CN117459575A

CN117459575A - Service data pushing method, device, computer equipment and storage medium

Info

Publication number: CN117459575A
Application number: CN202311125884.9A
Authority: CN
Inventors: 郭亮
Original assignee: Tencent Technology Shanghai Co Ltd
Current assignee: Tencent Technology Shanghai Co Ltd
Priority date: 2023-09-01
Filing date: 2023-09-01
Publication date: 2024-01-26

Abstract

The application relates to a service data pushing method, a service data pushing device, computer equipment, a storage medium and a computer program product. The method comprises the following steps: extracting data characteristics to be pushed corresponding to service data to be pushed, and extracting interacted data characteristics corresponding to interacted service data respectively; acquiring feature interaction parameters, carrying out feature interaction on each interacted data feature according to the feature interaction parameters to obtain each interaction feature, acquiring fusion guide parameters, and adjusting each interaction feature according to the fusion guide parameters to obtain each adjustment feature; fusing each adjustment feature with the data feature to be pushed to obtain a fusion feature; and extracting information based on the fusion characteristics to obtain push reference information of the service data to be pushed, and pushing the service data to be pushed to the object terminal corresponding to the object identifier when the push reference information accords with preset push conditions. By adopting the method, the accuracy of service data pushing is improved, and pushing resources are saved.

Description

Service data pushing method, device, computer equipment and storage medium

Technical Field

The present invention relates to the field of internet technologies, and in particular, to a service data pushing method, an apparatus, a computer device, a storage medium, and a computer program product.

Background

With the development of internet technology, an internet information pushing technology has emerged, through which internet service data can be pushed to each internet-use object, for example, video related to an internet service, news related to an internet service, pictures related to an internet service, text related to an internet service, and the like can be pushed to each internet-use object. Currently, when service data is pushed, service data is typically pushed to all usage objects related to the service. However, pushing service data to all internet usage objects related to the service has problems of low service data pushing accuracy and pushing resource waste.

Disclosure of Invention

In view of the foregoing, it is desirable to provide a service data pushing method, apparatus, computer device, computer readable storage medium, and computer program product that can improve the accuracy of service data pushing and save pushing resources.

In a first aspect, the present application provides a service data pushing method. The method comprises the following steps:

acquiring a service data pushing request, wherein the service data pushing request carries an object identifier, acquiring service data to be pushed based on the service data pushing request, and acquiring each interacted service data based on the object identifier;

extracting the corresponding characteristics of the service data to be pushed to obtain the characteristics of the data to be pushed, and extracting the corresponding characteristics of each interacted service data to obtain the characteristics of each interacted data;

the method comprises the steps of obtaining feature interaction parameters, carrying out feature interaction on each interacted data feature according to the feature interaction parameters to obtain interaction features corresponding to each interacted data feature, obtaining fusion guide parameters, and adjusting each interaction feature according to the fusion guide parameters to obtain adjustment features corresponding to each interaction feature;

fusing each adjustment characteristic with the data characteristic to be pushed to obtain a fusion characteristic corresponding to the service data to be pushed;

and extracting information based on the fusion characteristics to obtain pushing reference information of the service data to be pushed aiming at the object identifier, wherein the pushing reference information is used for pushing the service data to be pushed to the object terminal corresponding to the object identifier when the preset pushing condition is met.

In a second aspect, the present application further provides a service data pushing device. The device comprises:

the data acquisition module is used for acquiring a service data pushing request, wherein the service data pushing request carries an object identifier, acquiring service data to be pushed based on the service data pushing request, and acquiring each interacted service data based on the object identifier;

the feature extraction module is used for extracting features corresponding to the service data to be pushed to obtain the features of the data to be pushed, extracting the features corresponding to the interactive service data respectively and obtaining the features of the interactive data;

the feature adjustment module is used for acquiring feature interaction parameters, carrying out feature interaction on each interacted data feature according to the feature interaction parameters to obtain interaction features corresponding to each interacted data feature respectively, acquiring fusion guide parameters, and adjusting each interaction feature according to the fusion guide parameters to obtain adjustment features corresponding to each interaction feature respectively;

the feature fusion module is used for fusing each adjustment feature with the feature of the data to be pushed to obtain fusion features corresponding to the service data to be pushed;

the data pushing module is used for extracting information based on the fusion characteristics to obtain pushing reference information of the service data to be pushed aiming at the object identification, and the pushing reference information is used for pushing the service data to be pushed to the object terminal corresponding to the object identification when the preset pushing condition is met.

In a third aspect, the present application also provides a computer device. The computer device comprises a memory storing a computer program and a processor which when executing the computer program performs the steps of:

In a fourth aspect, the present application also provides a computer-readable storage medium. The computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of:

In a fifth aspect, the present application also provides a computer program product. The computer program product comprises a computer program which, when executed by a processor, implements the steps of:

According to the business data pushing method, the business data pushing device, the computer equipment, the storage medium and the computer program product, the features corresponding to the business data to be pushed are extracted to obtain the features of the data to be pushed, the features corresponding to the interactive business data are extracted to obtain the features of the interactive data, the interactive data features are adjusted through the feature interaction parameters and the fusion guide parameters to obtain the adjustment features, the adjustment features are fused with the features of the data to be pushed to obtain the fusion features, so that the obtained fusion features have more information, the pushing reference information is extracted according to the fusion features, the accuracy of the obtained pushing reference information is improved, and finally the business data to be pushed is prevented from being pushed to the object terminal which does not meet the preset pushing conditions when the pushing reference information meets the preset pushing conditions, so that the accuracy of pushing the business data is improved, and pushing resources can be saved.

Drawings

FIG. 1 is an application environment diagram of a service data push method in one embodiment;

fig. 2 is a flow chart of a service data pushing method in an embodiment;

FIG. 3 is a flowchart of obtaining target push reference information in an embodiment;

FIG. 4 is a flow chart of a method for obtaining fusion features and target fusion features in one embodiment;

fig. 5 is a schematic diagram of a network structure for obtaining target push reference information in a specific embodiment;

FIG. 6 is a schematic diagram of a network architecture for deriving various interacted data features in one embodiment;

FIG. 7 is a diagram of a network architecture incorporating feature extraction in one embodiment;

FIG. 8 is a flowchart of a method for obtaining push reference information in one embodiment;

FIG. 9 is a flowchart of a business data push model in one embodiment;

FIG. 10 is a schematic diagram of a model architecture of a business data push model in one embodiment;

FIG. 11 is a comparative schematic diagram of test performance indicators in one embodiment;

fig. 12 is a flowchart of a service data pushing method in an embodiment;

FIG. 13 is a block diagram illustrating a service data pushing device according to an embodiment;

FIG. 14 is an internal block diagram of a computer device in one embodiment;

fig. 15 is an internal structural view of a computer device in another embodiment.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the present application.

Machine Learning (ML) is a multi-domain interdisciplinary, involving multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, etc. It is specially studied how a computer simulates or implements learning behavior of a human to acquire new knowledge or skills, and reorganizes existing knowledge structures to continuously improve own performance. Machine learning is the core of artificial intelligence, a fundamental approach to letting computers have intelligence, which is applied throughout various areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, confidence networks, reinforcement learning, transfer learning, induction learning, teaching learning, and the like.

The scheme provided by the embodiment of the application relates to artificial neural network and other technologies of artificial intelligence, and is specifically described through the following embodiments:

the service data pushing method provided by the embodiment of the application can be applied to an application environment shown in fig. 1. Wherein the terminal 102 communicates with the server 104 via a network. The data storage system may store data that the server 104 needs to process. The data storage system may exist alone, may be integrated on the server 104, or may be located on a cloud or other network server. The server 104 obtains a service data pushing request sent by the object terminal 102, wherein the service data pushing request carries an object identifier, obtains service data to be pushed based on the service data pushing request, and obtains each interacted service data based on the object identifier; the server 104 extracts the corresponding characteristics of the service data to be pushed to obtain the characteristics of the data to be pushed, and extracts the corresponding characteristics of each interactive service data to obtain the characteristics of each interactive data; the server 104 obtains feature interaction parameters, performs feature interaction on each interacted data feature according to the feature interaction parameters to obtain interaction features corresponding to each interacted data feature, obtains fusion guide parameters, and adjusts each interaction feature according to the fusion guide parameters to obtain adjustment features corresponding to each interaction feature; the server 104 fuses each adjustment feature with the feature of the data to be pushed to obtain a fusion feature corresponding to the service data to be pushed; the server 104 performs information extraction based on the fusion characteristics to obtain push reference information of the service data to be pushed for the object identifier, where the push reference information is used for pushing the service data to be pushed to the object terminal 102 corresponding to the object identifier when the preset push condition is met. The terminal can be, but not limited to, various personal computers, notebook computers, smart phones, tablet computers, internet of things equipment and portable wearable equipment, and the internet of things equipment can be smart speakers, smart televisions, smart air conditioners, smart vehicle-mounted equipment and the like. The portable wearable device may be a smart watch, smart bracelet, headset, or the like. The server may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing cloud computing service. The terminal and the server may be directly or indirectly connected through wired or wireless communication, which is not limited herein.

In one embodiment, as shown in fig. 2, a service data pushing method is provided, and the method is taken as an example for describing that the method is applied to the server in fig. 1, it is to be understood that the method can also be applied to a terminal, and can also be applied to a system including the terminal and the server, and is implemented through interaction between the terminal and the server. In this embodiment, the method includes the steps of:

s202, a service data pushing request is acquired, the service data pushing request carries an object identifier, service data to be pushed is acquired based on the service data pushing request, and each interacted service data is acquired based on the object identifier.

