribute characteristics of the viewed object may include the author of the advertisement, the time the advertisement was published, etc. When the viewed object is a product, multiple attribute characteristics of the viewed object may include the product's listing time, product description, etc. When the viewed object is a video, multiple attribute characteristics of the viewed object may include the video description, video author, etc. When the viewed object is an article, multiple attribute characteristics of the viewed object may include the article description, article publication time, etc.

[0112] Step 62: The electronic device processes the browsing data using a click-through rate (CTR) model to obtain the CTR of the browsing object.

[0113] The click-through rate (CTR) model is derived from the CTR model generation method.

[0114] In some embodiments, the electronic device inputs browsing data into a trained click-through rate (CTR) model. This model performs binarization, activation processing, and normalization on multiple attribute features of the browsed objects contained in the browsing data to obtain an estimated CTR for the browsing data. The estimated CTR can then be used to determine whether to recommend the browsing data to the user.

[0115] The technical solution provided by the above embodiments can bring at least the following beneficial effects: since the click-through rate prediction model is obtained by training multiple times using browsing sample data, the click-through rate of the training data predicted by the click-through rate prediction model will be more accurate.

[0116] The above combinationFigures 1-6 The methods provided by the embodiments of this disclosure are described in detail. To achieve the above-described functions, the click-through rate (CTR) model generation apparatus includes hardware structures and / or software modules corresponding to each function, and these hardware structures and / or software modules can constitute a CTR model generation apparatus. Those skilled in the art will readily recognize that, based on the algorithm steps of the examples described in conjunction with the embodiments disclosed herein, this disclosure can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this disclosure.

[0117] This disclosure embodiment can divide the click-through rate (CTR) model generation device into functional modules based on the above method example. For example, the CTR model generation device can be divided into functional modules corresponding to each function, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware or as a software functional module. It should be noted that the module division in this disclosure embodiment is illustrative and only represents one logical functional division; other division methods may be used in actual implementation.

[0118] The following, combined with Figure 7 This disclosure provides a detailed description of the click-through rate model generation apparatus provided in the embodiments. It should be understood that the descriptions of the apparatus embodiments correspond to the descriptions of the method embodiments; therefore, any content not described in detail here can be found in the above method embodiments, and for the sake of brevity, will not be repeated here.

[0119] Figure 7 This is a logical structure block diagram of a click-through rate model generation device according to an exemplary embodiment. (Refer to...) Figure 7 The click-through rate (CTR) model generation device includes an acquisition module 710 and a processing module 720. The acquisition module 710 is configured to acquire multiple browsing sample data, each of which includes multiple attribute features of the viewed object and the click-through rate of the viewed object; for example, combining... Figure 1 The acquisition module 710 can be used to perform step 11.

[0120] Processing module 720 is configured to train a first prediction model based on multiple browsing sample data to obtain a second prediction model. During training, the input control layer of the first prediction model is invoked to binarize the initial first weight parameters to obtain an input control coefficient vector. The input control layer is then invoked to adjust multiple attribute features corresponding to each browsing sample data based on the elements in the input control coefficient vector, and the adjusted target attribute features are input to the input layer of the first prediction model. Multiple attribute features, elements in the first weight parameters, elements in the input control coefficient vector, and elements in the corresponding parameters of the second prediction model correspond one-to-one. For example, combining... Figure 1 The processing module 720 can be used to perform step 12.

[0121] Processing module 720 is further configured to determine target elements that satisfy preset conditions from the elements in the corresponding parameters of the second prediction model; for example, combining Figure 1 The processing module 720 can be used to execute step 13.

[0122] The processing module is also configured to retrain the second prediction model based on multiple attribute features corresponding to the target element, thereby obtaining a trained click-through rate model. For example, combining... Figure 1 The processing module 720 can be used to execute step 14.

[0123] Optionally, the input control layer is specifically used to input the first weight parameter into the input control function to obtain the input control coefficient vector.

[0124] Optionally, the processing module 720 is further configured to train the first prediction model based on multiple browsing sample data and a loss function to obtain a second prediction model; wherein, the loss function is used to adjust the parameters in the first prediction model so that the prediction result of the first prediction model meets the preset requirements, and in the process of adjusting the parameters in the first prediction model, the first weight parameters tend to be binarized. For example, combined with Figure 3 The processing module 720 can be used to execute step 121.

