Click rate correction method and device, electronic equipment and storage medium

Abstract

The present disclosure relates to a click rate correction method and device, electronic equipment and storage medium. The method comprises: obtaining object information of a target object; obtaining virtual space information of a to-be-predicted virtual space based on resource data corresponding to the to-be-predicted virtual space; obtaining an initial estimated click rate of the target object for the to-be-predicted virtual space from a click rate estimation module; and performing correction processing on the initial estimated click rate according to the object information and the virtual space information to obtain a predicted click rate of the target object for the to-be-predicted virtual space. The embodiments of the present disclosure can improve the accuracy of the predicted click rate of the target object for the to-be-predicted virtual space and the efficiency of resource interaction of the target object through the virtual space.

Description

Technical Field

[0001] This disclosure relates to the field of neural network technology, and in particular to a click rate correction method, apparatus, electronic device, and storage medium. Background Technology

[0002] With the popularization of the Internet, users can interact through virtual spaces, for example, users can interact with resources through virtual spaces.

[0003] In order to improve the efficiency of resource interaction during the process of users interacting with resources through virtual spaces, the click-through rate of users for each virtual space can be determined by the matching degree between users and each virtual space. Then, based on the click-through rate of users for each virtual space, virtual spaces recommended to users can be determined, prioritizing the recommendation of virtual spaces with higher click-through rates, thereby improving the efficiency of users interacting with resources through virtual spaces.

[0004] Currently, click-through rate is determined based on the degree of matching between users and virtual spaces. Due to the lack of representation of the potential relationship between users and virtual spaces, the accuracy of the matching degree between users and virtual spaces is poor, which in turn leads to poor accuracy in predicting the click-through rate of users for virtual spaces. Summary of the Invention

[0005] This disclosure provides a click-through rate (CTR) correction method, apparatus, electronic device, and storage medium to at least solve the problem of poor accuracy in predicting user click-through rates for virtual spaces in related technologies. The technical solution of this disclosure is as follows:

[0006] According to a first aspect of the present disclosure, a click-through rate correction method is provided, comprising:

[0007] Obtain object information of the target object;

[0008] Based on the resource data corresponding to the virtual space to be predicted, obtain the virtual space information of the virtual space to be predicted;

[0009] The initial estimated click-through rate of the target object for the virtual space to be predicted is obtained from the click-through rate prediction module;

[0010] The initial estimated click-through rate is corrected based on the object information and the virtual space information to obtain the predicted click-through rate of the target object for the virtual space to be predicted.

[0011] In one embodiment, the step of correcting the initial estimated click-through rate based on the object information and the virtual space information to obtain the predicted click-through rate of the target object for the virtual space to be predicted includes:

[0012] The initial estimated click-through rate is corrected using a correction model based on the object information and the virtual space information to obtain the predicted click-through rate of the target object for the virtual space to be predicted.

[0013] The correction model is pre-trained and determines the deviation of the initial estimated click-through rate based on the object information and the virtual space information, and eliminates the deviation of the initial estimated click-through rate.

[0014] In one embodiment, the correction model includes a correction value prediction unit and a correction unit. The step of correcting the initial estimated click-through rate based on the object information and the virtual space information using the correction model to obtain the predicted click-through rate of the target object for the virtual space to be predicted includes:

[0015] The correction value prediction unit performs prediction processing on the object information and the virtual space information to obtain the correction value;

[0016] The correction unit performs correction processing on the initial estimated click-through rate and the correction value to obtain the predicted click-through rate of the target object for the virtual space to be predicted.

[0017] In one embodiment, the correction unit includes a correction layer and a normalization layer. The step of correcting the initial estimated click-through rate and the correction value using the correction unit to obtain the predicted click-through rate of the target object for the virtual space to be predicted includes:

[0018] The initial estimated click-through rate is inversely normalized by the correction layer to obtain the inversely normalized initial estimated click-through rate. The correction value is then used to correct and adjust the inversely normalized initial estimated click-through rate to obtain the predicted target click value of the target object for the virtual space to be predicted.

[0019] The predicted target click value is normalized by the normalization layer to obtain the predicted click rate of the target object for the virtual space to be predicted.

[0020] In one embodiment, the step of correcting the initial estimated click-through rate using a correction model based on the object information and the virtual space information to obtain the predicted click-through rate of the target object for the virtual space to be predicted includes:

[0021] Based on the object information and the virtual space information, feature cross-referencing is performed to obtain cross-feature information;

[0022] The initial estimated click-through rate is corrected by a correction model based on the object information, the virtual space information, and the cross-feature information to obtain the predicted click-through rate of the target object for the virtual space to be predicted.

[0023] In one embodiment, the object information includes first object information, and obtaining the object information of the target object includes:

[0024] Obtain at least one search information for the target object, wherein the search information is record information generated when searching for resource data;

[0025] The search information is processed to identify categories, thereby obtaining category information corresponding to each search information.

[0026] The category information corresponding to each of the search information is used as the first object information.

[0027] In one embodiment, the object information includes second object information, and obtaining the object information of the target object includes:

[0028] Obtain at least one browsing information of the target object, wherein the browsing information is record information generated when browsing resource data;

[0029] For any of the browsing information, a target virtual space and a browsing time for the target virtual space are determined based on the browsing information, and first target resource data displayed in the target virtual space during the browsing time are determined;

[0030] The category information corresponding to each of the first target resource data is used as the second object information of the target object.

[0031] In one embodiment, the virtual space information includes first virtual space information, and the step of obtaining the virtual space information of the virtual space to be predicted based on the resource data corresponding to the virtual space to be predicted includes:

[0032] Obtain the resource list of the virtual space to be predicted;

[0033] Obtain the category information corresponding to each resource data in the resource list;

[0034] The category information corresponding to each resource data in the resource list is used as the first virtual space information corresponding to the virtual space to be predicted.

[0035] In one embodiment, the virtual space information includes second virtual space information, and the step of obtaining the virtual space information of the virtual space to be predicted based on the resource data corresponding to the virtual space to be predicted includes:

[0036] Obtain the second target resource data currently displayed in the virtual space to be predicted;

[0037] The category information corresponding to the second target resource data is used as the second virtual space information corresponding to the virtual space to be predicted.

[0038] In one embodiment, the method further includes:

[0039] Construct a training set, which includes multiple sample groups. Each sample group includes sample information and annotation information for the sample information. The sample information includes object information of the sample object, virtual space information of the sample virtual space, and the sample predicted click rate of the sample object for the sample virtual space. The annotation information is used to characterize the click situation of the sample object for the sample virtual space.

[0040] The initial correction model corrects the predicted click-through rate of the sample based on the object information of the sample object and the virtual space information of the sample virtual space, thereby obtaining the predicted click-through rate of the sample object for the sample virtual space.

[0041] Based on the predicted click-through rate of the sample object in the sample virtual space and the annotation information, the loss value of the initial correction model is determined;

[0042] The initial correction model is trained based on the loss value to obtain the correction model.

[0043] According to a second aspect of the present disclosure, a click-through rate correction device is provided, comprising:

[0044] The first acquisition unit is configured to acquire object information of the target object;

[0045] The second acquisition unit is configured to acquire virtual space information of the virtual space to be predicted based on the resource data corresponding to the virtual space to be predicted.

[0046] The third acquisition unit is configured to acquire the initial estimated click rate of the target object for the virtual space to be predicted from the click rate prediction module;

[0047] The correction unit is configured to perform correction processing on the initial estimated click-through rate based on the object information and the virtual space information to obtain the predicted click-through rate of the target object for the virtual space to be predicted.

