CN116361641A

CN116361641A - Interactive index recognition model training and object recommending method and device

Info

Publication number: CN116361641A
Application number: CN202310102424.8A
Authority: CN
Inventors: 李鲲鹏; 邵广翠; 杨乃君; 宋洋
Original assignee: Beijing Dajia Internet Information Technology Co Ltd
Current assignee: Beijing Dajia Internet Information Technology Co Ltd
Priority date: 2023-01-29
Filing date: 2023-01-29
Publication date: 2023-06-30

Abstract

The method comprises the steps of obtaining sample index association characteristics corresponding to a plurality of historical exposure accounts, sample interaction labels and historical resource interaction indexes corresponding to positive sample accounts, and carrying out label configuration on a plurality of preset index intervals based on the historical resource interaction indexes to obtain sample interaction index labels, wherein the sample interaction index labels represent the probability that the historical resource interaction indexes are larger than or equal to the lower limit values of the preset index intervals; based on the sample index association characteristics, the sample interaction index labels and the sample interaction labels, the interactive index recognition model to be trained and the interactive recognition model to be trained are subjected to joint training, and the target interactive index recognition model is obtained. By utilizing the method and the device, the accuracy of the interactive index identification can be improved.

Description

Interactive index recognition model training and object recommending method and device

Technical Field

The disclosure relates to the technical field of artificial intelligence, in particular to an interactive index recognition model training and object recommendation method and device.

Background

With the development of internet technology, object recommendation of commodities, applications, stores, living rooms, and the like is being performed through the internet, and has been a major form of object recommendation. GMV (Gross Merchandise Volume, commodity transaction total) and the like, which can reflect the benefits of the object provider, are important indicators of interest in the object recommendation process.

In the related art, the deep learning model is often combined to identify the indexes, in the modeling process, a plurality of sub-buckets are often arranged (a plurality of index intervals are arranged) from the classification angle, the sample index labels are arranged according to the condition of the sub-buckets of the indexes corresponding to the sample data (the sub-buckets of the indexes are 1, and vice versa, 0), and in the identification process, the regression model of the classification problem such as softmax is combined to predict the indexes; however, in the related art, the modeling manner from the classification perspective has no relationship among different categories, so that the size relationship among different index intervals cannot be identified, the model prediction distribution is too concentrated, the index identification accuracy is low, the effective power-assisted object recommendation cannot be realized, and the user experience is poor.

Disclosure of Invention

The invention provides an interactive index recognition model training and object recommending method and device, which at least solve the technical problems that the size relation of different index intervals cannot be recognized in the related technology, the model prediction distribution is too concentrated, the index recognition accuracy of the model is low, the effective power-assisted object recommendation cannot be realized, the user experience is poor and the like. The technical scheme of the present disclosure is as follows:

according to a first aspect of an embodiment of the present disclosure, there is provided an interactive index recognition model training method, including:

Acquiring sample index association characteristics corresponding to a plurality of historical exposure accounts, sample interaction labels corresponding to a plurality of historical exposure accounts and historical resource interaction indexes corresponding to positive sample accounts in a plurality of historical exposure accounts, wherein the historical resource interaction indexes are virtual resource amounts brought by corresponding object providers for object acquisition interaction operations triggered by the positive sample accounts; the sample index association features represent virtual resource consumption conditions corresponding to a plurality of historical exposure accounts and virtual resource acquisition conditions corresponding to object providers of the historical exposure accounts, and the sample interaction labels represent probabilities of triggering objects to acquire interaction operations of the historical exposure accounts;

performing label configuration on a plurality of preset index intervals based on the historical resource interaction indexes to obtain sample interaction index labels, wherein the sample interaction index labels represent the probability that the historical resource interaction indexes are larger than or equal to the lower limit values of the preset index intervals; the preset index intervals are adjacent index intervals;

and based on the sample index association characteristics, the sample interaction index labels and the sample interaction labels, carrying out joint training on the to-be-trained interaction index recognition model and the to-be-trained interaction recognition model to obtain a target interaction index recognition model.

In an alternative embodiment, the plurality of historical exposure accounts further comprises a negative sample account; based on the sample index association feature, the sample interaction index label and the sample interaction label, performing joint training on the to-be-trained interaction index recognition model and the to-be-trained interaction recognition model to obtain a target interaction index recognition model comprises the following steps:

randomly sampling a first index association feature corresponding to the negative sample account in the sample index association features to obtain a second index association feature; the difference between the data volume corresponding to the second index association feature and the data volume corresponding to a third index association feature is smaller than a preset threshold, and the third index association feature is an index association feature corresponding to the positive sample account in the sample index association features;

inputting the third index association characteristic into the interaction index recognition model to be trained to perform interaction index recognition processing, so as to obtain a predicted interaction index label corresponding to the positive sample account;

inputting the second index association characteristic and the third index association characteristic into the interactive recognition model to be trained for interactive recognition processing to obtain a predicted interactive label;

And carrying out joint training on the to-be-trained interactive index recognition model and the to-be-trained interactive recognition model according to the predicted interactive index label, the sample interactive index label, the predicted interactive label and the interactive label corresponding to the predicted interactive label in the sample interactive label to obtain the target interactive index recognition model.

In an alternative embodiment, the method further comprises:

acquiring associated task training features corresponding to a plurality of historical exposure accounts; the associated task training features are extracted task training features in the process of training at least one task model associated with the interactive index recognition model to be trained;

inputting the third index association characteristic into the to-be-trained interactive index recognition model to perform interactive index recognition processing, and obtaining the predicted interactive index label corresponding to the positive sample account comprises the following steps:

inputting the third index association characteristic and a first training characteristic corresponding to the positive sample account in the association task training characteristics into the interaction index recognition model to be trained for interaction index recognition processing, and obtaining the predicted interaction index label;

inputting the second index association feature and the third index association feature into the interaction recognition model to be trained for interaction recognition processing, and obtaining a predicted interaction label comprises the following steps:

Inputting the second index associated feature, the third index associated feature, the first training feature and the second training feature into the interactive recognition model to be trained for interactive recognition processing to obtain the predicted interactive label; the second training features are training features of the target negative sample account in the associated task training features; the target negative sample account is a negative sample account corresponding to the second index association feature.

In an alternative embodiment, the method further comprises:

acquiring sample identification features corresponding to a plurality of historical exposure accounts;

inputting the third index association characteristic and a first identification characteristic corresponding to the positive sample account number in the sample identification characteristic into the interaction index identification model to be trained for interaction index identification processing, and obtaining the predicted interaction index label;

Inputting the second index association feature, the third index association feature, the first identification feature and the second identification feature into the interactive identification model to be trained for interactive identification processing, and obtaining the predicted interactive label; the second identification feature is an identification feature of a target negative sample account in the sample identification features; the target negative sample account is a negative sample account corresponding to the second index association feature.

In an optional embodiment, the obtaining sample identification features corresponding to the plurality of historical exposure accounts includes:

acquiring sample identification information corresponding to a plurality of historical exposure accounts;

inputting the sample identification information into the identification feature extraction model to be trained to perform feature extraction processing to obtain the sample identification features;

the step of performing joint training on the to-be-trained interaction recognition model and the to-be-trained interaction index recognition model in the to-be-trained interaction index recognition model according to the predicted interaction index label, the sample interaction index label, the predicted interaction label and the interaction label corresponding to the predicted interaction label in the sample interaction label, and the step of obtaining the target interaction index recognition model includes:

And carrying out joint training on the identification feature extraction model to be trained, the interaction identification model to be trained and the identification feature extraction model to be trained in the identification model to be trained according to the prediction interaction index label, the sample interaction index label, the prediction interaction label and the interaction label corresponding to the prediction interaction label in the sample interaction label, so as to obtain the identification feature extraction model corresponding to the target interaction index identification model and the identification feature extraction model to be trained.

In an optional embodiment, the obtaining sample index association features corresponding to the plurality of historical exposure accounts includes:

acquiring sample index associated information corresponding to a plurality of historical exposure accounts, wherein the sample index associated information is virtual resource consumption condition information corresponding to a plurality of historical exposure accounts and virtual resource acquisition condition information corresponding to object providers of a plurality of historical exposure accounts;

inputting the sample index association information into an index feature extraction model to be trained to perform feature extraction processing, so as to obtain the sample index association features;

based on the sample index association feature, the sample interaction index label and the sample interaction label, performing joint training on the to-be-trained interaction index recognition model and the to-be-trained interaction recognition model to obtain a target interaction index recognition model comprises the following steps:

And based on the sample index association features, the sample interaction index labels and the sample interaction labels, carrying out joint training on the index feature extraction model to be trained, the interaction identification model to be trained and the interaction index identification model to be trained, and obtaining an index feature extraction model corresponding to the target interaction index identification model and the index feature extraction model to be trained.

In an alternative embodiment, the plurality of historical exposure accounts further comprises a negative sample account; the method further comprises the steps of:

acquiring associated task training features corresponding to a plurality of historical exposure accounts and/or sample identification features corresponding to a plurality of historical exposure accounts; the associated task training features are task training features extracted in the process of training at least one task model associated with the interactive index recognition model to be trained;

inputting the sample index associated features, the associated task training features and/or the sample identification features into a feature fusion model to be trained for fusion processing to obtain sample fusion features;

Based on the sample fusion features, the sample interaction index labels and the sample interaction labels, the feature fusion model to be trained, the interaction identification model to be trained and the interaction index identification model to be trained are jointly trained, and the feature fusion model corresponding to the target interaction index identification model and the feature fusion model to be trained is obtained.

In an optional embodiment, the performing label configuration on the plurality of preset index intervals based on the historical resource interaction index to obtain a sample interaction index label includes:

determining a target index interval in which the historical resource interaction index is located from a plurality of preset index intervals;

performing label configuration on the first index interval based on the first preset label to obtain a first interaction index label; the first index interval is an interval in which the upper limit value of the plurality of preset index intervals is smaller than the target index and is larger than or equal to the lower limit value of the plurality of preset index intervals;

performing label configuration on the target index interval and the second index interval based on a second preset label to obtain a second interaction index label; the second index section is a section in which the lower limit value of the preset index sections is larger than the upper limit value of the target index section;

And generating the sample interaction index label according to the first interaction index label and the interaction index label.

