Recommended content processing method and device, equipment, readable storage medium and product

Abstract

The embodiment of the invention provides a recommended content processing method and device, equipment, a readable storage medium and a product, and the method comprises the steps: obtaining a preset training data set, the training data comprises feature data and annotation data, the feature data comprises historical features and content features corresponding to each recommended content associated with a user, and the annotation data comprises the historical features and the content features corresponding to each recommended content associated with the user; the annotation data is a click parameter of the user for each recommended content, and the content features comprise a prediction score associated with the recommendation queue and a sorting parameter; a preset to-be-trained model is iteratively trained based on the training data set, a target recommendation model is obtained, the to-be-trained model comprises an attention estimation module, and the attention estimation module is used for determining predicted click parameters associated with the recommendation content when the multiple pieces of multi-body recommendation content are jointly displayed. Therefore, the click operation of the user on each piece of recommended content can be considered when multiple pieces of multi-body recommended content co-exist, and the accuracy and comprehensiveness of content recommendation in the special form of the ranking list distribution position are improved.

Description

Technical Field

[0001] The embodiments of the present disclosure relate to the field of data processing technology, and in particular, to a method, device, equipment, readable storage medium and product for processing recommended content. Background Art

[0002] The user can browse the media content based on the target application preset in the terminal device, wherein the media content includes but is not limited to book media content, audio media content, graphic media content, etc.

[0003] In the related art, in order to enable users to browse richer media content that better meets their actual needs in a target application, user-related information is generally determined based on the user's browsing operation on the media content, and media content is recalled and recommended based on the related information.

[0004] However, the recommended content generated by the above method is often not very accurate, and the genre types of the covered media content are relatively single. Summary of the invention

[0005] The embodiments of the present disclosure provide a recommended content processing method, apparatus, device, readable storage medium and product, which are used to solve the technical problem that the accuracy of content recommended by current content recommendation methods is not high.

[0006] In a first aspect, an embodiment of the present disclosure provides a method for processing recommended content, including:

[0007] Acquire a preset training data set, the training data set includes multiple sets of training data, the training data includes feature data and annotation data, the feature data includes historical features associated with the user, content features corresponding to each recommended content in the recommendation queue, the annotation data is a click parameter of the user for each recommended content, the content features include a predicted score and a ranking parameter of the recommendation queue during the screening process and an associated parameter of each recommended content, the recommendation queue includes a first preset number of recommended contents of multiple genres, the historical features and the click parameters are acquired after authorization by the user;

[0008] The preset model to be trained is iteratively trained based on the training data set until the model to be trained meets the preset convergence conditions, thereby obtaining a target recommendation model, wherein the model to be trained includes a cross-feature extraction module, a vectorization module and an attention estimation module, and the attention estimation module is used to determine the predicted click parameters associated with each recommended content when multiple recommended contents are displayed together.

[0009] In a second aspect, an embodiment of the present disclosure provides a method for processing recommended content, including:

[0010] In response to a recommendation operation triggered by a target user, obtaining recommendation data fed back by a preset screening model, wherein the recommendation data includes a second preset number of recommended contents, a prediction score associated with each recommended content, recommended content features, and a ranking parameter used by the recommendation model in the recommendation process;

[0011] Based on a preset target recommendation model and a preset optimization algorithm, data processing is performed on the recommendation data and the target historical features associated with the target user, and a target recommendation queue is determined in the recommendation data, wherein the target recommendation queue includes a first preset number of recommended contents, wherein the first preset number is less than the second preset number, the target recommendation model is obtained by training based on the recommended content processing method described in the first aspect, and the target historical features are obtained after authorization by the user;

[0012] The target recommendation queue is displayed in a display interface.

[0013] In a third aspect, an embodiment of the present disclosure provides a recommended content processing device, including:

[0014] an acquisition module, configured to acquire a preset training data set, wherein the training data set includes multiple sets of training data, wherein the training data includes feature data and annotation data, wherein the feature data includes historical features associated with the user and content features corresponding to each recommended content in the recommendation queue, wherein the annotation data is a click parameter of the user for each recommended content, wherein the content features include a predicted score and a sorting parameter of the recommendation queue during the screening process and an associated parameter of each recommended content, wherein the recommendation queue includes a first preset number of recommended contents of multiple genres, wherein the historical features and the click parameters are acquired after authorization by the user;

[0015] A training module is used to iteratively train a preset model to be trained based on the training data set until the model to be trained meets the preset convergence conditions to obtain a target recommendation model, wherein the model to be trained includes a cross-feature extraction module, a vectorization module and an attention estimation module, and the attention estimation module is used to determine the predicted click parameters associated with each recommended content when multiple recommended contents are displayed together.

[0016] In a fourth aspect, an embodiment of the present disclosure provides a recommended content processing device, including:

[0017] a data acquisition module, configured to acquire, in response to a recommendation operation triggered by a target user, recommendation data fed back by a preset screening model, wherein the recommendation data includes a second preset number of recommended contents, a prediction score associated with each recommended content, recommended content features, and a ranking parameter used by the recommendation model in the recommendation process;

[0018] a processing module, configured to perform data processing on the recommendation data and the target historical features associated with the target user based on a preset target recommendation model and a preset optimization algorithm, determine a target recommendation queue in the recommendation data, the target recommendation queue includes a first preset number of recommended contents, wherein the first preset number is less than the second preset number, the target recommendation model is obtained by training based on the recommended content processing device according to the third aspect, and the target historical features are obtained after authorization by the user;

[0019] A display module is used to display the target recommendation queue in a display interface.

[0020] In a fifth aspect, an embodiment of the present disclosure provides an electronic device, including: a processor and a memory;

[0021] The memory stores computer-executable instructions;

[0022] The processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the recommended content processing method described in the first aspect and various possible designs of the first aspect or the second aspect and various possible designs of the second aspect.

[0023] In a sixth aspect, an embodiment of the present disclosure provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer execution instructions. When a processor executes the computer execution instructions, the recommended content processing method described in the first aspect and various possible designs of the first aspect or the second aspect and various possible designs of the second aspect is implemented.

[0024] In a seventh aspect, an embodiment of the present disclosure provides a computer program product, including a computer program, which, when executed by a processor, implements the recommended content processing method described in the first aspect and various possible designs of the first aspect or the second aspect and various possible designs of the second aspect.

[0025] The recommended content processing method, device, equipment, readable storage medium and product provided in this embodiment perform model training operations based on historical features associated with users, content features associated with recommended content, and click parameters of users on multiple multi-genre recommended content in the recommendation queue during the model training process, so that when content recommendations are subsequently made based on the trained target recommendation model, the user's browsing operations on each recommended content when multiple recommended content co-occur can be considered, thereby improving the accuracy and comprehensiveness of content recommendations under the special form of ranking distribution positions. In addition, by introducing optimization algorithms and value algorithms, the optimal target recommendation queue can be more accurately determined in combination with historical features and recommended content features, thereby improving the accuracy of content recommendations. In addition, by using a multi-genre recommendation queue for model training operations, the trained target recommendation model can recommend recommended content of diverse genres to users based on recommendation data. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0027] Figure 1 A diagram of the system architecture on which the embodiments of the present disclosure are based;

[0028] Figure 2 A flowchart of a method for processing recommended content provided by an embodiment of the present disclosure;

[0029] Figure 3 A model architecture diagram of a target recommendation model provided in an embodiment of the present disclosure;

[0030] Figure 4 It is another system architecture diagram based on the embodiments of the present disclosure;

[0031] Figure 5 A flowchart of a method for processing recommended content provided by an embodiment of the present disclosure;

[0032] Figure 6 A flowchart of a method for processing recommended content provided by another embodiment of the present disclosure;

[0033] Figure 7 A model architecture diagram of a target recommendation model provided in an embodiment of the present disclosure;

[0034] Figure 8 A schematic diagram of the structure of a recommended content processing device provided in an embodiment of the present disclosure;

[0035] Fig. 9 A schematic diagram of the structure of a recommended content processing device provided in an embodiment of the present disclosure;

[0036] Fig.10 A schematic diagram of the structure of an electronic device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0038] It is understandable that before using the technical solutions disclosed in the embodiments of the present disclosure, the types, scope of use, usage scenarios, etc. of the personal information involved in the present disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0039] For example, in response to receiving an active request from a user, a prompt message is sent to the user to clearly prompt the user that the operation requested to be performed will require obtaining and using the user's personal information. Thus, the user can autonomously choose whether to provide personal information to software or hardware such as an electronic device, application, server, or storage medium that performs the operation of the technical solution of the present disclosure according to the prompt message.