The service data push request is used for requesting to push service data, and the service data refers to data of a service that needs to be pushed, for example, the service data may be a service video, a service picture, a service text, and the like. The service refers to a transaction in the internet, in which service data can be pushed, for example, the service can be an advertisement service of goods, a push service of music, a push service of news, a push service of video, and the like. The object identifier is used for uniquely identifying a corresponding object, which may be a real object, such as a person or an object, or may be a virtual object, such as a virtual artificial intelligence, a virtual even image, a virtual anchor, or the like. The service data to be pushed refers to the service data which needs to be judged whether to be pushed to the object terminal corresponding to the object identifier. The interactive service data refers to service data that the object corresponding to the object identifier has interacted with, i.e. there is interaction between the object corresponding to the object identifier and the service data, for example, browsing the service data, clicking the service data, interacting with the service data, and so on.

Specifically, the server may obtain a service data push request sent by the terminal, and then analyze the service data push request to obtain the object identifier. The server can acquire the service data pushing request through a triggering event, and the triggering event can be an event triggering the service data pushing. And then the server can acquire the service data to be pushed from the database according to the service data pushing request, can acquire the service data to be pushed from the server providing the service, can acquire the service data to be pushed from the Internet, and the like. The server may then obtain the respective interacted service data corresponding to the object identifier from the database, or may obtain the respective interacted service data corresponding to the object identifier from the server providing the service.

S204, extracting the corresponding characteristics of the service data to be pushed to obtain the characteristics of the data to be pushed, and extracting the corresponding characteristics of each interactive service data to obtain the characteristics of each interactive data.

The data to be pushed is a feature for representing the service data to be pushed, and the interactive data feature is a feature for representing the interactive service data.

Specifically, the server may directly perform vectorization on the service data to be pushed to obtain the feature of the data to be pushed, or may perform vectorization on each piece of interacted service data to obtain the feature of the data to be pushed. In one embodiment, the server may perform feature extraction using a neural network that performs feature extraction, which may be a pre-trained neural network, such as a convolutional neural network, a recurrent neural network, a feed forward neural network, or the like. Namely, the service data to be pushed and the interacted service data are input into the neural network to obtain the output data characteristics to be pushed and the interacted data characteristics.

S206, acquiring characteristic interaction parameters, carrying out characteristic interaction on each interacted data characteristic according to the characteristic interaction parameters to obtain interaction characteristics corresponding to each interacted data characteristic, acquiring fusion guide parameters, and adjusting each interaction characteristic according to the fusion guide parameters to obtain adjustment characteristics corresponding to each interaction characteristic.

The feature interaction parameters are parameters for feature interaction, and the feature interaction parameters can be obtained through pre-training or preset. The interactive features are features obtained after feature interaction is performed by using feature interaction parameters, and the obtained interactive features can improve information quantity and accuracy of the features through feature interaction. The fusion guide parameters are parameters for feature adjustment, and can be obtained through pre-training or preset. The adjustment feature is a feature obtained by fusing the guide parameters to perform feature adjustment. The feature can be adjusted before feature fusion by fusing the guide parameters, so that the accuracy of the subsequent feature fusion is improved.

Specifically, the server may obtain the feature interaction parameters from the database, or may obtain the feature interaction parameters from a server providing the data service. And then, carrying out nonlinear weighting on each interacted data characteristic by using the characteristic interaction parameter to obtain the interaction characteristic corresponding to each interacted data characteristic. The server may then obtain the fusion guidance parameters from the database, or may obtain the fusion guidance parameters from a server providing the data service. And then respectively carrying out nonlinear weighting on each interaction characteristic according to the fusion guide parameters to obtain adjustment characteristics respectively corresponding to each interaction characteristic.

And S208, fusing each adjustment characteristic with the data characteristic to be pushed to obtain the fusion characteristic corresponding to the service data to be pushed.

The fusion features are features obtained by fusing each adjustment feature with the data feature to be pushed.

Specifically, the server can fuse each adjustment feature with the data feature to be pushed through a pre-trained neural network to obtain a fusion feature. The server may calculate vector operation values between the vectors of the respective adjustment features and the vectors of the data features to be pushed, for example, calculate sums between the vectors, calculate products between the vectors, calculate average values between the vectors, calculate maximum or minimum values between the vectors, divide the vectors, subtract the vectors, and so on, to obtain the fusion feature. The server can also directly splice each adjustment feature with the data feature to be pushed to obtain a fusion feature.

S210, information extraction is carried out based on the fusion characteristics, push reference information of the service data to be pushed aiming at the object identification is obtained, and the push reference information is used for pushing the service data to be pushed to the object terminal corresponding to the object identification when the preset pushing condition is met.

The push reference information is reference information for characterizing whether the push service data is to be pushed to the object terminal corresponding to the object identifier, for example, the push degree may be the push degree, and when the push degree is higher, the push service data is required to be pushed to the object terminal corresponding to the object identifier. The preset pushing condition refers to a preset condition for pushing the service data to be pushed, and includes, but is not limited to, that the pushing reference information is consistent with preset pushing reference information or the pushing degree represented by the pushing reference information reaches a preset degree threshold. The object terminal refers to a terminal used by an object corresponding to the object identifier.

Specifically, the server performs information extraction on the fusion features to obtain push reference information of the service data to be pushed aiming at the object identifier, wherein the server can perform information extraction on the fusion features through a neural network to obtain the push reference information, can perform dimension reduction on the fusion features and normalization to obtain the push reference information, and can weight the fusion features by using information extraction parameters to obtain the push reference information, wherein the information extraction parameters are parameters for extracting the push reference information, can be pre-trained and can be pre-set. And then the server compares the push reference information with preset push conditions, and when the push reference information does not meet the preset push conditions, the server indicates that even if the service data to be pushed is pushed to the object terminal corresponding to the object identifier, the object corresponding to the object identifier does not interact with the service data to be pushed, and at the moment, the service data to be pushed does not need to be pushed to the object terminal corresponding to the object identifier. When the pushing reference information accords with the preset pushing condition, the object corresponding to the object identifier is explained to be likely to interact with the service data to be pushed, and at the moment, the server pushes the service data to be pushed to the object terminal corresponding to the object identifier.

In one embodiment, S206, namely, performing feature interaction on each interacted data feature according to the feature interaction parameter to obtain interaction features corresponding to each interacted data feature, includes the steps of:

Performing full-connection operation on each interacted data characteristic by using the characteristic interaction parameter to obtain each interaction full-connection characteristic; and carrying out nonlinear activation on each interactive full-connection feature to obtain interactive features corresponding to each interactive data feature.

The full connection operation means that each sub-parameter in the interacted data characteristic and the characteristic interaction parameter is weighted, so that an output result is obtained. The interactive full-connection feature is a feature obtained by performing full-connection operation on the interactive data feature by using the feature interaction parameter. Nonlinear activation refers to activation calculation using a nonlinear activation function, which may be an S-type function or a ReLU (linear rectification function) function or a tanh (hyperbolic tangent function) function, or the like.

Specifically, the server uses each sub-parameter in the feature interaction parameters to weight each interacted data feature respectively to obtain each weighted feature corresponding to each sub-parameter, and calculates the sum of each weighted feature corresponding to each sub-parameter to obtain each interaction full-connection feature. And then performing nonlinear activation on all interactive full-connection features by using a nonlinear activation function, so as to obtain interactive features corresponding to the interactive data features respectively.

In one embodiment, the server may splice each of the interacted data features, and input the spliced features into a fully connected neural network, where the fully connected neural network performs a fully connected operation through trained feature interaction parameters in the hidden layer, and performs nonlinear activation, so as to obtain interaction features corresponding to each of the outputted interacted data features.

In the embodiment, the feature interaction parameters are used for carrying out full-connection operation on each interacted data feature to obtain each interacted full-connection feature, and then nonlinear activation is carried out on each interacted full-connection feature to obtain the interaction feature corresponding to each interacted data feature, so that implicit feature interaction can be carried out, and the accuracy of the obtained interaction feature is improved.

In one embodiment, S206, that is, adjusts each interaction feature according to the fusion guiding parameter to obtain an adjustment feature corresponding to each interaction feature, includes the steps of:

performing full-connection operation on each interaction feature by using the fusion guide parameters to obtain each fusion guide full-connection feature; and carrying out nonlinear activation on all the fusion guide full-connection features to obtain adjustment features corresponding to all the interaction features.

The fusion guide full-connection feature is a feature obtained by performing full-connection operation on the interaction feature by using the fusion guide parameter.

Specifically, the server weights each interaction feature by using each sub-parameter in the fusion guide parameters to obtain each weighted feature corresponding to each sub-parameter, and calculates the sum of each weighted feature corresponding to each sub-parameter to obtain each fusion guide full-connection feature. And then nonlinear activation is carried out on all the fusion guide full-connection features by using a nonlinear activation function, so that adjustment features corresponding to all the interaction features are obtained.

In one embodiment, the server may splice each interactive feature, and input the spliced feature into a fully connected neural network, where the fully connected neural network performs a fully connected operation through the trained fusion guiding parameters in the hidden layer, and performs nonlinear activation, so as to obtain adjustment features corresponding to each output interactive feature.

In the above embodiment, the full connection operation is performed on each interactive feature by using the fusion guide parameter to obtain each fusion guide full connection feature, and then the nonlinear activation is performed on each fusion guide full connection feature to obtain the adjustment feature corresponding to each interactive feature, that is, the fusion guide parameter is introduced to further adjust each interactive feature, so that the accuracy of the obtained adjustment feature can be improved, and the accuracy of the subsequent feature fusion is further improved.