[0125] Optionally, the loss function includes a cross-entropy loss function and a target loss function, the target loss function being related to the first weight parameter.

[0126] Optionally, the processing module 720 is further configured to determine elements from the corresponding parameters in the second prediction model that are greater than or equal to a preset threshold, as target elements that satisfy preset conditions. For example, combined with Figure 3 The processing module 720 can be used to perform step 131.

[0127] Optionally, the processing module 720 is also configured to construct a third prediction model based on multiple attribute features corresponding to the target element and the second prediction model; for example, combining... Figure 4 The processing module 720 can be used to perform step 141.

[0128] Processing module 720 is also configured to retrain the third prediction model to obtain a trained click-through rate model. For example, combined with... Figure 4 The processing module 720 can be used to perform step 142.

[0129] Optionally, the processing module 720 is further configured to obtain an adjustment coefficient vector based on elements in the corresponding parameters of the second prediction model; the adjustment coefficient vector is used to filter multiple attribute features of the browsed object; for example, combined with Figure 5 The processing module 720 can be used to perform step 1411.

[0130] Processing module 720 is further configured to replace the corresponding parameters in the second prediction model with the target parameters to obtain the replaced second prediction model; the target parameters are composed of target elements; for example, combined with Figure 5 The processing module 720 can be used to perform step 1412.

[0131] Processing module 720 is also configured to add an adjustment layer before the input control layer of the replaced second prediction model to obtain a third prediction model. The adjustment layer is used to filter multiple attribute features of the browsed object using an adjustment coefficient vector. For example, combining... Figure 5 The processing module 720 can be used to execute step 1413.

[0132] Of course, the path selection device provided in this embodiment includes, but is not limited to, the modules described above. For example, the path selection device may also include a storage module 730. The storage module 730 may be used to store the program code of the write path selection device, and may also be used to store data generated by the write path selection device during operation, such as data in a write request.

[0133] Figure 8 This is a logic block diagram illustrating a click-through rate determination device according to an exemplary embodiment. (Refer to...) Figure 8 The click-through rate determination device includes an acquisition module 810 and a processing module 820.

[0134] Module 810 is configured to acquire browsing data; the browsing data includes multiple attribute characteristics of the browsing object; for example, combining... Figure 6 The acquisition module 810 can be used to execute step 61.

[0135] The processing module 820 is also configured to process the browsing data using a click-through rate (CTR) model to obtain the CTR corresponding to the browsed object. For example, combined with... Figure 6 The processing module 820 can be used to execute step 62.

[0136] Of course, the click-through rate determination device provided in this embodiment includes, but is not limited to, the modules described above. For example, the path selection device may also include a storage module 830. The storage module 830 may be used to store the program code of the write path selection device, and may also be used to store data generated by the write path selection device during operation, such as data in write requests.

[0137] Figure 9 A schematic diagram of a possible structure of the electronic device involved in the above embodiments is shown. For example... Figure 9 As shown, the electronic device 90 includes a processor 901 and a memory 902.

[0138] Understandable. Figure 9 The electronic device 90 shown can implement all the functions of the above-described click-through rate (CTR) model generation method, and also all the functions of the above-described CTR determination method. The functions of each module in the above-described CTR model generation device can be implemented in the processor 901 of the electronic device 90. The storage module of the CTR model generation device is equivalent to the memory 902 of the electronic device 90. Similarly, the functions of each module in the above-described CTR determination device can be implemented in the processor 901 of the electronic device 90. The storage module of the CTR determination device is equivalent to the memory 902 of the electronic device 90.

[0139] The processor 901 may include one or more processing cores, such as a 4-core processor or a 9-core processor. The processor 901 may include an application processor (AP), a modem processor, a graphics processing unit (GPU), an image signal processor (ISP), a controller, memory, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural network processing unit (NPU). Different processing units may be independent devices or integrated into one or more processors.

[0140] The memory 902 may include one or more computer-readable storage media, which may be non-transitory. The memory 902 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer-readable storage media in the memory 902 is used to store at least one instruction, which is executed by the processor 901 to implement the click-through rate model generation method provided in the embodiments of this disclosure.

