Data pushing method and data sorting model processing method and device

Abstract

The invention relates to a data pushing method, a data sorting model processing method and device, computer equipment, a storage medium and a computer program product. The method comprises the steps of obtaining object information of a target object and data information corresponding to each piece of to-be-processed data in a data pool; based on the object information and the data information corresponding to each piece of to-be-processed data, obtaining initial feature data of each piece of to-be-processed data in each feature extraction dimension; for each piece of to-be-processed data, determining at least one weight distribution feature on each prediction target according to the weight influence data corresponding to the current prediction target, and determining a weight corresponding to each feature extraction dimension based on the at least one weight distribution feature, and determining push data corresponding to the target object based on the initial feature data of the current to-be-processed data in each feature extraction dimension and the weight corresponding to each feature extraction dimension. And the pushing precision is improved.

Description

Technical Field

[0001] The present application relates to the field of Internet technology, and in particular to a data push method, a data sorting model processing method, an apparatus, a computer device, a storage medium, and a computer program product. Background Art

[0002] With the development of Internet technology, Internet platforms have become diversified, such as social platforms, video platforms, news platforms, shopping platforms, etc. Internet platforms can push data of interest to objects.

[0003] In traditional technology, a multi-target prediction method is used to determine the push priority of the data to be processed. In the process of multi-target prediction, multiple prediction targets share an expert network. For each prediction target, the prediction result of the current prediction target is determined based on the same implicit representation input. The prediction result obtained in this way is not accurate, resulting in low push accuracy and a waste of computer resources. Summary of the invention

[0004] Based on this, it is necessary to provide a data push method, data sorting model processing method, device, computer equipment, computer-readable storage medium and computer program product that can improve the push accuracy in response to the above technical problems.

[0005] In a first aspect, the present application provides a data push method. The method comprises:

[0006] Obtain object information of the target object and data information corresponding to each piece of data to be processed in the data pool;

[0007] Based on the object information and the data information corresponding to each of the data to be processed, obtaining preliminary feature data of each of the data to be processed in each feature extraction dimension;

[0008] For each data to be processed, on each prediction target, at least one of the model input feature, the preliminary extracted feature and the object coding feature is used as the weight influence data corresponding to the current prediction target, the model input feature is obtained based on the object information and the data information corresponding to the current data to be processed, the preliminary extracted feature is obtained based on the preliminary feature data of the current data to be processed in each feature extraction dimension, the object coding feature is obtained based on the object information encoding, at least one weight distribution feature is determined according to the weight influence data corresponding to the current prediction target, based on the at least one weight distribution feature, the weight corresponding to each feature extraction dimension is determined, based on the preliminary feature data of the current data to be processed in each feature extraction dimension and the weight corresponding to each feature extraction dimension, the prediction result corresponding to the current prediction target is obtained, and based on the prediction results corresponding to each prediction target, the sorting reference information of the current data to be processed is determined;

[0009] Based on the sorting reference information corresponding to each piece of data to be processed in the data pool, the push data corresponding to the target object is determined.

[0010] On the other hand, the present application also provides a data push device. The device includes:

[0011] An acquisition module is used to acquire object information of a target object and data information corresponding to each piece of data to be processed in a data pool; and is also used to acquire preliminary feature data of each piece of data to be processed in each feature extraction dimension based on the object information and the data information corresponding to each piece of data to be processed;

[0012] A processing module, for, for each data to be processed and on each prediction target, using at least one of the model input feature, the preliminary extraction feature and the object coding feature as the weight influence data corresponding to the current prediction target, the model input feature is obtained based on the object information and the data information corresponding to the current data to be processed, the preliminary extraction feature is obtained based on the preliminary feature data of the current data to be processed in each feature extraction dimension, the object coding feature is obtained based on the object information encoding, determining at least one weight distribution feature according to the weight influence data corresponding to the current prediction target, determining the weights corresponding to each feature extraction dimension based on the at least one weight distribution feature, obtaining the prediction result corresponding to the current prediction target based on the preliminary feature data of the current data to be processed in each feature extraction dimension and the weights corresponding to each feature extraction dimension, and determining the sorting reference information of the current data to be processed based on the prediction results corresponding to each prediction target;

[0013] The determination module is used to determine the push data corresponding to the target object based on the sorting reference information corresponding to each piece of data to be processed in the data pool.

[0014] On the other hand, the present application further provides a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the method provided in the first aspect when executing the computer program.

[0015] On the other hand, the present application further provides a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the method provided in the first aspect is implemented.

[0016] On the other hand, the present application further provides a computer program product, including a computer program, which implements the method provided in the first aspect when executed by a processor.

[0017] The above-mentioned data push method, device, computer equipment, storage medium and computer program product first obtain the object information of the target object and the data information corresponding to each data to be processed in the data pool; based on the object information and the data information corresponding to each data to be processed, obtain the preliminary feature data of each data to be processed in each feature extraction dimension; then, for each data to be processed, at each prediction target, use at least one of the model input feature, the preliminary extracted feature and the object coding feature as the weight influence data corresponding to the current prediction target, the model input feature is obtained based on the object information and the data information corresponding to the current data to be processed, the preliminary extracted feature is obtained based on the preliminary feature data of the current data to be processed in each feature extraction dimension, the object coding feature is obtained based on the object information encoding, at least one weight distribution feature is determined according to the weight influence data corresponding to the current prediction target, based on the at least one weight distribution feature, the weight corresponding to each feature extraction dimension is determined, based on the current data to be processed in each The preliminary feature data of the feature extraction dimensions and the weights corresponding to each feature extraction dimension are used to obtain the prediction result corresponding to the current prediction target, and the sorting reference information of the current data to be processed is determined based on the prediction results corresponding to each prediction target; because on each prediction target, in the process of determining the weights corresponding to each feature extraction dimension, at least one of the influence of object information, data information itself on weight distribution, preliminary feature data itself on weight distribution, and object personalized information on weight distribution is combined, so that the finally determined weight is more accurate, the accuracy of the prediction result corresponding to each prediction target is improved, and the improvement of the accuracy of the prediction result can further promote the improvement of the accuracy of the sorting reference information; finally, based on the sorting reference information corresponding to each data to be processed in the data pool, the push data corresponding to the target object is determined, and the improvement of the accuracy of the sorting reference information makes it more likely that the target object is interested in the push data, thereby improving the accuracy of data push and saving computer resources.

[0018] In a second aspect, the present application provides a data sorting model processing method. The method comprises:

[0019] Acquire a training sample, where the training sample includes sample object information of the sample object and sample data information of the sample data;

[0020] Based on the sample object information and the sample data information, preliminary feature data of the sample data in each feature extraction dimension is obtained through the feature extraction network corresponding to each feature extraction dimension in the data sorting model to be trained;

[0021] At each prediction target, weighted feature extraction is performed on the sample influence data through the weight extraction network corresponding to the current prediction target in the data sorting model to obtain the weights corresponding to each feature extraction dimension. The target feature data corresponding to the current prediction target is determined based on the preliminary feature data of the sample data in each feature extraction dimension and the weights corresponding to each feature extraction dimension. Based on the target feature data, the current prediction target is predicted through the prediction network corresponding to the current prediction target in the data sorting model to obtain the prediction result corresponding to the current prediction target. Based on the prediction result corresponding to the current prediction target and the label corresponding to the current prediction target, the model loss corresponding to the current prediction target is determined.

[0022] According to the model losses corresponding to each prediction target, the parameters of the data sorting model to be trained are adjusted until the stopping conditions of the model training are met, and the trained data sorting model is obtained. The trained data sorting model is used to determine the push data corresponding to the target object.

[0023] On the other hand, the present application also provides a data sorting model processing device, the device comprising:

[0024] an acquisition module, used to acquire training samples, the training samples including sample object information of sample objects and sample data information of sample data; and further used to acquire preliminary feature data of sample data in each feature extraction dimension through a feature extraction network corresponding to each feature extraction dimension in a data sorting model to be trained based on the sample object information and the sample data information;

[0025] A prediction module is used to perform weighted feature extraction on sample influence data on each prediction target through a weight extraction network corresponding to the current prediction target in a data sorting model, obtain weights corresponding to each feature extraction dimension, determine target feature data corresponding to the current prediction target based on preliminary feature data of the sample data in each feature extraction dimension and the weights corresponding to each feature extraction dimension, predict the current prediction target based on the target feature data through a prediction network corresponding to the current prediction target in the data sorting model, obtain a prediction result corresponding to the current prediction target, and determine a model loss corresponding to the current prediction target based on the prediction result corresponding to the current prediction target and a label corresponding to the current prediction target;

[0026] The training module is used to adjust the parameters of the data sorting model to be trained according to the model losses corresponding to each prediction target, until the stopping conditions of the model training are met, and the trained data sorting model is obtained. The trained data sorting model is used to determine the push data corresponding to the target object.

[0027] On the other hand, the present application further provides a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the method provided in the second aspect when executing the computer program.

[0028] On the other hand, the present application further provides a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the method provided in the second aspect is implemented.

[0029] On the other hand, the present application further provides a computer program product, including a computer program, which implements the method provided in the second aspect when executed by a processor.