According to a second aspect of the embodiments of the present disclosure, there is provided an object recommendation method, including:

acquiring target index association characteristics; the target index association features are features for representing virtual resource consumption conditions corresponding to the target account and virtual resource acquisition conditions of at least one target object provider, wherein the at least one target object provider is a provider of at least one object to be recommended;

inputting the target index association characteristics into a target interaction index recognition model obtained based on the interaction index recognition model training method provided by the first aspect to perform interaction index recognition processing to obtain a target interaction index label, wherein the target interaction index label represents the probability that the predicted resource interaction index corresponding to the target account is greater than or equal to the lower limit value of a plurality of preset index intervals; the predicted resource interaction index triggers object acquisition interaction aiming at least one object to be recommended for the target account number, and a predicted quantity of virtual resources is brought for at least one target object provider; the preset index intervals are adjacent index intervals;

Determining the predicted resource interaction index according to the target interaction index label;

and recommending the target object in at least one object to be recommended to the target account number based on the predicted resource interaction index.

In an alternative embodiment, the method further comprises:

acquiring target associated task characteristics corresponding to the target account and/or target identification characteristics corresponding to the target account; the target associated task features are features required by identifying task indexes associated with the predicted resource interaction indexes;

inputting the target index associated features, the target associated task features and/or the target identification features into a feature fusion model for fusion processing to obtain target fusion features; the feature fusion model is obtained by combined training with the target interaction index recognition model;

the step of inputting the target index association characteristic into the target interaction index recognition model obtained based on the interaction index recognition model training method provided by the first aspect to perform interaction index recognition processing, and the step of obtaining a target interaction index label comprises the following steps:

and inputting the target fusion characteristics into the target interaction index recognition model to perform interaction index recognition processing, so as to obtain the target interaction index label.

In an optional embodiment, the determining the predicted resource interaction indicator according to the target interaction indicator tag includes:

determining an index mean value corresponding to a plurality of preset index intervals and a target index difference corresponding to each preset index interval, wherein the target index difference corresponding to each preset index interval is a difference value between the probability corresponding to each preset index interval and the probability corresponding to the previous preset index interval, and the previous preset index interval is an index interval with an upper limit value adjacent to a lower limit value of each preset index interval; the prediction index data corresponding to the preset index region before the first preset index region is zero; the first preset index interval is an interval with the minimum upper limit value among a plurality of preset index intervals;

and generating the predicted resource interaction index according to the index mean value and the target index difference.

In an alternative embodiment, the obtaining the target-index-associated feature includes:

acquiring target index associated information corresponding to the target account, wherein the target index associated information is information representing virtual resource consumption conditions corresponding to the target account and virtual resource acquisition conditions corresponding to at least one object to be recommended;

And inputting the target index related information into an index feature extraction model to perform feature extraction processing to obtain the target index related features, wherein the index feature extraction model is obtained by combined training with the target interactive index recognition model.

According to a third aspect of embodiments of the present disclosure, there is provided an interactive index recognition model training apparatus, including:

the sample data acquisition module is configured to acquire sample index association characteristics corresponding to a plurality of historical exposure accounts, sample interaction labels corresponding to a plurality of historical exposure accounts and historical resource interaction indexes corresponding to positive sample accounts in a plurality of historical exposure accounts, wherein the historical resource interaction indexes are object acquisition interaction operations triggered by the positive sample accounts and are virtual resource amounts brought by corresponding object providers; the sample index association features represent virtual resource consumption conditions corresponding to a plurality of historical exposure accounts and virtual resource acquisition conditions corresponding to object providers of the historical exposure accounts, and the sample interaction labels represent probabilities of triggering objects to acquire interaction operations of the historical exposure accounts;

the tag configuration module is configured to perform tag configuration on a plurality of preset index intervals based on the historical resource interaction indexes to obtain sample interaction index tags, wherein the sample interaction index tags represent the probability that the historical resource interaction indexes are larger than or equal to the lower limit values of the preset index intervals; the preset index intervals are adjacent index intervals;

And the joint training module is configured to perform joint training on the interactive index recognition model to be trained and the interactive recognition model to be trained based on the sample index association characteristics, the sample interactive index labels and the sample interactive labels to obtain a target interactive index recognition model.

In an alternative embodiment, the plurality of historical exposure accounts further comprises a negative sample account; the joint training module comprises:

the random sampling unit is configured to perform random sampling on the first index association characteristic corresponding to the negative sample account in the sample index association characteristics to obtain a second index association characteristic; the difference between the data volume corresponding to the second index association feature and the data volume corresponding to a third index association feature is smaller than a preset threshold, and the third index association feature is an index association feature corresponding to the positive sample account in the sample index association features;

the interactive index recognition processing unit is configured to input the third index association characteristic into the interactive index recognition model to be trained to perform interactive index recognition processing, so as to obtain a predicted interactive index label corresponding to the positive sample account;

The interactive identification processing unit is configured to input the second index association characteristic and the third index association characteristic into the interactive identification model to be trained for interactive identification processing, so as to obtain a predicted interactive label;

and the joint training unit is configured to perform joint training on the to-be-trained interactive index recognition model and the to-be-trained interactive recognition model according to the predicted interactive index label, the sample interactive index label, the predicted interactive label and the interactive label corresponding to the predicted interactive label in the sample interactive label, so as to obtain the target interactive index recognition model.

In an alternative embodiment, the apparatus further comprises:

the associated task training feature acquisition module is configured to acquire associated task training features corresponding to a plurality of historical exposure accounts; the associated task training features are extracted task training features in the process of training at least one task model associated with the interactive index recognition model to be trained;

the first interactive index recognition processing module is specifically configured to perform inputting the third index association feature and a first training feature corresponding to the positive sample account number in the association task training features into the interactive index recognition model to be trained to perform interactive index recognition processing, so as to obtain the predicted interactive index label;

The interactive recognition processing module is specifically configured to perform interactive recognition processing by inputting the second index association feature, the third index association feature, the first training feature and the second training feature into the interactive recognition model to be trained, so as to obtain the predicted interactive label; the second training features are training features of the target negative sample account in the associated task training features; the target negative sample account is a negative sample account corresponding to the second index association feature.

In an alternative embodiment, the apparatus further comprises:

the sample identification feature acquisition module is configured to acquire sample identification features corresponding to a plurality of historical exposure accounts;

the first interactive index recognition processing module is specifically configured to perform inputting the third index association feature and a first identification feature corresponding to the positive sample account number in the sample identification features into the interactive index recognition model to be trained to perform interactive index recognition processing, so as to obtain the predicted interactive index label;

the interactive identification processing module is specifically configured to perform interactive identification processing by inputting the second index association feature, the third index association feature, the first identification feature and the second identification feature into the interactive identification model to be trained, so as to obtain the predicted interactive label; the second identification feature is an identification feature of a target negative sample account in the sample identification features; the target negative sample account is a negative sample account corresponding to the second index association feature.

In an alternative embodiment, the sample identification feature acquisition module includes:

the sample identification information acquisition unit is configured to acquire sample identification information corresponding to a plurality of historical exposure account numbers;

the first feature extraction processing unit is configured to input the sample identification information into the identification feature extraction model to be trained to perform feature extraction processing to obtain the sample identification features;

the interaction recognition processing module is specifically configured to perform joint training on the to-be-trained identification feature extraction model, the to-be-trained interaction recognition model and the to-be-trained interaction index recognition model in the to-be-trained interaction index recognition model according to the predicted interaction index label, the sample interaction index label, the predicted interaction label and the interaction label corresponding to the predicted interaction label in the sample interaction label, so as to obtain an identification feature extraction model corresponding to the target interaction index recognition model and the to-be-trained identification feature extraction model.

In an alternative embodiment, the sample data acquisition module includes:

a sample index association information obtaining unit configured to obtain sample index association information corresponding to a plurality of historical exposure accounts, where the sample index association information is virtual resource consumption condition information corresponding to a plurality of historical exposure accounts and virtual resource obtaining condition information corresponding to object providers of the historical exposure accounts;

The second feature extraction processing unit is configured to input the sample index association information into an index feature extraction model to be trained to perform feature extraction processing, so as to obtain the sample index association features;

the joint training module is specifically configured to perform joint training on the to-be-trained index feature extraction model, the to-be-trained interaction identification model and the to-be-trained interaction index identification model based on the sample index associated feature, the sample interaction index label and the sample interaction label, so as to obtain an index feature extraction model corresponding to the target interaction index identification model and the to-be-trained index feature extraction model.

In an alternative embodiment, the plurality of historical exposure accounts further comprises a negative sample account; the apparatus further comprises:

the characteristic acquisition module is configured to acquire associated task training characteristics corresponding to a plurality of historical exposure accounts and/or sample identification characteristics corresponding to a plurality of historical exposure accounts; the associated task training features are task training features extracted in the process of training at least one task model associated with the interactive index recognition model to be trained;

The first fusion processing module is configured to input the sample index associated features, the associated task training features and/or the sample identification features into a feature fusion model to be trained for fusion processing, so as to obtain sample fusion features;

the joint training module is specifically configured to perform joint training on the feature fusion model to be trained, the interactive identification model to be trained and the interactive index identification model to be trained based on the sample fusion feature, the sample interactive index label and the sample interactive label, so as to obtain a feature fusion model corresponding to the target interactive index identification model and the feature fusion model to be trained.

In an alternative embodiment, the tag configuration module includes:

a target index section determining unit configured to perform determining a target index section in which the history resource interaction index is located from a plurality of the preset index sections;

the first label configuration unit is configured to execute label configuration on the first index interval based on a first preset label to obtain a first interactive index label; the first index interval is an interval in which the upper limit value of the plurality of preset index intervals is smaller than the target index and is larger than or equal to the lower limit value of the plurality of preset index intervals;

The second label configuration unit is configured to perform label configuration on the target index interval and the second index interval based on a second preset label to obtain a second interactive index label; the second index section is a section in which the lower limit value of the preset index sections is larger than the upper limit value of the target index section;

and the sample interaction index label generating unit is configured to generate the sample interaction index label according to the first interaction index label and the interaction index label.

According to a fourth aspect of the embodiments of the present disclosure, there is provided an object recommendation apparatus, including:

the target index associated feature acquisition module is configured to acquire target index associated features; the target index association features are features for representing virtual resource consumption conditions corresponding to the target account and virtual resource acquisition conditions of at least one target object provider, wherein the at least one target object provider is a provider of at least one object to be recommended;

the second interactive index recognition processing module is configured to perform interactive index recognition processing on the target interactive index recognition model obtained by inputting the target index association characteristics based on the interactive index recognition model training method provided by the first aspect, so as to obtain a target interactive index label, wherein the target interactive index label represents the probability that the predicted resource interactive index corresponding to the target account is greater than or equal to the lower limit value of a plurality of preset index intervals; the predicted resource interaction index triggers object acquisition interaction aiming at least one object to be recommended for the target account number, and a predicted quantity of virtual resources is brought for at least one target object provider; the preset index intervals are adjacent index intervals;

A predicted resource interaction index determination module configured to perform determining the predicted resource interaction index according to the target interaction index tag;

and the object pushing module is configured to execute the recommendation of the target object in at least one object to be recommended to the target account number based on the predicted resource interaction index.