[0048] In one embodiment, the correction unit is further configured to perform:

[0049] The initial estimated click-through rate is corrected using a correction model based on the object information and the virtual space information to obtain the predicted click-through rate of the target object for the virtual space to be predicted.

[0050] The correction model is pre-trained and determines the deviation of the initial estimated click-through rate based on the object information and the virtual space information, and eliminates the deviation of the initial estimated click-through rate.

[0051] In one embodiment, the correction model includes a correction value prediction unit and a correction unit, the correction unit being further configured to perform:

[0052] The correction value prediction unit performs prediction processing on the object information and the virtual space information to obtain the correction value;

[0053] The correction unit performs correction processing on the initial estimated click-through rate and the correction value to obtain the predicted click-through rate of the target object for the virtual space to be predicted.

[0054] In one embodiment, the bias correction unit includes a bias correction layer and a normalization layer, and the bias correction unit is further configured to perform:

[0055] The initial estimated click-through rate is inversely normalized by the correction layer to obtain the inversely normalized initial estimated click-through rate. The correction value is then used to correct and adjust the inversely normalized initial estimated click-through rate to obtain the predicted target click value of the target object for the virtual space to be predicted.

[0056] The predicted target click value is normalized by the normalization layer to obtain the predicted click rate of the target object for the virtual space to be predicted.

[0057] In one embodiment, the correction unit is further configured to perform:

[0058] Based on the object information and the virtual space information, feature cross-referencing is performed to obtain cross-feature information;

[0059] The initial estimated click-through rate is corrected by a correction model based on the object information, the virtual space information, and the cross-feature information to obtain the predicted click-through rate of the target object for the virtual space to be predicted.

[0060] In one embodiment, the object information includes first object information, and the first acquisition unit is further configured to perform:

[0061] Obtain at least one search information for the target object, wherein the search information is record information generated when searching for resource data;

[0062] The search information is processed to identify categories, thereby obtaining category information corresponding to each search information.

[0063] The category information corresponding to each of the search information is used as the first object information.

[0064] In one embodiment, the object information includes second object information, and the first acquisition unit is further configured to perform:

[0065] Obtain at least one browsing information of the target object, wherein the browsing information is record information generated when browsing resource data;

[0066] For any of the browsing information, a target virtual space and a browsing time for the target virtual space are determined based on the browsing information, and first target resource data displayed in the target virtual space during the browsing time are determined;

[0067] The category information corresponding to each of the first target resource data is used as the second object information of the target object.

[0068] In one embodiment, the virtual space information includes first virtual space information, and the second acquisition unit is further configured to perform:

[0069] Obtain the resource list of the virtual space to be predicted;

[0070] Obtain the category information corresponding to each resource data in the resource list;

[0071] The category information corresponding to each resource data in the resource list is used as the first virtual space information corresponding to the virtual space to be predicted.

[0072] In one embodiment, the virtual space information includes second virtual space information, and the second acquisition unit is further configured to perform:

[0073] Obtain the second target resource data currently displayed in the virtual space to be predicted;

[0074] The category information corresponding to the second target resource data is used as the second virtual space information corresponding to the virtual space to be predicted.

[0075] In one embodiment, the device further includes:

[0076] The construction unit is configured to construct a training set, which includes multiple sample groups. Each sample group includes sample information and annotation information for the sample information. The sample information includes object information of the sample object, virtual space information of the sample virtual space, and the sample predicted click rate of the sample object for the sample virtual space. The annotation information is used to characterize the click situation of the sample object for the sample virtual space.

[0077] The processing unit is configured to perform bias correction processing on the predicted click-through rate of the sample based on the object information of the sample object and the virtual space information of the sample virtual space using an initial bias correction model, so as to obtain the predicted click-through rate of the sample object for the sample virtual space.

[0078] The determining unit is configured to determine the loss value of the initial correction model based on the predicted click-through rate of the sample object for the sample virtual space and the annotation information.

[0079] The training unit is configured to train the initial correction model based on the loss value to obtain the correction model.

[0080] According to a third aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement any of the click-through rate correction methods provided in the first aspect.

[0081] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided, wherein when instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform any of the click-through rate correction methods provided in the first aspect.

[0082] According to a fifth aspect of the present disclosure, a computer program product is provided, the computer program product including instructions that, when executed by a processor of an electronic device, enable the electronic device to perform any of the click-through rate correction methods provided in the first aspect.

[0083] The technical solutions provided by the embodiments of this disclosure have at least the following beneficial effects:

[0084] The click-through rate (CTR) correction method, apparatus, electronic device, and storage medium provided in this disclosure can acquire object information of a target object and, based on resource data corresponding to the virtual space to be predicted, acquire virtual space information of the virtual space to be predicted. After acquiring the initial estimated CTR of the target object for the virtual space to be predicted, the initial estimated CTR can be corrected based on object information and virtual space feature information to obtain the predicted CTR of the target object for the virtual space to be predicted. Based on the CTR correction method, apparatus, electronic device, and storage medium provided in this disclosure, after obtaining the initial estimated CTR of the target object for the virtual space to be predicted, further correction processing of the initial estimated CTR can be performed based on object information and virtual space information, which can improve the accuracy of the predicted CTR of the target object for the virtual space to be predicted, and further improve the efficiency of users interacting with resources through virtual space.

[0085] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0086] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0087] Figure 1 This is a flowchart illustrating a click-through rate correction method according to an exemplary embodiment.

[0088] Figure 2 This is a flowchart illustrating a click-through rate correction method according to an exemplary embodiment.

[0089] Figure 3 This is a flowchart of step 204 in a click-through rate correction method according to an exemplary embodiment.

[0090] Figure 4 This is a flowchart illustrating a click-through rate correction method according to an exemplary embodiment.

[0091] Figure 5 This is a flowchart of step 102 in a click-through rate correction method according to an exemplary embodiment.

[0092] Figure 6 This is a flowchart of step 102 in a click-through rate correction method according to an exemplary embodiment.

[0093] Figure 7 This is a flowchart of step 104 in a click-through rate correction method according to an exemplary embodiment.

[0094] Figure 8 This is a flowchart of step 104 in a click-through rate correction method according to an exemplary embodiment.

[0095] Figure 9 This is a flowchart illustrating a click-through rate correction method according to an exemplary embodiment.

[0096] Figure 10 This is a block diagram illustrating a click-rate correction device according to an exemplary embodiment.

[0097] Figure 11 This is a block diagram illustrating an electronic device according to an exemplary embodiment. Detailed Implementation

[0098] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0099] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0100] It should also 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 used for display, data used for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties.

[0101] Figure 1 This is a flowchart illustrating a click-through rate correction method according to an exemplary embodiment. This embodiment uses the application of this method to a server as an example for illustration. It is understood that this method can also be applied to a system including a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0102] In step 102, the object information of the target object is obtained.

[0103] In this embodiment of the disclosure, the target object is the user currently being recommended a virtual space. For example, the object information may include data related to the target object's interaction with the virtual space, the target object's behavioral data, and the target object's associated operations on resource data (e.g., browsing, searching, etc.). For example, the object information corresponding to the target object can be obtained based on the target object's logs.

[0104] In step 104, the virtual space information of the virtual space to be predicted is obtained based on the resource data corresponding to the virtual space to be predicted.