[0141] In some embodiments, the electronic device 90 may optionally include a peripheral device interface 903 and at least one peripheral device. The processor 901, memory 902, and peripheral device interface 903 can be connected via a bus or signal line. Each peripheral device can be connected to the peripheral device interface 903 via a bus, signal line, or circuit board. Specifically, the peripheral device includes at least one of the following: a radio frequency circuit 904, a touch display screen 905, a camera assembly 906, an audio circuit 907, a positioning assembly 908, and a power supply 909.

[0142] Peripheral device interface 903 can be used to connect at least one I / O (Input / Output) related peripheral device to processor 901 and memory 902. In some embodiments, processor 901, memory 902 and peripheral device interface 903 are integrated on the same chip or circuit board; in some other embodiments, any one or two of processor 901, memory 902 and peripheral device interface 903 can be implemented on separate chips or circuit boards, which is not limited in this embodiment.

[0143] The radio frequency (RF) circuit 904 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The RF circuit 904 communicates with communication networks and other communication devices via electromagnetic signals. The RF circuit 904 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals back into electrical signals. Optionally, the RF circuit 904 includes: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, etc. The RF circuit 904 can communicate with other click-through rate model generation devices via at least one wireless communication protocol. This wireless communication protocol includes, but is not limited to: metropolitan area networks (MANs), various generations of mobile communication networks (2G, 3G, 4G, and 5G), wireless local area networks (WLANs), and / or Wi-Fi (Wireless Fidelity) networks. In some embodiments, the RF circuit 904 may also include circuitry related to NFC (Near Field Communication), which is not limited in this disclosure.

[0144] Display screen 905 is used to display a UI (User Interface). This UI may include graphics, text, icons, video, and any combination thereof. When display screen 905 is a touch display screen, it also has the ability to collect touch signals on or above its surface. These touch signals can be input as control signals to processor 901 for processing. In this case, display screen 905 can also be used to provide virtual buttons and / or a virtual keyboard, also known as soft buttons and / or a soft keyboard. In some embodiments, display screen 905 can be a single unit, serving as the front panel of electronic device 90; display screen 905 can be made of materials such as LCD (Liquid Crystal Display) or OLED (Organic Light-Emitting Diode).

[0145] Camera assembly 906 is used to acquire images or videos. Optionally, camera assembly 906 includes a front-facing camera and a rear-facing camera. Typically, the front-facing camera is located on the front panel of the click-through rate model generation device, and the rear-facing camera is located on the back of the click-through rate model generation device. Audio circuit 907 may include a microphone and a speaker. The microphone is used to acquire sound waves from the user and the environment, and convert the sound waves into electrical signals that are input to processor 901 for processing, or input to radio frequency circuit 404 for voice communication. For stereo acquisition or noise reduction purposes, there may be multiple microphones, each located in a different part of the electronic device 90. The microphone may also be an array microphone or an omnidirectional acquisition microphone. The speaker is used to convert electrical signals from processor 901 or radio frequency circuit 904 into sound waves. The speaker may be a conventional diaphragm speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can convert electrical signals not only into sound waves that are audible to humans, but also into sound waves that are inaudible to humans for purposes such as ranging. In some embodiments, audio circuit 907 may also include a headphone jack.

[0146] Positioning component 908 is used to locate the current geographic location of electronic device 90 for navigation or LBS (Location Based Service). Positioning component 908 can be a positioning component based on the US GPS (Global Positioning System), China's BeiDou system, Russia's Granas system, or the European Union's Galileo system.

[0147] Power supply 909 is used to supply power to various components in electronic device 90. Power supply 909 can be AC ​​power, DC power, a disposable battery, or a rechargeable battery. When power supply 909 includes a rechargeable battery, the rechargeable battery can support wired or wireless charging. The rechargeable battery can also be used to support fast charging technology.

[0148] In some embodiments, the electronic device 90 further includes one or more sensors 910. The one or more sensors 910 include, but are not limited to, accelerometers, gyroscopes, pressure sensors, fingerprint sensors, optical sensors, and proximity sensors.