In an alternative embodiment, the apparatus further comprises:

the characteristic acquisition module is configured to acquire target associated task characteristics corresponding to the target account and/or target identification characteristics corresponding to the target account; the target associated task features are features required by identifying task indexes associated with the predicted resource interaction indexes;

the second fusion processing module is configured to perform fusion processing on the target index associated features, the target associated task features and/or the target identification feature input feature fusion model to obtain target fusion features; the feature fusion model is obtained by combined training with the target interaction index recognition model;

the second interactive index recognition processing module is specifically configured to perform interactive index recognition processing by inputting the target fusion feature into the target interactive index recognition model to obtain the target interactive index tag.

In an alternative embodiment, the prediction resource interaction index determination module includes:

a calculation unit configured to determine an index mean value corresponding to a plurality of preset index intervals and a target index difference corresponding to each preset index interval, wherein the target index difference corresponding to each preset index interval is a difference value between a probability corresponding to each preset index interval and a probability corresponding to a previous preset index interval, and the previous preset index interval is an index interval with an upper limit value adjacent to a lower limit value of each preset index interval; the prediction index data corresponding to the preset index region before the first preset index region is zero; the first preset index interval is an interval with the minimum upper limit value among a plurality of preset index intervals;

and the prediction resource interaction index generating unit is configured to generate the prediction resource interaction index according to the index mean value and the target index difference.

In an alternative embodiment, the target index association feature acquiring module includes:

the target index associated information acquisition unit is configured to acquire target index associated information corresponding to the target account, wherein the target index associated information is information representing virtual resource consumption conditions corresponding to the target account and virtual resource acquisition conditions corresponding to at least one object to be recommended;

And the third feature extraction processing unit is configured to perform feature extraction processing on the target index related information input into an index feature extraction model to obtain the target index related features, wherein the index feature extraction model is obtained by combined training with the target interactive index recognition model.

According to a fifth aspect of embodiments of the present disclosure, there is provided an electronic device, comprising: a processor; a memory for storing the processor-executable instructions; wherein the processor is configured to execute the instructions to implement the method of any one of the first or second aspects above.

According to a fourth aspect of embodiments of the present disclosure, there is provided a computer readable storage medium, which when executed by a processor of an electronic device, causes the electronic device to perform the method of any one of the above-described first or second aspects of embodiments of the present disclosure.

According to a fifth aspect of embodiments of the present disclosure, there is provided a computer program product comprising instructions which, when run on a computer, cause the computer to perform the method of any one of the first or second aspects described above.

The technical scheme provided by the embodiment of the disclosure at least brings the following beneficial effects:

in the training process of the interactive index recognition model, the object triggered by the positive sample account in the plurality of historical exposure accounts is combined to obtain interactive operation, the historical resource interactive index of virtual resource quantity brought by the corresponding object provider is configured, the adjacent plurality of preset index intervals are subjected to label configuration to generate a sample interactive index label, the problem that zero value index is introduced in the resource interactive index learning process to cause underestimation of the resource interactive index is avoided, the sample interactive index label can characterize the probability that the historical resource interactive index is greater than or equal to the lower limit value of the plurality of preset index intervals, the size relation of the plurality of preset index intervals is indicated by the sample interactive index label on the basis of the index interval in which the indication positive sample account corresponds to the historical resource interactive index, the distribution balance of the learned resource interactive index can be ensured, and the sample index association characteristic of the virtual resource consumption condition corresponding to the object provider and the sample interactive label are combined, the interactive index recognition model to be trained and the sample interactive index to be trained are combined to be trained, whether the interactive index recognition model to be trained is triggered in the learning resource interactive index learning process or not is enabled to trigger the corresponding to obtain the target interactive index, the interactive index is effectively estimated, the user can be prevented from being interacted by the fact that the user is high in accuracy and the user can be effectively prolonged, and the user can be effectively recognized by the user is prevented from the fact that the interactive index is interacted.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the disclosure and together with the description, serve to explain the principles of the disclosure and do not constitute an undue limitation on the disclosure.

FIG. 1 is a schematic diagram of an application environment shown in accordance with an exemplary embodiment;

FIG. 2 is a flowchart illustrating an interactive metric recognition model training method, according to an exemplary embodiment;

FIG. 3 is a flowchart illustrating a tag configuration for a plurality of preset index intervals based on historical resource interaction indexes to obtain sample interaction index tags according to an exemplary embodiment;

FIG. 4 is a flowchart illustrating a joint training of an interactive index recognition model to be trained and an interactive recognition model to be trained based on sample index association features, sample interactive index tags, and sample interactive tags to obtain a target interactive index recognition model according to an exemplary embodiment;

FIG. 5 is a schematic illustration of an interactive metric recognition model training process according to an exemplary provided;

FIG. 6 is a flowchart illustrating an object recommendation method, according to an example embodiment;

FIG. 7 is a block diagram of an interactive metric recognition model training apparatus, according to an exemplary embodiment;

FIG. 8 is a block diagram of an object recommendation device, according to an example embodiment;

FIG. 9 is a block diagram of an electronic device for interactive metric recognition model training, according to an exemplary embodiment;

FIG. 10 is a block diagram of an electronic device for object recommendation, according to an example embodiment.

Detailed Description

In order to enable those skilled in the art to better understand the technical solutions of the present disclosure, the technical solutions of the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.

It should be noted that the terms "first," "second," and the like in the description and claims of the present disclosure and in the foregoing figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the disclosure described herein may be capable of operation in sequences other than those illustrated or described herein. The implementations described in the following exemplary examples are not representative of all implementations consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with some aspects of the present disclosure as detailed in the accompanying claims.

It should be noted that, the user information (including, but not limited to, user equipment information, user personal information, etc.) and the data (including, but not limited to, data for presentation, analyzed data, etc.) related to the present disclosure are information and data authorized by the user or sufficiently authorized by each party.

Referring to fig. 1, fig. 1 is a schematic diagram illustrating an application environment that may include a terminal 100 and a server 200 according to an exemplary embodiment.

In an alternative embodiment, the terminal 100 may be used to provide an object recommendation service to any user. Specifically, the terminal 100 may include, but is not limited to, a smart phone, a desktop computer, a tablet computer, a notebook computer, a smart speaker, a digital assistant, an augmented reality (augmented reality, AR)/Virtual Reality (VR) device, a smart wearable device, or other type of electronic device, or may be software running on the electronic device, such as an application program, etc. Alternatively, the operating system running on the electronic device may include, but is not limited to, an android system, an IOS system, linux, windows, and the like.

In an alternative embodiment, server 200 may be configured to train an interactive metrics recognition model and provide background services to terminal 100 based on the trained interactive metrics recognition model. Specifically, the server 200 may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server that provides a cloud computing service.

In addition, it should be noted that, fig. 1 is only an application environment provided by the present disclosure, and in practical application, other application environments may also be included, for example, may include more terminals.

In the embodiment of the present disclosure, the terminal 100 and the server 200 may be directly or indirectly connected through a wired or wireless communication manner, which is not limited herein.

FIG. 2 is a flowchart illustrating an interactive index recognition model training method according to an exemplary embodiment, optionally, the interactive index recognition model training method may be applied to an electronic device such as a server or a terminal, and as shown in FIG. 2, the interactive index recognition model training method may include the following steps:

in step S201, sample index association features corresponding to a plurality of historical exposure accounts, sample interaction labels corresponding to a plurality of historical exposure accounts, and historical resource interaction indexes corresponding to positive sample accounts in a plurality of historical exposure accounts are obtained.

In a specific embodiment, the plurality of historical exposure accounts may be exposure accounts corresponding to recommended objects in the object recommendation platform in a preset historical time period, and optionally, the recommended objects may be commodities, stores, living broadcast rooms, and the like, and the objects may be user accounts; optionally, in the case of pushing multimedia data corresponding to a certain recommended object to any user account, the user account may be a historical exposure account corresponding to the recommended object, and correspondingly, the recommended object may be a historical recommended object corresponding to the user account, and the provider of the historical recommended object is an object provider corresponding to the user account. Specifically, the multimedia data corresponding to the recommended object may be data for introducing the recommended object, for example, details page content corresponding to the commodity, homepage content corresponding to the store, live stream corresponding to the live broadcasting room, and the like.

In a specific embodiment, the sample index association feature may represent virtual resource consumption conditions corresponding to a plurality of historical exposure accounts and virtual resource acquisition conditions corresponding to object providers corresponding to a plurality of historical exposure accounts. Specifically, the sample index association features may be obtained by extracting features of sample index association information based on a corresponding index feature extraction model. Optionally, the index feature extraction model may be independently trained in advance, or may be trained (jointly trained) with the interaction index recognition model to be trained, and optionally, taking the example of acquiring the sample index association feature by combining the index feature extraction model trained with the interaction index recognition model to be trained, the acquiring the sample index association feature corresponding to the plurality of historical exposure accounts may include:

acquiring sample index associated information corresponding to a plurality of historical exposure accounts;

inputting the sample index association information into an index feature extraction model to be trained to perform feature extraction processing, so as to obtain sample index association features;

in a specific embodiment, the sample index related information may be virtual resource consumption information corresponding to a plurality of historical exposure accounts and virtual resource acquisition condition information corresponding to an object provider corresponding to a plurality of historical exposure accounts. In practical application, each recommendation corresponds to a sample (a piece of virtual resource consumption condition information corresponding to a historical exposure account and virtual resource acquisition condition information corresponding to an object provider corresponding to the historical exposure account); optionally, the sample index association information may include virtual resource cumulative consumption information of the historical exposure account in at least one first preset historical period (for example, virtual resource consumption of last 7 days, 30 days, and 90 days), single virtual resource consumption information of the historical exposure account in at least one first preset historical period, an object provider of the historical recommended object in a second preset historical period, an average virtual resource obtained based on the recommended object in the live broadcast in the last 30 days (for example, an average virtual resource obtained based on the recommended object in the live broadcast in the last 30 days, a virtual resource corresponding to an average single recommended object in the live broadcast in the last 30 days, and the like), a cumulative virtual resource obtained based on the recommended object in the second preset historical period by the object provider of the historical recommended object, and the like.