[0105] In this embodiment, the virtual space to be predicted can be any virtual space currently in an open state. The resource data of the virtual space to be predicted may include resource data to be displayed and resource data currently being displayed. Resource data may include data such as virtual items, virtual resources, virtual services, and goods; this embodiment does not specifically limit the resource data. The virtual space information may include attribute data of the virtual space, attribute data of the resource data corresponding to the virtual space, user information of the user associated with the virtual space, etc.; this embodiment does not specifically limit the virtual space information.

[0106] In step 106, the initial estimated click-through rate of the target object for the virtual space to be predicted is obtained from the click-through rate prediction module.

[0107] In this embodiment of the disclosure, the click-through rate (CTR) prediction module may include a CTR prediction platform or a pre-trained neural network for predicting CTR. Specifically, the initial estimated CTR of the target object for the virtual space to be predicted can be obtained from the CTR prediction platform. For example, the initial estimated CTR of the target object for the virtual space to be predicted can be queried from the CTR prediction platform based on the object identifier of the target object and the virtual space identifier of the virtual space to be predicted; or, a pre-trained neural network for predicting CTR can be used to predict the CTR based on the object information and the virtual space information to obtain the initial estimated CTR of the target object for the virtual space to be predicted. The initial estimated CTR is the preliminary predicted probability of the target object clicking on the virtual space to be predicted.

[0108] This disclosure does not specifically limit the method of obtaining the initial estimated click-through rate. Any method that can obtain the initial estimated click-through rate of the target object for the virtual space to be predicted is applicable to this disclosure.

[0109] In step 108, the initial estimated click-through rate is corrected based on the object information and virtual space information to obtain the predicted click-through rate of the target object for the virtual space to be predicted.

[0110] In this embodiment of the disclosure, a correction value for the initial estimated click-through rate can be determined based on object information and virtual space information, and the initial estimated click-through rate can be further corrected based on the correction value to obtain the corrected predicted click-through rate, and the corrected predicted click-through rate can be used as the predicted click-through rate of the target object for the virtual space to be predicted.

[0111] The click-through rate (CTR) correction method provided in this disclosure can obtain object information of a target object and, based on resource data corresponding to the virtual space to be predicted, obtain virtual space information of the virtual space to be predicted. After obtaining the initial estimated CTR of the target object for the virtual space to be predicted, the initial estimated CTR can be corrected based on object information and virtual space feature information to obtain the predicted CTR of the target object for the virtual space to be predicted. Based on the CTR correction method provided in this disclosure, after obtaining the initial estimated CTR of the target object for the virtual space to be predicted, further correction processing of the initial estimated CTR can be performed based on object information and virtual space information, which can improve the accuracy of the predicted CTR of the target object for the virtual space to be predicted, and further improve the efficiency of users interacting with resources through virtual space.

[0112] In an exemplary embodiment, step 108 above, which involves correcting the initial estimated click-through rate based on object information and virtual space information to obtain the predicted click-through rate of the target object for the virtual space to be predicted, can be achieved through the following steps:

[0113] By using a correction model to correct the initial estimated click-through rate based on object information and virtual space information, the predicted click-through rate of the target object for the virtual space to be predicted is obtained.

[0114] Among them, the bias correction model is a pre-trained model that determines the deviation of the initial estimated click-through rate based on object information and virtual space information, and eliminates the deviation of the initial estimated click-through rate.

[0115] In this embodiment of the disclosure, the correction model is a pre-trained model used to correct the initial estimated click-through rate (CTR). Object information, virtual space information, and the initial estimated CTR can be input into the correction model. The model then performs predictive processing on the object information and virtual space information to determine the deviation of the initial estimated CTR. Based on this deviation, the initial estimated CTR is corrected. For example, the deviation is used to adjust the initial estimated CTR to eliminate the deviation, thereby obtaining the output of the correction model, which is the predicted CTR of the target object for the virtual space to be predicted.

[0116] By analogy, the predicted click-through rate for each virtual space to be predicted can be obtained for the target object. Then, the virtual spaces to be predicted can be sorted from highest to lowest according to the predicted click-through rate to obtain a virtual space recommendation list for the target object. Finally, virtual spaces can be pushed to the target object in order according to the virtual space recommendation list.

[0117] The click-through rate correction method provided in this embodiment, after obtaining the initial estimated click-through rate of the target object for the virtual space to be predicted, uses a correction model to perform correction processing on the initial estimated click-through rate based on object information and virtual space information. This can improve the accuracy of the predicted click-through rate of the target object for the virtual space to be predicted, and further improve the efficiency of users interacting with resources through the virtual space.

[0118] In an exemplary embodiment, the correction model includes a correction value prediction unit and a correction unit, referring to... Figure 2 As shown, by using a correction model to correct the initial estimated click-through rate based on object information and virtual space information, the predicted click-through rate of the target object for the virtual space to be predicted can be obtained through the following steps:

[0119] In step 202, the object information and virtual space information are predicted by the correction value prediction unit to obtain the correction value;

[0120] In step 204, the initial estimated click-through rate and the correction value are corrected by the correction unit to obtain the predicted click-through rate of the target object for the virtual space to be predicted.

[0121] In this embodiment of the disclosure, object information and virtual space information can be input into the bias correction prediction unit for prediction processing. The output of the bias correction prediction unit is the bias correction value, which can be used to characterize the deviation corresponding to the initial estimated click-through rate.

[0122] For example, the bias correction prediction unit may include a feature extraction module and a prediction module. For example, the feature extraction module may be a feature embedding layer, which can transform object information and virtual space information into dense features, and use these dense features as input to the prediction module. The prediction module may be a network layer constructed from multiple convolutional networks. After performing convolutional prediction processing on the dense features output by the feature extraction module through this layer, the bias correction value can be obtained.

[0123] In an exemplary embodiment, since too many network layers can lead to overfitting, while too few network layers can lead to underfitting, the prediction module in this embodiment may include a 3-layer network with [128, 32, 1] network nodes, so that the network can learn fully and converge quickly, that is, improve prediction accuracy and prediction efficiency.

[0124] After obtaining the correction value, the correction value and the initial estimated click-through rate can be input into the correction unit. The correction unit can use the correction value to adjust the initial estimated click-through rate in order to correct the initial estimated click-through rate and obtain the predicted click-through rate of the target object for the virtual space to be predicted.

[0125] The click-through rate correction method provided in this embodiment predicts the correction value corresponding to the initial estimated click-through rate based on object information and virtual space information by a correction value prediction unit. Then, the correction unit uses the correction value to correct the initial estimated click-through rate, which can improve the accuracy of the predicted click-through rate of the target object for the virtual space to be predicted, and further improve the efficiency of users interacting with resources through the virtual space.

[0126] In an exemplary embodiment, the correction unit includes a correction layer and a normalization layer. In step 204, the correction unit performs correction processing on the initial estimated click-through rate and the correction value to obtain the predicted click-through rate of the target object for the virtual space to be predicted. This can be achieved through the following steps:

[0127] In step 302, the initial estimated click-through rate is inversely normalized through the correction layer to obtain the inversely normalized initial estimated click-through rate. The correction value is then used to correct and adjust the inversely normalized initial estimated click-through rate to obtain the predicted target click value of the target object for the virtual space to be predicted.

[0128] In step 304, the predicted target click value is normalized by a normalization layer to obtain the predicted click rate of the target object for the virtual space to be predicted.

