CN114077701A

CN114077701A - Method and device for determining resource information, computer equipment and storage medium

Info

Publication number: CN114077701A
Application number: CN202010812722.2A
Authority: CN
Inventors: 郭正凯; 胡勇
Original assignee: Beijing Dajia Internet Information Technology Co Ltd
Current assignee: Beijing Dajia Internet Information Technology Co Ltd
Priority date: 2020-08-13
Filing date: 2020-08-13
Publication date: 2022-02-22

Abstract

The disclosure relates to a method and a device for determining resource information, a computer device and a storage medium, and belongs to the technical field of multimedia. According to the method, the fusion characteristic used for representing the characteristics of the user information and the characteristics of the resource information is obtained, the first behavior parameters of each interactive behavior are respectively and independently predicted based on the fusion characteristic, the fusion characteristic and the first behavior parameters of each interactive behavior are comprehensively considered, the second behavior parameters of each interactive behavior are obtained, the first behavior parameters of other interactive behaviors are referred in the process of obtaining the second behavior parameters of each interactive behavior, so that potential relation among different interactive behaviors is cooperatively considered, the second behavior parameters have higher accuracy compared with the first behavior parameters, and the accuracy of the resource recommendation process is greatly improved.

Description

Method and device for determining resource information, computer equipment and storage medium

Technical Field

The present disclosure relates to the field of multimedia technologies, and in particular, to a method and an apparatus for determining resource information, a computer device, and a storage medium.

Background

With the development of computer technology, more and more users can browse multimedia resources such as video, audio, image and text information and the like at any time through the terminal, and the server can recommend some multimedia resources which may be interested by the users from a massive multimedia database, so that the resource browsing requirements of the users can be better met.

Taking audio recommendation as an example, the server usually extracts the feature vector of the user and the feature vector of the audio to be recommended respectively, inputs the two into the neural network, and predicts the predicted values of various behavior tags of the audio to be recommended by the user through the neural network, for example, the behavior tags may include click behavior, collection behavior, use behavior, upload behavior, and the like.

However, in some scenarios, there is a certain correlation between behavior tags, for example, after a user collects an audio, the audio is used to photograph related works within one day with a high probability, that is, there is a potential relationship between the collection behavior and the usage behavior, and in such a scenario, if the predicted values of the behavior tags are still independently and respectively predicted through the neural network, the resource recommendation accuracy will be poor.

Disclosure of Invention

The disclosure provides a resource information determination method, a resource information determination device, a computer device and a storage medium, so as to at least improve the resource recommendation accuracy. The technical scheme of the disclosure is as follows:

according to a first aspect of the embodiments of the present disclosure, there is provided a method for determining resource information, including:

acquiring fusion characteristics based on user information of a target account and resource information of candidate resources, wherein the fusion characteristics are used for representing characteristics of the user information and characteristics of the resource information;

acquiring a plurality of first behavior parameters based on the fusion characteristics, wherein the first behavior parameters are used for representing the possibility of one interaction behavior of the target account on the candidate resources;

acquiring a plurality of second behavior parameters based on the fusion features and the plurality of first behavior parameters, wherein the second behavior parameters are used for representing the possibility of the one interactive behavior under the condition that other interactive behaviors except the one interactive behavior occur;

and recommending resources to the target account based on the plurality of second behavior parameters.

In one possible implementation manner, the acquiring the fusion feature based on the user information of the target account and the resource information of the candidate resource includes:

inputting the user information and the resource information into a first fully-connected network, and performing feature extraction on the user information and the resource information through the first fully-connected network to obtain the fusion feature, wherein the first fully-connected network is used for extracting the fusion feature.

In one possible implementation, the performing, by the first fully-connected network, feature extraction on the user information and the resource information to obtain the fusion feature includes:

and weighting the user information and the resource information through at least one full connection layer in the first full connection network, and acquiring the output characteristic of the last full connection layer as the fusion characteristic.

In one possible embodiment, the obtaining a plurality of first behavior parameters based on the fused feature includes:

and mapping the fusion features to a plurality of behavior tags to obtain a plurality of first behavior parameters, wherein one behavior tag corresponds to a first behavior parameter of an interaction behavior.

In one possible embodiment, the obtaining a plurality of second behavior parameters based on the fused feature and the plurality of first behavior parameters includes:

acquiring a plurality of behavior characteristics respectively corresponding to the plurality of value intervals based on the plurality of value intervals where the plurality of first behavior parameters are located, wherein the behavior characteristics are used for expressing parameter characteristics of the first behavior parameters;

splicing the fusion characteristics, the plurality of behavior characteristics, the account identification of the target account and the resource identification of the candidate resource to obtain splicing characteristics;

inputting the splicing characteristics into a second fully-connected network, and processing the splicing characteristics through the second fully-connected network to obtain the plurality of second behavior parameters, wherein the second fully-connected network is used for obtaining the second behavior parameters according to the splicing characteristics.