In a specific embodiment, the index feature extraction model to be trained may be a feature extraction network to be trained, for example, a convolutional neural network, etc., and specifically, the network structure and the layer number may be set in combination with the actual application requirement.

In a specific embodiment, the historical resource interaction index may acquire an interaction operation for an object triggered by the positive sample account, and is a virtual resource amount brought by a corresponding object provider; the sample interaction label characterizes the probability that a plurality of historical exposure accounts trigger an object to acquire interaction operation; specifically, the positive sample account number may include a plurality of objects. Specifically, the plurality of historical exposure objects further include a negative sample account, which may include a plurality of objects.

In practical application, under the condition that a recommended object is exposed to a user account, if the user account triggers an interaction operation (object acquisition interaction operation) for acquiring the recommended object, a historical resource interaction index is corresponding to the recommended object, and correspondingly, the user account can be a positive sample account, and a sample interaction label corresponding to the positive sample account is 1; otherwise, if the user account does not trigger the interactive operation of acquiring the recommended object, a corresponding historical resource interactive index (i.e., the historical resource interactive index is 0) is not generated, and accordingly, the user account may be a negative sample account, and a sample interactive label corresponding to the negative sample account is 0.

In step S203, tag configuration is performed on a plurality of preset index intervals based on the historical resource interaction index, so as to obtain a sample interaction index tag.

In a specific embodiment, the sample interaction index tag may represent a probability that a historical resource interaction index corresponding to the positive sample account is greater than or equal to a lower limit value of a plurality of preset index intervals; optionally, the sample interaction index label may be a numerical value, or may be a character representation of probability that the historical resource interaction index corresponding to the positive sample account is greater than or equal to the lower limit value of the multiple preset index intervals. The plurality of preset index intervals are a plurality of adjacent index intervals. Specifically, since the plurality of preset index sections are adjacent, and when the history resource interaction index is located in a certain preset index section (target index section), the probability that the history resource interaction index is greater than or equal to the target index section lower limit value is determined to be 1, and the probability that the history resource interaction index is greater than or equal to the target index section and then the index section (index section with the lower limit value greater than the target index section upper limit value) lower limit value is also determined to be 1; and the probability of the lower limit value of the index section (the index section with the upper limit value smaller than the lower limit value of the target index section) before the target index section with the history resource interaction index larger than or equal to is 0; correspondingly, under the condition that the sample interaction index label represents the probability that the historical resource interaction index corresponding to the positive sample account is larger than or equal to the lower limit value of the plurality of preset index intervals, the sample interaction index label can indicate the size relation of the plurality of preset index intervals on the basis of indicating the interval to which the historical resource interaction index belongs.

In an alternative embodiment, the plurality of preset index intervals may be set in association with a distribution range corresponding to the resource interaction index in the practical application, for example, the plurality of preset index intervals include (0, 50), (50, 100), (100, 150), (150, 200), (200, 500), (500, 1000), (1000, 10000), and the like.

In an optional embodiment, as shown in fig. 3, the configuring the label for the plurality of preset index intervals based on the historical resource interaction index to obtain the sample interaction index label may include:

in step S301, determining a target index interval in which the historical resource interaction index is located from a plurality of preset index intervals;

in step S303, performing label configuration on the first index interval based on the first preset label to obtain a first interaction index label;

in step S305, based on the second preset label, performing label configuration on the target index interval and the second index interval to obtain a second interaction index label;

in step S307, a sample interaction index tag is generated from the first interaction index tag and the interaction index tag.

In a specific embodiment, taking the sample interaction index label as a numerical value as an example, the first preset label is 0, and the second preset label is 1; the first index section is a section in which the upper limit value of the plurality of preset index sections is smaller than the lower limit value of the target index section and is larger than or equal to the lower limit value of the plurality of preset index sections; the second index section is a section in which a lower limit value of the plurality of preset index sections is greater than an upper limit value of the target index section.

In a specific embodiment, it is assumed that a plurality of preset index intervals may be set in combination with a resource benefit distribution range in an actual application, for example, the plurality of preset index intervals include (0, 10), (10, 20), (20, 30), (30, 50), (50, 100) and (100, 300), (alternatively, it is assumed that a historical resource interaction index corresponding to a certain object in the positive sample account is 45, the corresponding target index interval is (30, 50), the first index interval includes (0, 10), (10, 20) and (20, 30), the second index interval includes (50, 100) and (100, 300), and optionally, the label configuration may be sequentially performed in combination with a numerical range corresponding to the plurality of preset index intervals from large to small, the corresponding first interaction index label is 000, the second interaction index label is 111, and the sample interaction index label is 000111.

In the above embodiment, the second preset label is used to perform label configuration on the target index interval where the historical resource interaction index corresponding to the sample account is located and the second index interval where the lower limit value is greater than the upper limit value of the target index interval; and the first index region with the upper limit value smaller than the lower limit value of the target index region is configured by combining the first preset label, so that the obtained sample interaction index label indicates the size relation of a plurality of preset index regions on the basis of indicating the index region where the history resource interaction index corresponding to the positive sample account number is located, and further the distribution balance of the resource interaction index learned by the follow-up model can be ensured.

In step S205, based on the sample index association feature, the sample interaction index label and the sample interaction label, the interactive index recognition model to be trained and the interactive recognition model to be trained are jointly trained, so as to obtain a target interactive index recognition model.

In a specific embodiment, the third sample index association feature and the sample interaction index label corresponding to the positive sample account number in the sample index association feature may be used as training data of the to-be-trained interaction index recognition model, and the sample index association feature and the sample interaction label may be used as training data of the to-be-trained interaction recognition model.

In a specific embodiment, the interactive index recognition model to be trained may be a deep learning network to be trained, and specifically, the network structure and the layer number may be set in combination with actual application requirements. The interactive recognition model to be trained can be a deep learning network to be trained, and particularly can be used for setting a network structure and the layer number according to actual application requirements.

In an optional embodiment, the performing the joint training on the to-be-trained interactive index recognition model and the to-be-trained interactive recognition model based on the sample index association feature, the sample interactive index label and the sample interactive label to obtain the target interactive index recognition model may include:

Inputting the third index association characteristic into an interactive index recognition model to be trained to perform interactive index recognition processing, so as to obtain a predicted interactive index label corresponding to the positive sample account;

inputting the sample index association features into an interactive recognition model to be trained for interactive recognition processing to obtain a predicted interactive label;

and carrying out joint training on the interactive index recognition model to be trained and the interactive recognition model to be trained according to the predicted interactive index label, the sample interactive index label, the predicted interactive label and the sample interactive label to obtain a target interactive index recognition model.

In a specific embodiment, the predicted interactive index label may be an interactive index identification model to be trained, and the interactive index identification process is performed in combination with a third index association feature, where the predicted interactive index label, specifically, the predicted interactive index label characterizes a probability that a historical resource interactive index corresponding to the positive sample account is greater than or equal to a plurality of preset index interval lower limit values, which is predicted by the interactive index identification model to be trained;

in a specific embodiment, the predicted interactive label may be an interactive recognition model to be trained, and the interactive recognition processing is performed by combining with the sample index association feature, and the predicted interactive label may specifically represent the probability that the predicted interactive recognition model to be trained predicts that the object is triggered to acquire the interactive operation by the plurality of historical exposure accounts.

In a specific embodiment, performing joint training on the to-be-trained interactive index recognition model and the to-be-trained interactive recognition model according to the predicted interactive index label, the sample interactive index label, the predicted interactive label and the sample interactive label to obtain the target interactive index recognition model may include: determining first loss information according to the predicted interactive index label and the sample interactive index label; determining second loss information according to the predicted interactive label and the sample interactive label; determining first target loss information according to the first loss information and the second loss information; and carrying out back propagation training on the interactive index recognition model to be trained and the interactive recognition model to be trained based on the first target loss information to obtain a target interactive index recognition model.

In a specific embodiment, the first loss information may represent accuracy of the to-be-trained interactive index recognition model to recognize the interactive index tag; optionally, the predicted interactive index tag and the sample interactive index tag may be substituted into a corresponding preset loss function, so as to obtain the first loss information. The second loss information can represent the accuracy of the interactive identification model to be trained to identify the interactive label; optionally, the loss functions corresponding to the first loss information and the second loss information may be calculated and set in combination with actual application requirements. Alternatively, the predicted interactive label and the sample interactive label may be substituted into a corresponding preset loss function, so as to obtain the second loss information. Optionally, the first loss information and the second loss information may be added to obtain first target loss information, or the first loss information and the second loss information may be weighted and summed to obtain the first target loss information, and specifically, weights corresponding to the first loss information and the second loss information may be set in combination with actual application requirements.

In a specific embodiment, performing back propagation training on the interactive index recognition model to be trained and the interactive recognition model to be trained based on the first target loss information, and obtaining the target interactive index recognition model may include: updating the to-be-trained interactive index recognition model and model parameters in the to-be-trained interactive recognition model based on the first target loss information (specifically, the model parameters can be updated by combining a gradient descent method), repeating the interactive index recognition processing to the training iteration step of updating the model parameters based on the updated to-be-trained interactive index recognition model and the to-be-trained interactive recognition model until the preset convergence condition is met, and taking the to-be-trained interactive index recognition model corresponding to the preset convergence condition as the target interactive index recognition model.

In a specific embodiment, the meeting of the preset convergence condition may be that the first target loss information is less than or equal to a preset loss threshold, or the number of training iteration steps reaches a preset number of times, and the preset loss threshold and the preset number of times may be specifically set in combination with the model precision and the training speed requirement in practical application.