In one embodiment, S208, that is, fusing each adjustment feature with a feature of data to be pushed to obtain a fusion feature corresponding to service data to be pushed, includes the steps of:

calculating binary operation values of each adjustment feature and the data feature to be pushed respectively to obtain each binary operation value; and taking each binary operation value as a fusion characteristic corresponding to the service data to be pushed.

The binary operation value is a value obtained by binary operation, and the binary operation is an operation applied to two objects, and a rule such as addition and multiplication of a third element is formed by two elements.

Specifically, the server may calculate a binary operation value of each adjustment feature and the feature of the data to be pushed, for example, the server may calculate an inner product of each adjustment feature and the feature of the data to be pushed to obtain each binary operation value, and then use each obtained binary operation value as a feature vector to obtain a fusion feature corresponding to the service data to be pushed. The server can also calculate the sum of each adjustment feature and the data to be pushed to obtain the fusion feature corresponding to the service data to be pushed, and the server can calculate the average of each adjustment feature and the data to be pushed to obtain the fusion feature corresponding to the service data to be pushed.

Summarizing the above embodiments, calculating the binary operation values of each adjustment feature and the feature of the data to be pushed to obtain each binary operation value, and then using each binary operation value as the fusion feature corresponding to the service data to be pushed, thereby improving the accuracy of the fusion feature.

In one embodiment, as shown in fig. 3, S210, performing information extraction based on the fusion feature to obtain push reference information of the service data to be pushed for the object identifier, includes:

s302, obtaining each object attribute characteristic corresponding to the object identifier, and obtaining each service data attribute characteristic corresponding to the service data to be pushed.

The object attribute features refer to features for characterizing basic attributes corresponding to the object identifier, such as identity features, age features, gender features and the like of the object. The service data attribute features are features for characterizing data attributes corresponding to the service data, such as a time feature for creating the service data, a category feature for the service data, an identification feature for the service data, and so on. The data attribute corresponding to the service data is determined when the service data is created.

Specifically, the server may obtain each piece of basic attribute information corresponding to the object identifier from the database, and then perform feature conversion by using each piece of basic attribute information to obtain the object attribute feature corresponding to each piece of basic attribute information. The server can acquire each data attribute information corresponding to the service data to be pushed from the database, and then perform feature conversion by using each data attribute information to obtain the service data attribute feature corresponding to each data attribute information.

S304, fusing the attribute characteristics of each object with the attribute characteristics of each service data to obtain target fusion characteristics corresponding to the service data to be pushed.

The target fusion feature is used for representing the feature obtained by fusing the object attribute feature and the service data attribute feature.

Specifically, the server splices each object attribute feature to obtain a target object attribute feature, then splices each service data attribute feature to obtain a target service data attribute feature, and then fuses the target object attribute feature and the target service data attribute feature, wherein the server can directly splice the target object attribute feature and the target service data attribute feature to obtain a target fusion feature, the server can also calculate the sum of the target object attribute feature and the target service data attribute feature to obtain a target fusion feature, and the server can also calculate the product of the target object attribute feature and the target service data attribute feature to obtain the target fusion feature. The server can also calculate the average of the attribute characteristics of the target object and the attribute characteristics of the target service data to obtain the target fusion characteristics.

S306, splicing the target fusion features and the fusion features to obtain splicing features, and extracting information based on the splicing features to obtain target pushing reference information of the service data to be pushed aiming at the object identification.

The target push reference information refers to push reference information obtained by extracting information by using target fusion features and fusion features.

Specifically, the server performs head-to-tail splicing on the target fusion feature and the fusion feature, can splice the target fusion feature as a head part and the fusion feature as a tail part, can splice the fusion feature as a head part and the target fusion feature as a tail part to obtain a spliced feature, and then inputs the spliced feature into a pre-trained neural network for information extraction to extract information, so as to obtain target pushing reference information of the output business data to be pushed aiming at the object identification.

In the above embodiment, by acquiring each object attribute feature corresponding to the object identifier, acquiring each service data attribute feature corresponding to the service data to be pushed, then fusing each object attribute feature with each service data attribute feature to obtain a target fusion feature corresponding to the service data to be pushed, and finally extracting information by using the target fusion feature and the fusion feature to obtain target pushing reference information, that is, extracting information by using the target fusion feature and the fusion feature together, thereby improving the accuracy of the target pushing reference information.

In one embodiment, S304, that is, fusing each object attribute feature with each service data attribute feature to obtain a target fusion feature corresponding to the service data to be pushed, includes the steps of:

calculating binary operation of each object attribute characteristic and each business data attribute characteristic to obtain each target binary operation value; and taking each target binary operation value as a target fusion characteristic corresponding to the service data to be pushed.

Specifically, the server calculates a binary operation value of each object attribute feature and each service data attribute feature, for example, the server may calculate an inner product of the object attribute feature and the service data attribute feature to obtain a binary operation value, the server may calculate a sum of the object attribute feature and the service data attribute feature to obtain a binary operation value, and the server may calculate an average value of the object attribute feature and the service data attribute feature to obtain a binary operation value. And finally, the server takes each calculated target binary operation value as a target fusion characteristic corresponding to the service data to be pushed.

In the above embodiment, binary operation values of each target are obtained by calculating binary operation of attribute features of each object and attribute features of each service data; each target binary operation value is used as a target fusion characteristic corresponding to the service data to be pushed, so that the accuracy of the target fusion characteristic is improved

In a specific embodiment, as shown in fig. 4, in order to obtain a schematic diagram of the fusion feature and the target fusion feature, the object identifier corresponds to 314 object features, including object attribute features and object sequence features formed by each adjustment feature. The service data to be pushed corresponds to 64 features, including each service data attribute feature and the data feature to be pushed. Then, the server needs to calculate the inner product of the feature corresponding to each object identifier and the feature corresponding to each service data to be pushed, so as to obtain the inner product value between all the features, for example, the server needs to calculate the inner product between the object sequence feature and 64 features corresponding to the service data to be pushed respectively, and the server traverses to calculate the inner product between each object attribute feature and 64 features corresponding to the service data to be pushed. And finally, the server takes the inner product value among all the features as a final fusion feature, and then uses the final fusion feature to extract information, so that the extracted push reference information is more accurate. In a specific embodiment, in order to save computing performance, the server may also only calculate an inner product of the feature of the data to be pushed and the feature of the object sequence, so as to obtain a fusion feature, and then use the fusion feature to extract information.

In one embodiment, S304, that is, extracting information based on the splicing feature, obtains target push reference information of the service data to be pushed for the object identifier, including the steps of:

acquiring full connection parameters, and weighting the splicing characteristics by using the full connection parameters to obtain full connection characteristics; and activating the full-connection feature to obtain an activation feature, and normalizing the activation feature to obtain target pushing reference information of the service data to be pushed aiming at the object identification.

The full-connection parameters are parameters used in full-connection operation, and may be trained in advance, set in advance, or acquired from a server providing full-connection operation service.

Specifically, the server acquires the full-connection parameters, calculates the product of the full-connection parameters and the splicing characteristics to obtain full-connection characteristics, and then uses an activation function to perform activation calculation on the full-connection characteristics to obtain activation characteristics. And finally, normalizing the activated characteristic by using a normalization algorithm to obtain a normalized characteristic, and taking the normalized characteristic as target pushing reference information of the finally obtained service data to be pushed aiming at the object identifier by the server. In one embodiment, the server may input the splice feature into the trained fully connected neural network for calculation, and take the output result as the target push reference information.

In the above embodiment, the full connection parameters are obtained, the splicing characteristics are weighted by using the full connection parameters to obtain the full connection characteristics, then the full connection characteristics are activated to obtain the activation characteristics, and the activation characteristics are normalized to obtain the target push reference information of the service data to be pushed aiming at the object identifier, namely the target push parameter information is obtained through the full connection operation, so that the accuracy of the obtained target push reference information is improved.

In a specific embodiment, as shown in fig. 5, in order to obtain the network structure schematic diagram of the target push reference information, specifically: the server calculates the inner product of the features corresponding to the object identifiers and the features corresponding to the service data to be pushed, wherein the features corresponding to the object identifiers comprise object sequence features formed by the attribute features of each object and the adjustment features, the dimension size is B.314.64, namely 314 features corresponding to the object identifiers, the vector length of each feature is 64 dimensions, B represents the number of objects, and at the moment, only one object identifier is used, and B is 1. The features corresponding to the service data to be pushed comprise various service data attribute features and data features to be pushed, the dimension size is B88 by 64, namely 88 features corresponding to the service data to be pushed exist, and the vector length of each feature is 64 dimensions. Then, a fusion feature is calculated, and the dimension size of the fusion feature is B x 27632. And then inputting the fusion characteristics into a fully-connected neural network, and performing dimension reduction through fully-connected operation to obtain the characteristics after fully-connected operation, wherein the dimension is 1024. And finally, normalizing the features after the full-connection operation to obtain the pushing degree corresponding to the service data to be pushed, and carrying out the inner product of the features in pairs to improve the matching property of the service data to be pushed and the object corresponding to the object identifier, namely, the accuracy of the obtained fusion features is improved, and the accuracy of the pushing degree is further improved.

In one embodiment, S204, that is, extracting features corresponding to each of the interacted service data to obtain each of the interacted data features, includes the steps of:

extracting data characteristics corresponding to each interacted service data to obtain each initial data characteristic; and calculating the attention weights corresponding to the initial data features respectively according to the data features to be pushed, and weighting the corresponding initial data features according to the attention weights to obtain the interacted data features.