[0030] The data sorting model processing method, device, computer equipment, storage medium and computer program product first obtain training samples, and the training samples include sample object information of sample objects and sample data information of sample data; based on the sample object information and sample data information, the preliminary feature data of the sample data in each feature extraction dimension is obtained through the feature extraction network corresponding to each feature extraction dimension in the data sorting model to be trained; then, on each prediction target, the weight feature extraction is performed on the sample influence data through the weight extraction network corresponding to the current prediction target in the data sorting model to obtain the weight corresponding to each feature extraction dimension; according to the preliminary feature data of the sample data in each feature extraction dimension and the weight corresponding to each feature extraction dimension, the target feature data corresponding to the current prediction target is determined; based on the target feature data, the current prediction target is predicted through the prediction network corresponding to the current prediction target in the data sorting model to obtain the prediction result corresponding to the current prediction target; based on the prediction result corresponding to the current prediction target and the label corresponding to the current prediction target, the model loss corresponding to the current prediction target is determined; finally, according to the model loss corresponding to each prediction target, the parameters of the data sorting model to be trained are adjusted until the stop condition of the model training is met, and the training of the data sorting model is obtained. During the above model training process, the influence of object information, data information itself on weight distribution, the influence of preliminary feature data itself on weight distribution, and the influence of object personalized information on weight distribution were modeled, which improved the prediction accuracy of the data sorting model. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 An application environment diagram of a data push method in an embodiment;

[0032] Figure 2 A schematic diagram of a flow chart of a data push method in an embodiment;

[0033] Figure 3A schematic diagram of the structure of a data sorting model in one embodiment;

[0034] Figure 4 A schematic diagram of a data display entry in an embodiment;

[0035] Figure 5 A schematic diagram of a video display interface in one embodiment;

[0036] Figure 6 A schematic diagram of a flow chart of a data sorting model processing method in one embodiment;

[0037] Figure 7 A schematic diagram of a process for determining push data in one embodiment;

[0038] Figure 8 is a structural block diagram of a data push device in an embodiment;

[0039] Fig. 9 It is a structural block diagram of a data sorting model processing device in one embodiment;

[0040] Fig.10 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0042] The data push method provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other servers. The data push method provided in the embodiment of the present application can be executed by the terminal 102 or the server 104 alone, or it can be executed by the terminal 102 and the server 104 in collaboration, and the embodiment of the present application is not limited to this. During the execution process, first, the object information of the target object and the data information corresponding to each data to be processed in the data pool can be obtained; based on the object information and the data information corresponding to each data to be processed, the preliminary feature data of each data to be processed in each feature extraction dimension is obtained; then, for each data to be processed, at each prediction target, at least one of the model input feature, the preliminary extraction feature and the object coding feature is used as the weight influence data corresponding to the current prediction target, the model input feature is obtained based on the object information and the data information corresponding to the current data to be processed, the preliminary extraction feature is obtained based on the preliminary feature data of the current data to be processed in each feature extraction dimension, the object coding feature is obtained based on the object information encoding, at least one weight distribution feature is determined according to the weight influence data corresponding to the current prediction target, based on the at least one weight distribution feature, the weight corresponding to each feature extraction dimension is determined, based on the preliminary feature data of the current data to be processed in each feature extraction dimension and the weight corresponding to each feature extraction dimension, the prediction result corresponding to the current prediction target is obtained, based on the prediction result corresponding to each prediction target, the sorting reference information of the current data to be processed is determined; finally, based on the sorting reference information corresponding to each data to be processed in the data pool, the push data corresponding to the target object is determined.

[0043] The terminal 102 may be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, IoT devices, and portable wearable devices. The IoT devices may be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. The portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. The server 104 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services.

[0044] In some embodiments, Figure 2 As shown, a data push method is provided, which can be Figure 1 The method may be executed by a terminal or server alone, or by a system consisting of a terminal and a server. Taking the method executed by a system consisting of a terminal and a server as an example, the data push method provided in the embodiment of the present application includes the following steps:

[0045] Step 202: Obtain object information of the target object and data information corresponding to each piece of data to be processed in the data pool.

[0046] Among them, the data in the data push method provided in the embodiment of the present application can be any type of data. Exemplarily, the data in the data push method provided in the embodiment of the present application can be video, news, message, etc., which is not limited in the embodiment of the present application.

[0047] Among them, different applications (Application, referred to as APP) can be installed on the terminal, and the APP corresponding to the data push method provided in the embodiment of the present application can be called the target application. The target application can be a social APP, a video APP, or a news APP, etc., which is not limited in the embodiment of the present application.

[0048] The target object may be an object identifier that currently needs to push data, and the object identifier that currently needs to push data may be an object identifier that is in an online state.

[0049] In a specific implementation, after detecting that the object has performed a login operation, the terminal can generate a login request and send the login request to the server. The server can extract the login object identifier from the login request and use the login object identifier as the object identifier in an online state, that is, use the login object identifier as the target object; or, after detecting that the object has performed a login operation, the terminal verifies the login information carried in the login request. If the verification passes, the terminal extracts the login object identifier from the login request and uses the login object identifier as the object identifier in an online state, that is, uses the login object identifier as the target object.

[0050] Among them, the object information of the target object may include information used to describe the characteristics of the target object, the object information of the target object may include object attribute information and data interaction information, the object attribute information may include a unique identification code of the target object in the target application; the data interaction information may include a data sequence composed of data on which the target object has performed interactive operations.

[0051] The data pool can be used to store data uploaded by all objects through the target application. In a specific implementation, the target application provides a data upload portal, through which the object can upload data. Each data stored in the data pool can be used as data to be processed, and the data information corresponding to each data to be processed can be obtained.

[0052] Among them, the data information corresponding to the data to be processed may include information used to describe the characteristics of the data to be processed, and the data information corresponding to the data to be processed may include data attribute information and object interaction information. The data attribute information may include a unique identification code of the data to be processed in the target application, the number of times the data to be processed has been interacted, etc.; the object interaction information may include an object sequence consisting of objects that have performed interactive operations on the data to be processed.

[0053] Step 204 , based on the object information and the data information corresponding to each of the to-be-processed data, obtain preliminary feature data of each to-be-processed data in each feature extraction dimension.

[0054] Among them, for each data to be processed, the server can perform feature conversion processing on the object information of the target object and the data information of the data to be processed respectively to obtain object conversion features and data conversion features; based on the object conversion features and data conversion features, obtain the preliminary feature data of the data to be processed in each feature extraction dimension.

[0055] Among them, the object information of the target object may include discrete information, continuous information and indefinite-length information. The discrete information in the object information may include a unique identification code of the target object in the target application, the continuous information in the object information may include object attribute information with digital values, and the indefinite-length information in the object information may include a data sequence composed of data on which the target object has performed interactive operations.

[0056] Among them, the server's feature conversion processing of the target object's object information may include: performing vector mapping on discrete information in the object information to obtain corresponding mapping features; performing value transformation on continuous information in the object information to obtain corresponding mapping features; performing sampling processing on indefinite-length information in the object information to obtain fixed-length features; the server may use the mapping features corresponding to the discrete information, the mapping features corresponding to the continuous information, and the fixed-length features as object conversion features obtained through feature conversion processing.

[0057] Among them, the data information of the data to be processed may also include discrete information, continuous information and indefinite length information. The discrete information in the data information may include a unique identification code of the data to be processed in the target application, the continuous information in the data information may include the number of times the data to be processed has been interacted, etc. The indefinite length information in the data information may include an object sequence consisting of objects that have performed interactive operations on the data to be processed.

[0058] Among them, the server's feature conversion processing of the data information to be processed may include: performing vector mapping on discrete information in the data information to obtain corresponding mapping features; performing value transformation on continuous information in the data information to obtain corresponding mapping features; performing sampling processing on indefinite-length information in the data information to obtain fixed-length features; the server may use the mapping features corresponding to the discrete information, the mapping features corresponding to the continuous information, and the fixed-length features as data conversion features obtained through feature conversion processing.

[0059] Among them, for each data to be processed, after obtaining the object conversion feature and the data conversion feature in the above manner, the server can splice the object conversion feature and the data conversion feature to obtain the spliced ​​feature, and subsequently obtain the preliminary feature data of the data to be processed in each feature extraction dimension based on the spliced ​​feature.

[0060] Step 206, for each data to be processed, on each prediction target, at least one of the model input feature, the preliminary extracted feature and the object coding feature is used as the weight influence data corresponding to the current prediction target, the model input feature is obtained based on the object information and the data information corresponding to the current data to be processed, the preliminary extracted feature is obtained based on the preliminary feature data of the current data to be processed in each feature extraction dimension, the object coding feature is obtained based on the object information encoding, at least one weight distribution feature is determined according to the weight influence data corresponding to the current prediction target, based on the at least one weight distribution feature, the weight corresponding to each feature extraction dimension is determined, based on the preliminary feature data of the current data to be processed in each feature extraction dimension and the weight corresponding to each feature extraction dimension, the prediction result corresponding to the current prediction target is obtained, and based on the prediction results corresponding to each prediction target, the sorting reference information of the current data to be processed is determined.

[0061] Among them, for each data to be processed, the method provided in this step can be used to determine the sorting reference information. The following is explained using one of the data to be processed as an example. For the convenience of description, the data to be processed is called the current data to be processed.

[0062] Among them, for the current data to be processed, the server can use at least one of the model input features, preliminary extraction features and object coding features as the weight influence data corresponding to the current prediction target for each prediction target.

[0063] The server may obtain the model input feature based on the object information and the data information corresponding to the current data to be processed. In a specific implementation, the server may perform feature conversion processing on the object information of the target object and the data information of the current data to be processed respectively to obtain the object conversion feature and the data conversion feature; the object conversion feature and the data conversion feature are spliced ​​to obtain the spliced ​​feature, and the spliced ​​feature is used as the model input feature.

[0064] The server may obtain preliminary feature data of the current data to be processed in each feature extraction dimension, and use the obtained preliminary feature data as preliminary extracted features.

[0065] Among them, the object information may also include the object activity, which may include the number of clicks on the data display entrance on the target application by the object within a preset time period, the number of like operations performed, the total data display time, etc. In the case where the data in the data push method provided in the embodiment of the present application is a video, the total data display time is the total time the object plays the video through the data display entrance on the target application. The server can encode the object information to obtain the object coding characteristics.

[0066] In some embodiments, on each prediction target, when the weight influencing data includes model input features, preliminary extracted features, and object coding features, the server may respectively obtain the weight distribution features corresponding to the model input features, the weight distribution features corresponding to the preliminary extracted features, and the weight distribution features corresponding to the object coding features, and subsequently determine the weights corresponding to each feature extraction dimension on the current prediction target based on the three obtained weight distribution features.

[0067] It can be understood that the weight influence data may include only any one of the model input features, preliminary extraction features and object coding features, or only any two of them. The above description of including three features at the same time is only an example and this application does not limit this.

[0068] Among them, on each prediction target, after obtaining at least one weight distribution feature, the server can splice at least one weight distribution feature to obtain a weight splicing feature, compress the weight splicing feature to obtain a weight compression feature, stretch the weight compression feature to obtain a weight vector, and determine the weights corresponding to each feature extraction dimension based on the weight vector.

[0069] Among them, on each prediction target, after obtaining the weights corresponding to each feature extraction dimension, the server can multiply the preliminary feature data of the current data to be processed in each feature extraction dimension and the weights corresponding to each feature extraction dimension respectively, to obtain the multiplication results corresponding to each feature extraction dimension, and then add up the multiplication results to obtain the feature extraction result corresponding to the current prediction target. The feature extraction result can be input into the prediction network corresponding to the current prediction target, and the output of the prediction network is used as the prediction result of the current data to be processed on the current prediction target.

[0070] Among them, after obtaining the prediction results of the current data to be processed on each prediction target, the server can combine the prediction results on each prediction target to obtain the sorting reference information of the current data to be processed, and the sorting reference information is used to indicate the score of the push priority.