[0149] An accelerometer can detect the magnitude of acceleration along the three axes of a coordinate system established by the electronic device 90. A gyroscope sensor can detect the orientation and rotation angle of the electronic device 90; the gyroscope sensor can work in conjunction with the accelerometer to acquire 3D motion data from the user on the electronic device 90. A pressure sensor can be located on the side bezel of the electronic device 90 and / or on the lower layer of the touchscreen display 905. When the pressure sensor is located on the side bezel of the electronic device 90, it can detect the user's grip signal on the electronic device 90. A fingerprint sensor is used to acquire the user's fingerprint. An optical sensor is used to acquire ambient light intensity. A proximity sensor, also known as a distance sensor, is typically located on the front panel of the electronic device 90. The proximity sensor is used to acquire the distance between the user and the front of the electronic device 90.

[0150] This disclosure also provides a computer-readable storage medium storing instructions that, when executed by a processor of a click-through rate (CTR) model generation apparatus, enable the CTR model generation apparatus to perform the CTR model generation method described above. Similarly, when executed by a processor of a click-through rate (CTR) determination apparatus, the CTR determination apparatus enables the CTR determination apparatus to perform the CTR determination method described above.

[0151] This disclosure also provides a computer program product containing instructions that, when run on a click-through rate (CTR) model generation apparatus, causes the CTR model generation apparatus to execute the CTR model generation method provided in this disclosure. When executed on the processor of a CTR determination apparatus, it enables the CTR determination apparatus to execute the CTR determination apparatus method provided in this disclosure.

[0152] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0153] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for generating a click rate model, the method comprising: The method comprises the following steps: acquiring a plurality of browsing sample data, each piece of the browsing sample data comprising a plurality of attribute features of a browsing object and a click rate of the browsing object; the attribute features are used to represent values corresponding to the browsing sample data for a feature category corresponding to the attribute features; the feature category comprises all features after the feature category is exhausted; training a first estimation model based on the plurality of browsing sample data to obtain a second estimation model; in the training process, an input control layer in the first estimation model is called to perform binary processing on initial first weight parameters in the first estimation model to obtain an input control coefficient vector; the input control layer is called to multiply each element in the input control coefficient vector with a plurality of attribute features corresponding to each piece of browsing sample data to obtain target attribute features, the target attribute features are input into an input layer in the first estimation model, and activation processing and normalization processing are performed on the target attribute features to obtain the click rate of the browsing object in the plurality of browsing sample data; the plurality of browsing sample data are input into the first estimation model to obtain an estimated click rate of the browsing object in the plurality of browsing sample data, and a loss value between the estimated click rate and the click rate is calculated; each element in the first weight parameters in the first estimation model is iteratively trained according to the loss value and a loss function, so that each element in the first weight parameters tends to be binary, until the loss value is less than or equal to a preset loss value, to obtain the second estimation model; the plurality of attribute features, the elements in the first weight parameters, the elements in the input control coefficient vector, and the elements in the corresponding parameters in the second estimation model correspond one by one; determining elements greater than or equal to a preset threshold from the elements in the corresponding parameters in the second estimation model as target elements satisfying a preset condition; retraining the second estimation model based on a plurality of attribute features corresponding to the target elements to obtain a trained click rate model.

2. The method of claim 1, wherein, The input control layer is specifically configured to input the first weight parameters into an input control function to obtain the input control coefficient vector.

3. The method of claim 1, wherein, The loss function comprises a cross-entropy loss function and a target loss function, and the target loss function is related to the first weight parameters.

4. The method according to any one of claims 1 to 3, characterized in that, Retraining the second estimation model based on a plurality of attribute features corresponding to the target elements to obtain a trained click rate model comprises: constructing a third estimation model based on the plurality of attribute features corresponding to the target elements and the second estimation model; retraining the third estimation model to obtain a trained click rate model.

5. The method of claim 4, wherein, The method of constructing a third estimation model based on a plurality of attribute features corresponding to the target elements and the second estimation model comprises: obtaining an adjustment coefficient vector based on the elements in the corresponding parameters in the second estimation model; the adjustment coefficient vector is used to filter the plurality of attribute features of the browsing object; Replace the corresponding parameters in the second estimation model with the target parameters to obtain a second estimation model after replacement; the target parameters are composed of the target elements; Add an adjustment layer before the input control layer of the second estimation model after replacement to obtain the third estimation model, and the adjustment layer is used to filter the multiple attribute features of the browsing object by using the adjustment coefficient vector.