In practical application, the sample index association features are index association features corresponding to a plurality of historical exposure objects, namely, the sample index association features are sample data of an exposure space, but the interactive index recognition model to be trained is used for carrying out resource interactive index recognition, namely, the sample data of the exposure space is needed to carry out index recognition of an interactive (interactive) space, and because the account occupation ratio of a user account for the interactive operation obtained by an exposure space triggering object is often smaller, the problem of sample selection deviation exists, the index of the interactive space is underestimated, and the index recognition cannot be accurately carried out; if modeling is directly performed in the interaction space (i.e. the index association feature corresponding to the user account of the interaction operation is obtained by using the trigger image as sample data), the index identification inaccuracy caused by the inconsistent space exists due to the fact that prediction is required to be performed in the exposure space on the subsequent line. Accordingly, in another optional embodiment, in order to solve the problem of accuracy of index recognition caused by spatial inconsistency, as shown in fig. 4, the performing joint training on the to-be-trained interactive index recognition model and the to-be-trained interactive recognition model based on the sample index association feature, the sample interactive index label and the sample interactive label to obtain the target interactive index recognition model may include the following steps:

In step S401, randomly sampling a first index association feature corresponding to the negative sample account in the sample index association features to obtain a second index association feature;

in step S403, inputting the third index association feature into the interactive index recognition model to be trained to perform interactive index recognition processing, so as to obtain a predicted interactive index label corresponding to the positive sample account;

in step S405, inputting the second index association feature and the third index association feature into the interactive recognition model to be trained to perform interactive recognition processing, so as to obtain a predicted interactive label;

in step S407, the interactive index recognition model to be trained and the interactive recognition model to be trained are jointly trained according to the predicted interactive index label, the sample interactive index label, the interactive label corresponding to the predicted interactive label in the predicted interactive label and the sample interactive label, so as to obtain a target interactive index recognition model.

In a specific embodiment, a difference between the data amount corresponding to the second index-related feature and the data amount corresponding to the third index-related feature is smaller than a preset threshold, and the third index-related feature is an index-related feature corresponding to the positive sample account in the sample index-related features.

In a specific embodiment, the interaction label corresponding to the predicted interaction label in the sample interaction label may be an interaction label corresponding to a positive sample account number and a target negative sample account number in the sample interaction label; the target negative sample account is a negative sample account corresponding to the second index association feature.

In a specific embodiment, performing joint training on the to-be-trained interactive index recognition model and the to-be-trained interactive recognition model according to the predicted interactive index label, the sample interactive index label, the interactive label corresponding to the predicted interactive label in the predicted interactive label and the sample interactive label, and obtaining the target interactive index recognition model may include: determining third loss information according to the predicted interactive index label and the sample interactive index label; determining fourth loss information according to the predicted interactive label and the interactive label corresponding to the predicted interactive label in the sample interactive label; determining second target loss information according to the third loss information and the fourth loss information; and carrying out back propagation training on the interactive index recognition model to be trained and the interactive recognition model to be trained based on the second target loss information to obtain a target interactive index recognition model.

In a specific embodiment, the interactive labels corresponding to the predicted interactive labels in the predicted interactive index labels, the sample interactive index labels, the predicted interactive labels and the sample interactive labels are used for performing combined training on the to-be-trained interactive index recognition model and the to-be-trained interactive recognition model to obtain specific refinement of the refinement step of the target interactive index recognition model, which can be referred to the above-mentioned combined training on the to-be-trained interactive index recognition model and the to-be-trained interactive recognition model according to the predicted interactive index labels, the sample interactive index labels, the predicted interactive labels and the sample interactive labels, so as to obtain relevant refinement steps corresponding to the target interactive index recognition model, which are not described herein.

In the above embodiment, by randomly sampling the first index association feature corresponding to the negative sample account number in the sample index association features, the difference between the data amount of the index association feature corresponding to the positive sample account number and the data amount of the index association feature corresponding to the negative sample account number participating in the interactive identification process is smaller, so that the situation that the sample selection bias is brought by inconsistent exposure space where the sample data is located in the training stage and the interaction space where the model prediction stage is located, the index of the interaction space is underestimated, and the third index association feature corresponding to the positive sample account number and the sample interaction index label are combined, the third index association feature of the exposure space and the second index association feature obtained by negative sampling from the exposure space are combined while the interaction space is modeled, the problem that the space is inconsistent can be better relieved, and the index identification accuracy of the trained target interaction index identification model can be effectively improved, the effective power-assisted object recommendation is achieved, and the user experience is large.

In practical application, each time a recommendation is made, a corresponding action path (i.e. recommended operation information) is recorded, so that multi-target task learning (such as click rate, conversion rate, browsing duration and other recognition tasks) can be performed, optionally, task training features extracted by corresponding task models in the multi-target task learning task process can be used for enriching sample data in the resource interaction index recognition process, and correspondingly, the method further comprises the following steps:

acquiring associated task training features corresponding to a plurality of historical exposure accounts;

correspondingly, inputting the third index association feature into the to-be-trained interactive index recognition model to perform interactive index recognition processing, and obtaining the predicted interactive index label corresponding to the positive sample account may include:

inputting the first training features corresponding to the positive sample account in the third index association features and the associated task training features into an interactive index recognition model to be trained for interactive index recognition processing, and obtaining a predicted interactive index label;

correspondingly, inputting the second index association feature and the third index association feature into the to-be-trained interactive recognition model for interactive recognition processing, and obtaining the predicted interactive label may include:

Inputting the second index associated feature, the third index associated feature, the first training feature and the second training feature into an interactive recognition model to be trained for interactive recognition processing to obtain a predicted interactive label; the second training features are training features of target negative sample accounts in the associated task training features; the target negative sample account is a negative sample account corresponding to the second index association feature.

In a specific embodiment, the associated task training feature may be an extracted task training feature in a process of training at least one task model associated with the interaction index recognition model to be trained; alternatively, the at least one task model may include at least one of a click rate recognition model, a conversion rate recognition model, and a browsing duration recognition model. Specifically, the associated task training features may be training data required for training at least one task model, and optionally, the associated task training information may include account feature information corresponding to a plurality of historical exposure accounts (such as gender of an account corresponding to a user, historical interaction operation, etc.), sample operation information of a plurality of historical exposure accounts for each corresponding historical recommendation object, object feature information corresponding to a sample recommendation object, and interaction features of a sample recommendation object in a preset time period, where the preset time period includes a time point when the sample recommendation object is recommended to the corresponding historical exposure account; the sample recommended objects are historical recommended objects corresponding to the plurality of historical exposure accounts.

In an optional embodiment, after the associated task training features and the sample index associated features are fused, the features corresponding to the positive sample account number are selected from the fused features, and the interaction index recognition model to be trained is input for interaction index recognition processing, so that a predicted interaction index label is obtained; and selecting the characteristics corresponding to the positive sample account and the characteristics corresponding to the target negative sample account from the fusion characteristics, inputting the interaction recognition model to be trained for interaction recognition processing, and obtaining the predicted interaction label.

In the above embodiment, in the interactive index recognition process, the first training features corresponding to the positive sample account number in the associated task training features are merged, and in the interactive recognition processing process, the task training features corresponding to the positive sample account number and the target negative sample account number in the associated task training features are merged, so that sample data can be better enriched, further, the resource interactive index features learned in the model training process can be enriched, and the index recognition accuracy of the trained interactive index recognition model is improved.

In an optional embodiment, in order to enhance the memory of the model, sample identification features corresponding to the historical exposure account may be introduced in the model training process, and accordingly, the method further includes:

inputting the first identification feature corresponding to the positive sample account in the third index association feature and the sample identification feature into an interactive index identification model to be trained for interactive index identification processing, and obtaining a predicted interactive index label;

inputting the second index associated feature, the third index associated feature, the first identification feature and the second identification feature into an interactive identification model to be trained for interactive identification processing to obtain a predicted interactive label; the second identification feature is an identification feature of a target negative sample account in the sample identification features; the target negative sample account is a negative sample account corresponding to the second index association feature.

In a specific embodiment, the sample identification features include account identification features of a plurality of historical exposure accounts, object identification features of a plurality of historical recommendation objects corresponding to the historical exposure accounts, and account identification features of a release account corresponding to the historical recommendation objects.

In an optional embodiment, after the sample identification feature and the sample index association feature are fused, selecting a feature corresponding to the positive sample account from the fused features, and inputting an interactive index recognition model to be trained to perform interactive index recognition processing to obtain a predicted interactive index label; and selecting the characteristics corresponding to the positive sample account and the characteristics corresponding to the target negative sample account from the fusion characteristics, inputting the interaction recognition model to be trained for interaction recognition processing, and obtaining the predicted interaction label.

In the above embodiment, in the interactive index recognition process, the first identification feature corresponding to the positive sample account number in the sample identification feature is merged, and in the interactive recognition processing process, the identification features corresponding to the positive sample account number and the target negative sample account number in the sample identification feature are merged, so that the memory of the model can be enhanced, and further, the index recognition accuracy of the trained interactive index recognition model can be improved.

In an alternative embodiment, the sample identification feature may be obtained by feature extraction of sample identification information based on a corresponding identification feature extraction model. Optionally, the identification feature extraction model may be independently trained in advance, or may be trained with the interaction index recognition model to be trained (jointly trained), or, alternatively, taking the example of acquiring sample identification features by combining the identification feature extraction model trained with the interaction index recognition model to be trained, where the acquiring sample identification features corresponding to the plurality of historical exposure accounts includes:

inputting the sample identification information into an identification feature extraction model to be trained to perform feature extraction processing to obtain sample identification features;

correspondingly, the performing joint training on the to-be-trained interactive recognition model and the to-be-trained interactive index recognition model in the to-be-trained interactive index recognition model according to the predicted interactive index label, the sample interactive index label, the predicted interactive label and the interactive label corresponding to the predicted interactive label in the sample interactive label may include:

and carrying out joint training on the to-be-trained identification feature extraction model, the to-be-trained interaction identification model and the to-be-trained interaction index identification model in the to-be-trained interaction index identification model according to the predicted interaction index label, the sample interaction index label, the interaction label corresponding to the predicted interaction label in the predicted interaction label and the sample interaction label, so as to obtain an identification feature extraction model corresponding to the target interaction index identification model and the to-be-trained identification feature extraction model.

In a specific embodiment, the sample identification information includes account identification information of a plurality of historical exposure accounts, object identification information of a historical recommendation object corresponding to the plurality of historical exposure accounts, and account identification information of a release account corresponding to the historical recommendation object.

In a specific embodiment, referring to the above related step, second target loss information is determined based on the predicted interactive index tag, the sample interactive index tag, the predicted interactive tag, and the interactive tag corresponding to the predicted interactive tag in the sample interactive tag, and the target interactive index recognition model to be trained and the interactive recognition model to be trained are subjected to back propagation training based on the second target loss information, so that the target interactive index recognition model is replaced with the identification feature extraction model to be trained, the interactive recognition model to be trained, and the interactive index recognition model to be trained based on the second target loss information, so that the identification feature extraction model (the trained identification feature extraction model) corresponding to the identification feature extraction model to be trained and the target interactive index recognition model are obtained. Correspondingly, in the training process, the sample identification features are updated continuously along with the updating of the identification feature extraction model to be trained.