[0040] As an optional but non-limiting implementation, in response to receiving an active request from the user, the prompt information may be sent to the user in the form of a pop-up window, in which the prompt information may be presented in text form. In addition, the pop-up window may also carry a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0041] It is understandable that the above notification and the process of obtaining user authorization are merely illustrative and do not constitute a limitation on the implementation of the present disclosure. Other methods that meet the relevant laws and regulations may also be applied to the implementation of the present disclosure.

[0042] In order to solve the technical problem that the accuracy of content recommended by current content recommendation methods is not high, the present disclosure provides a recommended content processing method, device, equipment, readable storage medium and product.

[0043] It should be noted that the recommended content processing method, device, equipment, readable storage medium and product provided by the present disclosure can be applied in any content recommendation scenario.

[0044] In the related art, in order to facilitate users to browse more abundant media content in the target application, the user's browsing operation on a certain media content is generally obtained, and the user's interest in a certain type of media content is determined based on the browsing operation, and then the media content is recalled and recommended. For example, a user can browse books in a book application. When the user triggers the recommendation page, a preset number of recommended contents can be displayed on the recommendation page so that the user can discover more books for browsing.

[0045] However, content recommendations based on users' browsing operations on single media content are often not very accurate. The generated recommended content often covers fewer genres, and the media content categories are relatively single, which cannot meet the personalized needs of users.

[0046] In the process of solving the above technical problems, the inventors discovered through research that in order to improve the accuracy of content recommendation, data analysis can be performed based on users' click operations on multiple recommended contents, and the recommended contents can be determined based on the analysis results.

[0047] Optionally, a model training operation can be performed based on historical features, content features associated with the recommended content, and user click parameters on multiple recommended contents in the recommendation queue, so that when content recommendations are subsequently performed based on the trained target recommendation model, the user's browsing operations on each recommended content when multiple recommended contents co-occur can be considered, thereby improving the accuracy and comprehensiveness of content recommendations under the special form of ranking distribution positions.

[0048] In addition, after training the target recommendation model, the recommendation data can be input into the target recommendation model. After obtaining the cross-recommendation feature vector output by the target recommendation model, the cross-recommendation feature vector can be input into the optimization algorithm to obtain the optimization algorithm to select the target recommendation queue from the recommendation data.

[0049] As an implementable method, multiple rounds of sequence optimization operations can be performed through an optimization algorithm, and the value of each round of recommendation lists can be calculated in combination with a preset value function to select the optimal sequence based on the value for content recommendation operations.

[0050] Figure 1 This is a system architecture diagram based on the embodiments of the present disclosure, such as Figure 1As shown, the system architecture based on the present disclosure includes at least a data server 11 and a server 12, wherein the data server 11 may store a preset training data set. The training data set includes multiple sets of training data, and the training data includes feature data and annotation data. The feature data includes historical features, content features corresponding to each recommended content in the recommendation queue associated with the user, and the annotation data is the user's click parameters for each recommended content. The content features include the predicted score and sorting parameters of the recommendation queue during the screening process, and the recommendation queue includes a first preset number of recommended contents. The server 12 can obtain the training data set from the data server 11, and perform iterative training operations on the preset model to be trained based on the training data set to generate a target recommendation model.

[0051] Figure 2 A flowchart of a method for processing recommended content provided by an embodiment of the present disclosure is shown in FIG. Figure 2 As shown, the method includes:

[0052] Step 201, obtaining a preset training data set, wherein the training data set includes multiple groups of training data, the training data includes feature data and annotation data, the feature data includes historical features associated with the user, content features corresponding to each recommended content in the recommendation queue, the annotation data is the user's click parameters for each recommended content, the content features include the predicted score and sorting parameters of the recommendation queue during the screening process and the associated parameters of each recommended content, the recommendation queue includes a first preset number of recommended contents of multiple genres, and the historical features and the click parameters are obtained after the user is authorized.

[0053] The execution subject of this embodiment is a recommended content processing device. The recommended content processing device can be coupled to a server. The server can communicate with a data server, so as to obtain a training data set from the data server, and perform an iterative training operation on a preset model to be trained based on the training data set to obtain a target recommendation model.

[0054] In this embodiment, the user can browse media content on the target application. The media content includes but is not limited to book media content, audio media content, graphic media content, etc. In order to enable users to browse richer media content, content recommendation operations can also be performed based on the user's click operation on the media content. A preset number of recommended contents can be displayed on the recommendation page so that the user can view the recommended contents.

[0055] Taking practical application as an example, the recommendation page may include a content recommendation area, in which eight recommended contents may be displayed, and the user may switch to view the remaining eight recommended contents by sliding in a preset direction.

[0056] Alternatively, currently, the browsing preferences of users are generally analyzed based on the browsing operation of users on a single media content, and content recommendations are made based on the browsing preferences of users, but the recommendation accuracy is often not high. Since a preset number of recommended contents are often displayed on the recommendation page, content recommendation operations can be made based on the user's click operations on the preset number of recommended contents.

[0057] In order to achieve accurate content recommendation operations, a target recommendation model can be pre-trained. Therefore, after obtaining the recommendation data to be recommended, a target recommendation queue can be screened out from the recommendation data based on the target recommendation model. The recommendation data includes a second preset number of recommended contents, and the target recommendation queue includes a first preset number of recommended contents, and the first preset number is less than the second preset number.

[0058] For example, if the recommended content is a book content, the recommended data includes 20 book media contents to be recommended. Based on the target recommendation model, 8 book media contents that best meet the actual needs of the user can be selected from the 20 book media contents to form a target recommendation queue.

[0059] In order to implement the training operation of the target recommendation model, first, a preset training data set can be obtained, the training data set includes multiple sets of training data, the training data includes feature data and annotation data, the feature data includes historical features, content features corresponding to each recommended content in the recommendation queue associated with the user, wherein the recommendation queue associated with the user may include a first preset number of recommended contents recommended to the user in history. The recommended content is displayed based on the recommendation request triggered by the user. The user can click on the multiple recommended contents. The content features may include associated parameters of each recommended content, such as identification information, publication information, classification information, category label information, etc. of the recommended content.

[0060] The historical features include, but are not limited to, the user's consumption history of the recommended content, the user's historical activity, and other features.

[0061] It should be noted that the historical features and the click operation are obtained after obtaining full authorization from the user.

[0062] As an optional but non-limiting implementation, in response to receiving an active request from the user, the prompt information may be sent to the user in the form of a pop-up window, in which the prompt information may be presented in text form. In addition, the pop-up window may also carry a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0063] It is understandable that the above notification and the process of obtaining user authorization are merely illustrative and do not constitute a limitation on the implementation of the present disclosure. Other methods that meet the relevant laws and regulations may also be applied to the implementation of the present disclosure.

[0064] In addition, the recommendation queue can be obtained by rough sorting, fine sorting, and mixed sorting according to categories of the recalled content associated with the user. Therefore, in order to improve the accuracy of model training, the content features can also include the predicted score and sorting parameters of the recommendation queue during the screening process. Among them, the predicted score can be the estimated score of the fine sorting of the X associated with each recommended content. The sorting parameters can be the sorting formula name, formula score, and formula order corresponding to the genre queue in the mixed sorting process.

[0065] By introducing the prediction scores and sorting parameters in the refined and mixed sorting processes, the prediction accuracy of the trained target recommendation model can be further improved.

[0066] Furthermore, users can also click on the recommended content in the recommendation queue. Therefore, the labeled data is the user's click parameters for each recommended content. For example, the label data may include the CTR, second retention, second retention, ..., seventh retention values ​​associated with each recommended content. Among them, N retention means that the user clicks on a recommended content in this recommendation request, and consumes the recommended content after X days, which is recorded as X retention.

[0067] Step 202: iteratively train the preset model to be trained based on the training data set until the model to be trained meets the preset convergence conditions, thereby obtaining a target recommendation model, wherein the model to be trained includes a cross-feature extraction module, a vectorization module, and an attention estimation module, and the attention estimation module is used to determine the predicted click parameters associated with each recommended content when multiple recommended contents are displayed together.