The initial data feature refers to an initial feature of the interacted service data, and is used for representing the feature of the interacted service data. Attention weight is used for representing the interest degree of the object identification corresponding object in the interacted service data, and the higher the attention weight is, the higher the interest degree in the interacted service data is.

Specifically, the server may perform feature conversion, such as vectorization conversion, on each interacted service data to obtain each initial data feature, and may perform feature extraction on each interacted service data through a neural network for feature extraction to obtain each initial data feature. And then calculating the attention weights corresponding to the initial data features respectively according to the data features to be pushed, wherein the server can calculate the correlation between the data features to be pushed and the initial data features, and determine the attention weights corresponding to the initial data features according to the correlation. The server then calculates the product of the attention weight and the corresponding initial data feature to obtain a weighted feature, i.e. the interacted data feature. The server weights all initial data features using corresponding attention weights to obtain each interacted data feature.

In the above embodiment, each initial data feature is obtained by extracting the data feature corresponding to each interacted service data; the attention weights corresponding to the initial data features are calculated according to the data features to be pushed, the corresponding initial data features are weighted according to the attention weights, and the interacted data features are obtained, namely, the initial data features are weighted through the attention weights, so that the interacted data features are obtained, and the accuracy of the interacted data features is improved.

In one embodiment, the calculating the attention weights corresponding to the initial data features according to the data features to be pushed includes the steps of:

splicing all the initial data features to obtain initial data sequence features, and calculating the transposition of the initial data sequence features to obtain transposition features; multiplying the transposed feature with the data feature to be pushed to obtain a correlation feature, carrying out mean pooling based on the correlation feature to obtain pooled features, and normalizing the pooled features to obtain the attention weights respectively corresponding to the initial data features.

The initial data sequence features are features obtained by splicing the initial data features in sequence according to a preset sequence. The correlation feature refers to a feature for characterizing the correlation between the feature of the data to be pushed and the feature of the initial data.

Specifically, the server sequentially splices all initial data features to obtain initial data sequence features, then calculates transposition of the initial data sequence features to obtain transposition features, calculates the product of the transposition features and the data features to be pushed to obtain correlation features, obtains the length corresponding to the initial data sequence features, carries out mean pooling on the correlation features according to the obtained length to obtain pooling features, and finally carries out normalization to obtain attention weights corresponding to all the initial data features.

In a specific embodiment, the following formulas may be used to derive the various interacted data characteristics.

Where u refers to the sequence feature of each interacted data feature. Q is the feature of the data to be pushed, V is the feature of the initial data sequence, d _V Refers to the vector length of the original data sequence features. softmax refers to the normalization function.

In one embodiment, the server may calculate the corresponding attention weights using the respective initial data features, specifically: the server determines a current initial data feature and initial data features except the current initial data feature from the initial data features, splices the initial data features except the current initial data feature to obtain a current sequence feature, calculates the transposition of the current sequence feature to obtain a current transposition feature, calculates the product of the current initial data feature and the current transposition feature to obtain a current correlation feature, and carries out mean value pooling and normalization operation on the current correlation feature to obtain the attention weight corresponding to the current initial data feature. And then the server traverses each initial data feature in turn to obtain the attention weight corresponding to each initial data feature.

In the above embodiment, by splicing the initial data features, an initial data sequence feature is obtained, and a transpose of the initial data sequence feature is calculated, so as to obtain a transpose feature; multiplying the transposed feature with the data feature to be pushed to obtain a correlation feature, carrying out mean pooling based on the correlation feature to obtain pooled features, normalizing the pooled features to obtain attention weights respectively corresponding to the initial data features, namely determining the attention weights by calculating the correlation feature, thereby improving the accuracy of the obtained attention weights.

splicing the data features to be pushed with each initial data feature to obtain data splicing features; acquiring a correlation feature extraction parameter, and carrying out feature extraction on the data splicing features according to the correlation feature extraction parameter to obtain data correlation features; and normalizing the data correlation characteristics to obtain the attention weights respectively corresponding to the initial data characteristics.

The correlation feature extraction parameter refers to a parameter for extracting correlation between features, and the correlation feature extraction parameter may be trained in advance, set in advance, or acquired from a service provider providing a data service in advance. The data correlation feature refers to a feature which is extracted by using a correlation feature extraction parameter and is used for representing the correlation between the data feature to be pushed and the initial data feature.

Specifically, the server splices the data feature to be pushed with each initial data feature, wherein the data feature to be pushed can be spliced with each initial data feature as a head part, or can be spliced with each initial data feature as a tail part, so as to obtain a data splicing feature. The server may then obtain the correlation feature extraction parameters from the database, the correlation feature extraction parameters from the server providing the data service, and the correlation feature extraction parameters uploaded by the terminal. And then the server uses the correlation characteristic extraction parameters to perform correlation calculation on the data characteristics to be pushed in the data splicing characteristics and each initial data characteristic, so as to obtain data correlation characteristics. And finally, normalizing the data correlation characteristics by the server to obtain the attention weights respectively corresponding to the initial data characteristics. In one embodiment, the server may splice the data feature to be pushed with each initial data feature to obtain each spliced feature, sequentially extract each spliced feature by using a correlation feature extraction parameter to obtain a correlation feature corresponding to each spliced feature, and normalize all the correlation features to obtain the attention weight corresponding to each initial data feature.

In one embodiment, the server may splice the data feature to be pushed with each initial data feature to obtain each spliced feature, and then input each spliced feature into an attention weight extraction network, where the attention weight extraction network is a neural network that is trained in advance and is used for extracting attention weights. And extracting correlation characteristics corresponding to each spliced characteristic through correlation characteristic extraction parameters in the attention weight extraction network, and normalizing all the correlation characteristics to obtain the attention weight corresponding to each initial data characteristic.

In a specific embodiment, as shown in fig. 6, a network architecture diagram for obtaining each of the interacted data features is specifically: the server inputs the respective initial data features and the data features to be pushed into an attention unit, which may be a pre-trained neural network unit for extracting attention weights. The attention unit obtains each interacted data feature by calculating the correlation between the data feature to be pushed and each initial data feature, determining the attention weight corresponding to each initial data feature based on the correlation, and then carrying out attention weighting on the corresponding initial data feature by using the attention weight corresponding to each initial data feature.

In the above embodiment, the data correlation feature is obtained by obtaining the correlation feature extraction parameter and performing feature extraction on the data splicing feature according to the correlation feature extraction parameter, and the data correlation feature is normalized to obtain the attention weight corresponding to each initial data feature, that is, the attention weight is extracted by using the correlation feature extraction parameter, so that the attention weight obtaining efficiency is improved.

In a specific embodiment, as shown in fig. 7, a network architecture diagram of fusion feature extraction is provided, where N (N is a positive integer) initial data features are acquired, and each initial data feature is weighted by a corresponding attention weight to obtain N interacted data features. The N interacted data features are then input into a feature interaction network, which is a pre-trained neural network for implicit feature interaction of the input features, which may be a fully connected neural network, such as an MLP (Multi-layer perceptron) neural network. The server obtains N interactive features output by the feature interactive network, and then inputs the N interactive features into a fusion guide network, wherein the fusion guide network is a pre-trained neural network and is used for adjusting the input features by using network parameters, so that the obtained adjustment has more information. The converged guidance network may be a fully connected neural network, for example, an MLP neural network. The server obtains N adjustment features output by the fusion guide network, then calculates an inner product value between each adjustment feature and data adjustment to be pushed, so as to obtain N inner product values, obtains fusion features according to the N inner product values, and enables the obtained fusion features to have rich information, so that accuracy of the obtained fusion features is improved.

In one embodiment, the service data pushing method further includes the steps of:

inputting service data to be pushed and each interacted service data into a service data pushing model, extracting the characteristics corresponding to the service data to be pushed through the service data pushing model to obtain the characteristics of the data to be pushed, and extracting the characteristics corresponding to each interacted service data to obtain the characteristics of each interacted data;

acquiring characteristic interaction parameters through a service data push model, carrying out characteristic interaction on each interacted data characteristic according to the characteristic interaction parameters to obtain interaction characteristics respectively corresponding to each interacted data characteristic, acquiring fusion guide parameters, and adjusting each interaction characteristic according to the fusion guide parameters to obtain adjustment characteristics respectively corresponding to each interaction characteristic;

fusing each adjustment feature with the data feature to be pushed through the service data pushing model to obtain a fusion feature corresponding to the service data to be pushed;

and extracting information based on the fusion characteristics through a service data pushing model to obtain pushing reference information of the service data to be pushed aiming at the object identification.

The service data pushing model is a neural network model for determining pushing parameter information corresponding to service data to be pushed, and the service data pushing model can be obtained by training an initialized neural network by using training data.

Specifically, the server trains the training data in advance to obtain a service data pushing model, and then deploys the service data pushing model. When the service data pushing model needs to be used, the deployed service data pushing model can be called for use. The method comprises the steps that a server inputs service data to be pushed and interactive service data into a service data pushing model, the service data pushing model extracts characteristics of input data, the extracted characteristics are updated by using trained characteristic interaction parameters and fusion guide parameters to obtain updated characteristics, finally the updated characteristics are fused to obtain fusion characteristics, and information extraction is carried out by using the fusion characteristics to obtain pushing reference information of the service data to be pushed for object identification. And finally, the server judges whether the service data to be pushed needs pushing or not according to the pushing reference information. In one embodiment, the server may send the service data to be pushed and each interacted service data to a service side providing the model service, and then the service side providing the model service performs forward calculation through the deployed service data pushing model to obtain pushing reference information of the service data to be pushed for the object identifier. The server acquires push reference information of the service data to be pushed, which is returned by a server side providing the model service, aiming at the object identification. In one embodiment, the server may implement the steps of any one of the foregoing embodiments through a service data push model to obtain push parameter information corresponding to service data to be pushed.