[0129] In this embodiment, the bias correction layer can perform inverse normalization on the initial estimated click-through rate (CTR) to obtain the inverse normalized initial CTR. Then, a bias correction value is used to adjust the inverse normalized initial CTR to obtain the predicted target click value for the target object in the virtual space to be predicted. Finally, the normalization layer normalizes the predicted target click value to obtain the predicted CTR for the target object in the virtual space to be predicted. This embodiment does not specifically limit the bias correction algorithm and the normalization layer; any algorithm that can achieve inverse normalization and any method of achieving normalization is applicable to this embodiment. For example, the bias correction layer can be deployed with a bias correction algorithm, which can refer to the following formula (I), and the normalization layer can refer to the following formula (II).

[0130]

[0131] Where 'a' represents the initial estimated click-through rate (CTR), 'b' represents the correction value, and 'b' can be any real number. When 'b' is negative, it indicates that the initial estimated CTR is too high and needs to be adjusted downwards. When 'b' is positive, it indicates that the initial estimated CTR is too low and needs to be adjusted upwards. 'f(a,b)' represents the predicted target click value.

[0132] Formula (II) = σ(x) = 1 / (1 + exp(-x))

[0133] Where x = f(a,b), the normalization function normalizes the predicted target click value to the range of 0 to 1, and the final output value σ(x) is the predicted click rate after correction.

[0134] The click-through rate correction method provided in this embodiment performs inverse normalization on the initial estimated click-through rate through a correction layer to fully amplify the difference in the initial estimated click-through rate. After summing the difference with the correction value, normalization is performed again, thereby correcting the initial estimated click-through rate. This improves the accuracy of the predicted click-through rate of the target object for the virtual space to be predicted, and further improves the efficiency of users interacting with resources through the virtual space.

[0135] In one exemplary embodiment, reference is made to Figure 4 As shown, by using a correction model to correct the initial estimated click-through rate based on object information and virtual space information, the predicted click-through rate of the target object for the virtual space to be predicted can be obtained through the following steps:

[0136] In step 402, feature cross-referencing is performed based on object information and virtual space information to obtain cross-feature information;

[0137] In step 404, the initial estimated click-through rate is corrected by the correction model based on object information, virtual space information and cross feature information to obtain the predicted click-through rate of the target object for the virtual space to be predicted.

[0138] In this embodiment of the disclosure, after obtaining object information and virtual space information, feature cross-referencing can be performed on the object information and virtual space information to obtain cross-feature information. This embodiment of the disclosure does not specifically limit the method of feature cross-referencing, and any method that can achieve feature cross-referencing is applicable to this embodiment of the disclosure.

[0139] For example, feature crossing can be achieved by multiplying object information and virtual space information. For instance, assuming that the object information includes first object information and second object information, and the virtual space information includes first virtual space information and second virtual space information, then the cross-feature information obtained after feature crossing includes: first object information × first virtual space information, first object information × second virtual space information, second object information × first virtual space information, and second object information × first virtual space information.

[0140] After obtaining the cross-feature information, the object information, virtual space information, and cross-feature information can be input into the correction model. The correction model can then perform prediction processing on the object information, virtual space information, and cross-feature information to obtain the correction value corresponding to the initial estimated click-through rate. This correction value can then be used to correct the initial estimated click-through rate to obtain the predicted click-through rate of the target object for the virtual space to be predicted.

[0141] The click-through rate correction method provided in this embodiment can obtain cross-feature information by cross-analyzing object information and virtual space information. This cross-feature data characterizes the potential feature relationship between the object and the virtual space, enriching the feature information and further improving the accuracy of correction and the precision of click-through rate prediction.

[0142] In one exemplary embodiment, the object information includes first object information, referred to... Figure 5 As shown, in step 102, obtaining the object information of the target object can be achieved through the following steps:

[0143] In step 502, at least one search information for the target object is obtained, wherein the search information is record information generated when searching for resource data;

[0144] In step 504, category identification processing is performed on each search information to obtain the category information corresponding to each search information;

[0145] In step 506, the category information corresponding to each search result is used as the first object information.

[0146] In this embodiment of the disclosure, the first object information is used to characterize the category features corresponding to the searched resources when the target object performs a search operation. In one example, the first object information can be the user's search category features.

[0147] For example, when a user performs a search operation on resource data on the platform, a search log is generated. The search log records search information, which may include keyword information, text information, voice information, image information, etc., corresponding to the search operation. At least one search piece of information can be obtained from the search log of the target object, and category identification processing is performed on each search piece of information to determine the category information of the resource data corresponding to the search piece of information. For example, at least one search piece of information can be obtained from the search log corresponding to a preset time period, or a preset number of search pieces of information can be obtained from the search log of the target object. In this embodiment of the disclosure, the method of obtaining at least one search piece of information of the target object is not specifically limited.

[0148] For example, taking search information as keyword information, text information, or voice information, natural language processing technology can be used to extract the category information corresponding to the resource data in the search information as the first object information; or, taking search information as image information, image processing and recognition technology can be used to extract the category information corresponding to the resource data in the search information as the first object information.

[0149] The category information corresponding to the resource data is used to identify the resource category to which the resource data belongs. For example, if a user's search information is the keyword "refrigerator," the corresponding resource data is "refrigerator," and the category information for "refrigerator" may include a first-level category: "home appliances," and a second-level category: "refrigerator." In this embodiment, category division can be based on requirements, such as coarser-grained or finer-grained category and category hierarchy division. This embodiment does not specifically limit the category information corresponding to the resource data.

[0150] The click-through rate correction method provided in this embodiment combines first object information to predict the correction value. Since the first object information represents the category characteristics of the resource data searched when the target object performs a search operation, it can reflect the target object's preferences to a certain extent and more significantly reflect the difference between the virtual space to be predicted and the target object, thereby improving the prediction accuracy of the correction value.

[0151] In one exemplary embodiment, the object information includes second object information, referred to... Figure 6 As shown, in step 102, obtaining the object information of the target object can be achieved through the following steps:

[0152] In step 602, at least one browsing information of the target object is obtained, and the browsing information is the record information generated when browsing resource data;

[0153] In step 604, for any browsing information, the target virtual space and the browsing time for the target virtual space are determined based on the browsing information, and the first target resource data displayed in the target virtual space during the browsing time are determined.

[0154] In step 606, the category information corresponding to each first target resource data is used as the second object information of the target object.

[0155] In this embodiment, the second object information is used to characterize the category features corresponding to the resource data displayed in the virtual space when the target object historically viewed the virtual space. In one example, the second object information can be the user's viewed category features. When a user views the virtual space, a corresponding browsing log is generated. The browsing log is used to record browsing information, which can be used to record the target virtual space viewed by the target object and the browsing time for the target virtual space. For example, at least one piece of browsing information can be obtained from the browsing log corresponding to a preset time period, or a preset number of pieces of browsing information can be obtained from the target object's browsing log. In this embodiment, the method of obtaining at least one piece of browsing information from the target object's browsing log is not specifically limited.

[0156] After determining the target virtual space and the browsing time for the target virtual space by browsing information, the first target resource data displayed by the target virtual space during that browsing time can be obtained. This first target resource data refers to the resource data that the target virtual space emphasizes and explains during that browsing time. For example, assuming that the target virtual space 1 and the browsing time (10:00 to 10:30) can be determined based on browsing information 1, and the resource data displayed by the target virtual space 1 during 10:00 to 10:30 includes refrigerators and fans, then it can be determined that the first target resource data may include refrigerators and fans.

[0157] After determining the primary target resource data, the corresponding category information can be determined. For example, the category information can be retrieved from a resource data information database. For instance, the category information includes first-level directories and second-level categories. After determining the category information for each primary target resource data, this category information can be used as the secondary object information for the target object.