[0068] In this embodiment, after the training data set is obtained, the preset model to be trained can be iteratively trained based on the training data set until the model to be trained meets the preset convergence condition, thereby obtaining a target recommendation model.

[0069] In which, during each round of training, the loss value corresponding to the model to be trained can be calculated, and whether the model to be trained meets the preset convergence condition can be determined based on the loss value. Alternatively, whether the model to be trained meets the preset convergence condition can be determined based on the training duration or the number of training rounds, etc. The present disclosure does not impose any restrictions on this.

[0070] Optionally, the model to be trained includes a cross-feature extraction module, a vectorization module, and an attention estimation module. The cross-feature extraction module is used to extract and cross-link hidden features associated with content features and historical features to obtain cross-features. The vectorization module is used to vectorize cross-features to obtain cross-feature vectors. The attention estimation module is used to predict the predicted click parameters associated with each recommended content when multiple recommended contents are displayed together based on the cross-feature vector.

[0071] The trained target recommendation model can predict the predicted click parameters associated with each recommended content when multiple recommended contents co-occur in the recommended data after obtaining the recommended data to be recommended. The optimal recommended content sequence can be determined based on the predicted click parameters and the preset optimization algorithm and value algorithm to obtain the target recommendation queue.

[0072] The recommended content processing method provided in this embodiment performs model training operations based on historical features associated with users, content features associated with recommended content, and click parameters of users on multiple recommended content of multiple genres in the recommendation queue during the model training process. Thus, when content recommendations are subsequently made based on the trained target recommendation model, the user's browsing operations on each recommended content when multiple recommended content of different genres co-occur can be considered, thereby improving the accuracy and comprehensiveness of content recommendations in the special form of ranking distribution positions. In addition, by introducing optimization algorithms and value algorithms, the optimal and diverse target recommendation queue can be more accurately determined in combination with historical features and recommended content features, thereby improving the accuracy of content recommendations.

[0073] Further, based on any of the above embodiments, step 202 includes:

[0074] The training data is input into the model to be trained, and a prediction result output by the model to be trained is obtained, wherein the prediction result includes a predicted click parameter associated with each recommended content.

[0075] The prediction result and the labeled data associated with the training data are calculated using a preset loss function to determine the loss value corresponding to the model to be trained.

[0076] Determine whether the model to be trained satisfies the convergence condition based on the loss value.

[0077] If it is determined that the model to be trained meets the convergence condition, the current model to be trained is determined as the target recommendation model.

[0078] In this embodiment, after obtaining the training data set, the training data can be input into the model to be trained to obtain the prediction result output by the model to be trained, and the prediction result includes the predicted click parameter associated with each recommended content. For example, the training data includes eight book media contents, and the prediction result may include the predicted click parameter associated with each book media content.

[0079] Furthermore, after obtaining the prediction result output by the model to be trained, the prediction result and the label data associated with the training data can be calculated by a preset loss function to determine the loss value corresponding to the model to be trained. Any loss function can be used to calculate the loss value corresponding to the model to be trained, and the present disclosure does not limit this.

[0080] After obtaining the loss value corresponding to the model to be trained, it can be determined whether the model to be trained meets the convergence condition based on the loss value. Alternatively, it can be determined whether the model to be trained meets the convergence condition based on the current number of training rounds and training duration.

[0081] Furthermore, if it is determined that the model to be trained meets the convergence condition, the current model to be trained is determined as the target recommendation model. Otherwise, the parameters of the model to be trained can be adjusted based on the loss value, and the step of inputting the training data into the model to be trained and obtaining the prediction result output by the model to be trained is returned until the model to be trained meets the convergence condition and the target recommendation model is obtained.

[0082] Further, based on any of the above embodiments, determining whether the model to be trained meets the convergence condition based on the loss value includes:

[0083] Determine whether the loss value is less than a preset loss value threshold.

[0084] or,

[0085] Determine whether the difference in loss between two rounds of iterative training is less than a preset difference threshold.

[0086] In this embodiment, after obtaining the loss value corresponding to the model to be trained, it can be determined based on the loss value whether the model to be trained meets the convergence condition.

[0087] Optionally, a loss value threshold may be preset. It is determined whether the loss value is less than a preset loss value threshold. If so, it can be determined that the model to be trained meets the preset convergence condition. Alternatively, a difference threshold may be preset to determine whether the difference in loss between two rounds of iterative training is less than a preset difference threshold. If so, it indicates that the prediction result of the model to be trained is relatively stable, and it can be determined that the model to be trained meets the preset convergence condition.

[0088] Alternatively, it may be determined whether the model to be trained meets the convergence condition based on the current number of training rounds and the training duration. For example, when the number of training rounds reaches a preset round number threshold, the model to be trained is determined to be converged. Alternatively, when the training duration reaches a preset duration, the model to be trained is determined to be converged. The present disclosure does not limit this.

[0089] The recommended content processing method provided in this embodiment determines the loss value of the model to be trained based on the prediction results and the labeled data, so that the parameters of the model to be trained can be adjusted based on the loss value to improve the prediction accuracy of the model. In addition, it can be accurately determined whether the model to be trained has converged based on the loss value and the convergence condition, and then the training operation of the model to be trained can be accurately implemented based on the convergence condition, thereby improving the efficiency and accuracy of model training.

[0090] Further, based on any of the above embodiments, obtaining the prediction result output by the model to be trained includes:

[0091] The first hidden layer features associated with the historical features and the second hidden layer features associated with the content features are extracted by a cross-feature extraction module corresponding to the model to be trained, and the cross-features associated with the training data are determined based on the first hidden layer features and the second hidden layer features.

[0092] The cross-features are vectorized by the vectorization module to obtain a cross-feature vector.

[0093] The cross feature vector is processed based on the bidirectional conversion network associated with the attention estimation module to obtain the prediction result.

[0094] In this embodiment, the model to be trained includes a cross-feature extraction module, a vectorization module, and an attention estimation module. The cross-feature extraction module is used to extract and cross the hidden features associated with the content features and historical features to obtain cross-features. The vectorization module is used to vectorize the cross-features to obtain a cross-feature vector. The attention estimation module is used to predict the predicted click parameters associated with each recommended content when multiple recommended contents are displayed together based on the cross-feature vector.

[0095] Therefore, after the training data is input into the model to be trained, the first hidden layer feature associated with the historical feature and the second hidden layer feature associated with the content feature corresponding to each recommended content can be extracted through the cross-feature extraction module corresponding to the model to be trained, and the cross-feature associated with the training data is determined based on the first hidden layer feature and the second hidden layer feature. The cross-feature is vectorized by the vectorization module to obtain a cross-feature vector. The bidirectional transformer network (Bi-directional Transformer) associated with the attention estimation module performs data processing on the cross-feature vector to obtain a prediction result.

[0096] Figure 3 The model architecture diagram of the target recommendation model provided in the embodiment of the present disclosure is as follows: Figure 3 As shown, the target recommendation model 31 may include a cross-feature extraction module 32, a vectorization module 33 and an attention estimation module 34. The cross-feature extraction module 32 includes a first submodule 35 for extracting hidden features on the user side, a second submodule 36 for extracting hidden features on the recommended content side and a feature cross submodule 37. After the cross-feature extraction module 32 performs a cross-feature extraction operation, the cross-features may be vectorized by the vectorization module 33 to obtain a cross-feature vector. The cross-feature vector is processed based on the bidirectional conversion network associated with the attention estimation module 34 to obtain a prediction result.

[0097] The recommended content processing method provided in this embodiment extracts the latent features associated with the historical features and the latent features associated with the content features respectively, and determines the cross-features associated with the training data based on the first latent features and the second latent features, so as to comprehensively consider both the content features and the historical features, perform attention prediction based on the cross-features, and improve the accuracy of the prediction results.

[0098] Figure 4 Another system architecture diagram based on the embodiments of the present disclosure is shown in FIG. Figure 4 As shown, the system architecture on which the present disclosure is based includes at least a terminal device 41 and a server 42. Among them, a target application is set in the terminal device 41, and the user can browse the media content in the target application and trigger the recommendation operation. A target recommendation model can be set in the server 42. Therefore, after obtaining the recommendation operation sent by the terminal device, the target recommendation queue can be determined based on the recommendation data and the target historical features associated with the target user. And the target recommendation queue is fed back to the terminal device for display.