[0071] Among them, the prediction target may include executing a trigger operation on the data display entrance and the data interaction meeting the preset conditions. The data interaction meeting the preset conditions may include: the object executing a like operation on the current data to be processed, the object executing a forwarding operation on the current data to be processed, the object executing a sharing operation on the current data to be processed, the object executing a collection operation on the current data to be processed, the display time of the current data to be processed exceeds the preset time, etc.

[0072] Step 208: Determine the push data corresponding to the target object based on the sorting reference information corresponding to each piece of data to be processed in the data pool.

[0073] Among them, after obtaining the sorting reference information corresponding to each of the to-be-processed data in the data pool, the server can use the to-be-processed data whose sorting reference information meets the preset conditions as the push data corresponding to the target object. Alternatively, in the embodiment of the present application, the data in the data pool is the data that is ranked in front of the preset number in the rough sorting result. After obtaining the sorting reference information corresponding to each of the to-be-processed data in the data pool, the server can find the to-be-processed data whose sorting reference information meets the preset conditions as the refined sorting result, and can adopt the preset strategy to further screen the refined sorting result to obtain the push data corresponding to the target object.

[0074] In the above embodiment, firstly, the object information of the target object and the data information corresponding to each data to be processed in the data pool are obtained; based on the object information and the data information corresponding to each data to be processed, the preliminary feature data of each data to be processed in each feature extraction dimension is obtained; then, for each data to be processed, at each prediction target, at least one of the model input feature, the preliminary extracted feature and the object coding feature is used as the weight influence data corresponding to the current prediction target, the model input feature is obtained based on the object information and the data information corresponding to the current data to be processed, the preliminary extracted feature is obtained based on the preliminary feature data of the current data to be processed in each feature extraction dimension, the object coding feature is obtained based on the object information encoding, at least one weight distribution feature is determined according to the weight influence data corresponding to the current prediction target, based on the at least one weight distribution feature, the weight corresponding to each feature extraction dimension is determined, based on the preliminary features of the current data to be processed in each feature extraction dimension The weights corresponding to the data and each feature extraction dimension are used to obtain the prediction results corresponding to the current prediction target, and the sorting reference information of the current data to be processed is determined based on the prediction results corresponding to each prediction target; because on each prediction target, in the process of determining the weights corresponding to each feature extraction dimension, at least one of the influence of object information, data information itself on weight distribution, preliminary feature data itself on weight distribution, and object personalized information on weight distribution is combined, so that the final determined weight is more accurate, and the accuracy of the prediction results corresponding to each prediction target is improved. The improvement in the accuracy of the prediction results can further promote the improvement in the accuracy of the sorting reference information; finally, based on the sorting reference information corresponding to each data to be processed in the data pool, the push data corresponding to the target object is determined. The improvement in the accuracy of the sorting reference information makes it more likely that the target object is interested in the pushed data, thereby improving the accuracy of data push and saving computer resources.

[0075] In some embodiments, based on the object information and the data information corresponding to each data to be processed, preliminary feature data of each data to be processed in each feature extraction dimension is obtained, including: for each data to be processed, feature conversion processing is performed on the object information and the data information corresponding to the current data to be processed to obtain object conversion features and data conversion features; the object conversion features and the data conversion features are spliced ​​to obtain spliced ​​features; feature extraction is performed on the spliced ​​features through the feature extraction networks corresponding to each feature extraction dimension to obtain preliminary feature data of the current data to be processed in each feature extraction dimension.

[0076] Among them, the server can perform vector mapping on the discrete information in the object information to obtain the corresponding mapping features, perform value transformation on the continuous information in the object information to obtain the corresponding mapping features, and perform sampling processing on the indefinite-length information in the object information to obtain fixed-length features. The server can use all of the above features as object conversion features.

[0077] Among them, for each data to be processed, the server can perform vector mapping on the data information of the data to be processed to obtain the corresponding mapping features, and perform value transformation on the continuous information in the data information to obtain the corresponding mapping features; sample the indefinite-length information in the data information to obtain the fixed-length features, and the server can use the above features as data conversion features.

[0078] Among them, for each data to be processed, after obtaining the object conversion feature and the data conversion feature in the above manner, the server can splice the object conversion feature and the data conversion feature to obtain the splicing feature.

[0079] In the specific implementation, it is assumed that the object conversion feature and the data conversion feature contain N features in total, and the nth feature is recorded as x n , x n The dimension is d n , these N features are concatenated to obtain the concatenated feature X:

[0080] X=[…,x n ,…]

[0081] Among them, the dimension of the splicing feature X is

[0082] Among them, the server can input the splicing features into the feature extraction networks corresponding to each feature extraction dimension respectively, and the feature extraction networks corresponding to each feature extraction dimension respectively extract features from the splicing features to obtain preliminary feature data of the current data to be processed in each feature extraction dimension.

[0083] In the specific implementation, assuming that there are M feature extraction dimensions, for the mth feature extraction dimension, the preliminary feature data extracted by the feature extraction network corresponding to the feature extraction dimension can be recorded as e m (X), e m The dimension of (X) is denoted by d e .

[0084] Exemplarily, the feature extraction network corresponding to each feature extraction dimension may adopt a deep neural network (DNN for short).

[0085] In the above embodiment, for each data to be processed, first, feature conversion processing is performed on the object information and the data information corresponding to the current data to be processed, respectively, to obtain object conversion features and data conversion features; the object conversion features and the data conversion features are spliced ​​to obtain spliced ​​features; then, feature extraction is performed on the spliced ​​features through the feature extraction networks corresponding to each feature extraction dimension, respectively, to obtain preliminary feature data of the current data to be processed in each feature extraction dimension. The preliminary feature data can be used to determine the weight distribution feature, so that the weights corresponding to each feature extraction dimension finally determined are more accurate, thereby improving the accuracy of data push.

[0086] In some embodiments, the step of acquiring model input features includes: performing feature conversion processing on the object information and the data information corresponding to the current data to be processed, respectively, to obtain object conversion features and data conversion features; splicing the object conversion features and the data conversion features to obtain splicing features, and using the splicing features as model input features.

[0087] Among them, for each data to be processed, the server can perform feature conversion processing on the object information and the data information corresponding to the current data to be processed respectively through the method provided by the aforementioned embodiment to obtain object conversion features and data conversion features; and splice the object conversion features and the data conversion features to obtain splicing features, and the server can use the splicing features as model input features.

[0088] Among them, for each data to be processed, after the server obtains the preliminary feature data of the data to be processed in each feature extraction dimension, the preliminary feature data of the data to be processed in each feature extraction dimension can be used as the preliminary extracted features.

[0089] In the above embodiment, the server uses the concatenated features obtained by concatenating the object conversion features and the data conversion features as the model input features. In the process of determining the weights corresponding to each feature extraction dimension, the influence of the model input features on the weight distribution is combined, so that the final determined weights are more accurate, and the accuracy of the prediction results corresponding to each prediction target is improved.

[0090] In some embodiments, the weight influence data corresponding to the current prediction target includes model input features, preliminary extracted features and object coding features; at least one weight distribution feature is determined based on the weight influence data corresponding to the current prediction target, including: obtaining a weight extraction matrix, performing weight extraction on the model input features through the weight extraction matrix, and obtaining weight distribution features corresponding to the model input features; performing statistical processing on the preliminary feature data corresponding to each feature extraction dimension in the preliminary extracted features to obtain the statistical processing results corresponding to each feature extraction dimension, and determining the weight distribution features corresponding to the preliminary extracted features based on the statistical processing results corresponding to each feature extraction dimension; performing scene feature extraction on the object coding features to obtain object scene features, and determining the weight distribution features corresponding to the object coding features based on the preliminary extracted features and the object scene features.

[0091] Among them, for each data to be processed, on each prediction target, the server can use any one of the model input features, preliminary extraction features and object coding features as weight influence data, or can use any two as weight influence data, or can use all three features as weight influence data. The weight influence data on each prediction target can be the same or different. In the embodiment of the present application, the weight influence data on each prediction target includes model input features, preliminary extraction features and object coding features as an example for explanation.

[0092] Among them, on each prediction target, the server can obtain the weight extraction matrix corresponding to the current prediction target, multiply the model input features and the weight extraction matrix, and obtain the weight distribution features corresponding to the model input features on the current prediction target.

[0093] In the specific implementation, assuming that there are T prediction targets in total, for the prediction target t, the weight distribution characteristics corresponding to the model input features on the prediction target t can be calculated using the following formula:

[0094]

[0095] in, represents the weight distribution characteristics corresponding to the model input features on the predicted target t, X represents the model input features, Represents the weight extraction matrix corresponding to the predicted target t. The weight extraction matrix is ​​a trainable parameter matrix that is continuously updated during the model training process and fixed after the model training is completed. The dimension is d X ×M.

[0096] Among them, the server can perform statistical processing on the preliminary feature data corresponding to each feature extraction dimension in the preliminary extracted features, obtain the statistical processing results corresponding to each feature extraction dimension, and construct the weight distribution features corresponding to the preliminary extracted features based on the statistical processing results corresponding to each feature extraction dimension.

[0097] Among them, the server can extract scene features from the object coding features to obtain object scene features, and fuse the preliminary extracted features and the object scene features to obtain weight distribution features corresponding to the object coding features.

[0098] In the above embodiments, a method for obtaining weight distribution features corresponding to model input features, a method for obtaining weight distribution features corresponding to preliminary extracted features, and a method for obtaining weight distribution features corresponding to object coding features are provided. The weight distribution features can be used to determine the weights corresponding to each feature extraction dimension. The weights determined in this way are more accurate, thereby improving the accuracy of the prediction results corresponding to each prediction target.

[0099] In some embodiments, statistical processing is performed on preliminary feature data corresponding to each feature extraction dimension in the preliminary extracted features to obtain statistical processing results corresponding to each feature extraction dimension, including: for the preliminary feature data corresponding to each feature extraction dimension in the preliminary extracted features, aggregation processing is performed on each element contained in the preliminary feature data corresponding to the current feature extraction dimension to obtain an aggregation processing result, and the statistical processing result corresponding to the current feature extraction dimension is obtained based on the number of elements contained in the preliminary feature data corresponding to the current feature extraction dimension and the aggregation processing result.

[0100] Among them, for the preliminary feature data corresponding to each feature extraction dimension in the preliminary extracted features, each element contained in the preliminary feature data corresponding to the current feature extraction dimension can be added to obtain an aggregate processing result, and then the aggregate processing result is divided by the number of elements contained in the preliminary feature data corresponding to the current feature extraction dimension to obtain a statistical processing result corresponding to the current feature extraction dimension. The server can construct a weight distribution feature corresponding to the preliminary extracted features based on the statistical processing results corresponding to each feature extraction dimension.