In a specific embodiment, the performing the back propagation training on the identification feature extraction model to be trained, the interaction identification model to be trained and the interaction index identification model to be trained based on the second target loss information to obtain specific refinements of the identification feature extraction model to be trained and the target interaction index identification model corresponding to the identification feature extraction model to be trained, and the performing the back propagation training on the interaction index identification model to be trained and the interaction identification model to be trained based on the second target loss information to obtain specific refinements of the target interaction index identification model are not described herein.

In the above embodiment, the model for extracting the sample identification features is trained together with the interactive index recognition model to be trained, so that the accuracy and the effectiveness of the extracted sample identification features can be greatly improved, the interactive index can be better recognized, and the interactive index recognition accuracy of the trained interactive index recognition model is improved.

sample index associated information corresponding to a plurality of historical exposure accounts is obtained, wherein the sample index associated information is virtual resource consumption condition information corresponding to the plurality of historical exposure accounts and virtual resource obtaining condition information corresponding to object providers of the plurality of historical exposure accounts;

correspondingly, the performing the joint training on the to-be-trained interactive index recognition model and the to-be-trained interactive recognition model based on the sample index association feature, the sample interactive index label and the sample interactive label to obtain the target interactive index recognition model may include:

Based on the sample index association features, the sample interaction index labels and the sample interaction labels, carrying out joint training on the to-be-trained index feature extraction model, the to-be-trained interaction identification model and the to-be-trained interaction index identification model to obtain an index feature extraction model corresponding to the target interaction index identification model and the to-be-trained index feature extraction model.

In a specific embodiment, the sample-index-related feature, the sample-interaction-index-label and the sample-interaction-label are used for performing joint training on the to-be-trained-index-feature extraction model, the to-be-trained-interaction-identification model and the to-be-trained-interaction-index-identification model to obtain specific refinement of the target-interaction-index identification model and the to-be-trained-index-feature extraction model corresponding to the to-be-trained-index-feature extraction model, and the to-be-trained-interaction-index identification model and the to-be-trained-interaction-identification model are used for performing joint training, so that relevant refinement of the target-interaction-index identification model is obtained, which is not repeated herein, and in particular, back propagation training of the to-be-trained-index-feature extraction model can be increased in the back propagation training process.

In a specific embodiment, in the case of training the model for extracting the sample index-related features together with the interactive index recognition model to be trained, the sample index-related features may be updated continuously with the update of the index feature extraction model to be trained.

In the above embodiment, the model for extracting the sample index association features is trained together with the interactive index recognition model to be trained, so that the accuracy and the effectiveness of the extracted sample index association features can be greatly improved, the interactive index can be better recognized, and the interactive index recognition accuracy of the trained interactive index recognition model is improved.

In an optional embodiment, the associated task training features corresponding to the plurality of historical exposure accounts and the sample identification features corresponding to the plurality of historical exposure accounts may be introduced in the model training process, and accordingly, the method may further include:

acquiring associated task training features corresponding to a plurality of historical exposure accounts and/or sample identification features corresponding to a plurality of historical exposure accounts; the associated task training features are task training features extracted in the process of training at least one task model associated with the interaction index recognition model to be trained;

correspondingly, the performing joint training on the interactive index recognition model to be trained and the interactive recognition model to be trained based on the sample index association features, the sample interactive index labels and the sample interactive labels to obtain the target interactive index recognition model comprises the following steps:

based on the sample fusion characteristics, the sample interaction index labels and the sample interaction labels, carrying out joint training on the to-be-trained characteristic fusion model, the to-be-trained interaction identification model and the to-be-trained interaction index identification model to obtain a characteristic fusion model corresponding to the target interaction index identification model and the to-be-trained characteristic fusion model.

In a specific embodiment, the feature fusion model to be trained may include a stitching module, an MLP (Multi-Layer preference Multi-Layer perceptron), and a nonlinear activation Layer.

In a specific embodiment, based on the sample fusion feature, the sample interaction index label and the sample interaction label, the feature fusion model to be trained, the interaction identification model to be trained and the interaction index identification model to be trained are jointly trained, so that the feature fusion model corresponding to the target interaction index identification model and the feature fusion model to be trained can be obtained, and the feature fusion model to be trained and the interaction index identification model to be trained can be jointly trained based on the sample index association feature, the sample interaction index label and the sample interaction label, so that corresponding detailed description of the target interaction index identification model is obtained, which is not repeated herein. Specifically, the relevant sample index associated features in the joint training process can be replaced by corresponding sample fusion features, and in the process of back propagation, back propagation training of the feature fusion model to be trained is increased. Correspondingly, the sample fusion features are updated continuously along with the updating of the model in the training process.

In the above embodiment, in the process of performing joint training on the interactive index recognition model to be trained and the interactive recognition model to be trained, the associated task training features and the sample identification features are introduced, so that the model memory can be enhanced on the basis of enriching sample data, and the interactive index recognition accuracy of the trained interactive index recognition model can be better improved.

In a specific embodiment, as shown in FIG. 5, FIG. 5 is a schematic diagram of an interactive metric recognition model training process according to an exemplary implementation. Optionally, assuming that the task model associated with the interaction index recognition model to be trained comprises a click rate recognition model and a browsing duration recognition model, optionally, in the training process of the click rate recognition model and the browsing duration recognition model, the associated task training features extracted from the associated task training information, the sample index associated features extracted from the sample index associated information based on the to-be-trained index feature extraction model, and the sample identification features extracted from the sample identification information based on the under-band identification feature extraction model are input into a to-be-trained feature fusion model to be subjected to fusion processing, so as to obtain sample fusion features, and then, the first fusion features corresponding to positive sample account numbers in the sample fusion features can be input into the to-be-trained interaction index recognition model to be subjected to interaction index recognition processing; inputting a first fusion feature in the sample fusion features and a second fusion feature corresponding to a part of negative sample account numbers obtained based on random negative sampling (the data difference corresponding to the first fusion feature and the second fusion feature is smaller than a preset threshold value) into an interactive recognition model to be trained for interactive recognition processing; and then, carrying out counter propagation training on the interactive recognition model to be trained, the interactive index recognition model to be trained, the feature fusion model to be trained, the index feature extraction model to be trained and the identification feature extraction model to be trained by combining the sample interactive index label, the sample interactive label, the prediction interactive index label output by the interactive index recognition model to be trained and the prediction interactive label output by the interactive recognition model to be trained until corresponding convergence conditions are reached, so that a trained interactive recognition model, a target interactive index recognition model, a feature fusion model, an index feature extraction model and an identification feature extraction model can be obtained.

In addition, in the back propagation process, the back propagation training is not performed on the feature extraction model to be trained for extracting the relevant task training features, so that the surface affects the relevant task model.

According to the technical scheme provided by the embodiment of the specification, in the process of training the interactive index recognition model, the interactive operation is acquired by combining the object triggered by the positive sample account in the plurality of historical exposure accounts, the historical resource interactive index of the virtual resource quantity brought by the corresponding object provider is acquired, the adjacent plurality of preset index intervals are subjected to label configuration to generate the sample interactive index label, the problem that zero value index is introduced in the process of learning the resource interactive index, the resource interactive index is underestimated is avoided, the sample interactive index label can represent the probability that the historical resource interactive index is larger than or equal to the lower limit value of the plurality of preset index intervals, so that the sample interactive index label indicates the size relation of the plurality of preset index intervals on the basis of indicating the index interval where the positive sample account corresponds to the historical resource interactive index, the distribution balance of the learned resource interactive index can be ensured, and the sample index association characteristic of the virtual resource consumption condition corresponding to the corresponding object provider and the virtual resource acquisition condition are combined, and the sample interactive label can be avoided, the interactive index recognition model to be trained and the interactive index to be combined, the interactive index to be effectively estimated is effectively prevented from being interacted, and the interactive index is effectively estimated, and the interactive index is effectively prevented from being interacted, and the user is effectively interacted with the user to be effectively and the user is prevented from being interacted with the training model.

The following describes an object recommendation method based on the target interaction index recognition model, as shown in fig. 6, fig. 6 is a flowchart illustrating an object recommendation method according to an exemplary embodiment, and optionally, the object recommendation method may be applied to an electronic device such as a server or a terminal, and specifically, the object recommendation method may include the following steps:

in step S601, target index association features are acquired.

In a specific embodiment, the target index association feature is a feature that characterizes a virtual resource consumption condition corresponding to the target account and a virtual resource acquisition condition of at least one target object provider, where the at least one target object provider is a provider of at least one object to be recommended; specifically, the target account number may be an account number to which an object is currently required to be exposed (recommended). The at least one object to be recommended may be an object within the object recommendation platform. Optionally, the objects in the platform can be filtered in advance by combining with the application requirement and combining with the click rate recognition model, the conversion rate recognition model, the browsing duration recognition model and the like, so as to determine at least one object to be recommended.

In an optional embodiment, the acquiring the target index association feature may include:

Acquiring target index association information corresponding to a target account;

and inputting the target index associated information into an index feature extraction model to perform feature extraction processing to obtain target index associated features.

In a specific embodiment, the target index association information may include virtual resource cumulative consumption information of the target account number in at least one first preset history period, single virtual resource consumption information of the target account number in at least one first preset history period, average virtual resources obtained based on the recommended object by the at least one target object provider in a second preset history period (e.g., average virtual resource amount obtained based on the recommended object in a live broadcast room in approximately 30 days, virtual resource amount corresponding to an average single recommended object in a live broadcast room in approximately 30 days, etc.), cumulative virtual resources obtained based on the recommended object by the at least one target object provider in the second preset history period, etc.

In an alternative embodiment, the index feature extraction model may be independently trained, or may be trained with the target interactive index recognition model (i.e., the index feature extraction model is trained in conjunction with the target interactive index recognition model).

In the above embodiment, the target index association features are extracted from the target index association information by combining the index feature extraction model obtained by training together with the target interaction index recognition model, so that the accuracy and effectiveness of the extracted target index association features can be greatly improved, and further, the interaction index can be better recognized, and the interaction index recognition accuracy is improved.

In step S603, the target-index-related features are input into a target-interaction-index recognition model to perform interaction-index recognition processing, so as to obtain a target interaction index tag.

In a specific embodiment, the target interaction index tag may represent a probability that a predicted resource interaction index corresponding to the target account is greater than or equal to a lower limit value of a plurality of preset index intervals; the predicted resource interaction index may trigger object acquisition interaction for at least one object to be recommended for the target account number, and bring a predicted amount of virtual resources for at least one target object provider; the plurality of preset index intervals are a plurality of adjacent index intervals.