[0099] Figure 5 A flowchart of a method for processing recommended content provided by an embodiment of the present disclosure is shown in FIG. Figure 5 As shown, the method includes:

[0100] Step 501: In response to a recommendation operation triggered by a target user, obtain recommendation data fed back by a preset screening model, wherein the recommendation data includes a second preset number of recommended contents, a prediction score associated with each recommended content, recommended content features, and a sorting parameter used by the recommendation model in the recommendation process.

[0101] The execution subject of this embodiment is a recommended content processing device. The recommended content processing device can be coupled to a server. The server can be connected to a terminal device in communication, so as to obtain a recommendation request sent by the terminal device, obtain recommendation data based on the recommendation request, perform data processing on the recommendation data and the target historical features associated with the target user through a target recommendation model and a preset optimization algorithm, and determine a target recommendation queue in the recommendation data.

[0102] In this embodiment, the user can browse media content in the target application preset in the terminal device, wherein the media content includes but is not limited to book media content, audio media content, graphic media content, etc. In order to enable the user to browse richer media content, a target recommendation list can also be displayed in the recommendation page.

[0103] Optionally, in order to realize the generation of the target recommendation list, in response to the recommendation operation triggered by the target user, the recommendation data fed back by the preset screening model is obtained. Among them, the recommendation data can be obtained after data recall, rough sorting, fine sorting and mixed sorting according to the user's associated information. The recommendation data includes a second preset number of recommended contents. For example, 20 recommended contents can be obtained after rough sorting, fine sorting and mixed sorting of the recalled media content, so the recommendation data can be formed according to the 20 recommended contents. In addition, the recommendation data also includes the predicted scores associated with each recommended content, the recommended content features, and the sorting parameters used by the recommendation model in the recommendation process. The predicted score can be the estimated score of the fine sorting of the X associated with each recommended content. The sorting parameters can be the sorting formula name, formula score and formula order corresponding to the genre queue in the mixed sorting process. The recommended content features can include the identification information, publishing information, classification information, category label information, etc. of the recommended content associated with each recommended content.

[0104] Step 502: perform data processing on the recommendation data and the target historical features associated with the target user based on a preset target recommendation model and a preset optimization algorithm, determine a target recommendation queue in the recommendation data, the target recommendation queue includes a first preset number of multi-genre recommended contents, wherein the first preset number is less than the second preset number, the target recommendation model is obtained by training based on the recommendation content processing method described in any of the above embodiments, and the target historical features are obtained after user authorization.

[0105] In this embodiment, after the recommendation data is acquired, the recommendation data and the target historical features associated with the target user may be processed based on a preset target recommendation model and a preset optimization algorithm to determine a target recommendation queue in the recommendation data.

[0106] The recommendation data may be input into a target recommendation model, which may predict the click parameters of the user on each recommended content when multiple recommended contents co-occur in the recommendation data. The click parameter may be generated when the user clicks on the recommended content again within X days in the current recommendation operation. For example, the three-retention click parameter may be generated when the user clicks on the recommended content again within 3 days after clicking on the recommended content.

[0107] Since the recommended data includes the second preset number of recommended contents, the first preset number of recommended contents can be further determined from the second preset number of recommended contents to form a target recommendation queue. Therefore, after the target recommendation model extracts hidden features from the recommended data and the target historical features, the hidden feature vector can be input into the optimization algorithm, and the optimization algorithm can further determine the first preset number of recommended contents from the second preset number of recommended contents based on the hidden feature vector.

[0108] Furthermore, the first preset number of recommended contents screened by the optimization algorithm can be input into the target recommended contents, and the target recommendation model can predict the click parameters of the user on each recommended content when the first preset number of recommended contents co-occur. Then, the recommended content sequence can be accurately determined based on the prediction results to obtain the target recommendation queue.

[0109] Among them, during the training process of the target recommendation model, the user's click parameters for different recommended contents can be comprehensively considered, so that the target recommendation model can more accurately predict the user's click parameters for each recommended content when multiple recommended contents co-occur.

[0110] The target history features include, but are not limited to, the target user's consumption history of the recommended content, the target user's historical activity, etc. The target history features are obtained after obtaining full authorization from the user.

[0111] Step 503: Display the target recommendation queue in a display interface.

[0112] In this embodiment, after the target recommendation queue is predicted, the target recommendation queue can be displayed in the display interface, so that the user can view the recommended content in the target recommendation queue.

[0113] The recommended content processing method provided in this embodiment predicts the target recommendation queue based on a pre-trained target recommendation model and an optimization algorithm, so that a better quality recommended content sequence can be obtained through sequence optimization operations based on comprehensive consideration of the user's click operations on different recommended contents, thereby improving the accuracy of content recommendations.

[0114] Optionally, based on any of the above embodiments, step 501 includes:

[0115] In response to the target user's triggering operation on the recommendation page, a plurality of recall contents associated with the target user are acquired from a preset content database through a preset recall model.

[0116] A preliminary screening operation is performed on the plurality of recalled contents based on the first associated data of the target user through a preset first screening model to obtain a first screening result.

[0117] The first screening result is screened again based on the second associated data of the target user through a preset second screening model to obtain a second screening result, wherein the data volume of the second associated data is greater than the data volume of the first associated data, and the screening accuracy of the second screening model is greater than the screening accuracy of the first screening model.

[0118] The second screening result is screened according to a preset mixed screening condition to obtain the recommended data.

[0119] In this embodiment, the recommended data can be obtained by data recall, rough sorting, fine sorting and mixed sorting according to the user's related information. Different data screening stages use different amounts of user data and different model processing accuracies.

[0120] Optionally, the user can trigger the recommendation page in the target application. In response to the target user's triggering operation on the recommendation page, a plurality of recall contents associated with the target user is obtained from a preset content database through a preset recall model. A preliminary screening operation is performed on the plurality of recall contents based on the first associated data of the target user through a preset first screening model to obtain a first screening result. A second screening operation is performed on the first screening result based on the second associated data of the target user through a preset second screening model to obtain a second screening result, wherein the amount of data of the second associated data is greater than the amount of data of the first associated data, and the screening accuracy of the second screening model is greater than the screening accuracy of the first screening model. The second screening result is screened according to a preset mixed screening condition to obtain recommended data. The mixed screening condition can be screening according to the category of the recommended content and the degree of association with the user.

[0121] By screening the recalled contents layer by layer, a second preset number of recommended contents can be finally obtained, for example, the second preset number can be 20. By determining the target recommendation queue based on the second preset number of recommended contents, the possibility of recommending recommended contents with low ranking scores can be reduced, and the content quality of the recommended contents can be improved. And the accuracy of content recommendation can be improved.

[0122] The recommended content processing method provided in this embodiment determines the multiple recommended contents obtained after data recall, rough sorting, fine sorting and mixed sorting as recommended data, thereby reducing the calculation amount of the target recommendation model and avoiding recommending media content with low fine sorting scores.

[0123] Figure 6 A flowchart of a method for processing recommended content provided by another embodiment of the present disclosure is provided. Based on any of the above embodiments, Figure 6 As shown, step 502 includes:

[0124] Step 601: Input the recommendation data and the target historical features into the target recommendation model to obtain a cross-recommendation feature vector output by the target recommendation model.

[0125] Step 602: Perform a preset number of sequence optimization operations on the cross-recommendation feature vectors using a preset optimization algorithm, determine a first preset number of recommended contents from the second preset number of recommended contents, and obtain screening results associated with each round of sequence optimization operations.

[0126] Step 603: Input each screening result into the attention estimation module in the target recommendation model to obtain an estimation result corresponding to each screening result, wherein the estimation result includes the predicted click parameters of the target user for each recommended content when a first preset number of recommended contents co-occur in the screening results.

[0127] Step 604: Determine the value information corresponding to each estimation result.

[0128] Step 605: Determine the target recommendation queue based on the value information corresponding to each estimation result.

[0129] In this embodiment, the recommendation data and the target historical features may be input into the target recommendation model, and the target recommendation model may be used to extract and cross-reference the hidden features of the recommendation data and the target historical features to obtain a cross-reference feature vector output by the target recommendation model.