In the embodiment, the service data to be pushed and the interacted service data are input into the service data pushing model, and then the service data pushing model performs forward calculation to obtain the output pushing reference information of the service data to be pushed aiming at the object identifier, so that the efficiency of obtaining the pushing reference information is improved. And then judging whether the service data to be pushed needs pushing or not according to the pushing reference information, namely, using the trained service data pushing model to push the service data, thereby improving the service data pushing efficiency.

In one embodiment, the traffic data push model includes: the system comprises a feature extraction network, a feature interaction network, a convergence guiding network, a feature convergence network and an information extraction network; as shown in fig. 8, the service data pushing method further includes:

s802, extracting the corresponding features of the service data to be pushed through a feature extraction network to obtain the features of the data to be pushed, and extracting the corresponding features of each interacted service data to obtain the features of each interacted data;

s804, carrying out feature interaction on each interacted data feature through feature interaction parameters in a feature interaction network to obtain interaction features respectively corresponding to each interacted data feature, and adjusting each interaction feature through fusion guidance parameters in a fusion guidance network to obtain adjustment features respectively corresponding to each interaction feature;

S806, fusing each adjustment feature with the feature of the data to be pushed through a feature fusion network to obtain fusion features corresponding to the service data to be pushed;

s808, extracting information from the fusion features through an information extraction network to obtain push reference information of the service data to be pushed aiming at the object identification.

Wherein the feature extraction network is a neural network for performing feature extraction. The feature interaction network is a neural network for implicit feature interactions. The fusion guidance network is a neural network for guiding feature fusion, and the feature fusion network is a neural network for fusing features. The information extraction network is a neural network for extracting push reference information. The neural network may be a convolutional neural network, a recurrent neural network, a time-sequential neural network, a graph neural network, or the like.

Specifically, the service data pushing model of the server may include a feature extraction network, a feature interaction network, a convergence guiding network, a feature convergence network and an information extraction network. And then the feature extraction network is used for extracting the features of the output data, namely the features of the data to be pushed and the features of the interacted data. In one embodiment, the feature extraction network may include an initial feature extraction sub-network and an attention feature extraction sub-network, where the initial feature extraction sub-network is used to extract the data features to be pushed and initial data features corresponding to each interacted service data, and then the initial data features are input into the attention feature extraction sub-network to perform attention weighting, so as to extract the weighted attention features, that is, each interacted data feature. And then the server performs feature interaction on each extracted interacted data feature through a feature interaction network, adjusts through a fusion guiding network to obtain each adjustment feature, fuses each adjustment feature and the data feature to be pushed through a feature fusion network to obtain a fusion feature, and finally inputs the fusion feature into an information extraction network to extract information to obtain output pushing reference information.

In the above embodiment, the push reference information of the service data to be pushed for the object identifier is extracted through the feature extraction network, the feature interaction network, the fusion guiding network, the feature fusion network and the information extraction network in the service data push model, so that the accuracy of the obtained push reference information is improved.

In one embodiment, as shown in fig. 9, the training of the business data push model includes the steps of:

s902, acquiring training data to be pushed, pushing labels and training interacted service data corresponding to the training object identifiers.

The training data to be pushed refers to data to be pushed used in training, and the pushing label refers to a label which corresponds to the training data to be pushed and is pushed or not, wherein the label comprises a pushed label and an un-pushed label. The training object identification refers to an object identification corresponding to training data to be pushed. Training interacted service data refers to training interacted service data corresponding to the object identifier.

Specifically, the server may obtain the training data to be pushed and the training interacted service data corresponding to the training object identifier from the database, and obtain the interaction behavior information of the training object identifier on the training data to be pushed to determine the push label, for example, when the object corresponding to the training object identifier does not have interaction behavior on the training data to be pushed, the push label may be an unpuncrushed label, and when the object corresponding to the training object identifier does have interaction behavior on the training data to be pushed, the push label may be a pushed label. The server may also obtain the push tag directly from the database. The server can also acquire the training data to be pushed, the push label and the training interactive service data corresponding to the training object identifier from the service side providing the push service. The server can also acquire training data to be pushed, a pushing label and training interactive service data corresponding to the training object identifier uploaded by the terminal.

S904, inputting the training data to be pushed and the interactive service data of each training into an initial service data pushing model for forward calculation to obtain training pushing reference information corresponding to the training data to be pushed.

The initial service data push model refers to a service data push model initialized by model parameters, and the model parameter initialization may include zero initialization, random initialization, gaussian distribution initialization and the like. The training pushing reference information refers to pushing reference information of training data to be pushed, which is obtained during training, aiming at the training object identification.

Specifically, the server inputs training data to be pushed and each training interactive service data into an initial service data pushing model, the initial service data pushing model extracts the characteristics of the training data to be pushed and the characteristics of each training interactive service data to obtain the training data to be pushed and the training interactive data characteristics, the training interactive data characteristics are adjusted and interacted by using initialized characteristic interaction parameters to obtain the training interactive characteristics, the training interactive characteristics are adjusted by using initialized fusion guide parameters to obtain the training adjustment characteristics, the training adjustment characteristics are fused with the training data characteristics to be pushed to obtain the training fusion characteristics, and finally information extraction is carried out by using the training fusion characteristics to obtain the output training pushing reference information.

S906, carrying out loss calculation based on the training pushing reference information and the pushing label to obtain training loss information, and reversely updating the initial service data pushing model according to the training loss information to obtain an updated service data pushing model.

The training loss information is used for training errors between the push reference information and the push label, and when the training loss information is smaller, the training loss information indicates that the closer the training push reference information is to the push label, the more accurate the service data push model obtained through training is. The updated service data pushing model refers to the service data pushing model with updated model parameters.

Specifically, the server calculates an error between the training push reference information and the push label by using a preset loss function, so as to obtain training loss information, wherein the loss function can be a cross entropy loss function, a mean square error loss function, a minimum absolute value error loss function and the like. The server then uses a gradient descent algorithm, which may be a random place descent algorithm, a batch gradient descent algorithm, a momentum gradient descent algorithm, or the like, to reverse update the initialization parameters in the initial business data push model according to the training loss information, thereby obtaining an updated business data push model.

S908, taking the updated service data pushing model as an initial service data pushing model, and returning to the step of obtaining the training data to be pushed, the pushing label and the training interactive service data corresponding to the training object identifier for iterative execution until the training completion condition is reached, and obtaining the service data pushing model.

The training completion condition refers to a condition that a service data push model is obtained through training, and the training completion condition comprises, but is not limited to, the maximum number of training iterations, no change of model parameters, the fact that model loss information reaches a preset threshold value, and the like.

Specifically, the server judges whether the model loss information reaches a preset threshold, when the model loss information does not reach the preset threshold, the server takes the updated service data pushing model as an initial service data pushing model, and returns to the iterative execution of the steps of acquiring the training to-be-pushed data, the pushing label and the interactive service data of each training corresponding to the training object identifier, until the model loss information reaches the preset threshold, the initial service data pushing module when the model loss information reaches the preset threshold is taken as a final training to obtain the service data pushing model.

In the above embodiment, through obtaining the training data to be pushed, the push label and the training interactive service data corresponding to the training object identifier, then training the initial service data push model until reaching the training completion condition, obtaining the service data push model, that is, obtaining the service data push model through pre-training, so that the service data push model is convenient for subsequent direct use, and the service data push efficiency is improved.

In a specific embodiment, as shown in fig. 10, a schematic model architecture of a service data push model is provided, specifically: the method comprises the steps that a server acquires all initial data features and data features to be pushed, all initial data features are input into an attention network of a fusion neural network to carry out attention weighting, all interacted data features after weighting are obtained, all interacted data features are input into a feature interaction network to carry out feature interaction, the feature interaction network can be an MLP network, all output interaction features are obtained, namely the features are subjected to implicit intersection through the feature interaction network, the learning capacity of a model on the features is improved, and information loss is reduced. And then inputting each interaction characteristic into a fusion guidance network for adjustment, wherein the fusion guidance network can be an MLP network to obtain each output adjustment characteristic, namely, model parameters can be introduced before characteristic fusion is carried out through the fusion guidance network, and extraction guidance is carried out on the characteristic fusion through the model parameters, namely, the interaction characteristic is adjusted to obtain each adjustment characteristic, so that the accuracy of the characteristic fusion can be improved. And then calculating inner products between each adjustment feature and the data feature to be pushed respectively to obtain fusion features, obtaining each object attribute feature and each service data attribute feature, and calculating inner products between each object attribute feature and each service data attribute feature respectively to obtain target fusion features. And then the server splices the fusion characteristic and the target fusion characteristic to obtain a splicing characteristic, inputs the splicing characteristic into a fully-connected neural network to perform fully-connected operation, normalizes the splicing characteristic to obtain the pushing degree of the service data to be pushed aiming at the object identifier, and pushes the service data to be pushed to the object terminal corresponding to the object identifier when the pushing degree exceeds a preset pushing degree threshold. In a specific embodiment, the server may perform a comparison test on the performance of the service data push model. FIG. 11 is a graph showing comparison of test performance metrics at various time points, wherein the test performance metrics can be evaluated using an AUC (area under the curve) metric. The solid line represents the AUC index test value of the service data pushing model for carrying out service data pushing at different time points, and the dotted line represents the AUC index test value for directly fusing each initial data characteristic and the data characteristic to be pushed and carrying out service data pushing. The AUC index value of the service data pushing by using the service data pushing model is obviously improved, namely the service data pushing is performed by using the service data pushing model, so that the accuracy of the service data pushing is improved, and the waste of pushing resources is avoided.