In step S605, a predicted resource interaction index is determined according to the target interaction index tag;

in an optional embodiment, determining the predicted resource interaction indicator according to the target interaction indicator label may include:

Determining index mean values corresponding to a plurality of preset index intervals and target index differences corresponding to each preset index interval; and generating a predicted resource interaction index according to the index mean value and the target index difference.

In a specific embodiment, the target index difference corresponding to each preset index interval is a difference between a probability corresponding to each preset index interval (a probability that the predicted resource interaction index is greater than the lower limit value of the preset index interval) and a probability corresponding to a previous preset index interval, where the previous preset index interval is an index interval with an upper limit value adjacent to the lower limit value of each preset index interval; the prediction index data corresponding to the previous preset index region of the first preset index region is zero; the first preset index interval is the interval with the minimum upper limit value among a plurality of preset index intervals. Because in an ideal state, the target index difference between the preset index interval where the predicted resource interaction index is located and the previous preset index interval is 1, and the other intervals and the previous interval are 0, and accordingly, the preset resource interaction index is determined by combining the target index difference and the index mean value, even if the probability that the actually predicted resource interaction index is greater than the lower limit value of the preset index interval has a certain deviation from the actual, the index mean value of the preset index interval where the mainly combined predicted resource interaction index is located can be ensured to determine the preset resource interaction index, and the identification accuracy of the preset resource interaction index is effectively ensured.

In an optional embodiment, generating the predicted resource interaction indicator according to the index mean value and the target index difference may include summing products of the index mean value corresponding to each preset index interval and the target index difference corresponding to each preset index interval to obtain the predicted resource interaction indicator.

In the above embodiment, the predicted resource interaction index is determined by combining the index average value corresponding to the plurality of preset index intervals and the target index difference between the probability corresponding to each preset index interval and the probability corresponding to the previous preset index interval, so that the preset resource interaction index is determined by mainly combining the index average value of the preset index interval where the predicted resource interaction index is located, and the identification accuracy of the preset resource interaction index is effectively ensured.

In step S607, a target object of the at least one object to be recommended is recommended to the target account number based on the predicted resource interaction index.

In a specific embodiment, the object to be recommended with the highest corresponding predicted resource interaction index is taken as the target object, and the target object is sent to the terminal corresponding to the target account, so that the target object is recommended to the target account.

In an alternative embodiment, the method may further include:

acquiring target associated task characteristics corresponding to a target account and/or target identification characteristics corresponding to the target account;

inputting the target index associated features, the target associated task features and/or the target identification features into a feature fusion model for fusion processing to obtain target fusion features;

correspondingly, inputting the target index association characteristic into the target interaction index recognition model to perform interaction index recognition processing, and obtaining the target interaction index label includes:

and inputting the target fusion characteristics into a target interaction index recognition model to perform interaction index recognition processing, so as to obtain a target interaction index label.

In a specific embodiment, the target associated task feature may be a feature required for identifying a task index associated with a predicted resource interaction index; specifically, the task index associated with the predicted resource interaction index may be an index required to be identified by a task model associated with a target interaction index identification model such as click rate, conversion rate, browsing duration, and the like. The feature fusion model is obtained by combined training with a target interaction index recognition model; specifically, the target associated task feature may be task data required in the process of identifying the corresponding index for at least one task model, and optionally, the target associated task feature may include account feature information corresponding to the target account (such as gender of the user corresponding to the account, historical interaction operation, etc.), historical operation information of the target account corresponding to the object to be recommended, object feature information corresponding to the object to be recommended, and interaction features of the object to be recommended in a preset current time period, where the preset current time period includes a current time point.

In an alternative embodiment, the target identification feature may be obtained by performing feature extraction on the target identification information based on an identification feature extraction model. Optionally, the identification feature extraction model can be obtained based on combined training with the target interaction index recognition model, or can be obtained through independent training; optionally, the target identification information may include account identification information of the target account, object identification information of the history recommended object corresponding to the target account, and account identification information of the publishing account corresponding to the history recommended object. Accordingly, the target identification feature may include an account identification feature of the target account, an object identification feature of the target account corresponding to the historical recommendation object, and an account identification feature of the historical recommendation object corresponding to the publishing account.

In the above embodiment, in the process of identifying the interactive index, the target associated task feature and/or the target identification feature are introduced, so that the model memory can be enhanced on the basis of enriching the data for learning the interactive index, and the interactive index identification accuracy of the target interactive index identification model can be better improved.

As can be seen from the technical solutions provided in the embodiments of the present disclosure, in an object recommendation process, a target interaction index tag is obtained by inputting a target index association feature corresponding to a target account into a target interaction index recognition model to perform interaction index recognition processing, where the target interaction index tag may represent a probability that a predicted resource interaction index corresponding to the target account is greater than or equal to a lower limit value of a plurality of preset index intervals, so that the target interaction index tag indicates a size relationship of the plurality of preset index intervals on the basis of indicating an index interval where the predicted resource interaction index corresponding to the target account is located, and distribution balance of learned resource interaction indexes may be ensured; and then, combining the target interaction index label to determine and reflect the target account to trigger the object acquisition interaction aiming at least one object to be recommended, so as to bring the predicted resource interaction index of the virtual resource for at least one target object provider, effectively ensure the index identification accuracy, effectively assist the object recommendation and greatly improve the user experience.

FIG. 7 is a block diagram illustrating an interactive metric recognition model training apparatus, according to an example embodiment. Referring to fig. 7, the apparatus includes:

the sample data obtaining module 710 is configured to perform obtaining sample index association features corresponding to a plurality of historical exposure accounts, sample interaction labels corresponding to a plurality of historical exposure accounts, and historical resource interaction indexes corresponding to positive sample accounts in a plurality of historical exposure accounts, where the historical resource interaction indexes are object obtaining interaction operations triggered by the positive sample accounts, and are virtual resource amounts brought by corresponding object providers; the sample index association features are features for representing virtual resource consumption conditions corresponding to a plurality of historical exposure accounts and virtual resource acquisition conditions corresponding to object providers of the plurality of historical exposure accounts, and the sample interaction labels represent the probability of triggering objects to acquire interaction operations by the plurality of historical exposure accounts;

the tag configuration module 720 is configured to perform tag configuration on a plurality of preset index intervals based on the historical resource interaction indexes to obtain sample interaction index tags, wherein the sample interaction index tags represent the probability that the historical resource interaction indexes are larger than or equal to the lower limit values of the preset index intervals; the plurality of preset index intervals are a plurality of adjacent index intervals;

The joint training module 730 is configured to perform joint training on the to-be-trained interactive index recognition model and the to-be-trained interactive recognition model based on the sample index association feature, the sample interactive index label and the sample interactive label, so as to obtain a target interactive index recognition model.

In an alternative embodiment, the plurality of historical exposure accounts further includes a negative sample account; the joint training module 730 includes:

the random sampling unit is configured to perform random sampling on the first index association characteristic corresponding to the negative sample account in the sample index association characteristics to obtain a second index association characteristic; the difference between the data volume corresponding to the second index association feature and the data volume corresponding to the third index association feature is smaller than a preset threshold, and the third index association feature is an index association feature corresponding to a positive sample account in the sample index association features;

the interactive identification processing unit is configured to execute the interactive identification processing of inputting the second index association characteristic and the third index association characteristic into the interactive identification model to be trained to obtain a predicted interactive label;

The joint training unit is configured to perform joint training on the interactive index recognition model to be trained and the interactive recognition model to be trained according to the predicted interactive index label, the sample interactive index label, the interactive label corresponding to the predicted interactive label in the predicted interactive label and the sample interactive label, so as to obtain a target interactive index recognition model.

In an alternative embodiment, the apparatus further comprises:

the associated task training feature acquisition module is configured to acquire associated task training features corresponding to a plurality of historical exposure accounts; the associated task training features are extracted task training features in the process of training at least one task model associated with the interaction index recognition model to be trained;

the first interactive index recognition processing module is specifically configured to execute the steps of inputting a first training feature corresponding to a positive sample account in the third index association feature and the associated task training feature into an interactive index recognition model to be trained for interactive index recognition processing, and obtaining a predicted interactive index label;

the interactive recognition processing module is specifically configured to execute the steps of inputting the second index association feature, the third index association feature, the first training feature and the second training feature into an interactive recognition model to be trained to perform interactive recognition processing, so as to obtain a predicted interactive label; the second training features are training features of the target negative sample account in the associated task training features; the target negative sample account is a negative sample account corresponding to the second index association feature.

In an alternative embodiment, the apparatus further comprises:

the first interactive index recognition processing module is specifically configured to execute the steps of inputting a first identification feature corresponding to a positive sample account number in the third index association feature and the sample identification feature into an interactive index recognition model to be trained for interactive index recognition processing, and obtaining a predicted interactive index label;

the interactive identification processing module is specifically configured to execute the steps of inputting the second index association feature, the third index association feature, the first identification feature and the second identification feature into an interactive identification model to be trained for interactive identification processing, and obtaining a predicted interactive label; the second identification feature is an identification feature of a target negative sample account in the sample identification features; the target negative sample account is a negative sample account corresponding to the second index association feature.

In an alternative embodiment, the sample identification feature acquisition module comprises:

the first feature extraction processing unit is configured to input sample identification information into an identification feature extraction model to be trained to perform feature extraction processing, so as to obtain sample identification features;

The interactive recognition processing module is specifically configured to execute joint training according to the predicted interactive index label, the sample interactive index label, the interactive label corresponding to the predicted interactive label in the predicted interactive label and the sample interactive label, the to-be-trained identification feature extraction model in the to-be-trained interactive index recognition model, the to-be-trained interactive recognition model and the to-be-trained interactive index recognition model, and obtain an identification feature extraction model corresponding to the target interactive index recognition model and the to-be-trained identification feature extraction model.

In an alternative embodiment, the sample data acquisition module 710 includes:

the sample index associated information acquisition unit is configured to acquire sample index associated information corresponding to a plurality of historical exposure accounts, wherein the sample index associated information is virtual resource consumption condition information corresponding to the plurality of historical exposure accounts and virtual resource acquisition condition information of an object provider corresponding to the plurality of historical exposure accounts;

the second feature extraction processing unit is configured to input sample index association information into an index feature extraction model to be trained to perform feature extraction processing, so as to obtain sample index association features;

the joint training module 730 is specifically configured to perform joint training on the to-be-trained index feature extraction model, the to-be-trained interactive recognition model and the to-be-trained interactive index recognition model based on the sample index associated feature, the sample interactive index label and the sample interactive label, so as to obtain an index feature extraction model corresponding to the target interactive index recognition model and the to-be-trained index feature extraction model.