[0130] Furthermore, the cross-recommendation feature vector can be output to an optimization algorithm. The optimization algorithm can perform a preset number of sequence optimization operations based on the cross-recommendation feature vector, determine the first preset number of recommended contents from the second preset number of recommended contents, and obtain the screening results associated with each round of sequence optimization operations. The optimization algorithm can be a beam search algorithm. It can perform a sequence optimization operation and determine the first preset number of recommended contents from the second preset number of recommended contents as the optimal sequence.

[0131] Each screening result is input into the attention estimation module in the target recommendation model, and the attention estimation module can estimate the click parameters of the user on each recommended content when the first preset number of recommended contents in the screening result co-occur, and obtain the estimated result. The estimated result is processed based on the preset value algorithm to obtain the value information.

[0132] Among them, the value algorithm can be shown in formula 1:

[0133] Value(rep)=CTR×Re1×…×Re7

[0134] Among them, CTR is the probability that the user clicks on at least one recommended content in this recommendation operation, and ReX represents the probability that at least one recommended content X is retained.

[0135] Furthermore, the target recommendation queue may be determined based on value information corresponding to each estimation result.

[0136] Further, based on any of the above embodiments, step 605 includes:

[0137] The target value information satisfying the preset conditions is determined from the value information associated with the preset number of sequence optimization operations.

[0138] A first preset number of recommended contents in the screening results associated with the target value information are determined as the target recommendation queue.

[0139] In this embodiment, the optimization algorithm can determine a set of screening results in the recommendation data based on the cross feature vectors each time. After a preset number of sequence optimization operations, multiple sets of screening results can be obtained.

[0140] The value information associated with each group of screening results may be calculated separately. The target value information that meets the preset conditions may be determined from the value information associated with the preset number of sequence optimization operations. For example, the value information with the highest value may be determined as the target value information. The first preset number of recommended contents in the screening results associated with the target value information may be determined as the target recommendation queue.

[0141] Further, based on any of the above embodiments, step 601 includes:

[0142] The hidden features associated with the target historical features and the hidden features associated with the recommended data are extracted through a cross feature extraction module corresponding to the target recommendation model.

[0143] The cross recommendation feature is determined based on the hidden layer feature associated with the target historical feature and the hidden layer feature associated with the recommendation data.

[0144] The cross-recommendation feature is vectorized by a vectorization module corresponding to the target recommendation model to obtain the cross-recommendation feature vector.

[0145] In this embodiment, the target recommendation model includes a cross-feature extraction module, a vectorization module, and an attention estimation module. The cross-feature extraction module is used to extract and cross the hidden features associated with the recommendation data and the target historical features to obtain cross-features. The vectorization module is used to vectorize the cross-features to obtain a cross-feature vector. The attention estimation module is used to predict the predicted click parameters associated with each recommended content when multiple recommended contents are displayed together based on the cross-feature vector.

[0146] Therefore, after the recommendation data is input into the target recommendation model, the hidden features associated with the target historical features and the hidden features associated with the recommendation data can be extracted by the cross-feature extraction module corresponding to the target recommendation model. The cross-recommendation features are determined based on the hidden features associated with the target historical features and the hidden features associated with the recommendation data. The cross-recommendation features are vectorized by the vectorization module corresponding to the target recommendation model to obtain a cross-recommendation feature vector.

[0147] Figure 7 The model architecture diagram of the target recommendation model provided in the embodiment of the present disclosure is as follows: Figure 7As shown, the target recommendation model 71 may include a cross-feature extraction module 72, a vectorization module 73 and an attention estimation module 74. Among them, the cross-feature extraction module 72 includes a first submodule 75 for extracting hidden layer features on the user side, a second submodule 76 for extracting hidden layer features on the recommended content side, and a feature cross submodule 77. After the cross-feature extraction module 72 is obtained to perform a cross-feature extraction operation, the cross-feature can be vectorized by the vectorization module 73 to obtain a cross-feature vector. The cross-feature vector can be input into a preset optimization algorithm 78 to obtain the screening result obtained by the optimization algorithm after the sequence optimization operation. The screening result is input into the attention estimation module 74, so the attention estimation module 74 can perform data processing on the screening result based on the preset bidirectional conversion network to obtain an estimated result. The estimated result is input into the preset value algorithm 79, so the value of the current estimated result can be calculated. In order to achieve the selection of the optimal sequence.

[0148] The recommended content processing method provided in this embodiment can obtain multiple rounds of screening results through multiple rounds of sequence optimization operations by setting an optimization algorithm and a value algorithm respectively, and considers the value information of the recommended contents associated with the multiple rounds of screening results when they co-occur based on the value algorithm, and performs screening of the optimal sequence based on the value information to obtain the target recommendation queue, thereby improving the accuracy of content recommendation.

[0149] Figure 8 A structural diagram of a recommended content processing device provided in an embodiment of the present disclosure, such as Figure 8 As shown, the device includes: an acquisition module 81 and a training module 82. The acquisition module 81 is used to acquire a preset training data set, the training data set includes multiple groups of training data, the training data includes feature data and annotation data, the feature data includes historical features associated with the user, content features corresponding to each recommended content in the recommendation queue, the annotation data is the user's click parameter for each recommended content, the content features include the predicted score and sorting parameters of the recommendation queue during the screening process and the associated parameters of each recommended content, the recommendation queue includes a first preset number of multi-genre recommended contents, the historical features and the click parameters are obtained after the user is authorized. The training module 82 is used to iteratively train the preset model to be trained based on the training data set until the model to be trained meets the preset convergence conditions to obtain the target recommendation model, wherein the model to be trained includes a cross-feature extraction module, a vectorization module and an attention estimation module, and the attention estimation module is used to determine the predicted click parameters associated with each recommended content when multiple recommended contents are displayed together.

[0150] Further, based on any of the above embodiments, the training module is used to: input the training data into the model to be trained, obtain the prediction results output by the model to be trained, and the prediction results include the predicted click parameters associated with each recommended content. The prediction results and the labeled data associated with the training data are calculated using a preset loss function to determine the loss value corresponding to the model to be trained. Based on the loss value, determine whether the model to be trained meets the convergence condition. If it is determined that the model to be trained meets the convergence condition, the current model to be trained is determined as the target recommendation model.

[0151] Further, based on any of the above embodiments, the training module is used to: extract the first hidden layer feature associated with the historical feature and the second hidden layer feature associated with the content feature through the cross-feature extraction module corresponding to the model to be trained, and determine the cross-feature associated with the training data based on the first hidden layer feature and the second hidden layer feature. Perform vectorization operation on the cross-feature through the vectorization module to obtain a cross-feature vector. Perform data processing on the cross-feature vector based on the bidirectional conversion network associated with the attention estimation module to obtain the prediction result.

[0152] Further, based on any of the above embodiments, the training module is used to: determine whether the loss value is less than a preset loss value threshold, or determine whether the difference in loss between two rounds of iterative training is less than a preset difference threshold.

[0153] Fig. 9 A structural diagram of a recommended content processing device provided in an embodiment of the present disclosure, such as Fig. 9 As shown, the device includes: a data acquisition module 91, a processing module 92 and a display module 93. The data acquisition module 91 is used to respond to the recommendation operation triggered by the target user, and obtain the recommendation data fed back by the preset screening model. The recommendation data includes a second preset number of recommended contents, the prediction score associated with each recommended content, the recommended content features, and the sorting parameters used by the recommendation model in the recommendation process. The processing module 92 is used to perform data processing on the recommendation data and the target historical features associated with the target user based on a preset target recommendation model and a preset optimization algorithm, and determine a target recommendation queue in the recommendation data. The target recommendation queue includes a first preset number of recommended contents, wherein the first preset number is less than the second preset number, and the target recommendation model is obtained by training based on the recommended content processing device described in any of the above embodiments, and the target historical features are obtained after user authorization. The display module 93 is used to display the target recommendation queue in the display interface.