[0101] In the specific implementation, assuming that there are M feature extraction dimensions, for the mth feature extraction dimension, the preliminary feature data extracted by the feature extraction network corresponding to the feature extraction dimension can be recorded as e m (X), e m The dimension of (X) is denoted by d e , the following formula can be used to determine the weight distribution characteristics corresponding to the initial extracted features:

[0102]

[0103] in, Represents the weight distribution characteristics corresponding to the initial extracted features, The dimension is also M, e m (X) represents the preliminary feature data extracted by the feature extraction network corresponding to the mth feature extraction dimension, [e m (X)] u Indicates e m The i-th element of (X).

[0104] In the above embodiment, a specific method for determining the weight distribution characteristics corresponding to the preliminary extracted features is provided. The weight distribution characteristics can be used to determine the weights corresponding to each feature extraction dimension. The weights determined in this way are more accurate, thereby improving the accuracy of the prediction results corresponding to each prediction target.

[0105] In some embodiments, based on the preliminary extracted features and the object scene features, the weight distribution features corresponding to the object coding features are determined, including: fusing the preliminary feature data corresponding to each feature extraction dimension in the preliminary extracted features with the object scene features respectively to obtain the fusion processing results corresponding to each feature extraction dimension; based on the fusion processing results corresponding to each feature extraction dimension, the weight distribution features corresponding to the object coding features are determined.

[0106] Among them, the server can perform dot product operations on the preliminary feature data corresponding to each feature extraction dimension in the preliminary extracted features and the object scene features respectively to obtain the fusion processing results corresponding to each feature extraction dimension; based on the fusion processing results corresponding to each feature extraction dimension, construct the weight distribution features corresponding to the object coding features.

[0107] The object coding feature may be a personalized coding feature (Personalization Code) obtained by encoding the object information. For each prediction target, the scene enhancement network POSO corresponding to the current prediction target may be used to extract scene features from the object coding feature to obtain the object scene feature.

[0108] In the specific implementation, the following formula can be used to extract scene features from object coding features:

[0109] Q t =q t (X PC )

[0110] Among them, X PC Encode features for the object, q t () is the scene reinforcement network corresponding to the current prediction target, Q t The dimension is d eThe scene reinforcement network parameters corresponding to each prediction target are independent of each other, and the parameters of the scene reinforcement network corresponding to each prediction target are determined through model training.

[0111] Among them, the server can use the following formula to determine the weight distribution characteristics corresponding to the object encoding characteristics:

[0112]

[0113] in, Represents the weight distribution feature corresponding to the object encoding feature, represents the object scene feature corresponding to the current prediction target, e m (x) represents the preliminary feature data extracted by the feature extraction network corresponding to the mth feature extraction dimension, [e m (x)] i Indicates e m The i-th element of (X), e m The dimension of (X) is denoted by d e , The dimension of is also M.

[0114] In the above embodiment, a specific method for determining the weight distribution feature corresponding to the object coding feature is provided. The weight distribution feature can be used to determine the weights corresponding to each feature extraction dimension. The weights determined in this way are more accurate, thereby improving the accuracy of the prediction results corresponding to each prediction target.

[0115] In some embodiments, at least one weight distribution feature includes a weight distribution feature corresponding to a model input feature, a weight distribution feature corresponding to a preliminary extracted feature, and a weight distribution feature corresponding to an object coding feature; based on at least one weight distribution feature, the weights corresponding to each feature extraction dimension are determined, including: concatenating the weight distribution features corresponding to the model input feature, the weight distribution features corresponding to the preliminary extracted feature, and the weight distribution features corresponding to the object coding feature to obtain a weight concatenated feature; compressing the weight concatenated feature through a weight compression matrix and a nonlinear activation function to obtain a weight compression feature; stretching the weight compression feature through a weight stretching matrix and a normalization function to obtain a weight vector, and determining the weights corresponding to each feature extraction dimension based on the weight vector.

[0116] Wherein, when at least one weight distribution feature includes a weight distribution feature corresponding to a model input feature, a weight distribution feature corresponding to a preliminary extracted feature, and a weight distribution feature corresponding to an object coding feature, the server may concatenate the weight distribution feature corresponding to the model input feature, the weight distribution feature corresponding to the preliminary extracted feature, and the weight distribution feature corresponding to the object coding feature using the following formula to obtain a weight concatenation feature:

[0117]

[0118] in, Represents the weight distribution characteristics corresponding to the model input features, Represents the weight distribution characteristics corresponding to the initial extracted features, Represents the weight distribution feature corresponding to the object encoding feature.

[0119] Among them, after obtaining the weighted splicing features, the server can use the following formula to compress the weighted splicing features:

[0120]

[0121] in, represents the weight compression feature, The dimension is r represents the compression factor, a1 represents the nonlinear activation function, Z t represents the weighted concatenation feature, Represents the weight compression matrix corresponding to the current prediction target. The weight compression matrix is ​​a trainable parameter matrix that is continuously updated during the model training process and fixed after the model training is completed. The weight compression matrices corresponding to each prediction target are independent of each other.

[0122] Among them, after obtaining the weight compression feature, the server can use the following formula to pull up the weight splicing feature:

[0123]

[0124] in, represents the weight vector of dimension M obtained by the pull-up process, represents the weight compression feature, Represents the weight stretching matrix corresponding to the current prediction target. The weight stretching matrix is ​​a trainable parameter matrix that is continuously updated during the model training process and fixed after the model training is completed. The weight stretching matrices corresponding to each prediction target are independent of each other.

[0125] Among them, after obtaining the weight vector, the server can use the elements in the weight vector as the weights of the corresponding feature extraction dimensions, thereby obtaining the weights corresponding to each feature extraction dimension on the current prediction target.

[0126] In the above embodiment, a method is provided for determining the weights corresponding to each feature extraction dimension based on at least one weight distribution feature. The weights determined in this way are more accurate, thereby improving the accuracy of the prediction results corresponding to each prediction target.

[0127] In some embodiments, based on the object information and the data information corresponding to each data to be processed, preliminary feature data of each data to be processed in each feature extraction dimension is obtained, including: for each data to be processed, based on the object information and the data information corresponding to the current data to be processed, the preliminary feature data of the current data to be processed in each feature extraction dimension is obtained through the feature extraction network corresponding to each feature extraction dimension in the data sorting model; at least one weight distribution feature is determined according to the weight influence data corresponding to the current prediction target, and based on the at least one weight distribution feature, the weight corresponding to each feature extraction dimension is determined, including: through the weight extraction network in the data sorting model, weight feature extraction is performed on the weight influence data corresponding to the current prediction target to obtain the weight corresponding to each feature extraction dimension.

[0128] The data sorting model includes a feature processing layer, a feature input layer, a preliminary extraction layer, a weight extraction layer and a prediction layer. The feature processing layer is used to perform feature conversion processing on the object information and the data information corresponding to the current data to be processed, respectively, to obtain object conversion features and data conversion features; the object conversion features and the data conversion features are spliced ​​to obtain splicing features. The feature input layer is used to receive the splicing features input by the feature processing layer. The preliminary extraction layer includes a feature extraction network corresponding to each feature extraction dimension, and the feature extraction network corresponding to each feature extraction dimension is used to receive the splicing features input by the feature input layer, and perform feature extraction on the splicing features to obtain preliminary feature data of the current data to be processed in each feature extraction dimension. The weight extraction layer includes a weight extraction network corresponding to each prediction target. For each prediction target, the weight extraction network corresponding to the prediction target is used to receive the preliminary feature data input by each feature extraction network, and based on the weight influence data and the preliminary feature data of the current data to be processed in each feature extraction dimension, the weight corresponding to each feature extraction dimension on the prediction target is determined. Based on the weight corresponding to each feature extraction dimension on the prediction target and the preliminary feature data of the current data to be processed in each feature extraction dimension, the feature extraction result corresponding to the current prediction target is determined. The prediction layer includes a prediction network corresponding to each prediction target. For each prediction target, the prediction network corresponding to the prediction target is used to receive the feature extraction result input by the weight extraction network corresponding to the prediction target, and predict the prediction target based on the feature extraction result to obtain the prediction result corresponding to the current prediction target.

[0129] Among them, the weight extraction network corresponding to each prediction target includes: an input influence processing layer, an extraction influence processing layer, a personalized influence processing layer and a dot product processing layer. The input influence processing layer is used to receive the splicing features input by the feature input layer, and multiply the weight extraction matrix corresponding to the current prediction target and the splicing features to obtain the weight distribution features corresponding to the model input features; the extraction influence processing layer is used to receive the preliminary feature data corresponding to each feature extraction dimension, and determine the weight distribution features corresponding to the preliminary extraction features based on the preliminary feature data corresponding to each feature extraction dimension; the personalized influence processing layer is used to receive the object coding features, and use the scene enhancement network POSO corresponding to the current prediction target to extract the scene features of the object coding features to obtain the object scene features; the dot product processing layer is used to perform dot product operations on the preliminary feature data corresponding to each feature extraction dimension in the preliminary extraction features and the object scene features, respectively, to obtain the fusion processing results corresponding to each feature extraction dimension, and based on the fusion processing results corresponding to each feature extraction dimension, construct the weight distribution features corresponding to the object coding features. The specific processing process of each layer can refer to the description of the aforementioned embodiment.

[0130] In the specific implementation, the weight extraction network corresponding to each prediction target can use the following formula to determine the feature extraction result:

[0131]

[0132] in, Indicates the feature extraction result corresponding to the current prediction target, e m (X) represents the preliminary feature data corresponding to the mth feature extraction dimension, Represents the weight corresponding to the mth feature extraction dimension.

[0133] Among them, the network structure of the feature extraction network corresponding to each feature extraction dimension can be a deep neural network (Deep Neural Networks, referred to as DNN), and the prediction network corresponding to each prediction target can be a tower structure. In the specific implementation, the prediction network corresponding to each prediction target can use the following formula to determine the prediction result:

[0134]

[0135] Among them, p t Indicates the prediction result corresponding to the current prediction target, h t Represents the tower structure processing of the current prediction target.

[0136] It should be noted that: in addition to inputting the object information of the target object and the data information of the current data to be processed into the feature processing layer, context information can also be input into the feature processing layer. The context information may include information such as whether the time when the target object performs a trigger operation on the data display portal in the target application is a weekend.