[0158] The click-through rate correction method provided in this embodiment combines second object information to predict the correction value. Since the second object information represents the category characteristics of the resource data displayed in the virtual space when the target object viewed the virtual space in the past, it can reflect the target object's preferences to a certain extent and can more significantly reflect the difference between the virtual space to be predicted and the target object, thereby improving the prediction accuracy of the correction value.

[0159] In one exemplary embodiment, the virtual space information includes first virtual space information, referred to... Figure 7As shown, in step 104, the virtual space information of the virtual space to be predicted is obtained based on the resource data corresponding to the virtual space to be predicted. This can be achieved through the following steps:

[0160] In step 702, a list of resources for the virtual space to be predicted is obtained;

[0161] In step 704, the category information corresponding to each resource data in the resource list is obtained;

[0162] In step 706, the category information corresponding to each resource data in the resource list is used as the first virtual space information corresponding to the virtual space to be predicted.

[0163] In this embodiment of the disclosure, the resource list includes all resource data to be displayed in the virtual space to be predicted. This list is used to display information to users entering the virtual space to be predicted; that is, users entering the virtual space can obtain all the resource data to be displayed through this resource list. For example, if the resource data to be displayed in the virtual space this time includes refrigerators of brand A, fans of brand B, and sunscreens of brand C, then the resource list of the virtual space to be predicted will include refrigerators of brand A, fans of brand B, and sunscreens of brand C.

[0164] For example, a resource list for the virtual space to be predicted can be obtained from the backend based on the identifier of the virtual space to be predicted. After obtaining the resource list, the resource data to be displayed can be retrieved from the resource list, along with the category information corresponding to each resource data. For example, the category information corresponding to each resource data can be searched from a resource data information database. For example, the category information includes a first-level directory and a second-level category. After determining the category information corresponding to each resource data, this category information can be used as the first virtual space information corresponding to the virtual space to be predicted. This first virtual space information is a category feature used to characterize the resource data to be displayed in the virtual space to be predicted. In one example, this first virtual space information can be a virtual space shelf category feature.

[0165] The click-through rate correction method provided in this embodiment combines first virtual space information to predict the correction value. Since the first virtual space information is used to characterize the category features of the resource data to be displayed in the virtual space to be predicted, using it in combination with object information to predict the deviation value can more significantly reflect the difference between the virtual space to be predicted and the target object, thereby improving the prediction accuracy of the correction value.

[0166] In one exemplary embodiment, the virtual space information includes second virtual space information, referred to Figure 8As shown, in step 104, the virtual space information of the virtual space to be predicted is obtained based on the resource data corresponding to the virtual space to be predicted. This can be achieved through the following steps:

[0167] In step 802, the second target resource data currently displayed in the virtual space to be predicted is obtained;

[0168] In step 804, the category information corresponding to the second target resource data is used as the second virtual space information corresponding to the virtual space to be predicted.

[0169] In this embodiment, the second target resource data currently displayed in the virtual space to be predicted is the resource data highlighted and explained in the virtual space at the current moment. The second target resource data can be obtained by recognizing and processing the image and audio information of the virtual space at the current moment, and then the category information corresponding to the second target resource data can be searched from the resource data information database. Alternatively, the second target resource data and its corresponding category information can be directly obtained by recognizing and processing the image and audio information of the virtual space at the current moment. The specific recognition and processing can be implemented using image processing and recognition technology and natural language processing technology; this embodiment does not specifically limit the specific method of recognition and processing.

[0170] After obtaining the category information corresponding to the second target resource data, the category information corresponding to the second target resource data can be used as the second virtual space information. That is, the second virtual space information is used to characterize the category features of the virtual resources currently displayed in the virtual space to be predicted. In one example, the second virtual space information can explain the category features of the virtual space.

[0171] The click-through rate correction method provided in this embodiment combines second virtual space information to predict the correction value. Since the second virtual space information is used to characterize the category characteristics of the virtual resources currently displayed in the virtual space to be predicted, using it in combination with object information to predict the deviation value can more significantly reflect the difference between the virtual space to be predicted and the target object, thereby improving the prediction accuracy of the correction value.

[0172] In one exemplary embodiment, reference is made to Figure 9 As shown, the click-through rate correction method provided in this embodiment may further include the following steps:

[0173] In step 902, a training set is constructed, which includes multiple sample groups. Each sample group includes sample information and annotation information for the sample information. The sample information includes object information of the sample object, virtual space information of the sample virtual space, and sample predicted click rate of the sample object for the sample virtual space. The annotation information is used to characterize the click situation of the sample object for the sample virtual space.

[0174] In step 904, the predicted click-through rate of the sample is corrected by the initial correction model based on the object information of the sample object and the virtual space information of the sample virtual space, so as to obtain the predicted click-through rate of the sample user for the sample virtual space.

[0175] In step 906, the loss value of the initial correction model is determined based on the predicted click-through rate and annotation information of the sample object for the sample virtual space.

[0176] In step 908, an initial correction model is trained based on the loss value to obtain the correction model.

[0177] In this embodiment, sample information can be pre-constructed based on historical data. The virtual space involved in the historical data serves as the sample virtual space, and the objects involved are sample objects. The following uses one sample object and one sample virtual space as an example to illustrate the process of constructing a sample group. The virtual space information corresponding to the sample virtual space and the object information of the sample object can be obtained based on historical data.

[0178] For example, the virtual space information may include first virtual space information and second virtual space information. Historical data may include the opening data of the sample virtual space. Since the virtual space displays the resource data to be displayed through a resource list when it is opened, the first virtual space information corresponding to the sample virtual space can be obtained based on the historical data. The historical data may include replay data, so the second virtual space information corresponding to the sample virtual space at each time can be determined from the replay data. The process of determining the first virtual space information and the second virtual space information can be referred to the relevant description in the foregoing embodiments, and will not be repeated here in the embodiments of this disclosure.

[0179] For example, the object information may include first object information and second object information. The historical data may include the sample object's historical browsing logs for virtual space and the sample object's historical search logs for resource data. Therefore, the first object information of the sample object can be obtained from the historical search logs, and the second object information of the sample object can be obtained from the historical browsing logs. The specific process of determining the first object information and the second object information can be referred to the relevant description in the foregoing embodiments, and will not be repeated here in the embodiments of this disclosure.

[0180] In an exemplary embodiment, after obtaining object information and virtual space information, each feature data in the object information and each feature data in the virtual space information can be multiplied pairwise to obtain cross-feature information.

[0181] Historical data may also include the predicted click-through rate (CTR) of sample objects for each sample virtual space, as predicted by the CTR prediction platform. The predicted CTR of sample objects for each sample virtual space can then be obtained from the CTR prediction platform as the sample predicted CTR.

[0182] A sample can be defined as a sample object, a sample virtual space, the object information of the sample object, the virtual space information of the sample virtual space, and the sample predicted click-through rate of the sample object relative to the sample virtual space.

[0183] Historical data can also include the actual click behavior of sample objects in the sample virtual space (i.e., whether they clicked or not). Therefore, the actual click behavior of sample objects in the sample virtual space can be obtained from historical data as annotation information for sample information. Based on the sample information and its annotation information, a sample group can be constructed.

[0184] By analogy, multiple sample groups can be obtained. These sample groups can be used to construct a training set, which can then be used to train an initial correction model to obtain the final correction model. The training process for the initial correction model can include two parts: forward computation and backpropagation.