[0154] Further, on the basis of any of the above embodiments, the processing module is used to: input the recommendation data and the target historical features into the target recommendation model to obtain a cross-recommendation feature vector output by the target recommendation model. Perform a preset number of sequence optimization operations on the cross-recommendation feature vector through a preset optimization algorithm, determine a first preset number of recommended contents from the second preset number of recommended contents, and obtain screening results associated with each round of sequence optimization operations. Input each screening result into the attention estimation module in the target recommendation model to obtain an estimated result corresponding to each screening result, wherein the estimated result includes the predicted click parameters of the target user for each recommended content when the first preset number of recommended contents co-occur in the screening result. Determine the value information corresponding to each estimated result. Determine the target recommendation queue based on the value information corresponding to each estimated result.

[0155] Further, based on any of the above embodiments, the processing module is used to: determine target value information that meets preset conditions in the value information associated with the preset number of sequence optimization operations, and determine the first preset number of recommended contents in the screening results associated with the target value information as the target recommendation queue.

[0156] Further, based on any of the above embodiments, the processing module is used to: extract the hidden features associated with the target historical features and the hidden features associated with the recommended data through the cross-feature extraction module corresponding to the target recommendation model. Determine the cross-recommendation feature based on the hidden features associated with the target historical features and the hidden features associated with the recommended data. Vectorize the cross-recommendation feature through the vectorization module corresponding to the target recommendation model to obtain the cross-recommendation feature vector.

[0157] Further, based on any of the above embodiments, the data acquisition module is used to: in response to the trigger operation of the target user on the recommendation page, obtain multiple recall contents associated with the target user from a preset content database through a preset recall model. Perform a preliminary screening operation on multiple recall contents based on the first associated data of the target user through a preset first screening model to obtain a first screening result. Perform a second screening operation on the first screening result based on the second associated data of the target user through a preset second screening model to obtain a second screening result, wherein the data volume of the second associated data is greater than the data volume of the first associated data, and the screening accuracy of the second screening model is greater than the screening accuracy of the first screening model. Perform a screening operation on the second screening result according to the preset mixed screening condition to obtain the recommended data.

[0158] The device provided in this embodiment can be used to execute the technical solution of the above method embodiment. Its implementation principle and technical effect are similar, and this embodiment will not be repeated here.

[0159] In order to implement the above embodiments, the embodiments of the present disclosure further provide a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the recommended content processing method as described in any of the above embodiments is implemented.

[0160] In order to implement the above embodiments, the embodiments of the present disclosure further provide a computer program product, including a computer program, and when the computer program is executed by a processor, the method for processing recommended content as described in any of the above embodiments is implemented.

[0161] In order to implement the above embodiment, the present disclosure also provides an electronic device, including: a processor and a memory;

[0162] The memory stores computer-executable instructions;

[0163] The processor executes the computer-executable instructions stored in the memory, so that the processor executes the recommended content processing method as described in any of the above embodiments.

[0164] Fig.10 The electronic device 1000 is a schematic diagram of the structure of an electronic device provided in an embodiment of the present disclosure, and the electronic device 1000 may be a terminal device or a server. The terminal device may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, personal digital assistants (PDAs), tablet computers, portable multimedia players (PMPs), vehicle terminals (such as vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Fig.10 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0165] like Fig.10As shown, the electronic device 1000 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 1001, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1008 to a random access memory (RAM) 1003. Various programs and data required for the operation of the electronic device 1000 are also stored in the RAM 1003. The processing device 1001, the ROM 1002, and the RAM 1003 are connected to each other via a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0166] Typically, the following devices may be connected to the I / O interface 1005: input devices 1006 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 1007 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1008 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 may allow the electronic device 1000 to communicate with other devices wirelessly or by wire to exchange data. Although Fig.10 The electronic device 1000 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead.

[0167] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device 1009, or installed from a storage device 1008, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment of the present disclosure are executed.

[0168] It should be noted that the computer-readable medium disclosed above may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0169] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0170] The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device executes the method shown in the above embodiment.

[0171] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0172] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0173] The units involved in the embodiments described in the present disclosure may be implemented by software or hardware. The name of a unit does not limit the unit itself in some cases. For example, the first acquisition unit may also be described as a "unit for acquiring at least two Internet Protocol addresses".

[0174] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0175] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0176] In a first aspect, according to one or more embodiments of the present disclosure, a method for processing recommended content is provided, including:

[0177] A preset training data set is obtained, wherein the training data set includes multiple sets of training data, wherein the training data includes feature data and annotation data, wherein the feature data includes historical features associated with the user, content features corresponding to each recommended content in the recommendation queue, wherein the annotation data is a click parameter of the user for each recommended content, wherein the content features include a predicted score and a sorting parameter of the recommendation queue during the screening process, and an associated parameter of each recommended content, wherein the recommendation queue includes a first preset number of recommended contents of multiple genres, wherein the historical features and the click parameters are obtained after authorization by the user.

[0178] The preset model to be trained is iteratively trained based on the training data set until the model to be trained meets the preset convergence conditions, thereby obtaining a target recommendation model, wherein the model to be trained includes a cross-feature extraction module, a vectorization module and an attention estimation module, and the attention estimation module is used to determine the predicted click parameters associated with each recommended content when multiple recommended contents are displayed together.

[0179] According to one or more embodiments of the present disclosure, displaying the first default prompt information corresponding to the input box in a target area outside the input box includes: adjusting at least one size attribute of the first default prompt information so that the adjusted first default prompt information is suitable for display in the target area.

[0180] According to one or more embodiments of the present disclosure, the iterative training of a preset model to be trained based on the training data set until the model to be trained meets a preset convergence condition to obtain a target recommendation model includes:

[0181] Inputting the training data into the model to be trained, obtaining a prediction result output by the model to be trained, wherein the prediction result includes a predicted click parameter associated with each recommended content;

[0182] Calculating the prediction result and the labeled data associated with the training data using a preset loss function to determine a loss value corresponding to the model to be trained;

[0183] Determine whether the model to be trained meets the convergence condition based on the loss value;

[0184] If it is determined that the model to be trained meets the convergence condition, the current model to be trained is determined as the target recommendation model.

[0185] According to one or more embodiments of the present disclosure, obtaining the prediction result output by the model to be trained includes:

[0186] Extracting the first hidden layer feature associated with the historical feature and the second hidden layer feature associated with the content feature through the cross-feature extraction module corresponding to the model to be trained, and determining the cross-feature associated with the training data based on the first hidden layer feature and the second hidden layer feature;

[0187] Performing a vectorization operation on the cross-features by the vectorization module to obtain a cross-feature vector;

[0188] The cross feature vector is processed based on the bidirectional conversion network associated with the attention estimation module to obtain the prediction result.

[0189] According to one or more embodiments of the present disclosure, determining whether the model to be trained satisfies the convergence condition based on the loss value includes:

[0190] Determining whether the loss value is less than a preset loss value threshold;

[0191] or,

[0192] Determine whether the difference in loss between two rounds of iterative training is less than a preset difference threshold.

[0193] In a second aspect, according to one or more embodiments of the present disclosure, a method for processing recommended content is provided, including:

[0194] In response to a recommendation operation triggered by a target user, obtaining recommendation data fed back by a preset screening model, wherein the recommendation data includes a second preset number of recommended contents, a prediction score associated with each recommended content, recommended content features, and a ranking parameter used by the recommendation model in the recommendation process;

[0195] Based on a preset target recommendation model and a preset optimization algorithm, data processing is performed on the recommendation data and the target historical features associated with the target user, and a target recommendation queue is determined in the recommendation data, wherein the target recommendation queue includes a first preset number of recommended contents, wherein the first preset number is less than the second preset number, the target recommendation model is obtained by training based on the recommended content processing method described in any of the above embodiments, and the target historical features are obtained after authorization by the user;

[0196] The target recommendation queue is displayed in a display interface.

[0197] According to one or more embodiments of the present disclosure, the step of performing data processing on the recommendation data and the target history features associated with the target user based on a preset target recommendation model and a preset optimization algorithm, and determining a target recommendation queue in the recommendation data, includes:

[0198] Inputting the recommendation data and the target historical features into the target recommendation model to obtain a cross-recommendation feature vector output by the target recommendation model;

[0199] Performing a preset number of sequence optimization operations on the cross-recommendation feature vectors by using a preset optimization algorithm, determining a first preset number of recommended contents from the second preset number of recommended contents, and obtaining screening results associated with each round of sequence optimization operations;

[0200] Input each screening result into the attention estimation module in the target recommendation model to obtain an estimation result corresponding to each screening result, wherein the estimation result includes a predicted click parameter of the target user for each recommended content when a first preset number of recommended contents co-occur in the screening result;

[0201] Determine the value information corresponding to each estimated result;

[0202] The target recommendation queue is determined based on the value information corresponding to each estimation result.