acquiring a commodity video pushing request, wherein the commodity video pushing carries an object identifier, acquiring commodity videos to be pushed based on the commodity video pushing request, and acquiring each interacted commodity video based on the object identifier; extracting features corresponding to the commodity video to be pushed to obtain data features to be pushed, and extracting features corresponding to each interacted commodity video to obtain each interacted data feature; the method comprises the steps of obtaining feature interaction parameters, carrying out feature interaction on each interacted data feature according to the feature interaction parameters to obtain interaction features corresponding to each interacted data feature, obtaining fusion guide parameters, and adjusting each interaction feature according to the fusion guide parameters to obtain adjustment features corresponding to each interaction feature; fusing each adjustment feature with the data feature to be pushed to obtain a fusion feature corresponding to the commodity video to be pushed; and extracting information based on the fusion characteristics to obtain pushing reference information of the commodity video to be pushed aiming at the object identifier, wherein the pushing reference information is used for sending the commodity video to be pushed to the object terminal corresponding to the object identifier when the preset pushing condition is met.

The commodity video pushing request refers to a request for pushing a commodity video, the commodity video to be pushed refers to a commodity video which needs to be judged whether to be pushed to an object terminal corresponding to an object identifier, and the commodity video can be an introduction video of a commodity. The interactive commodity video refers to a video of a commodity of which the object identifies the corresponding object to which the interaction occurs, and for example, the video can be a browsed commodity video, a commented commodity video, a praise commodity video, a forwarded commodity video and the like.

Specifically, the server acquires a commodity video pushing request, the commodity video pushing carries an object identifier, acquires commodity videos to be pushed based on the commodity video pushing request, and acquires each interacted commodity video based on the object identifier. And then the server can input the commodity video to be pushed and each interacted commodity video into the business data push type to perform forward calculation, obtain the push degree of the outputted commodity video to be pushed aiming at the object identification, and then send the commodity video to be pushed to the object terminal corresponding to the object identification when the push degree exceeds a preset threshold.

In the embodiment, by acquiring the commodity video pushing request, acquiring the commodity video to be pushed based on the commodity video pushing request, acquiring each interacted commodity video based on the object identification, extracting the fusion feature by using the commodity video to be pushed and each interacted commodity video, improving the accuracy of the obtained fusion feature, acquiring the pushing reference information of the commodity video to be pushed for the object identification by using the fusion feature, and when the pushing reference information accords with the preset pushing condition, sending the commodity video to be pushed to the object terminal corresponding to the object identification, thereby improving the accuracy of commodity video pushing.

In a specific embodiment, as shown in fig. 12, a flow chart of service data pushing is provided, and the flow chart is executed by a computer device, and specifically includes the following steps:

s1202, acquiring a service data pushing request, wherein the service data pushing request carries an object identifier, acquiring service data to be pushed based on the service data pushing request, acquiring each interacted service data based on the object identifier, and inputting the service data to be pushed and each interacted service data into a service data pushing model.

And S1204, extracting the corresponding features of the service data to be pushed through a feature extraction network to obtain the features of the data to be pushed, extracting the corresponding data features of each interacted service data to obtain each initial data feature, calculating the attention weights corresponding to each initial data feature, and weighting the corresponding initial data features according to the attention weights to obtain each interacted data feature.

S1206, performing full-connection operation on each interacted data feature through the feature interaction parameter in the feature interaction network to obtain each interaction full-connection feature, and performing nonlinear activation on each interaction full-connection feature to obtain interaction features respectively corresponding to each interacted data feature.

S1208, performing full-connection operation on each interaction feature through fusion guide parameters in the fusion guide network to obtain each fusion guide full-connection feature, and performing nonlinear activation on each fusion guide full-connection feature to obtain adjustment features respectively corresponding to each interaction feature.

S1210, calculating binary operation values of each adjustment feature and the feature of the data to be pushed through a feature fusion network to obtain each binary operation value, and taking each binary operation value as a fusion feature corresponding to the service data to be pushed.

S1212, each object attribute feature corresponding to the object identifier is obtained, and each service data attribute feature corresponding to the service data to be pushed is obtained. And calculating binary operation of each object attribute feature and each service data attribute feature through a feature fusion network to obtain each target binary operation value, and taking each target binary operation value as a target fusion feature corresponding to the service data to be pushed.

S1214, splicing the target fusion feature and the fusion feature to obtain a splicing feature, obtaining a full-connection parameter, weighting the splicing feature by using the full-connection parameter to obtain the full-connection feature, activating the full-connection feature to obtain an activation feature, and normalizing the activation feature to obtain target pushing reference information of the service data to be pushed aiming at the object identifier.

S1216, when the target pushing reference information accords with a preset pushing condition, pushing the service data to be pushed to the object terminal corresponding to the object identifier.

In the embodiment, the service data to be pushed and the interacted service data are input into the service data pushing model, and the service data pushing model is used for calculating the target pushing reference information of the service data to be pushed aiming at the object identification, so that the accuracy of the obtained target pushing reference information is improved. And then pushing the service data to be pushed to the object terminal corresponding to the object identifier when the target pushing reference information accords with a preset pushing condition, thereby improving the accuracy of service data pushing.

In a specific embodiment, the service data pushing method is applied to a video platform, specifically: the video platform acquires a video pushing request sent by the user terminal, and acquires each video to be pushed according to the video pushing request, wherein the video to be pushed can be an advertisement video, a commodity video, a film video, a short video and the like. And then acquiring each push video interacted with in history according to the user identification. The video platform inputs each historical interacted push video and each video to be pushed into a video push model deployed by the video platform, extracts each interacted video feature by using feature interaction parameters and fusion guide parameters, fuses each interacted video feature with the video feature to be pushed, and finally extracts the push degree corresponding to the video to be pushed by using the fused feature, so as to obtain the push degree corresponding to the output video to be pushed aiming at the user identifier. And then comparing the pushing degree of each video to be pushed with a preset threshold value by the video platform, acquiring the video with the pushing degree exceeding the preset threshold value from each video to be pushed, and finally pushing the video with the pushing degree exceeding the preset threshold value to a user terminal by the video platform, wherein the user terminal receives and displays the pushed video, so that the accuracy of pushing the video by the video platform can be improved, and the pushing resources of the video platform can be saved.

In a specific embodiment, the service data pushing method is applied to an online shopping platform, specifically: the online shopping platform acquires an interested article pushing request sent by a user terminal, acquires various article data to be pushed according to the interested article pushing request, wherein the article data can be saving information of articles, such as pictures of the articles, videos of the articles, selling prices of the articles and the like, and acquires various interacted article data according to user identification, such as purchased article data, browsed article data and the like. And inputting the commodity data to be pushed and the interacted commodity data into a commodity data pushing model deployed by the online shopping platform in sequence, extracting the characteristics of the interacted commodity data by using characteristic interaction parameters and fusion guide parameters, fusing the characteristics of the interacted commodity data with the characteristics of the commodity data to be pushed, extracting the pushing degree corresponding to the commodity data to be pushed by using the fused characteristics, namely obtaining the pushing degree corresponding to the user identification of the commodity data to be pushed, which is output by the commodity data pushing model, comparing the pushing degree of the commodity data to be pushed with a preset threshold value, acquiring the commodity data with the pushing degree exceeding the preset threshold value from the commodity data to be pushed, pushing the commodity data with the pushing degree exceeding the preset threshold value to a user terminal, and receiving and displaying the pushed commodity data by the user terminal, so that the accuracy of the commodity data pushed by the online shopping platform can be improved, and the pushing resources of the online shopping platform can be saved.

It should be understood that, although the steps in the flowcharts related to the embodiments described above are sequentially shown as indicated by arrows, these steps are not necessarily sequentially performed in the order indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps in the flowcharts described in the above embodiments may include a plurality of steps or a plurality of stages, which are not necessarily performed at the same time, but may be performed at different times, and the order of the steps or stages is not necessarily performed sequentially, but may be performed alternately or alternately with at least some of the other steps or stages.

Based on the same inventive concept, the embodiment of the application also provides a service data pushing device for implementing the service data pushing method. The implementation of the solution provided by the device is similar to the implementation described in the above method, so the specific limitation in the embodiments of one or more service data pushing devices provided below may refer to the limitation of the service data pushing method in the above description, which is not repeated here.

In one embodiment, as shown in fig. 13, there is provided a service data pushing apparatus 1300, including: a data acquisition module 1302, a feature extraction module 1304, a feature adjustment module 1306, a feature fusion module 1308, and a data push module 1310, wherein:

the data acquisition module 1302 is configured to acquire a service data push request, where the service data push request carries an object identifier, acquire service data to be pushed based on the service data push request, and acquire each interacted service data based on the object identifier;

the feature extraction module 1304 is configured to extract features corresponding to the service data to be pushed, obtain features of the data to be pushed, and extract features corresponding to each piece of interacted service data, so as to obtain features of each piece of interacted data;

the feature adjustment module 1306 is configured to obtain feature interaction parameters, perform feature interaction on each interacted data feature according to the feature interaction parameters, obtain interaction features corresponding to each interacted data feature, obtain fusion guide parameters, and adjust each interaction feature according to the fusion guide parameters to obtain adjustment features corresponding to each interaction feature;

the feature fusion module 1308 is configured to fuse each adjustment feature with a feature of data to be pushed, so as to obtain a fusion feature corresponding to the service data to be pushed;

The data pushing module 1310 is configured to extract information based on the fusion feature, obtain pushing reference information of the to-be-pushed service data for the object identifier, where the pushing reference information is used to push the to-be-pushed service data to the object terminal corresponding to the object identifier when the to-be-pushed service data meets a preset pushing condition.