6. A click rate determination method characterized by, Comprise: Obtain browsing data; the browsing data comprises multiple attribute features of a browsing object; Process the browsing data by using a click rate model to obtain the click rate corresponding to the browsing object, and the click rate model is obtained according to the click rate model generation method in any one of claims 1-5. 7.A click rate model generation apparatus, characterized by comprising: Comprise: An acquisition module configured to acquire a plurality of browsing sample data, each piece of browsing sample data in the plurality of browsing sample data comprising multiple attribute features of a browsing object and a click rate of the browsing object; the attribute features are used to represent the values corresponding to the browsing sample data for the corresponding feature categories of the attribute features; the feature categories comprise all features after the feature categories are exhausted; A processing module configured to train a first estimation model based on the plurality of browsing sample data to obtain a second estimation model; wherein, in the training process, the input control layer in the first estimation model is called to perform binary processing on the initial first weight parameters in the first estimation model to obtain an input control coefficient vector; the input control layer is called to multiply each element in the input control coefficient vector with the corresponding multiple attribute features in each piece of browsing sample data to obtain target attribute features, and the target attribute features are input into the input layer in the first estimation model; and the target attribute features are subjected to activation processing and normalization processing, thereby obtaining the click rates of the browsing objects in the plurality of browsing sample data; the plurality of browsing sample data are input into the first estimation model to obtain the estimated click rates of the browsing objects in the plurality of browsing sample data, and a loss value between the estimated click rates and the click rates is calculated; each element in the first weight parameters in the first estimation model is iteratively trained according to the loss value and a loss function, so that each element in the first weight parameters tends to be binary, until the loss value is less than or equal to a preset loss value, to obtain the second estimation model; the multiple attribute features, the elements in the first weight parameters, the elements in the input control coefficient vector, and the elements in the corresponding parameters in the second estimation model correspond one by one; The processing module is further configured to determine elements greater than or equal to a preset threshold from the elements in the corresponding parameters in the second estimation model as target elements that satisfy a preset condition. The processing module is further configured to retrain the second estimation model based on the multiple attribute features corresponding to the target elements and the second estimation model to obtain a trained click rate model.

8. The apparatus of claim 7, wherein, The input control layer is specifically used to input the first weight parameters into an input control function to obtain the input control coefficient vector.

9. The apparatus of claim 7, wherein, The loss function comprises a cross-entropy loss function and a target loss function, and the target loss function is related to the first weight parameter.

10. The apparatus of any one of claims 7-9, wherein, The processing module is further configured to construct a third estimation model based on the plurality of attribute features corresponding to the target element and the second estimation model. The processing module is further configured to retrain the third estimation model to obtain a trained click rate model.

11. The apparatus of claim 10, wherein, The processing module is further configured to obtain an adjustment coefficient vector based on elements in corresponding parameters in the second estimation model, and the adjustment coefficient vector is used to filter the plurality of attribute features of the browsing object. The processing module is further configured to replace the corresponding parameters in the second estimation model with target parameters to obtain a replaced second estimation model, and the target parameters are composed of the target element. The processing module is further configured to add an adjustment layer before an input control layer of the replaced second estimation model to obtain the third estimation model, and the adjustment layer is used to filter the plurality of attribute features of the browsing object using the adjustment coefficient vector.

12. A click rate determining apparatus characterized by comprising: comprising: The acquisition module is configured to acquire browsing data, and the browsing data comprises a plurality of attribute features of a browsing object. The processing module is further configured to process the browsing data using a click rate model to obtain a click rate corresponding to the browsing object, and the click rate model is obtained according to the click rate model generation method of any one of claims 1-5.

13. An electronic device, comprising: The electronic device comprises: a processor; a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions to implement the click rate model generation method of any one of claims 1-5 and the click rate determination method of claim 6.

14. A computer-readable storage medium having stored thereon instructions, the instructions comprising, When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device can execute the click rate model generation method of any one of claims 1-5 and the click rate determination method of claim 6.

15. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the electronic device, the click rate model generation method of any one of claims 1-5 and the click rate determination method of claim 6 are implemented.

Images