In one possible embodiment, the processing the splicing feature through the second fully-connected network to obtain the plurality of second behavior parameters includes:

and performing weighting processing on the splicing characteristics through at least one fully-connected layer in the second fully-connected network, and mapping the output characteristics of the last fully-connected layer to a plurality of behavior labels to obtain a plurality of second behavior parameters.

In one possible embodiment, the recommending the resource to the target account based on the plurality of second behavior parameters includes:

carrying out weighted summation on the plurality of second behavior parameters to obtain global parameters;

recommending the candidate resources to the target account based on the fact that the global parameter is larger than a parameter threshold.

According to a second aspect of the embodiments of the present disclosure, there is provided an apparatus for determining resource information, including:

a first obtaining unit configured to execute resource information of a candidate resource and user information based on a target account, and obtain a fusion feature, where the fusion feature is used to represent a feature of the user information and a feature of the resource information;

a second obtaining unit, configured to perform obtaining, based on the fusion feature, a plurality of first behavior parameters, where the first behavior parameters are used to indicate a possibility that the target account performs an interaction behavior on the candidate resource;

a third obtaining unit configured to obtain a plurality of second behavior parameters indicating a possibility of occurrence of the one interactive behavior in a case where other interactive behaviors than the one interactive behavior occur, based on the fusion feature and the plurality of first behavior parameters;

and the resource recommending unit is configured to perform resource recommendation on the target account based on the plurality of second behavior parameters.

In one possible implementation, the first obtaining unit includes:

an extraction subunit, configured to perform inputting the user information and the resource information into a first fully-connected network, and perform feature extraction on the user information and the resource information through the first fully-connected network to obtain the fusion feature, where the first fully-connected network is used to extract the fusion feature.

In one possible implementation, the extraction subunit is configured to perform:

In one possible implementation, the second obtaining unit is configured to perform:

In one possible implementation, the third obtaining unit includes:

the obtaining subunit is configured to perform, based on a plurality of value intervals where the plurality of first behavior parameters are located, obtaining a plurality of behavior characteristics corresponding to the plurality of value intervals, respectively, where the behavior characteristics are used to represent parameter characteristics of the first behavior parameters;

the splicing subunit is configured to splice the fusion feature, the plurality of behavior features, the account identifier of the target account and the resource identifier of the candidate resource to obtain a spliced feature;

and the processing subunit is configured to input the splicing characteristic into a second fully-connected network, and process the splicing characteristic through the second fully-connected network to obtain the plurality of second behavior parameters, where the second fully-connected network is configured to obtain the second behavior parameters according to the splicing characteristic.

In one possible embodiment, the processing subunit is configured to perform:

In one possible embodiment, the resource recommendation unit is configured to perform:

According to a third aspect of embodiments of the present disclosure, there is provided a computer device comprising:

one or more processors;

one or more memories for storing the one or more processor-executable program codes;

wherein the one or more processors are configured to perform the method for determining resource information of any of the above first aspect and possible implementations of the first aspect.

According to a fourth aspect of embodiments of the present disclosure, there is provided a storage medium, wherein at least one program code of the storage medium, when executed by one or more processors of a computer device, enables the computer device to perform the method for determining resource information of any one of the above first aspect and possible implementations of the first aspect.

According to a fifth aspect of embodiments of the present disclosure, there is provided a computer program product comprising one or more program codes executable by one or more processors of a computer device, such that the computer device is capable of performing the method of determining resource information of any one of the above-mentioned first aspect and possible implementations of the first aspect.

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

the method comprises the steps of obtaining a fusion characteristic used for representing characteristics of user information and characteristics of resource information, respectively and independently predicting first behavior parameters of each interactive behavior based on the fusion characteristic, comprehensively considering the fusion characteristic and the first behavior parameters of each interactive behavior, and obtaining second behavior parameters of each interactive behavior.

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 present disclosure and, together with the description, serve to explain the principles of the disclosure and are not to be construed as limiting the disclosure.

FIG. 1 is a schematic diagram of an implementation environment of a method for determining resource information according to an example embodiment;

FIG. 2 is a flow diagram illustrating a method of determining resource information in accordance with an example embodiment;

FIG. 3 is a flow diagram illustrating a method for determining resource information in accordance with an exemplary embodiment;

FIG. 4 is a schematic diagram of a fully-connected layer provided by embodiments of the present disclosure;

fig. 5 is a schematic diagram of a method for determining resource information according to an embodiment of the present disclosure;

FIG. 6 is a schematic flow chart diagram of a resource recommendation process provided by the disclosed embodiments;

fig. 7 is a block diagram illustrating a logical structure of a resource information determination apparatus according to an exemplary embodiment;

fig. 8 is a schematic structural diagram of a computer device according to an embodiment of the present disclosure.