[0203] According to one or more embodiments of the present disclosure, determining the target recommendation queue based on the value information corresponding to each estimation result includes:

[0204] Determining target value information that meets preset conditions from value information associated with a preset number of sequence optimization operations;

[0205] A first preset number of recommended contents in the screening results associated with the target value information are determined as the target recommendation queue.

[0206] According to one or more embodiments of the present disclosure, the step of inputting the recommendation data and the target historical features into the target recommendation model to obtain a cross-recommendation feature vector output by the target recommendation model includes:

[0207] Extracting hidden features associated with the target historical features and hidden features associated with the recommendation data through a cross feature extraction module corresponding to the target recommendation model;

[0208] Determining the cross-recommendation feature based on the hidden layer feature associated with the target historical feature and the hidden layer feature associated with the recommendation data;

[0209] The cross-recommendation feature is vectorized by a vectorization module corresponding to the target recommendation model to obtain the cross-recommendation feature vector.

[0210] According to one or more embodiments of the present disclosure, in response to a recommendation operation triggered by a target user, obtaining recommendation data fed back by a preset screening model includes:

[0211] In response to a triggering operation of the target user on the recommendation page, a plurality of recall contents associated with the target user are acquired from a preset content database through a preset recall model;

[0212] Performing a preliminary screening operation on the plurality of recalled contents based on the first associated data of the target user by using a preset first screening model to obtain a first screening result;

[0213] Performing a second screening operation on the first screening result based on the second associated data of the target user by using a preset second screening model to obtain a second screening result, wherein the amount of data of the second associated data is greater than the amount of data of the first associated data, and the screening accuracy of the second screening model is greater than the screening accuracy of the first screening model;

[0214] The second screening result is screened according to a preset mixed screening condition to obtain the recommended data.

[0215] In a third aspect, according to one or more embodiments of the present disclosure, there is provided a recommended content processing device, including:

[0216] an acquisition module, configured to acquire a preset training data set, wherein the training data set includes multiple sets of training data, wherein the training data includes feature data and annotation data, wherein the feature data includes historical features associated with the user and content features corresponding to each recommended content in the recommendation queue, wherein the annotation data is a click parameter of the user for each recommended content, wherein the content features include a predicted score and a sorting parameter of the recommendation queue during the screening process and an associated parameter of each recommended content, wherein the recommendation queue includes a first preset number of recommended contents of multiple genres, wherein the historical features and the click parameters are acquired after authorization by the user;

[0217] A training module is used to iteratively train a preset model to be trained based on the training data set until the model to be trained meets the preset convergence conditions to obtain a target recommendation model, wherein the model to be trained includes a cross-feature extraction module, a vectorization module and an attention estimation module, and the attention estimation module is used to determine the predicted click parameters associated with each recommended content when multiple recommended contents are displayed together.

[0218] According to one or more embodiments of the present disclosure, the training module is used to:

[0219] Inputting the training data into the model to be trained, obtaining a prediction result output by the model to be trained, wherein the prediction result includes a predicted click parameter associated with each recommended content;

[0220] Calculating the prediction result and the labeled data associated with the training data using a preset loss function to determine a loss value corresponding to the model to be trained;

[0221] Determine whether the model to be trained meets the convergence condition based on the loss value;

[0222] If it is determined that the model to be trained meets the convergence condition, the current model to be trained is determined as the target recommendation model.

[0223] According to one or more embodiments of the present disclosure, the training module is used to:

[0224] Extracting the first hidden layer feature associated with the historical feature and the second hidden layer feature associated with the content feature through the cross-feature extraction module corresponding to the model to be trained, and determining the cross-feature associated with the training data based on the first hidden layer feature and the second hidden layer feature;

[0225] Performing a vectorization operation on the cross-features by the vectorization module to obtain a cross-feature vector;

[0226] The cross feature vector is processed based on the bidirectional conversion network associated with the attention estimation module to obtain the prediction result.

[0227] According to one or more embodiments of the present disclosure, the training module is used to:

[0228] Determining whether the loss value is less than a preset loss value threshold;

[0229] or,

[0230] Determine whether the difference in loss between two rounds of iterative training is less than a preset difference threshold.

[0231] In a fourth aspect, according to one or more embodiments of the present disclosure, a recommended content processing device is provided, including:

[0232] a data acquisition module, configured to acquire, in response to a recommendation operation triggered by a target user, recommendation data fed back by a preset screening model, wherein the recommendation data includes a second preset number of recommended contents, a prediction score associated with each recommended content, recommended content features, and a ranking parameter used by the recommendation model in the recommendation process;

[0233] a processing module, configured to perform data processing on the recommendation data and the target historical features associated with the target user based on a preset target recommendation model and a preset optimization algorithm, determine a target recommendation queue in the recommendation data, the target recommendation queue includes a first preset number of recommended contents, wherein the first preset number is less than the second preset number, the target recommendation model is obtained by training based on the recommended content processing device described in any of the above embodiments, and the target historical features are obtained after authorization by the user;

[0234] A display module is used to display the target recommendation queue in a display interface.

[0235] According to one or more embodiments of the present disclosure, the processing module is used to:

[0236] Inputting the recommendation data and the target historical features into the target recommendation model to obtain a cross-recommendation feature vector output by the target recommendation model;

[0237] Performing a preset number of sequence optimization operations on the cross-recommendation feature vectors by using a preset optimization algorithm, determining a first preset number of recommended contents from the second preset number of recommended contents, and obtaining screening results associated with each round of sequence optimization operations;

[0238] Input each screening result into the attention estimation module in the target recommendation model to obtain an estimation result corresponding to each screening result, wherein the estimation result includes a predicted click parameter of the target user for each recommended content when a first preset number of recommended contents co-occur in the screening result;

[0239] Determine the value information corresponding to each estimated result;

[0240] The target recommendation queue is determined based on the value information corresponding to each estimation result.

[0241] According to one or more embodiments of the present disclosure, the processing module is used to:

[0242] Determining target value information that meets preset conditions from value information associated with a preset number of sequence optimization operations;

[0243] A first preset number of recommended contents in the screening results associated with the target value information are determined as the target recommendation queue.

[0244] According to one or more embodiments of the present disclosure, the processing module is used to:

[0245] Extracting hidden features associated with the target historical features and hidden features associated with the recommendation data through a cross feature extraction module corresponding to the target recommendation model;

[0246] Determining the cross-recommendation feature based on the hidden layer feature associated with the target historical feature and the hidden layer feature associated with the recommendation data;

[0247] The cross-recommendation feature is vectorized by a vectorization module corresponding to the target recommendation model to obtain the cross-recommendation feature vector.

[0248] According to one or more embodiments of the present disclosure, the data acquisition module is used to:

[0249] In response to a triggering operation of the target user on the recommendation page, a plurality of recall contents associated with the target user are acquired from a preset content database through a preset recall model;

[0250] Performing a preliminary screening operation on the plurality of recalled contents based on the first associated data of the target user by using a preset first screening model to obtain a first screening result;

[0251] Performing a second screening operation on the first screening result based on the second associated data of the target user by using a preset second screening model to obtain a second screening result, wherein the amount of data of the second associated data is greater than the amount of data of the first associated data, and the screening accuracy of the second screening model is greater than the screening accuracy of the first screening model;

[0252] The second screening result is screened according to a preset mixed screening condition to obtain the recommended data.

[0253] In a fifth aspect, according to one or more embodiments of the present disclosure, there is provided an electronic device, comprising: at least one processor and a memory;

[0254] The memory stores computer-executable instructions;

[0255] The at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor performs the recommended content processing method described in the first aspect and various possible designs of the first aspect or the second aspect and various possible designs of the second aspect.

[0256] In a sixth aspect, according to one or more embodiments of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer execution instructions. When a processor executes the computer execution instructions, the recommended content processing method described in the first aspect and various possible designs of the first aspect or the second aspect and various possible designs of the second aspect is implemented.