[0137] For example, see Figure 3 As shown in the figure, the data sorting model includes two prediction targets, namely TaskA and TaskB, the preliminary extraction layer includes feature extraction networks corresponding to three feature extraction dimensions, namely DNN1, DNN2 and DNN3, the weight extraction layer includes weight extraction network A corresponding to TaskA and weight extraction network B corresponding to TaskB, and the prediction layer includes prediction network A corresponding to TaskA and prediction network B corresponding to TaskB. For each data to be processed, the feature processing layer can perform relevant processing on the object information of the target object, the data information of the current data to be processed, and the context information to obtain the spliced ​​features, and input the spliced ​​features to the feature input layer, and the feature input layer further inputs the spliced ​​features to DNN1, DNN2 and DNN3 respectively, and DNN1, DNN2 and DNN3 input their respective preliminary feature data to the weight extraction network A and the weight extraction network B. The feature input layer inputs the concatenated features to the input influence processing layers in the two weight extraction networks. Taking the weight extraction network A as an example, the input influence processing layer in the weight extraction network A performs weight extraction on the model input features based on the weight extraction matrix corresponding to TaskA, and obtains the weight distribution features corresponding to the model input features; the extraction influence processing layer in the weight extraction network A determines the weight distribution features corresponding to the preliminary extracted features based on the preliminary feature data input by DNN1, DNN2 and DNN3; the personalized influence processing layer in the weight extraction network A determines the object scene features based on the preliminary feature data input by DNN1, DNN2 and DNN3 and the object coding features; the dot product processing layer in the weight extraction network A determines the weight distribution features corresponding to the object coding features based on the preliminary feature data input by DNN1, DNN2 and DNN3 and the object scene features. The weight extraction network A further determines the corresponding weights of the three feature extraction dimensions based on the three weight distribution characteristics, and determines the feature extraction results corresponding to TaskA based on the corresponding weights of the three feature extraction dimensions and the preliminary feature data output by DNN1, DNN2 and DNN3 respectively. The weight extraction network A outputs the feature extraction results to the prediction network A, and the prediction network A outputs the prediction results corresponding to TaskA. The processing process of TaskB is similar to that of TaskA, and can specifically participate in the processing process of TaskA above.

[0138] It should be noted that: Figure 3In the model structure shown, there are only two prediction targets, which is only an example. There may be more prediction targets, and the embodiment of the present application does not limit the number of prediction targets.

[0139] In the above embodiment, in the process of determining the weights corresponding to each feature extraction dimension, at least one of the influence of object information, data information itself on weight distribution, preliminary feature data itself on weight distribution, and object personalized information on weight distribution is combined, so that the final determined weights are more accurate, and the accuracy of the prediction results corresponding to each prediction target is improved. The improvement in the accuracy of the prediction results can further promote the improvement in the accuracy of the sorting reference information; finally, based on the sorting reference information corresponding to each data to be processed in the data pool, the push data corresponding to the target object is determined. The improvement in the accuracy of the sorting reference information makes it more likely that the target object is interested in the pushed data, thereby improving the accuracy of data push.

[0140] In some embodiments, the training step of the data sorting model includes: obtaining training samples, the training samples include sample object information of the sample object and sample data information of the sample data; based on the sample object information and the sample data information, obtaining preliminary feature data of the sample data in each feature extraction dimension through the feature extraction network corresponding to each feature extraction dimension in the data sorting model to be trained; at each prediction target, performing weight feature extraction on the sample influence data through the weight extraction network corresponding to the current prediction target in the data sorting model to obtain the weights corresponding to each feature extraction dimension, determining the target feature data corresponding to the current prediction target according to the preliminary feature data of the sample data in each feature extraction dimension and the weights corresponding to each feature extraction dimension, predicting the current prediction target based on the target feature data through the prediction network corresponding to the current prediction target in the data sorting model to obtain the prediction result corresponding to the current prediction target, and determining the model loss corresponding to the current prediction target based on the prediction result corresponding to the current prediction target and the label corresponding to the current prediction target; adjusting the parameters of the data sorting model to be trained according to the model losses corresponding to each prediction target until the stop condition of the model training is met, thereby obtaining a trained data sorting model.

[0141] Among them, different applications (Application, APP for short) can be installed on the terminal, and the APP corresponding to the data push method provided in the embodiment of the present application can be called the target application. Each object that has completed registration on the target application may or may not use the data display function in the process of using the target application. In the process of using the data display function, the object needs to first click on the data display entrance on the target application. At this time, the target application will jump to the data display interface and start playing the data. For each data played by the target application, the object may have different interactive operations on it, such as watching for a few seconds and then swiping away, liking, forwarding, etc. The object's interactive operations on each data can be saved as historical data.

[0142] The sample objects can be obtained from the objects registered on the target application, the objects registered on the target application can upload data through the target application, and the sample data can be obtained from all the uploaded data. The obtained sample objects and sample data can constitute training samples.

[0143] Among them, the prediction target may include: executing a trigger operation on the data display entrance and data interaction meeting a preset condition. For the sample object and sample data in the training sample, it can be determined based on the historical data corresponding to the sample object whether the sample object has executed a trigger operation on the data display entrance. If the sample object has not executed a trigger operation on the data display entrance, it can be determined that there is no interaction with the sample data. At this time, it can be determined that the label corresponding to the target of "executing a trigger operation on the data display entrance" is 0, and it can also be determined that the label corresponding to the target of "data interaction meets the preset condition" is also 0. If the sample object has executed a trigger operation on the data display entrance, it can be determined that the label corresponding to the target of "executing a trigger operation on the data display entrance" is 1. It can be further determined based on the historical data corresponding to the sample object whether the data interaction between the sample object and the sample data meets the preset condition. If the preset condition is met, the label corresponding to the target of "data interaction meets the preset condition" is 1. If the preset condition is not met, the label corresponding to the target of "data interaction meets the preset condition" is 0. The above use of 0 and 1 as labels is only an example, and other numerical values ​​or characters can also be used, which is not limited in the embodiments of the present application.

[0144] Similar to the object information of the target object, the sample object information of the sample object may include information used to describe the characteristics of the sample object. For detailed description, please refer to the relevant description of the object information of the target object. Similar to the data information of the data to be processed, the sample data information of the sample data may include information used to describe the characteristics of the sample data. For detailed description, please refer to the relevant description of the data information of the data to be processed.

[0145] Among them, sample object information, sample data information and context information can be input into the feature processing layer, and the feature processing layer processes the input information to obtain sample splicing features. The sample splicing features can be input into the feature extraction networks corresponding to each feature extraction dimension in the preliminary extraction layer respectively to obtain preliminary feature data of the sample data in each feature extraction dimension.

[0146] Among them, at each prediction target in the data sorting model to be trained, the weight extraction network corresponding to the prediction target receives the preliminary feature data of the sample data in each feature extraction dimension, and based on the weight extraction matrix to be trained corresponding to the prediction target, performs weight extraction on the sample input feature to obtain the weight distribution feature corresponding to the sample input feature, and determines the weight distribution feature of the sample extraction feature based on the preliminary feature data of the sample data in each feature extraction dimension, and determines the weight distribution feature corresponding to the sample encoding feature based on the preliminary feature data of the sample data in each feature extraction dimension and the sample encoding feature, and determines the sample weight vector based on the weight distribution feature corresponding to the sample input feature, the weight distribution feature of the sample extraction feature and the weight distribution feature corresponding to the sample encoding feature, and determines the weight corresponding to each feature extraction dimension on the current prediction target according to the sample weight vector. The weight extraction network can further determine the target feature data corresponding to the current prediction target according to the preliminary feature data of the sample data in each feature extraction dimension and the weight corresponding to each feature extraction dimension on the current prediction target.

[0147] Among them, the method for obtaining the weight distribution characteristics corresponding to the above-mentioned sample input characteristics can refer to the method for obtaining the weight distribution characteristics corresponding to the model input characteristics in the aforementioned embodiment; the method for obtaining the weight distribution characteristics of the sample extraction characteristics can refer to the method for obtaining the weight distribution characteristics corresponding to the preliminary extraction characteristics in the aforementioned embodiment; the method for obtaining the weight distribution characteristics corresponding to the sample coding characteristics can refer to the method for obtaining the weight distribution characteristics corresponding to the object coding characteristics in the aforementioned embodiment.

[0148] Among them, on each prediction target in the data sorting model to be trained, after the weight extraction network obtains the target feature data corresponding to the current prediction target, the target feature data corresponding to the current prediction target can be input into the prediction network corresponding to the current prediction target, and the prediction network corresponding to the current prediction target can process the target feature data corresponding to the current prediction target to obtain the prediction result corresponding to the current prediction target.

[0149] Among them, for each prediction target in the data sorting model to be trained, the server can determine the model loss corresponding to the current prediction target based on the prediction result corresponding to the current prediction target and the label corresponding to the current prediction target. In the specific implementation, the following formula can be used to calculate the model loss corresponding to the current prediction target:

[0150] L t =-y t log(p t )-(1-y t )log(1-p t )

[0151] Among them, L t Indicates the model loss corresponding to the current prediction target, y t Indicates the label corresponding to the current prediction target, p t Indicates the prediction result corresponding to the current prediction target.

[0152] Among them, after obtaining the model loss corresponding to each prediction target, the server can calculate the total loss of the model losses corresponding to each prediction target, and adjust the parameters of the data sorting model to be trained based on the total loss until the stopping condition of the model training is met, thereby obtaining a trained data sorting model.

[0153] In a specific implementation, the server can use the following formula to calculate the total loss:

[0154]

[0155] Among them, L t represents the model loss corresponding to the t-th predicted target, and T is the number of predicted targets.

[0156] In the above embodiment, a training step of a data sorting model is provided. First, a training sample is obtained. The training sample includes sample object information of a sample object and sample data information of a sample data. Based on the sample object information and the sample data information, preliminary feature data of the sample data in each feature extraction dimension is obtained through the feature extraction network corresponding to each feature extraction dimension in the data sorting model to be trained. Then, on each prediction target, a weight feature extraction is performed on the sample influence data through the weight extraction network corresponding to the current prediction target in the data sorting model to obtain the weight corresponding to each feature extraction dimension. According to the preliminary feature data of the sample data in each feature extraction dimension and the weight corresponding to each feature extraction dimension, the target feature data corresponding to the current prediction target is determined. Based on the target feature data, the current prediction target is predicted through the prediction network corresponding to the current prediction target in the data sorting model to obtain the prediction result corresponding to the current prediction target. Based on the prediction result corresponding to the current prediction target and the label corresponding to the current prediction target, the model loss corresponding to the current prediction target is determined. Finally, according to the model loss corresponding to each prediction target, the parameters of the data sorting model to be trained are adjusted until the stop condition of the model training is met, and the training of the data sorting model is obtained. During the above model training process, the influence of object information, data information itself on weight distribution, the influence of preliminary feature data itself on weight distribution, and the influence of object personalized information on weight distribution were modeled, which improved the prediction accuracy of the data sorting model.