In an alternative embodiment, the plurality of historical exposure accounts further includes a negative sample account; the device further comprises:

the characteristic acquisition module is configured to acquire associated task training characteristics corresponding to the plurality of historical exposure accounts and/or sample identification characteristics corresponding to the plurality of historical exposure accounts; the associated task training features are task training features extracted in the process of training at least one task model associated with the interaction index recognition model to be trained;

the first fusion processing module is configured to execute the process of inputting the sample index associated features, the associated task training features and/or the sample identification features into the feature fusion model to be trained for fusion processing to obtain sample fusion features;

the joint training module 730 is specifically configured to perform joint training on the feature fusion model to be trained, the interactive identification model to be trained, and the interactive index identification model to be trained based on the sample fusion feature, the sample interactive index label, and the sample interactive label, so as to obtain a feature fusion model corresponding to the target interactive index identification model and the feature fusion model to be trained.

In an alternative embodiment, the tag configuration module 720 includes:

a target index section determining unit configured to perform determination of a target index section in which the history resource interaction index is located from among a plurality of preset index sections;

The first label configuration unit is configured to execute label configuration on the first index interval based on a first preset label to obtain a first interactive index label; the first index interval is an interval in which the upper limit value of the plurality of preset index intervals is smaller than the lower limit value of the plurality of preset index intervals, and the target index is larger than or equal to the lower limit value of the plurality of preset index intervals;

the second label configuration unit is configured to perform label configuration on the target index interval and the second index interval based on a second preset label to obtain a second interactive index label; the second index section is a section in which the lower limit value of the plurality of preset index sections is larger than the upper limit value of the target index section;

and the sample interaction index label generating unit is configured to generate a sample interaction index label according to the first interaction index label and the interaction index label.

The specific manner in which the various modules perform the operations in the apparatus of the above embodiments have been described in detail in connection with the embodiments of the method, and will not be described in detail herein.

FIG. 8 is a block diagram of an object recommendation device, according to an example embodiment. Referring to fig. 8, the apparatus includes:

a target-index-associated feature acquisition module 810 configured to perform acquisition of target-index-associated features; the target index association features are features for representing virtual resource consumption conditions corresponding to the target account and virtual resource acquisition conditions of at least one target object provider, wherein the at least one target object provider is at least one provider of an object to be recommended;

The second interactive index recognition processing module 820 is configured to perform interactive index recognition processing on the target interactive index recognition model obtained by inputting the target index association features based on the interactive index recognition model training method provided in the first aspect, so as to obtain a target interactive index label, wherein the target interactive index label characterizes the probability that the predicted resource interactive index corresponding to the target account is greater than or equal to the lower limit value of a plurality of preset index intervals; the predicted resource interaction index is a target account number, the object acquisition interaction aiming at least one object to be recommended is triggered, and the predicted quantity of the virtual resource is brought to at least one target object provider; the plurality of preset index intervals are a plurality of adjacent index intervals;

a predicted resource interaction index determination module 830 configured to perform determining a predicted resource interaction index based on the target interaction index tag;

the object pushing module 840 is configured to perform recommendation of a target object of the at least one object to be recommended to the target account number based on the predicted resource interaction index.

In an alternative embodiment, the apparatus further comprises:

the characteristic acquisition module is configured to acquire target associated task characteristics corresponding to the target account and/or target identification characteristics corresponding to the target account; the target associated task features are features required for identifying task indexes associated with predicted resource interaction indexes;

The second fusion processing module is configured to execute fusion processing on the target index associated feature, the target associated task feature and/or the target identification feature input feature fusion model to obtain a target fusion feature; the feature fusion model is obtained by combined training with the target interaction index recognition model;

the second interactive index recognition processing module 820 is specifically configured to perform interactive index recognition processing by inputting the target fusion feature into the target interactive index recognition model, so as to obtain a target interactive index tag.

In an alternative embodiment, the predictive resource interaction indicator determination module 830 includes:

a calculation unit configured to perform determining an index mean value corresponding to a plurality of preset index sections and a target index difference corresponding to each preset index section, the target index difference corresponding to each preset index section being a difference between a probability corresponding to each preset index section and a probability corresponding to a previous preset index section, the previous preset index section being an index section having an upper limit value adjacent to a lower limit value of each preset index section; the prediction index data corresponding to the previous preset index region of the first preset index region is zero; the first preset index interval is an interval with the minimum upper limit value among a plurality of preset index intervals;

And the prediction resource interaction index generating unit is configured to generate a prediction resource interaction index according to the index mean value and the target index difference.

In an alternative embodiment, the target-index-associated-feature acquisition module 810 includes:

the third feature extraction processing unit is configured to perform feature extraction processing by inputting the target index association information into an index feature extraction model, so as to obtain target index association features, wherein the index feature extraction model is obtained by combined training with a target interactive index recognition model.

FIG. 9 is a block diagram of an electronic device, which may be a server, with an internal structure diagram as shown in FIG. 9, for interactive metric recognition model training, according to an example embodiment. The electronic device includes a processor, a memory, and a network interface connected by a system bus. Wherein the processor of the electronic device is configured to provide computing and control capabilities. The memory of the electronic device includes a nonvolatile 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 network interface of the electronic device is used for communicating with an external terminal through a network connection. The computer program, when executed by a processor, implements an interactive index recognition model training method.

Fig. 10 is a block diagram illustrating an electronic device for object recommendation, which may be a terminal, according to an exemplary embodiment, and an internal structure diagram thereof may be as shown in fig. 10. The electronic device includes a processor, a memory, a network interface, a display screen, and an input device connected by a system bus. Wherein the processor of the electronic device is configured to provide computing and control capabilities. The memory of the electronic device includes a nonvolatile 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 network interface of the electronic device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement an object recommendation method. The display screen of the electronic equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the electronic 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 electronic 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. 9 or 10 are merely block diagrams of portions of structures related to the disclosed aspects and do not constitute limitations of the electronic devices to which the disclosed aspects may be applied, and that a particular electronic device may include more or fewer components than shown, or may combine certain components, or have a different arrangement of components.

In an exemplary embodiment, there is also provided an electronic device including: a processor; a memory for storing the processor-executable instructions; wherein the processor is configured to execute the instructions to implement an interactive index recognition model training method or an object recommendation method as in embodiments of the present disclosure.

In an exemplary embodiment, a computer readable storage medium is also provided, which when executed by a processor of an electronic device, causes the electronic device to perform the interactive index recognition model training method or the object recommendation method in the embodiments of the present disclosure.

In an exemplary embodiment, a computer program product containing instructions that, when run on a computer, cause the computer to perform the interactive index recognition model training method or the object recommendation method in the embodiments of the present disclosure is also provided.

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, storage, database, or other medium used in the various embodiments provided herein may include non-volatile and/or volatile memory. The nonvolatile memory can include Read Only Memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), memory bus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), among others.

Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any adaptations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.

It is to be understood that the present disclosure is not limited to the precise arrangements and instrumentalities shown in the drawings, and that various modifications and changes may be effected without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. The interactive index recognition model training method is characterized by comprising the following steps of:

2. The interactive index recognition model training method of claim 1, wherein a plurality of the historical exposure accounts further comprises negative sample accounts; based on the sample index association feature, the sample interaction index label and the sample interaction label, performing joint training on the to-be-trained interaction index recognition model and the to-be-trained interaction recognition model to obtain a target interaction index recognition model comprises the following steps:

3. The interactive metrics recognition model training method of claim 2, further comprising:

4. The interactive metrics recognition model training method of claim 2, further comprising:

5. The method for training an interactive index recognition model according to claim 4, wherein the obtaining sample identification features corresponding to a plurality of historical exposure accounts comprises:

6. The method for training an interactive index recognition model according to any one of claims 1 to 5, wherein the obtaining sample index association features corresponding to a plurality of historical exposure accounts comprises:

7. The interactive index recognition model training method according to any one of claims 1 to 5, wherein a plurality of the historical exposure accounts further comprises negative sample accounts; the method further comprises the steps of:

8. The method for training an interactive index recognition model according to any one of claims 1 to 5, wherein the performing label configuration on a plurality of preset index intervals based on the historical resource interactive index to obtain a sample interactive index label includes:

9. An object recommendation method, comprising:

Inputting the target index association characteristics into a target interaction index recognition model obtained based on the interaction index recognition model training method according to any one of claims 1 to 8 for interaction index recognition processing to obtain a target interaction index label, wherein the target interaction index label represents the probability that the predicted resource interaction index corresponding to the target account is greater than or equal to the lower limit value of a plurality of preset index intervals; the predicted resource interaction index triggers object acquisition interaction aiming at least one object to be recommended for the target account number, and a predicted quantity of virtual resources is brought for at least one target object provider; the preset index intervals are adjacent index intervals;

10. The object recommendation method of claim 9, further comprising:

inputting the target index association features into a target interaction index recognition model obtained based on the interaction index recognition model training method according to any one of claims 1 to 8 for interaction index recognition processing, and obtaining a target interaction index label comprises the following steps:

11. The method according to any one of claims 9 or 10, wherein determining the predicted resource interaction indicator according to the target interaction indicator label comprises:

12. The method for recommending objects according to any one of claims 9 or 10, wherein the obtaining the target index association feature comprises:

13. An interactive index recognition model training device, characterized by comprising:

14. An object recommendation device, characterized by comprising:

the second interactive index recognition processing module is configured to perform interactive index recognition processing on the target interactive index recognition model obtained based on the interactive index recognition model training method according to any one of claims 1 to 8 by inputting the target index association characteristics, so as to obtain a target interactive index label, wherein the target interactive index label represents the probability that the predicted resource interactive index corresponding to the target account is greater than or equal to the lower limit value of a plurality of preset index intervals; the predicted resource interaction index triggers object acquisition interaction aiming at least one object to be recommended for the target account number, and a predicted quantity of virtual resources is brought for at least one target object provider; the preset index intervals are adjacent index intervals;

15. An electronic device, comprising:

a processor;

a memory for storing the processor-executable instructions;

wherein the processor is configured to execute the instructions to implement the interactive index recognition model training method of any one of claims 1 to 8 or the object recommendation method of any one of claims 9 to 12.

16. A computer readable storage medium, characterized in that instructions in the storage medium, when executed by a processor of an electronic device, enable the electronic device to perform the interactive index recognition model training method of any one of claims 1 to 8 or the object recommendation method of any one of claims 9 to 12.