[0185] During the forward computation, the object information of the sample object (first object information and second object information), the virtual space information of the sample virtual space (first virtual space information and second virtual space information), and the cross-feature information can be input into the correction value prediction unit in the initial correction model. The output of the correction value prediction unit and the sample predicted click rate are then corrected by the correction layer in the initial correction model. After the corrected value is normalized by the normalization layer, the predicted click rate of the sample object for the sample virtual space is obtained.

[0186] During backpropagation, the predicted click-through rate and labeled information of the sample object in the sample virtual space are used to calculate the loss value using a loss function. Simultaneously, stochastic gradient descent is employed to solve for the gradient of the loss function based on the loss value, minimizing the loss value in the forward computation. Backpropagation updates the network parameters of the initial bias correction model through reverse, layer-by-layer gradient propagation, causing the network parameters to gradually converge.

[0187] For example, the initial bias correction model may include a bias correction value prediction unit and a bias correction unit. The bias correction value prediction unit includes a feature extraction module (Embedding Layer) and a prediction module (Neural Network), and the bias correction unit includes a bias correction layer and a normalization layer.

[0188] In the feature extraction module, the object information of the sample object, the virtual space information of the sample virtual space, and the cross-feature information generated by their intersection are transformed into dense features, which are then input into the prediction module. In the prediction module, a multi-layer convolutional network is used to perform convolutional prediction processing on the dense features to obtain the bias correction value corresponding to the predicted click-through rate of the sample. Since too many network layers can lead to overfitting, while too few layers can lead to underfitting, in one example, a 3-layer network with [128, 32, 1] network nodes can be constructed as the prediction module, allowing the network to learn fully and converge quickly.

[0189] For the bias correction algorithm, refer to Formula (I). During training, 'a' can represent the predicted click-through rate (CTR) of the sample, and 'b' can represent the bias correction value corresponding to the predicted CTR. After the bias correction algorithm corrects the predicted CTR and the corresponding bias correction value (the specific process can be referred to the relevant description in the foregoing embodiments, which will not be repeated here), the predicted target click value of the sample object in the sample virtual space can be obtained. In the normalization layer, the processing of the normalization layer can refer to Formula (II) above. During training, x = f(a,b), f(a,b) represents the predicted target click value of the sample object in the sample virtual space. The normalization layer compresses the predicted target click value to the range of 0 to 1, and the final output value is used as the predicted CTR of the sample object in the sample virtual space after correction by the initial bias correction model.

[0190] After obtaining the predicted click-through rate of the sample object for the sample virtual space, the loss value of the initial correction model can be determined based on the annotation information corresponding to the sample information (positive samples with click behavior and negative samples without click behavior) using a loss algorithm. This disclosure does not specifically limit the loss algorithm; any loss algorithm that can be used for loss calculation is applicable to this disclosure. For example, the loss algorithm can refer to the following formula (III).

[0191] l t =-y t logp t -(1-y t log(1-p) t Formula (III)

[0192] Where, p tUsed to represent the predicted click-through rate of a sample object within the sample virtual space. t ∈{0,1} represents the annotation information corresponding to the sample information, t represents one sample information, and l t This represents the loss value.

[0193] After determining the loss value, if the loss value does not meet the training conditions (e.g., the loss value is greater than or equal to the loss threshold), the network parameters of the initial correction model can be adjusted based on the loss value. The initial correction model with adjusted network parameters can then be iteratively trained based on the training set until the loss value meets the training conditions (e.g., the loss value is less than the loss threshold). At this point, training stops, and the current initial correction model is used as the final correction model.

[0194] After training, a correction process is performed based on the correction model. For example, when a user opens the platform, the click-through rate (CTR) can be estimated based on the user's log information in the platform's database and the opening information of various virtual spaces, generating an initial estimated CTR. The value ranges from 0 to 1, with a higher value indicating a greater likelihood that the model believes the user clicked on a virtual space. By using object information, virtual space information, cross-feature information, and the initial estimated CTR as input to the correction model, a more accurate predicted CTR can be obtained after the correction model corrects the initial estimated CTR.

[0195] The click-through rate correction method provided in this disclosure can construct a sample group by using the object information of the sample object, the virtual space information of the sample virtual space, the sample predicted click-through rate of the sample object for the sample virtual space, and the click situation of the sample object for the sample virtual space. A training set is constructed based on the sample group, and an initial correction model is trained based on the training set to obtain the correction model. Then, the correction model is used to correct the initial estimated click-through rate, which can improve the accuracy of the predicted click-through rate of the target object for the virtual space to be predicted, and further improve the efficiency of users interacting with resources through the virtual space.

[0196] This disclosure provides a click-through rate (CTR) correction method. By correcting the initial CTR prediction of the recommendation model, it provides a more accurate prediction of user CTR. The correction model provided in this disclosure primarily models the category features of resource data, effectively improving the category diversity of resource data displayed in the virtual spaces recommended to users. This CTR correction method, without improving existing recommendation models, effectively predicts the probability of users clicking on virtual spaces, enabling personalized virtual space recommendations, improving the matching efficiency between users and virtual spaces, and increasing the efficiency of user resource interaction through virtual spaces.

[0197] It should be understood that, although Figures 1-9The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figures 1-9 At least some of the steps in the process may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but may be executed at different times. The execution order of these steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least some of the steps or stages in other steps.

[0198] It is understood that the same / similar parts between the various embodiments of the methods described above in this specification can be referred to each other. Each embodiment focuses on the differences from other embodiments, and relevant parts can be referred to the description of other method embodiments.

[0199] Figure 10 This is a block diagram illustrating a click-through rate correction device according to an exemplary embodiment. (Refer to...) Figure 10 The device includes a first acquisition unit 1002, a second acquisition unit 1004, a third acquisition unit 1006, and a correction unit 1008.

[0200] The first acquisition unit 1002 is configured to acquire object information of the target object;

[0201] The second acquisition unit 1004 is configured to acquire virtual space information of the virtual space to be predicted based on the resource data corresponding to the virtual space to be predicted.

[0202] The third acquisition unit 1006 is configured to acquire the initial estimated click rate of the target object for the virtual space to be predicted from the click rate prediction module;

[0203] The correction unit 1008 is configured to perform correction processing on the initial estimated click-through rate based on the object information and the virtual space information to obtain the predicted click-through rate of the target object for the virtual space to be predicted.

[0204] The click-through rate correction device provided in this embodiment can acquire object information of a target object and, based on resource data corresponding to the virtual space to be predicted, acquire virtual space information of the virtual space to be predicted. After acquiring the initial estimated click-through rate of the target object for the virtual space to be predicted, the initial estimated click-through rate can be corrected based on object information and virtual space feature information to obtain the predicted click-through rate of the target object for the virtual space to be predicted. Based on the click-through rate correction device provided in this embodiment, after obtaining the initial estimated click-through rate of the target object for the virtual space to be predicted, further correction processing of the initial estimated click-through rate can be performed based on object information and virtual space information, which can improve the accuracy of the predicted click-through rate of the target object for the virtual space to be predicted, and further improve the efficiency of users interacting with resources through virtual space.

[0205] In one embodiment, the correction unit is further configured to perform:

[0206] The initial estimated click-through rate is corrected using a correction model based on the object information and the virtual space information to obtain the predicted click-through rate of the target object for the virtual space to be predicted.

[0207] The correction model is pre-trained and determines the deviation of the initial estimated click-through rate based on the object information and the virtual space information, and eliminates the deviation of the initial estimated click-through rate.