In one embodiment, the feature adjustment module 1306 is further configured to perform a full-connection operation on each of the interacted data features using the feature interaction parameters to obtain each of the interacted full-connection features; and carrying out nonlinear activation on each interactive full-connection feature to obtain interactive features corresponding to each interactive data feature.

In one embodiment, the feature adjustment module 1306 is further configured to perform a full-connection operation on each interaction feature using the fusion guidance parameter to obtain each fusion guidance full-connection feature; and carrying out nonlinear activation on all the fusion guide full-connection features to obtain adjustment features corresponding to all the interaction features.

In one embodiment, the feature fusion module 1308 is further configured to calculate binary operation values of each adjustment feature and the feature of the data to be pushed, so as to obtain each binary operation value; and taking each binary operation value as a fusion characteristic corresponding to the service data to be pushed.

In one embodiment, the data pushing module 1310 is further configured to obtain each object attribute feature corresponding to the object identifier, and obtain each service data attribute feature corresponding to the service data to be pushed; fusing the attribute characteristics of each object with the attribute characteristics of each service data to obtain target fusion characteristics corresponding to the service data to be pushed; and splicing the target fusion characteristics with the fusion characteristics to obtain splicing characteristics, and extracting information based on the splicing characteristics to obtain target pushing reference information of the service data to be pushed aiming at the object identification.

In one embodiment, the data pushing module 1310 is further configured to calculate binary operations of each object attribute feature and each service data attribute feature, so as to obtain each target binary operation value; and taking each target binary operation value as a target fusion characteristic corresponding to the service data to be pushed.

In one embodiment, the data pushing module 1310 is further configured to obtain a full connection parameter, and weight the spliced feature by using the full connection parameter to obtain a full connection feature; and activating the full-connection feature to obtain an activation feature, and normalizing the activation feature to obtain target pushing reference information of the service data to be pushed aiming at the object identification.

In one embodiment, feature extraction module 1304 includes:

the initial feature extraction unit is used for extracting data features corresponding to the interacted service data to obtain initial data features;

the attention feature extraction unit is used for calculating the attention weights corresponding to the initial data features respectively according to the data features to be pushed, and weighting the corresponding initial data features according to the attention weights to obtain the interacted data features.

In one embodiment, the attention feature extraction unit is further configured to splice each initial data feature to obtain an initial data sequence feature, and calculate a transpose of the initial data sequence feature to obtain a transpose feature; multiplying the transposed feature with the data feature to be pushed to obtain a correlation feature, carrying out mean pooling based on the correlation feature to obtain pooled features, and normalizing the pooled features to obtain the attention weights respectively corresponding to the initial data features.

In one embodiment, the attention feature extraction unit is further configured to splice the data feature to be pushed with each initial data feature to obtain a data splice feature; acquiring a correlation feature extraction parameter, and carrying out feature extraction on the data splicing features according to the correlation feature extraction parameter to obtain data correlation features; and normalizing the data correlation characteristics to obtain the attention weights respectively corresponding to the initial data characteristics.

In one embodiment, the service data pushing apparatus 1300 further includes:

the model pushing module is used for inputting the service data to be pushed and the interacted service data into the service data pushing model, extracting the corresponding characteristics of the service data to be pushed through the service data pushing model to obtain the characteristics of the data to be pushed, and extracting the corresponding characteristics of the interacted service data to obtain the characteristics of the interacted data; acquiring characteristic interaction parameters through a service data push model, carrying out characteristic interaction on each interacted data characteristic according to the characteristic interaction parameters to obtain interaction characteristics respectively corresponding to each interacted data characteristic, acquiring fusion guide parameters, and adjusting each interaction characteristic according to the fusion guide parameters to obtain adjustment characteristics respectively corresponding to each interaction characteristic; fusing each adjustment feature with the data feature to be pushed through the service data pushing model to obtain a fusion feature corresponding to the service data to be pushed; and extracting information based on the fusion characteristics through a service data pushing model to obtain pushing reference information of the service data to be pushed aiming at the object identification.

In one embodiment, the traffic data push model includes: the system comprises a feature extraction network, a feature interaction network, a convergence guiding network, a feature convergence network and an information extraction network; the service data pushing apparatus 1300 further includes:

The network pushing module is used for extracting the characteristics corresponding to the service data to be pushed through the characteristic extraction network to obtain the characteristics of the data to be pushed, extracting the characteristics corresponding to each interactive service data respectively and obtaining the characteristics of each interactive data; performing feature interaction on each interacted data feature through feature interaction parameters in a feature interaction network to obtain interaction features corresponding to each interacted data feature, and adjusting each interaction feature through fusion guidance parameters in a fusion guidance network to obtain adjustment features corresponding to each interaction feature; fusing each adjustment feature with the feature of the data to be pushed through a feature fusion network to obtain fusion features corresponding to the service data to be pushed; and extracting information from the fusion characteristics through an information extraction network to obtain push reference information of the service data to be pushed aiming at the object identification.

In one embodiment, the service data pushing apparatus 1300 further includes:

the model training module is used for acquiring training data to be pushed, pushing labels and training interactive service data corresponding to the training object identifiers; inputting training data to be pushed and each training interacted service data into an initial service data pushing model to perform forward calculation, and obtaining training pushing reference information corresponding to the training data to be pushed; performing loss calculation based on the training pushing reference information and the pushing label to obtain training loss information, and reversely updating an initial service data pushing model according to the training loss information to obtain an updated service data pushing model; and taking the updated service data pushing model as an initial service data pushing model, and returning to the step of obtaining the training data to be pushed, the pushing label and the interactive service data of each training corresponding to the training object identifier for iterative execution until the training completion condition is reached, so as to obtain the service data pushing model.

In one embodiment, the service data pushing apparatus 1300 further includes:

the commodity video pushing module is used for acquiring commodity video pushing requests, pushing commodity videos to be pushed based on the commodity video pushing requests, and acquiring all interacted commodity videos based on the object identifications, wherein the commodity video pushing requests carry object identifications; extracting features corresponding to the commodity video to be pushed to obtain data features to be pushed, and extracting features corresponding to each interacted commodity video to obtain each interacted data feature; the method comprises the steps of obtaining feature interaction parameters, carrying out feature interaction on each interacted data feature according to the feature interaction parameters to obtain interaction features corresponding to each interacted data feature, obtaining fusion guide parameters, and adjusting each interaction feature according to the fusion guide parameters to obtain adjustment features corresponding to each interaction feature; fusing each adjustment feature with the data feature to be pushed to obtain a fusion feature corresponding to the commodity video to be pushed; and extracting information based on the fusion characteristics to obtain pushing reference information of the commodity video to be pushed aiming at the object identifier, wherein the pushing reference information is used for sending the commodity video to be pushed to the object terminal corresponding to the object identifier when the preset pushing condition is met.

The modules in the service data pushing device can be realized in whole or in part by software, hardware and a combination thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.

In one embodiment, a computer device is provided, which may be a server, and the internal structure of which may be as shown in fig. 14. The computer device includes a processor, a memory, and a network interface connected by a system bus. Wherein the processor of the computer device is configured 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, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database of the computer device is used for storing service data to be pushed, interacted service data, object identification and the like. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program, when executed by a processor, implements a service data push method.

In one embodiment, a computer device is provided, which may be a terminal, and an internal structure diagram thereof may be as shown in fig. 15. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected by a system bus. Wherein the processor of the computer device is configured 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 and a computer program. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The communication interface of the computer device is used for carrying out wired or wireless communication with an external terminal, and the wireless mode can be realized through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. The computer program, when executed by a processor, implements a service data push method. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, can also be keys, a track ball or a touch pad arranged on the shell of the computer equipment, and can also be an external keyboard, a touch pad or a mouse and the like.

It will be appreciated by those skilled in the art that the structures shown in fig. 14 or 15 are merely block diagrams of portions of structures related to the aspects of the present application and are not intended to limit the computer devices to which the aspects of the present application may be applied, and that a particular computer device may include more or fewer components than shown, or may combine certain components, or may have a different arrangement of components.

In an embodiment, there is also provided a computer device comprising a memory and a processor, the memory having stored therein a computer program, the processor implementing the steps of the method embodiments described above when the computer program is executed.

In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored which, when executed by a processor, carries out the steps of the method embodiments described above.

In an embodiment, a computer program product is provided, comprising a computer program which, when executed by a processor, implements the steps of the method embodiments described above.

It should be noted that, user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, presented data, etc.) referred to in the present application are information and data authorized by the user or sufficiently authorized by each party.

Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, database, or other medium used in the various embodiments provided herein may include at least one of non-volatile and volatile memory. The nonvolatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, high density embedded nonvolatile Memory, resistive random access Memory (ReRAM), magnetic random access Memory (Magnetoresistive Random Access Memory, MRAM), ferroelectric Memory (Ferroelectric Random Access Memory, FRAM), phase change Memory (Phase Change Memory, PCM), graphene Memory, and the like. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory, and the like. By way of illustration, and not limitation, RAM can be in the form of a variety of forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), and the like. The databases referred to in the various embodiments provided herein may include at least one of relational databases and non-relational databases. The non-relational database may include, but is not limited to, a blockchain-based distributed database, and the like. The processors referred to in the embodiments provided herein may be general purpose processors, central processing units, graphics processors, digital signal processors, programmable logic units, quantum computing-based data processing logic units, etc., without being limited thereto.