[0157] In some embodiments, before obtaining the object information of the target object and the data information corresponding to each piece of data to be processed in the data pool, it also includes: in response to a login request of the target application, extracting a login object identifier from the login request, and using the login object identifier as the target object; after determining the push data corresponding to the target object based on the sorting reference information corresponding to each piece of data to be processed in the data pool, it also includes: in response to a trigger operation on the data display entrance in the target application, displaying the push data corresponding to the target object in sequence.

[0158] Among them, the APP corresponding to the data push method provided in the embodiment of the present application can be called the target application. The target application can be a social APP, a video APP, or a news APP, etc., which is not limited in the embodiment of the present application.

[0159] Among them, the object can perform a trigger operation on the target application on the terminal. In response to the trigger operation, the terminal can display the target application login interface. The target application login interface includes an identification input box, a password input box, a login control and a cancel control. The object can input the login object identification through the identification input box, and input the password corresponding to the login object identification through the password input box, and then perform a trigger operation on the login control. The terminal responds to the trigger operation, generates a login request, and sends the login request to the server. After receiving the login request, the server extracts the login object identification from the login request, uses the login object identification as the target object, and then starts to execute the steps in the above embodiment: obtaining the object information of the target object and the data information corresponding to each data to be processed in the data pool, obtaining preliminary feature data, fusion processing, determining the push data corresponding to the target object, etc. The detailed process can be found in the description of the above embodiment.

[0160] After determining the push data corresponding to the target object, the server can save the data identifier of the push data corresponding to the target object. The target application provides a data display entry. When the object performs a trigger operation on the data display entry, the terminal can send a push request to the server. After receiving the push request, the server sends the push data corresponding to each of the saved data identifiers to the terminal in sequence, and the terminal displays the received push data in sequence.

[0161] Wherein, there are multiple push data corresponding to the target object, and the multiple push data have sorting reference information, and the server can send the push data to the terminal in descending order according to the sorting reference information. When the data in the data push method provided in the embodiment of the present application is a video, the terminal plays each received video in sequence according to the order in which the video is received.

[0162] It should be noted that: while watching a video played by the terminal, the subject may not be interested in the video and thus swipe away the video. In response to the swipe away operation, the terminal may directly start playing the next video.

[0163] For example, the data in the data push method provided in the embodiment of the present application is a video, and the target application may be a social application. Figure 4 is an interface in the target application, which includes two entrances, namely, entrance 1 and entrance 2. Entrance 1 can be a display entrance for a list of objects with whom social interaction has recently taken place, and entrance 2 can be a video display entrance mentioned in the embodiment of the present application. After the object performs a trigger operation on the video display entrance, the terminal will jump to the video display interface, such as Figure 5As shown, the subject can watch videos on the video display interface. After the video display interface finishes playing one video, it will continue to play the next video. In the process of watching a video, the subject may finish watching the video, or may swipe away after watching for a period of time. The preset time the subject watches the video can be used as an interaction prediction target. For each video to be processed in the video pool, it is possible to predict whether the subject will perform a trigger operation on the above-mentioned video display entrance, and obtain a prediction result corresponding to the goal of "performing a trigger operation on the video display entrance". It is also possible to predict whether the subject's viewing time of the video to be processed can reach a preset time, and obtain a prediction result corresponding to the goal of "watching the video for a preset time". Based on these two prediction results, the sorting reference information of the videos to be processed is obtained.

[0164] In the above embodiment, in response to the login request of the target application, the login object identifier is extracted from the login request, and the login object identifier is used as the target object. Then, the push data corresponding to the target object is determined by the method provided in the embodiment of the present application, and in response to the triggering operation of the data display entrance in the target application, the push data corresponding to the target object is displayed in sequence. This method of embedding the data display entrance in the target application makes the functions of the target application more diversified and improves the object's experience of using the target application.

[0165] In some embodiments, a data sorting model processing method is provided, see Figure 6 | As shown in the figure, the data sorting model processing method includes:

[0166] Step 602: Acquire training samples, where the training samples include sample object information of sample objects and sample data information of sample data.

[0167] Among them, different applications (Application, referred to as APP) can be installed on the terminal, and the APP corresponding to the data push method provided in the embodiment of the present application can be called the target application. The server can obtain sample objects from the objects registered on the target application, and the objects registered on the target application can upload data through the target application, and the sample data can be obtained from all the uploaded data. The obtained sample objects and sample data can constitute training samples. For the detailed process, please refer to the description of the aforementioned embodiment.

[0168] Step 604, based on the sample object information and the sample data information, the preliminary feature data of the sample data in each feature extraction dimension is obtained through the feature extraction network corresponding to each feature extraction dimension in the data sorting model to be trained.

[0169] The data sorting model to be trained includes a feature processing layer, a feature input layer and a preliminary extraction layer, and preliminary feature data of the sample data in each feature extraction dimension can be obtained through these three layers. For details, please refer to the description of the above embodiment.

[0170] Step 606, on each prediction target, weighted feature extraction is performed on the sample influence data through the weight extraction network corresponding to the current prediction target in the data sorting model to obtain the weights corresponding to each feature extraction dimension, and the target feature data corresponding to the current prediction target is determined according to the preliminary feature data of the sample data in each feature extraction dimension and the weights corresponding to each feature extraction dimension. Based on the target feature data, the current prediction target is predicted through the prediction network corresponding to the current prediction target in the data sorting model to obtain the prediction result corresponding to the current prediction target, and the model loss corresponding to the current prediction target is determined based on the prediction result corresponding to the current prediction target and the label corresponding to the current prediction target.

[0171] Among them, at each prediction target, the weight distribution characteristics corresponding to the sample input features, the weight distribution characteristics of the sample extraction features and the weight distribution characteristics corresponding to the sample encoding features can be obtained through the weight extraction network corresponding to the current prediction target in the data sorting model, and the weights corresponding to each feature extraction dimension are determined based on these three characteristics. According to the preliminary feature data of the sample data in each feature extraction dimension and the weights corresponding to each feature extraction dimension, the target feature data corresponding to the current prediction target is determined, and based on the prediction result corresponding to the current prediction target and the label corresponding to the current prediction target, the model loss corresponding to the current prediction target is determined. For details, please refer to the description of the aforementioned embodiment.

[0172] Step 608, according to the model losses corresponding to each prediction target, the parameters of the data sorting model to be trained are adjusted until the stopping condition of the model training is met, and the trained data sorting model is obtained. The trained data sorting model is used to determine the push data corresponding to the target object.

[0173] Among them, after obtaining the model loss corresponding to each prediction target, the server can calculate the total loss of the model losses corresponding to each prediction target, and adjust the parameters of the data sorting model to be trained based on the total loss until the stopping condition of the model training is met, thereby obtaining the trained data sorting model. For details, please refer to the description of the aforementioned embodiment.

[0174] In the above embodiment, firstly, a training sample is obtained, and the training sample includes sample object information of the sample object and sample data information of the sample data; based on the sample object information and the sample data information, preliminary feature data of the sample data in each feature extraction dimension is obtained through the feature extraction network corresponding to each feature extraction dimension in the data sorting model to be trained; then, on each prediction target, weight feature extraction is performed on the sample influence data through the weight extraction network corresponding to the current prediction target in the data sorting model to obtain the weight corresponding to each feature extraction dimension; according to the preliminary feature data of the sample data in each feature extraction dimension and the weight corresponding to each feature extraction dimension, the target feature data corresponding to the current prediction target is determined; based on the target feature data, the current prediction target is predicted through the prediction network corresponding to the current prediction target in the data sorting model to obtain the prediction result corresponding to the current prediction target; based on the prediction result corresponding to the current prediction target and the label corresponding to the current prediction target, the model loss corresponding to the current prediction target is determined; finally, according to the model loss corresponding to each prediction target, the parameters of the data sorting model to be trained are adjusted until the stop condition of the model training is met, and the trained data sorting model is obtained. During the above model training process, the influence of object information, data information itself on weight distribution, the influence of preliminary feature data itself on weight distribution, and the influence of object personalized information on weight distribution were modeled, which improved the prediction accuracy of the data sorting model.

[0175] In some embodiments, see Figure 7As shown, the data push method provided in the embodiment of the present application can be applied in a video push scenario. A target application with a video playback function corresponding to the data push method provided in the embodiment of the present application is installed on the terminal. After detecting that the object has performed a login operation on the target application, the terminal can generate a login request and send it to the server. The server extracts the login object identifier from the login request, uses the login object identifier as the target object, and obtains the object information of the target object. Based on the object information, thousands of videos are recalled from millions of candidate videos, and then hundreds of videos are screened from thousands of videos in a rough sorting manner. The hundreds of videos are stored in a video pool, and each video in the video pool is used as a video to be processed. The video information corresponding to each video to be processed can be obtained. Based on the object information and the video information corresponding to each video to be processed, the preliminary feature data of each video to be processed in each feature extraction dimension is obtained. It is assumed that in the above-mentioned video push scenario, there are two prediction targets, namely, executing a trigger operation on the data display entrance and the video playback time exceeding the preset time. For each video to be processed, at the prediction target of "performing a trigger operation on the data display entrance", on the one hand, obtain the weight extraction matrix corresponding to this prediction target, multiply the weight extraction matrix and the model input feature to obtain the weight distribution feature corresponding to the model input feature; on the other hand, perform statistical processing on the preliminary feature data corresponding to each feature extraction dimension to obtain the statistical processing result corresponding to each feature extraction dimension, and determine the weight distribution feature corresponding to the preliminary extracted feature according to the statistical processing result corresponding to each feature extraction dimension; on the other hand, obtain the scene extraction function corresponding to this prediction target, use the scene extraction function to extract scene features from the object coding features, obtain the object scene features, fuse the preliminary feature data corresponding to each feature extraction dimension with the object scene features, and obtain the weight distribution feature corresponding to the object coding features; based on Based on the above three weight distribution characteristics, the weights corresponding to each feature extraction dimension on the prediction target of "performing a trigger operation on the data display entrance" are determined; based on the preliminary feature data of the current video to be processed in each feature extraction dimension and the weights corresponding to each feature extraction dimension, the prediction result corresponding to the prediction target of "performing a trigger operation on the data display entrance" is obtained; the same processing principle is performed on the prediction target of "the video playback time exceeds the preset time" to obtain the prediction result corresponding to the prediction target of "the video playback time exceeds the preset time"; the prediction results corresponding to the two prediction targets are merged to obtain the sorting reference information of the current video to be processed; the video to be processed whose sorting reference information meets the preset conditions is used as the refined ranking result, and the pre-set strategy is used to further screen the pushed video corresponding to the target object from the refined ranking result.After the object performs a trigger operation on the video display entrance on the target application, the terminal sends a push request to the server. After receiving the push request, the server sends the push videos corresponding to the target object to the terminal in sequence, and the terminal plays each push video in sequence according to the order of receipt.