[0257] In the seventh aspect, according to one or more embodiments of the present disclosure, a computer program product is provided, including a computer program, which, when executed by a processor, implements the recommended content processing method described in the first aspect and various possible designs of the first aspect or the second aspect and various possible designs of the second aspect.

[0258] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present disclosure is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in the present disclosure (but not limited to) by each other to form a technical solution.

[0259] In addition, although each operation is described in a specific order, this should not be understood as requiring these operations to be performed in the specific order shown or in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Some features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination mode.

[0260] Although the subject matter has been described in language specific to structural features and / or methodological logical actions, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. On the contrary, the specific features and actions described above are merely example forms of implementing the claims.

Claims

1. A method for processing recommended content, characterized in that: include: Acquire a preset training data set, the training data set includes multiple sets of training data, the training data includes feature data and annotation data, the feature data includes historical features associated with the user, content features corresponding to each recommended content in the recommendation queue, the annotation data is a click parameter of the user for each recommended content, the content features include a predicted score and a ranking parameter of the recommendation queue during the screening process and an associated parameter of each recommended content, the recommendation queue includes a first preset number of recommended contents of multiple genres, the historical features and the click parameters are acquired after authorization by the user; The preset model to be trained is iteratively trained based on the training data set until the model to be trained meets the preset convergence conditions, thereby obtaining a target recommendation model, wherein the model to be trained includes a cross-feature extraction module, a vectorization module and an attention estimation module, and the attention estimation module is used to determine the predicted click parameters associated with each recommended content when multiple recommended contents are displayed together.

2. The method according to claim 1, characterized in that: The iterative training of the preset model to be trained based on the training data set until the model to be trained meets the preset convergence condition to obtain the target recommendation model includes: Inputting the training data into the model to be trained, obtaining a prediction result output by the model to be trained, wherein the prediction result includes a predicted click parameter associated with each recommended content; Calculating the prediction result and the labeled data associated with the training data using a preset loss function to determine a loss value corresponding to the model to be trained; Determine whether the model to be trained meets the convergence condition based on the loss value; If it is determined that the model to be trained meets the convergence condition, the current model to be trained is determined as the target recommendation model.

3. The method according to claim 2, characterized in that The obtaining of the prediction result output by the to-be-trained model comprises: Extracting the first hidden layer feature associated with the historical feature and the second hidden layer feature associated with the content feature through the cross-feature extraction module corresponding to the model to be trained, and determining the cross-feature associated with the training data based on the first hidden layer feature and the second hidden layer feature; Performing a vectorization operation on the cross-features by the vectorization module to obtain a cross-feature vector; The cross feature vector is processed based on the bidirectional conversion network associated with the attention estimation module to obtain the prediction result.

4. The method according to claim 3, characterized in that: The determining whether the model to be trained satisfies the convergence condition based on the loss value includes: Determining whether the loss value is less than a preset loss value threshold; or, Determine whether the difference in loss between two rounds of iterative training is less than a preset difference threshold.

5. A method for processing recommended content, characterized in that: include: In response to a recommendation operation triggered by a target user, obtaining recommendation data fed back by a preset screening model, wherein the recommendation data includes a second preset number of recommended contents, a prediction score associated with each recommended content, recommended content features, and a ranking parameter used by the recommendation model in the recommendation process; Based on a preset target recommendation model and a preset optimization algorithm, data processing is performed on the recommendation data and the target historical features associated with the target user, and a target recommendation queue is determined in the recommendation data, wherein the target recommendation queue includes a first preset number of multi-genre recommended contents, wherein the first preset number is less than the second preset number, the target recommendation model is obtained by training based on the recommended content processing method according to any one of claims 1 to 4, and the target historical features are obtained after authorization by the user; The target recommendation queue is displayed in a display interface.

6. The method according to claim 5, characterized in that The step of performing data processing on the recommendation data and the target historical features associated with the target user based on a preset target recommendation model and a preset optimization algorithm, and determining a target recommendation queue in the recommendation data, includes: Inputting the recommendation data and the target historical features into the target recommendation model to obtain a cross-recommendation feature vector output by the target recommendation model; Performing a preset number of sequence optimization operations on the cross-recommendation feature vectors by using a preset optimization algorithm, determining a first preset number of recommended contents from the second preset number of recommended contents, and obtaining screening results associated with each round of sequence optimization operations; Input each screening result into the attention estimation module in the target recommendation model to obtain an estimation result corresponding to each screening result, wherein the estimation result includes a predicted click parameter of the target user for each recommended content when a first preset number of recommended contents co-occur in the screening result; Determine the value information corresponding to each estimated result; The target recommendation queue is determined based on the value information corresponding to each estimation result.

7. The method according to claim 6, characterized in that The determining the target recommendation queue based on the value information corresponding to each estimation result includes: Determining target value information that meets preset conditions from value information associated with a preset number of sequence optimization operations; A first preset number of recommended contents in the screening results associated with the target value information are determined as the target recommendation queue.

8. The method according to claim 6, characterized in that The step of inputting the recommendation data and the target historical features into the target recommendation model to obtain a cross-recommendation feature vector output by the target recommendation model includes: Extracting hidden features associated with the target historical features and hidden features associated with the recommendation data through a cross feature extraction module corresponding to the target recommendation model; Determining the cross-recommendation feature based on the hidden layer feature associated with the target historical feature and the hidden layer feature associated with the recommendation data; The cross-recommendation feature is vectorized by a vectorization module corresponding to the target recommendation model to obtain the cross-recommendation feature vector.

9. The method according to any one of claims 5 to 8, characterized in that: The step of obtaining the recommendation data fed back by the preset screening model in response to the recommendation operation triggered by the target user includes: In response to a triggering operation of the target user on the recommendation page, a plurality of recall contents associated with the target user are acquired from a preset content database through a preset recall model; Performing a preliminary screening operation on the plurality of recalled contents based on the first associated data of the target user by using a preset first screening model to obtain a first screening result; Performing a second screening operation on the first screening result based on the second associated data of the target user by using a preset second screening model to obtain a second screening result, wherein the amount of data of the second associated data is greater than the amount of data of the first associated data, and the screening accuracy of the second screening model is greater than the screening accuracy of the first screening model; The second screening result is screened according to a preset mixed screening condition to obtain the recommended data.

10. A recommended content processing device, characterized in that: include: an acquisition module, configured to acquire a preset training data set, wherein the training data set includes multiple sets of training data, wherein the training data includes feature data and annotation data, wherein the feature data includes historical features associated with the user and content features corresponding to each recommended content in the recommendation queue, wherein the annotation data is a click parameter of the user for each recommended content, wherein the content features include a predicted score and a sorting parameter of the recommendation queue during the screening process and an associated parameter of each recommended content, wherein the recommendation queue includes a first preset number of recommended contents of multiple genres, wherein the historical features and the click parameters are acquired after authorization by the user; A training module is used to iteratively train a preset model to be trained based on the training data set until the model to be trained meets the preset convergence conditions to obtain a target recommendation model, wherein the model to be trained includes a cross-feature extraction module, a vectorization module and an attention estimation module, and the attention estimation module is used to determine the predicted click parameters associated with each recommended content when multiple recommended contents are displayed together.

11. A recommended content processing device, characterized in that: include: a data acquisition module, configured to acquire, in response to a recommendation operation triggered by a target user, recommendation data fed back by a preset screening model, wherein the recommendation data includes a second preset number of recommended contents, a prediction score associated with each recommended content, recommended content features, and a ranking parameter used by the recommendation model in the recommendation process; a processing module, configured to perform data processing on the recommendation data and the target historical features associated with the target user based on a preset target recommendation model and a preset optimization algorithm, determine a target recommendation queue in the recommendation data, the target recommendation queue includes a first preset number of recommended contents, wherein the first preset number is less than the second preset number, the target recommendation model is obtained by training based on the recommended content processing device according to claim 10, and the target historical features are obtained after authorization by the user; A display module is used to display the target recommendation queue in a display interface.

12. An electronic device, characterized in that: include: Processor and memory; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory, so that the processor executes the recommended content processing method according to any one of claims 1 to 4 or 5 to 9.

13. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions. When the processor executes the computer-executable instructions, the recommended content processing method according to any one of claims 1 to 4 or 5 to 9 is implemented.

14. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the recommended content processing method according to any one of claims 1 to 4 or 5 to 9 is implemented.

Images