[0208] In one embodiment, the correction model includes a correction value prediction unit and a correction unit, wherein the correction unit 1008 is further configured to perform:

[0209] The correction value prediction unit performs prediction processing on the object information and the virtual space information to obtain the correction value;

[0210] The correction unit performs correction processing on the initial estimated click-through rate and the correction value to obtain the predicted click-through rate of the target object for the virtual space to be predicted.

[0211] In one embodiment, the correction unit includes a correction layer and a normalization layer, and the correction unit 1008 is further configured to perform:

[0212] The initial estimated click-through rate is inversely normalized by the correction layer to obtain the inversely normalized initial estimated click-through rate. The correction value is then used to correct and adjust the inversely normalized initial estimated click-through rate to obtain the predicted target click value of the target object for the virtual space to be predicted.

[0213] The predicted target click value is normalized by the normalization layer to obtain the predicted click rate of the target object for the virtual space to be predicted.

[0214] In one embodiment, the correction unit 1008 is further configured to perform:

[0215] Based on the object information and the virtual space information, feature cross-referencing is performed to obtain cross-feature information;

[0216] The initial estimated click-through rate is corrected by a correction model based on the object information, the virtual space information, and the cross-feature information to obtain the predicted click-through rate of the target object for the virtual space to be predicted.

[0217] In one embodiment, the object information includes first object information, and the first acquisition unit 1002 is further configured to perform:

[0218] Obtain at least one search information for the target object, wherein the search information is record information generated when searching for resource data;

[0219] The search information is processed to identify categories, thereby obtaining category information corresponding to each search information.

[0220] The category information corresponding to each of the search information is used as the first object information.

[0221] In one embodiment, the object information includes second object information, and the first acquisition unit 1002 is further configured to perform:

[0222] Obtain at least one browsing information of the target object, wherein the browsing information is record information generated when browsing resource data;

[0223] For any of the browsing information, a target virtual space and a browsing time for the target virtual space are determined based on the browsing information, and first target resource data displayed in the target virtual space during the browsing time are determined;

[0224] The category information corresponding to each of the first target resource data is used as the second object information of the target object.

[0225] In one embodiment, the virtual space information includes first virtual space information, and the second acquisition unit 1004 is further configured to perform:

[0226] Obtain the resource list of the virtual space to be predicted;

[0227] Obtain the category information corresponding to each resource data in the resource list;

[0228] The category information corresponding to each resource data in the resource list is used as the first virtual space information corresponding to the virtual space to be predicted.

[0229] In one embodiment, the virtual space information includes second virtual space information, and the second acquisition unit 1004 is further configured to perform:

[0230] Obtain the second target resource data currently displayed in the virtual space to be predicted;

[0231] The category information corresponding to the second target resource data is used as the second virtual space information corresponding to the virtual space to be predicted.

[0232] In one embodiment, the device further includes:

[0233] The construction unit is configured to construct a training set, which includes multiple sample groups. Each sample group includes sample information and annotation information for the sample information. The sample information includes object information of the sample object, virtual space information of the sample virtual space, and the sample predicted click rate of the sample object for the sample virtual space. The annotation information is used to characterize the click situation of the sample object for the sample virtual space.

[0234] The processing unit is configured to perform bias correction processing on the predicted click-through rate of the sample based on the object information of the sample object and the virtual space information of the sample virtual space using an initial bias correction model, so as to obtain the predicted click-through rate of the sample object for the sample virtual space.

[0235] The determining unit is configured to determine the loss value of the initial correction model based on the predicted click-through rate of the sample object for the sample virtual space and the annotation information.

[0236] The training unit is configured to train the initial correction model based on the loss value to obtain the correction model.

[0237] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0238] Figure 11 This is a block diagram illustrating an electronic device 1100 for click-through rate correction according to an exemplary embodiment. For example, the electronic device 1100 may be a mobile phone, computer, digital broadcasting terminal, messaging device, game console, tablet device, medical device, fitness equipment, personal digital assistant, etc.

[0239] Reference Figure 11The electronic device 1100 may include one or more of the following components: processing component 1102, memory 1104, power supply component 1106, multimedia component 1108, audio component 1110, input / output (I / O) interface 1112, sensor component 1114, and communication component 1116.

[0240] Processing component 1102 typically controls the overall operation of electronic device 1100, such as operations associated with display, telephone calls, data communication, camera operation, and recording operations. Processing component 1102 may include one or more processors 1120 to execute instructions to perform all or part of the steps of the methods described above. Furthermore, processing component 1102 may include one or more modules to facilitate interaction between processing component 1102 and other components. For example, processing component 1102 may include a multimedia module to facilitate interaction between multimedia component 1108 and processing component 1102.

[0241] Memory 1104 is configured to store various types of data to support the operation of electronic device 1100. Examples of such data include instructions for any application or method operating on electronic device 1100, contact data, phonebook data, messages, pictures, videos, etc. Memory 1104 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, optical disk, or graphene memory.

[0242] Power supply component 1106 provides power to various components of electronic device 1100. Power supply component 1106 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to electronic device 1100.

[0243] Multimedia component 1108 includes a screen that provides an output interface between the electronic device 1100 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touchscreen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors may sense not only the boundaries of the touch or swipe action but also the duration and pressure associated with the touch or swipe operation. In some embodiments, multimedia component 1108 includes a front-facing camera and / or a rear-facing camera. When the electronic device 1100 is in an operating mode, such as a shooting mode or a video mode, the front-facing camera and / or the rear-facing camera may receive external multimedia data. Each front-facing camera and rear-facing camera may be a fixed optical lens system or have focal length and optical zoom capabilities.

[0244] Audio component 1110 is configured to output and / or input audio signals. For example, audio component 1110 includes a microphone (MIC) configured to receive external audio signals when electronic device 1100 is in an operating mode, such as call mode, recording mode, and voice recognition mode. The received audio signals may be further stored in memory 1104 or transmitted via communication component 1116. In some embodiments, audio component 1110 also includes a speaker for outputting audio signals.

[0245] I / O interface 1112 provides an interface between processing component 1102 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to, home buttons, volume buttons, power buttons, and lock buttons.

[0246] Sensor assembly 1114 includes one or more sensors for providing state assessments of various aspects of electronic device 1100. For example, sensor assembly 1114 may detect the on / off state of electronic device 1100, the relative positioning of components such as the display and keypad of electronic device 1100, changes in position of electronic device 1100 or its components, the presence or absence of user contact with electronic device 1100, orientation or acceleration / deceleration of device 1100, and temperature changes of electronic device 1100. Sensor assembly 1114 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. Sensor assembly 1114 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, sensor assembly 1114 may also include an accelerometer, gyroscope, magnetometer, pressure sensor, or temperature sensor.

[0247] Communication component 1116 is configured to facilitate wired or wireless communication between electronic device 1100 and other devices. Electronic device 1100 can access wireless networks based on communication standards, such as WiFi, carrier networks (such as 2G, 3G, 4G, or 5G), or combinations thereof. In one exemplary embodiment, communication component 1116 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, communication component 1116 also includes a near-field communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on radio frequency identification (RFID) technology, Infrared Data Association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0248] In an exemplary embodiment, the electronic device 1100 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the methods described above.

[0249] In an exemplary embodiment, a computer-readable storage medium including instructions is also provided, such as a memory 1104 including instructions, which can be executed by a processor 1120 of an electronic device 1100 to perform the above-described method. For example, the computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0250] In an exemplary embodiment, a computer program product is also provided, which includes instructions that can be executed by a processor 1120 of an electronic device 1100 to perform the above-described method.