The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.

The above examples only represent a few embodiments of the present application, which are described in more detail and are not to be construed as limiting the scope of the present application. It should be noted that it would be apparent to those skilled in the art that various modifications and improvements could be made without departing from the spirit of the present application, which would be within the scope of the present application. Accordingly, the scope of protection of the present application shall be subject to the appended claims.

Claims

1. A method for pushing service data, the method comprising:

extracting the corresponding characteristics of the service data to be pushed to obtain the characteristics of the data to be pushed, and extracting the corresponding characteristics of each interactive service data to obtain the characteristics of each interactive data;

Acquiring feature interaction parameters, carrying out feature interaction on the interacted data features according to the feature interaction parameters to obtain interaction features corresponding to the interacted data features, acquiring fusion guide parameters, and adjusting the interaction features according to the fusion guide parameters to obtain adjustment features corresponding to the interaction features;

fusing each adjustment feature with the data feature to be pushed to obtain a fusion feature corresponding to the service data to be pushed;

and extracting information based on the fusion characteristics to obtain push reference information of the service data to be pushed aiming at the object identifier, wherein the push reference information is used for pushing the service data to be pushed to the object terminal corresponding to the object identifier when a preset push condition is met.

2. The method according to claim 1, wherein the performing feature interaction on the respective interacted data features according to the feature interaction parameters to obtain interaction features respectively corresponding to the respective interacted data features includes:

performing full connection operation on the interactive data features by using the feature interaction parameters to obtain interactive full connection features;

And performing nonlinear activation on the interactive full-connection features to obtain interactive features respectively corresponding to the interactive data features.

3. The method according to claim 1, wherein the adjusting each interaction feature according to the fused guiding parameter to obtain an adjustment feature corresponding to each interaction feature includes:

performing full-connection operation on the interaction features by using the fusion guide parameters to obtain full-connection features of the fusion guide;

and carrying out nonlinear activation on the fusion guide full-connection features to obtain adjustment features respectively corresponding to the interaction features.

4. The method of claim 1, wherein the fusing each adjustment feature with the feature of the data to be pushed to obtain a fused feature corresponding to the service data to be pushed, includes:

calculating binary operation values of each adjustment feature and the data feature to be pushed respectively to obtain each binary operation value;

and taking each binary operation value as a fusion characteristic corresponding to the service data to be pushed.

5. The method of claim 1, wherein the extracting information based on the fusion feature to obtain the push reference information of the service data to be pushed for the object identifier includes:

Acquiring each object attribute characteristic corresponding to the object identifier, and acquiring each service data attribute characteristic corresponding to the service data to be pushed;

fusing the object attribute characteristics with the service data attribute characteristics to obtain target fusion characteristics corresponding to the service data to be pushed;

and splicing the target fusion characteristics with the fusion characteristics to obtain splicing characteristics, and extracting information based on the splicing characteristics to obtain target pushing reference information of the service data to be pushed aiming at the object identification.

6. The method of claim 5, wherein the fusing the object attribute features with the service data attribute features to obtain target fusion features corresponding to the service data to be pushed comprises:

calculating binary operation of the attribute characteristics of each object and the attribute characteristics of each service data to obtain each target binary operation value;

and taking each target binary operation value as a target fusion characteristic corresponding to the service data to be pushed.

7. The method of claim 5, wherein the extracting information based on the stitching feature to obtain the target push reference information of the service data to be pushed for the object identifier includes:

Acquiring full connection parameters, and weighting the splicing characteristics by using the full connection parameters to obtain full connection characteristics;

activating the full connection feature to obtain an activation feature, and normalizing the activation feature to obtain target pushing reference information of the service data to be pushed aiming at the object identifier.

8. The method of claim 1, wherein extracting the features corresponding to the respective interacted service data to obtain the respective interacted data features comprises:

extracting data characteristics corresponding to the interacted service data to obtain initial data characteristics;

and calculating the attention weights corresponding to the initial data features respectively according to the data features to be pushed, and weighting the corresponding initial data features according to the attention weights to obtain the interacted data features.

9. The method according to claim 8, wherein calculating the attention weights respectively corresponding to the initial data features according to the data features to be pushed comprises:

splicing the initial data features to obtain initial data sequence features, and calculating the transposition of the initial data sequence features to obtain transposition features;

Multiplying the transposed feature with the data feature to be pushed to obtain a correlation feature, carrying out mean pooling based on the correlation feature to obtain a pooled feature, and normalizing the pooled feature to obtain the attention weight corresponding to each initial data feature.

10. The method according to claim 1, wherein calculating the attention weights respectively corresponding to the initial data features according to the data features to be pushed comprises:

splicing the data features to be pushed with the initial data features to obtain data splicing features;

acquiring a correlation feature extraction parameter, and carrying out feature extraction on the data splicing features according to the correlation feature extraction parameter to obtain data correlation features;

normalizing the data correlation characteristics to obtain the attention weights respectively corresponding to the initial data characteristics.

11. The method according to claim 1, characterized in that the method further comprises:

inputting the service data to be pushed and the interactive service data into a service data pushing model, extracting the corresponding characteristics of the service data to be pushed through the service data pushing model to obtain the characteristics of the data to be pushed, and extracting the corresponding characteristics of the interactive service data to obtain the characteristics of the interactive data;

Acquiring feature interaction parameters through the service data push model, carrying out feature interaction on the interacted data features according to the feature interaction parameters to obtain interaction features corresponding to the interacted data features respectively, acquiring fusion guide parameters, and adjusting the interaction features according to the fusion guide parameters to obtain adjustment features corresponding to the interaction features respectively;

and extracting information based on the fusion characteristics through the service data pushing model to obtain pushing reference information of the service data to be pushed aiming at the object identification.

12. The method of claim 11, wherein the traffic data push model comprises: the system comprises a feature extraction network, a feature interaction network, a convergence guiding network, a feature convergence network and an information extraction network; the method further comprises the steps of:

extracting the characteristics corresponding to the service data to be pushed through the characteristic extraction network to obtain the characteristics of the data to be pushed, and extracting the characteristics corresponding to each interactive service data to obtain the characteristics of each interactive data;

Performing feature interaction on the interacted data features through feature interaction parameters in the feature interaction network to obtain interaction features corresponding to the interacted data features, and adjusting the interaction features through fusion guide parameters in the fusion guide network to obtain adjustment features corresponding to the interaction features;

fusing each adjustment feature with the data feature to be pushed through the feature fusion network to obtain a fusion feature corresponding to the service data to be pushed;

and extracting information from the fusion characteristics through the information extraction network to obtain push reference information of the service data to be pushed aiming at the object identifier.

13. The method according to claim 11, wherein the training of the traffic data push model comprises the steps of:

acquiring training data to be pushed, a pushing label and training interactive service data corresponding to a training object identifier;

inputting the training data to be pushed and the interactive service data of each training into an initial service data pushing model for forward calculation to obtain training pushing reference information corresponding to the training data to be pushed;

Performing loss calculation based on the training push reference information and the push label to obtain training loss information, and reversely updating the initial service data push model according to the training loss information to obtain an updated service data push model;

and taking the updated service data pushing model as an initial service data pushing model, and returning to the step of acquiring the training data to be pushed, the pushing label and the training interactive service data corresponding to the training object identifier for iterative execution until the training completion condition is reached, so as to obtain the service data pushing model.

14. The method according to claim 1, characterized in that the method further comprises:

acquiring a commodity video pushing request, wherein the commodity video pushing carries an object identifier, acquiring a commodity video to be pushed based on the commodity video pushing request, and acquiring each interacted commodity video based on the object identifier;

extracting the corresponding features of the commodity video to be pushed to obtain data features to be pushed, and extracting the corresponding features of each interacted commodity video to obtain each interacted data feature;

Fusing each adjustment feature with the data feature to be pushed to obtain a fusion feature corresponding to the commodity video to be pushed;

and extracting information based on the fusion characteristics to obtain push reference information of the commodity video to be pushed aiming at the object identifier, wherein the push reference information is used for sending the commodity video to be pushed to the object terminal corresponding to the object identifier when a preset push condition is met.

15. A traffic data pushing apparatus, the apparatus comprising:

the feature extraction module is used for extracting features corresponding to the service data to be pushed to obtain the features of the data to be pushed, extracting the features corresponding to the interactive service data respectively to obtain the features of the interactive data;

the feature adjustment module is used for acquiring feature interaction parameters, carrying out feature interaction on the interacted data features according to the feature interaction parameters to obtain interaction features corresponding to the interacted data features respectively, acquiring fusion guide parameters, and adjusting the interaction features according to the fusion guide parameters to obtain adjustment features corresponding to the interaction features respectively;

The feature fusion module is used for fusing each adjustment feature with the feature of the data to be pushed to obtain a fusion feature corresponding to the service data to be pushed;

the data pushing module is used for extracting information based on the fusion characteristics to obtain pushing reference information of the service data to be pushed aiming at the object identifier, and the pushing reference information is used for pushing the service data to be pushed to the object terminal corresponding to the object identifier when a preset pushing condition is met.

16. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any one of claims 1 to 14 when the computer program is executed.

17. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 14.

18. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any one of claims 1 to 14.