[0176] In some embodiments, the data push method provided by the embodiment of the present application can be applied in the commodity push scenario. A target application with a commodity browsing function corresponding to the data push method provided by the embodiment of the present application is installed on the terminal. After detecting that the object has performed a login operation on the target application, the terminal can generate a login request and send it to the server. The server extracts the login object identifier from the login request and uses the login object identifier as the target object. Assume that in the above commodity push scenario, there are two predicted targets, namely, performing a trigger operation on the commodity browsing entrance and performing a collection operation on the commodity. For each unprocessed product in the product pool, for the prediction target of "performing a trigger operation on the product browsing entrance", on the one hand, the weight extraction matrix corresponding to this prediction target is obtained, and the weight extraction matrix is ​​multiplied by the model input feature to obtain the weight distribution feature corresponding to the model input feature; on the other hand, the preliminary feature data corresponding to each feature extraction dimension is statistically processed to obtain the statistical processing result corresponding to each feature extraction dimension, and the weight distribution feature corresponding to the preliminary extracted feature is determined according to the statistical processing result corresponding to each feature extraction dimension; on the other hand, the scene extraction function corresponding to this prediction target is obtained, and the scene extraction function is used to extract scene features from the object coding features to obtain the object scene features, and the preliminary feature data corresponding to each feature extraction dimension are respectively fused with the object scene features to obtain the object coding features. weight distribution characteristics corresponding to the code features; based on the above three weight distribution characteristics, determine the weights corresponding to each feature extraction dimension on the prediction target of "performing a trigger operation on the product browsing entrance", and obtain the prediction result corresponding to the prediction target of "performing a trigger operation on the data display entrance" based on the preliminary feature data of the current product to be processed in each feature extraction dimension and the weights corresponding to each feature extraction dimension; perform the same processing principle on the prediction target of "performing a collection operation on the product" to obtain the prediction result corresponding to the prediction target of "performing a collection operation on the product"; merge the prediction results corresponding to the two prediction targets to obtain the sorting reference information of the current product to be processed; based on the sorting reference information corresponding to each product to be processed in the product pool, filter and obtain the pushed products corresponding to the target object.

[0177] It should be noted that the application scenarios provided above are only used to illustrate the data push method of this application, and the application of the data push method provided by this application is not limited to the application scenarios provided above. For example, it can also be applied to product push scenarios, news push scenarios, etc., which are not limited in this embodiment of the application.

[0178] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0179] Based on the same inventive concept, the embodiment of the present application also provides a data push device for implementing the data push method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more data push device embodiments provided below can refer to the limitations on the data push method above, and will not be repeated here.

[0180] In one embodiment, Figure 8 As shown, a data push device is provided, comprising:

[0181] The acquisition module 801 is used to acquire the object information of the target object and the data information corresponding to each of the to-be-processed data in the data pool; and is also used to acquire the preliminary feature data of each to-be-processed data in each feature extraction dimension based on the object information and the data information corresponding to each of the to-be-processed data;

[0182] Processing module 802 is used to, for each data to be processed and on each prediction target, use at least one of the model input feature, the preliminary extraction feature and the object coding feature as the weight influence data corresponding to the current prediction target, the model input feature is obtained based on the object information and the data information corresponding to the current data to be processed, the preliminary extraction feature is obtained based on the preliminary feature data of the current data to be processed in each feature extraction dimension, the object coding feature is obtained based on the object information encoding, determine at least one weight distribution feature according to the weight influence data corresponding to the current prediction target, determine the weight corresponding to each feature extraction dimension based on the at least one weight distribution feature, obtain the prediction result corresponding to the current prediction target based on the preliminary feature data of the current data to be processed in each feature extraction dimension and the weight corresponding to each feature extraction dimension, and determine the sorting reference information of the current data to be processed based on the prediction results corresponding to each prediction target;

[0183] The determination module 803 is used to determine the push data corresponding to the target object based on the sorting reference information corresponding to each piece of data to be processed in the data pool.

[0184] In some embodiments, the acquisition module 801 is specifically used for: for each data to be processed, performing feature conversion processing on the object information and the data information corresponding to the current data to be processed, to obtain object conversion features and data conversion features; splicing the object conversion features and the data conversion features to obtain splicing features; extracting features from the spliced ​​features through the feature extraction networks corresponding to each feature extraction dimension, to obtain preliminary feature data of the current data to be processed in each feature extraction dimension.

[0185] In some embodiments, the processing module 802 is also used to: perform feature conversion processing on the object information and the data information corresponding to the current data to be processed, respectively, to obtain object conversion features and data conversion features; splice the object conversion features and the data conversion features to obtain splicing features, and use the splicing features as model input features.

[0186] In some embodiments, the weight influencing data corresponding to the current prediction target includes model input features, preliminary extracted features and object coding features; the processing module 802 is specifically used to: obtain a weight extraction matrix, perform weight extraction on the model input features through the weight extraction matrix, and obtain weight distribution features corresponding to the model input features; perform statistical processing on the preliminary feature data corresponding to each feature extraction dimension in the preliminary extracted features to obtain the statistical processing results corresponding to each feature extraction dimension, and determine the weight distribution features corresponding to the preliminary extracted features based on the statistical processing results corresponding to each feature extraction dimension; perform scene feature extraction on the object coding features to obtain object scene features, and determine the weight distribution features corresponding to the object coding features based on the preliminary extracted features and the object scene features.

[0187] In some embodiments, the processing module 802 is specifically used to: for the preliminary feature data corresponding to each feature extraction dimension in the preliminary extracted features, perform aggregation processing on each element contained in the preliminary feature data corresponding to the current feature extraction dimension to obtain an aggregation processing result; and obtain a statistical processing result corresponding to the current feature extraction dimension based on the number of elements contained in the preliminary feature data corresponding to the current feature extraction dimension and the aggregation processing result.

[0188] In some embodiments, the processing module 802 is specifically used to: fuse the preliminary feature data corresponding to each feature extraction dimension in the preliminary extracted features with the object scene features respectively to obtain the fusion processing results corresponding to each feature extraction dimension; based on the fusion processing results corresponding to each feature extraction dimension, determine the weight distribution characteristics corresponding to the object coding feature.

[0189] In some embodiments, at least one weight distribution feature includes a weight distribution feature corresponding to a model input feature, a weight distribution feature corresponding to a preliminary extracted feature, and a weight distribution feature corresponding to an object coding feature; the processing module 802 is specifically used to: concatenate the weight distribution features corresponding to the model input feature, the weight distribution features corresponding to the preliminary extracted feature, and the weight distribution features corresponding to the object coding feature to obtain a weight concatenated feature; compress the weight concatenated feature through a weight compression matrix and a nonlinear activation function to obtain a weight compressed feature; stretch the weight compressed feature through a weight stretching matrix and a normalization function to obtain a weight vector, and determine the weights corresponding to each feature extraction dimension based on the weight vector.

[0190] In some embodiments, the acquisition module 801 is specifically used to obtain, for each data to be processed, preliminary feature data of the current data to be processed in each feature extraction dimension through the feature extraction network corresponding to each feature extraction dimension in the data sorting model based on the object information and the data information corresponding to the current data to be processed; the processing module 802 is specifically used to perform weight feature extraction on the weight influence data corresponding to the current prediction target through the weight extraction network in the data sorting model, and obtain the weight corresponding to each feature extraction dimension.

[0191] In some embodiments, the data push device also includes a training module, which is used to obtain training samples, where the training samples include sample object information of the sample object and sample data information of the sample data; based on the sample object information and the sample data information, the preliminary feature data of the sample data in each feature extraction dimension is obtained through the feature extraction network corresponding to each feature extraction dimension in the data sorting model to be trained; at each prediction target, the sample influence data is subjected to weight feature extraction through the weight extraction network corresponding to the current prediction target in the data sorting model to obtain the weight corresponding to each feature extraction dimension; according to the preliminary feature data of the sample data in each feature extraction dimension and the weight corresponding to each feature extraction dimension, the target feature data corresponding to the current prediction target is determined; based on the target feature data, the current prediction target is predicted through the prediction network corresponding to the current prediction target in the data sorting model to obtain the prediction result corresponding to the current prediction target; based on the prediction result corresponding to the current prediction target and the label corresponding to the current prediction target, the model loss corresponding to the current prediction target is determined; according to the model loss corresponding to each prediction target, the parameters of the data sorting model to be trained are adjusted until the stop condition of the model training is met, and the trained data sorting model is obtained.

[0192] In some embodiments, the processing module 802 is also used to: respond to a login request from a target application, extract a login object identifier from the login request, and use the login object identifier as the target object; respond to a trigger operation on a data display entry in the target application, and display push data corresponding to the target object in sequence.

[0193] In one embodiment, Fig. 9 As shown, a data sorting model processing device is provided, comprising:

[0194] The acquisition module 901 is used to acquire training samples, where the training samples include sample object information of the sample object and sample data information of the sample data; and is also used to acquire preliminary feature data of the sample data in each feature extraction dimension through the feature extraction network corresponding to each feature extraction dimension in the data sorting model to be trained based on the sample object information and the sample data information;

[0195] Prediction module 902 is used to perform weight feature extraction on sample influence data on each prediction target through a weight extraction network corresponding to the current prediction target in the data sorting model, obtain the weights corresponding to each feature extraction dimension, determine the target feature data corresponding to the current prediction target according to the preliminary feature data of the sample data in each feature extraction dimension and the weights corresponding to each feature extraction dimension, predict the current prediction target based on the target feature data through a prediction network corresponding to the current prediction target in the data sorting model, obtain the prediction result corresponding to the current prediction target, and determine the model loss corresponding to the current prediction target based on the prediction result corresponding to the current prediction target and the label corresponding to the current prediction target;

[0196] The training module 903 is used to adjust the parameters of the data sorting model to be trained according to the model losses corresponding to each prediction target, until the stopping condition of the model training is met, and obtain the trained data sorting model. The trained data sorting model is used to determine the push data corresponding to the target object.

[0197] Each module in the above-mentioned data push device and data sorting model processing device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0198] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig.10As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store object information and data information. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a data push method is implemented.