[0251] It should be noted that the above-mentioned apparatus, electronic equipment, computer-readable storage medium, computer program product, etc., may also include other implementation methods according to the description of the method embodiments. For specific implementation methods, please refer to the description of the relevant method embodiments, which will not be elaborated here.

[0252] 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 claims.

[0253] 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 click-through rate correction method, characterized in that, include: Obtain object information of the target object, wherein the target object is the user currently to be recommended in the virtual space, and the object information includes first object information and second object information. The first object information includes user search category features, which are used to characterize the category features corresponding to the searched resources when the target object performs a search operation. The second object information includes user viewing category features, which are used to characterize the category features corresponding to the resource data displayed in the virtual space when the target object has historically viewed the virtual space. Both the first object information and the second object information adopt a multi-level directory hierarchy structure. Based on the resource data corresponding to the virtual space to be predicted, the virtual space information of the virtual space to be predicted is obtained. The virtual space information includes first virtual space information and second virtual space information. The first virtual space information includes virtual space shelf category features, which are used to characterize the category features of the resource data to be displayed in the virtual space to be predicted. The second virtual space information includes virtual space explanation category features, which are used to characterize the category features of the resource data currently displayed in the virtual space to be predicted. Both the first virtual space information and the second virtual space information adopt a multi-level directory hierarchy structure. The initial estimated click-through rate of the target object for the virtual space to be predicted is obtained from the click-through rate prediction module; The initial estimated click-through rate is corrected based on the object information and the virtual space information to obtain the predicted click-through rate of the target object for the virtual space to be predicted.

2. The method according to claim 1, characterized in that, The step of correcting the initial estimated click-through rate based on the object information and the virtual space information to obtain the predicted click-through rate of the target object for the virtual space to be predicted includes: The initial estimated click-through rate is corrected using a correction model based on the object information and the virtual space information to obtain the predicted click-through rate of the target object for the virtual space to be predicted. The correction model is pre-trained and determines the deviation of the initial estimated click-through rate based on the object information and the virtual space information, and eliminates the deviation of the initial estimated click-through rate.

3. The method according to claim 2, characterized in that, The correction model includes a correction value prediction unit and a correction unit. The step of correcting the initial estimated click-through rate based on the object information and the virtual space information using the correction model to obtain the predicted click-through rate of the target object for the virtual space to be predicted includes: The correction value prediction unit performs prediction processing on the object information and the virtual space information to obtain the correction value; The correction unit performs correction processing on the initial estimated click-through rate and the correction value to obtain the predicted click-through rate of the target object for the virtual space to be predicted.

4. The method according to claim 3, characterized in that, The correction unit includes a correction layer and a normalization layer. The step of correcting the initial estimated click-through rate and the correction value using the correction unit to obtain the predicted click-through rate of the target object for the virtual space to be predicted includes: The initial estimated click-through rate is inversely normalized by the correction layer to obtain the inversely normalized initial estimated click-through rate. The correction value is then used to correct and adjust the inversely normalized initial estimated click-through rate to obtain the predicted target click value of the target object for the virtual space to be predicted. The predicted target click value is normalized by the normalization layer to obtain the predicted click rate of the target object for the virtual space to be predicted.

5. The method according to any one of claims 2 to 4, characterized in that, The step of correcting the initial estimated click-through rate using a correction model based on the object information and the virtual space information to obtain the predicted click-through rate of the target object for the virtual space to be predicted includes: Based on the object information and the virtual space information, feature cross-referencing is performed to obtain cross-feature information; The initial estimated click-through rate is corrected by a correction model based on the object information, the virtual space information, and the cross-feature information to obtain the predicted click-through rate of the target object for the virtual space to be predicted.

6. The method according to claim 1, characterized in that, The object information includes first object information, and obtaining the object information of the target object includes: Obtain at least one search information for the target object, wherein the search information is record information generated when searching for resource data; The search information is processed to identify categories, thereby obtaining category information corresponding to each search information. The category information corresponding to each of the search information is used as the first object information.

7. The method according to claim 1 or 6, characterized in that, The object information includes second object information, and obtaining the object information of the target object includes: Obtain at least one browsing information of the target object, wherein the browsing information is record information generated when browsing resource data; For any of the browsing information, a target virtual space and a browsing time for the target virtual space are determined based on the browsing information, and first target resource data displayed in the target virtual space during the browsing time are determined; The category information corresponding to each of the first target resource data is used as the second object information of the target object.

8. The method according to claim 1, characterized in that, The virtual space information includes first virtual space information. The step of obtaining the virtual space information of the virtual space to be predicted based on the resource data corresponding to the virtual space to be predicted includes: Obtain the resource list of the virtual space to be predicted; Obtain the category information corresponding to each resource data in the resource list; The category information corresponding to each resource data in the resource list is used as the first virtual space information corresponding to the virtual space to be predicted.

9. The method according to claim 1 or 8, characterized in that, The virtual space information includes second virtual space information. The step of obtaining the virtual space information of the virtual space to be predicted based on the resource data corresponding to the virtual space to be predicted includes: Obtain the second target resource data currently displayed in the virtual space to be predicted; The category information corresponding to the second target resource data is used as the second virtual space information corresponding to the virtual space to be predicted.

10. The method according to any one of claims 2 to 4, characterized in that, The method further includes: Construct a training set, which includes multiple sample groups. Each sample group includes sample information and annotation information for the sample information. The sample information includes object information of the sample object, virtual space information of the sample virtual space, and the sample predicted click rate of the sample object for the sample virtual space. The annotation information is used to characterize the click situation of the sample object for the sample virtual space. The initial correction model corrects the predicted click-through rate of the sample based on the object information of the sample object and the virtual space information of the sample virtual space, thereby obtaining the predicted click-through rate of the sample object for the sample virtual space. Based on the predicted click-through rate of the sample object in the sample virtual space and the annotation information, the loss value of the initial correction model is determined; The initial correction model is trained based on the loss value to obtain the correction model.

11. A click-through rate correction device, characterized in that, include: The first acquisition unit is configured to acquire object information of the target object, which is the user currently being recommended in the virtual space. The object information includes first object information and second object information. The first object information includes user search category features, which are used to characterize the category features corresponding to the searched resources when the target object performs a search operation. The second object information includes user viewing category features, which are used to characterize the category features corresponding to the resource data displayed in the virtual space when the target object has historically viewed the virtual space. Both the first object information and the second object information adopt a multi-level directory hierarchy structure. The second acquisition unit is configured to acquire virtual space information of the virtual space to be predicted based on the resource data corresponding to the virtual space to be predicted. The virtual space information includes first virtual space information and second virtual space information. The first virtual space information includes virtual space shelf category features, which are used to characterize the category features of the resource data to be displayed in the virtual space to be predicted. The second virtual space information includes virtual space explanation category features, which are used to characterize the category features of the resource data currently displayed in the virtual space to be predicted. Both the first virtual space information and the second virtual space information adopt a multi-level directory hierarchy structure. The third acquisition unit is configured to acquire the initial estimated click rate of the target object for the virtual space to be predicted from the click rate prediction module; The correction unit is configured to perform correction processing on the initial estimated click-through rate based on the object information and the virtual space information to obtain the predicted click-through rate of the target object for the virtual space to be predicted.

12. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the click-through rate correction method as described in any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that, When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is enabled to perform the click-rate correction method as described in any one of claims 1 to 10.

Images