[0199] Those skilled in the art will understand that Fig.10 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0200] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0201] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0202] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0203] It should be noted that the object information (including but not limited to object device information, object personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the object or fully authorized by all parties, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0204] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0205] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0206] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A data push method, characterized in that: The method comprises: Obtain object information of the target object and data information corresponding to each piece of data to be processed in the data pool; Based on the object information and the data information corresponding to each of the data to be processed, obtaining preliminary feature data of each of the data to be processed in each feature extraction dimension; For each data to be processed, on each prediction target, at least one of the model input feature, the preliminary extracted feature and the object coding feature is used as the weight influence data corresponding to the current prediction target, the model input feature is obtained based on the object information and the data information corresponding to the current data to be processed, the preliminary extracted feature is obtained based on the preliminary feature data of the current data to be processed in each feature extraction dimension, and the object coding feature is obtained based on the object information encoding, at least one weight distribution feature is determined according to the weight influence data corresponding to the current prediction target, based on the at least one weight distribution feature, the weight corresponding to each feature extraction dimension is determined, based on the preliminary feature data of the current data to be processed in each feature extraction dimension and the weight corresponding to each feature extraction dimension, the prediction result corresponding to the current prediction target is obtained, and based on the prediction results corresponding to each prediction target, the sorting reference information of the current data to be processed is determined; Based on the sorting reference information corresponding to each piece of to-be-processed data in the data pool, the push data corresponding to the target object is determined.

2. The method according to claim 1, characterized in that The obtaining of preliminary feature data of each to-be-processed data in each feature extraction dimension based on the object information and the data information corresponding to each to-be-processed data includes: For each data to be processed, feature conversion processing is performed on the object information and the data information corresponding to the current data to be processed, so as to obtain object conversion features and data conversion features; the object conversion features and the data conversion features are spliced ​​to obtain spliced ​​features; feature extraction is performed on the spliced ​​features through the feature extraction networks corresponding to each feature extraction dimension, so as to obtain preliminary feature data of the current data to be processed in each feature extraction dimension.

3. The method according to claim 1, characterized in that The step of obtaining the model input features includes: The object information and the data information corresponding to the current data to be processed are respectively subjected to feature conversion processing to obtain object conversion features and data conversion features; the object conversion features and the data conversion features are spliced ​​to obtain splicing features, and the splicing features are used as model input features.

4. The method according to claim 1, characterized in that: The weight influence data corresponding to the current prediction target includes model input features, preliminary extraction features and object coding features; and determining at least one weight distribution feature according to the weight influence data corresponding to the current prediction target includes: Obtain a weight extraction matrix, perform weight extraction on the model input features through the weight extraction matrix, and obtain weight distribution features corresponding to the model input features; Performing statistical processing on preliminary feature data corresponding to each feature extraction dimension in the preliminary extracted features to obtain statistical processing results corresponding to each feature extraction dimension, and determining weight distribution features corresponding to the preliminary extracted features according to the statistical processing results corresponding to each feature extraction dimension; Perform scene feature extraction on the object coding feature to obtain the object scene feature, and determine the weight distribution feature corresponding to the object coding feature based on the preliminary extracted feature and the object scene feature.

5. The method according to claim 4, characterized in that The statistical processing of the preliminary feature data corresponding to each feature extraction dimension in the preliminary extracted features to obtain the statistical processing results corresponding to each feature extraction dimension includes: For the preliminary feature data corresponding to each feature extraction dimension in the preliminary extracted features, aggregation processing is performed on each element contained in the preliminary feature data corresponding to the current feature extraction dimension to obtain an aggregation processing result, and based on the number of elements contained in the preliminary feature data corresponding to the current feature extraction dimension and the aggregation processing result, a statistical processing result corresponding to the current feature extraction dimension is obtained.

6. The method according to claim 4, characterized in that The determining, based on the preliminary extracted features and the object scene features, weight distribution features corresponding to the object coding features comprises: The preliminary feature data corresponding to each feature extraction dimension in the preliminary extracted features are respectively fused with the object scene features to obtain fusion processing results corresponding to each feature extraction dimension; Based on the fusion processing results corresponding to each feature extraction dimension, the weight distribution characteristics corresponding to the object coding features are determined.

7. The method according to claim 1, characterized in that The at least one weight distribution feature includes a weight distribution feature corresponding to a model input feature, a weight distribution feature corresponding to a preliminary extracted feature, and a weight distribution feature corresponding to an object coding feature; and determining the weights corresponding to each feature extraction dimension based on the at least one weight distribution feature includes: The weight distribution feature corresponding to the model input feature, the weight distribution feature corresponding to the preliminary extracted feature, and the weight distribution feature corresponding to the object coding feature are concatenated to obtain a weight concatenated feature; The weight concatenation feature is compressed by a weight compression matrix and a nonlinear activation function to obtain a weight compression feature; the weight compression feature is stretched by a weight stretching matrix and a normalization function to obtain a weight vector, and based on the weight vector, the weights corresponding to each feature extraction dimension are determined.

8. The method according to claim 1, characterized in that The obtaining of preliminary feature data of each to-be-processed data in each feature extraction dimension based on the object information and the data information corresponding to each to-be-processed data includes: For each data to be processed, based on the object information and the data information corresponding to the current data to be processed, the preliminary feature data of the current data to be processed in each feature extraction dimension is obtained through the feature extraction network corresponding to each feature extraction dimension in the data sorting model; The determining at least one weight distribution feature according to the weight influence data corresponding to the current prediction target, and determining the weights corresponding to the respective feature extraction dimensions based on the at least one weight distribution feature, includes: Through the weight extraction network in the data sorting model, weight feature extraction is performed on the weight influencing data corresponding to the current prediction target to obtain the weights corresponding to each feature extraction dimension.

9. The method according to claim 8, characterized in that The training steps of the data sorting model include: Acquire a training sample, wherein the training sample includes sample object information of a sample object and sample data information of a sample data; Based on the sample object information and the sample data information, obtaining preliminary feature data of the sample data in each feature extraction dimension respectively through the feature extraction network corresponding to each feature extraction dimension in the data sorting model to be trained; At each prediction target, weighted feature extraction is performed on the sample influence data through the weight extraction network corresponding to the current prediction target in the data sorting model to obtain the weights corresponding to each feature extraction dimension, and the target feature data corresponding to the current prediction target is determined according to the preliminary feature data of the sample data in each feature extraction dimension and the weights corresponding to each feature extraction dimension. Based on the target feature data, the current prediction target is predicted through the prediction network corresponding to the current prediction target in the data sorting model to obtain the prediction result corresponding to the current prediction target; based on the prediction result corresponding to the current prediction target and the label corresponding to the current prediction target, the model loss corresponding to the current prediction target is determined; According to the model losses corresponding to the respective prediction targets, the parameters of the data sorting model to be trained are adjusted until the stopping condition of the model training is met, thereby obtaining a trained data sorting model.

10. The method according to claim 1, characterized in that Before obtaining the object information of the target object and the data information corresponding to each of the to-be-processed data in the data pool, the method further includes: In response to a login request of a target application, extracting a login object identifier from the login request, and using the login object identifier as a target object; After determining the push data corresponding to the target object based on the sorting reference information corresponding to each of the to-be-processed data in the data pool, the method further includes: In response to a triggering operation on a data display entry in the target application, push data corresponding to the target object is displayed in sequence.

11. A data sorting model processing method, characterized in that: The method comprises: Acquire a training sample, where the training sample includes sample object information of a sample object and sample data information of a sample data; Based on the sample object information and the sample data information, obtaining preliminary feature data of the sample data in each feature extraction dimension respectively through the feature extraction network corresponding to each feature extraction dimension in the data sorting model to be trained; At each prediction target, weighted feature extraction is performed on the sample influence data through the weight extraction network corresponding to the current prediction target in the data sorting model to obtain the weights corresponding to each feature extraction dimension, and the target feature data corresponding to the current prediction target is determined according to the preliminary feature data of the sample data in each feature extraction dimension and the weights corresponding to each feature extraction dimension. Based on the target feature data, the current prediction target is predicted through the prediction network corresponding to the current prediction target in the data sorting model to obtain the prediction result corresponding to the current prediction target; based on the prediction result corresponding to the current prediction target and the label corresponding to the current prediction target, the model loss corresponding to the current prediction target is determined; According to the model losses corresponding to each prediction target, the parameters of the data sorting model to be trained are adjusted until the stopping conditions of the model training are met, so as to obtain a trained data sorting model, which is used to determine the push data corresponding to the target object.

12. A data push device, characterized in that: The device comprises: An acquisition module is used to acquire object information of a target object and data information corresponding to each piece of data to be processed in a data pool; and is also used to acquire preliminary feature data of each piece of data to be processed in each feature extraction dimension based on the object information and the data information corresponding to each piece of data to be processed; A processing module, for, for each data to be processed and on each prediction target, using at least one of a model input feature, a preliminary extraction feature and an object coding feature as weight influence data corresponding to the current prediction target, wherein the model input feature is obtained based on the object information and data information corresponding to the current data to be processed, the preliminary extraction feature is obtained based on preliminary feature data of the current data to be processed in each feature extraction dimension, and the object coding feature is obtained based on the object information encoding, determining at least one weight distribution feature according to the weight influence data corresponding to the current prediction target, determining the weights corresponding to each feature extraction dimension based on the at least one weight distribution feature, obtaining a prediction result corresponding to the current prediction target based on the preliminary feature data of the current data to be processed in each feature extraction dimension and the weights corresponding to each feature extraction dimension, and determining sorting reference information of the current data to be processed based on the prediction results corresponding to each prediction target; The determination module is used to determine the push data corresponding to the target object based on the sorting reference information corresponding to each piece of to-be-processed data in the data pool.

13. A data sorting model processing device, characterized in that: The device comprises: an acquisition module, used to acquire training samples, the training samples including sample object information of sample objects and sample data information of sample data; and further used to acquire preliminary feature data of the sample data in each feature extraction dimension through a feature extraction network corresponding to each feature extraction dimension in a data sorting model to be trained based on the sample object information and the sample data information; A prediction module is used to perform weighted feature extraction on sample influence data on each prediction target through a weight extraction network corresponding to the current prediction target in a data sorting model, obtain weights corresponding to each feature extraction dimension, determine target feature data corresponding to the current prediction target based on preliminary feature data of the sample data in each feature extraction dimension and the weights corresponding to each feature extraction dimension, predict the current prediction target based on the target feature data through a prediction network corresponding to the current prediction target in the data sorting model, obtain a prediction result corresponding to the current prediction target, and determine a model loss corresponding to the current prediction target based on the prediction result corresponding to the current prediction target and a label corresponding to the current prediction target; The training module is used to adjust the parameters of the data sorting model to be trained according to the model losses corresponding to each prediction target, until the stopping condition of the model training is met, so as to obtain a trained data sorting model, and the trained data sorting model is used to determine the push data corresponding to the target object.

14. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 11 are implemented.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 11 are implemented.

16. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 11 are implemented.

Images