Detailed Description

In order to make the technical solutions of the present disclosure better understood by those of ordinary skill in the art, the technical solutions in 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 above-described drawings are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the disclosure described herein are capable of operation in sequences other than those illustrated or otherwise described herein. The implementations described in the exemplary embodiments below are not intended to represent all implementations consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure, as detailed in the appended claims.

The user information to which the present disclosure relates may be information authorized by the user or sufficiently authorized by each party.

Fig. 1 is a schematic diagram of an implementation environment of a resource information determination method according to an exemplary embodiment, referring to fig. 1, in which at least one terminal 101 and a server 102 may be included, and the following details are described below:

at least one terminal 101 is used for browsing multimedia resources, each of the at least one terminal 101 may have an application installed thereon, the application may be any client capable of providing a multimedia resource browsing service, and a user may browse the multimedia resources by starting the application, the application may be at least one of a shopping application, a takeaway application, a travel application, a game application or a social application, and the multimedia resources may include at least one of video resources, audio resources, picture resources, text resources or web resources.

The server 102 is a computer device for providing the multimedia resource recommendation service to the at least one terminal 101. The server 102 may include at least one of a server, a plurality of servers, a cloud computing platform, or a virtualization center. Alternatively, the server 102 may undertake primary computational work and the at least one terminal 101 may undertake secondary computational work; alternatively, the server 102 may undertake secondary computing work and the at least one terminal 101 may undertake primary computing work; alternatively, the server 102 and the at least one terminal 101 perform cooperative computing by using a distributed computing architecture.

In some embodiments, the server 102 may collect a behavior log of the platform user on the application program, select at least one sample user account from a total number of platform users in a random sampling manner, and obtain sample data of the at least one sample user account according to the behavior log of the at least one sample user account, so as to perform offline training on the first initial network and the second initial network, respectively, to obtain the first fully connected network and the second fully connected network. Wherein the first fully connected network and the second fully connected network are used for performing online evaluation of the resource recommendation process.

On the basis, the server 102 can extract a fusion feature for representing the global characteristic between the user information of the target account and the resource information of the candidate resource based on the first fully-connected network, predict a plurality of first behavior parameters according to the fusion feature, respectively extract a plurality of behavior features corresponding to the plurality of first behavior parameters, splice the fusion feature, the plurality of behavior features, the account identifier of the target account and the resource identifier of the candidate resource to obtain a spliced feature, call a second fully-connected network to process the spliced feature, predict a plurality of second behavior parameters, and then perform resource recommendation according to the plurality of second behavior parameters.

Optionally, the server 102 pushes any multimedia resource of which the second behavior parameter is greater than the parameter threshold to the terminal of the target account when recommending, optionally, the server performs weighted summation on the plurality of second behavior parameters to obtain a global parameter, and pushes the multimedia resource of which the global parameter is greater than the parameter threshold to the terminal of the target account, optionally, the server 102 sorts the multimedia resources according to a descending order of any second behavior parameter (or global parameter), and pushes the sorted multimedia resource located in the top N digits to the terminal of the user corresponding to the target account, where N is an integer greater than or equal to 1. The server 102 transmits the selected multimedia resource to the corresponding terminal so that the terminal can present the recommended multimedia resource in the application.

In an exemplary scenario, taking multimedia resources as audio resources (hereinafter referred to as "audio") for example, a short video application may be installed on a terminal of a user corresponding to a target account, the server 102 provides a short video service platform to the terminal through the short video application, the user may browse short videos or upload short videos through the short video application, the server 102 screens out a plurality of audios to be recommended corresponding to the target account from a massive audio library in a coarse ranking (i.e., coarse ranking) stage, predicts a plurality of second behavior parameters of the target account for any audio to be recommended in a fine ranking (i.e., fine ranking) stage based on the method for determining resource information provided by the embodiments of the present disclosure, and then can select different second behavior parameters as reference indexes to recommend resources according to requirements in different scenes based on different recommendation targets, for example, the plurality of second behavior parameters include a click rate parameter, a collection rate parameter, a confirmation usage rate parameter, an upload rate parameter, and the like, in a scenario in which only a click rate condition is considered, the server 102 sends, to a terminal in which the target account is located, audio to be recommended that is located in the top N bits after the click rate parameters are sorted in a descending order, in a scenario in which a comprehensive condition of multiple indexes is considered in a demand, the server 102 obtains a global parameter obtained by weighted summation of the plurality of second behavior parameters, and sends, to the terminal in which the target account is located, audio to be recommended that is located in the top N bits after the global parameter is sorted in a descending order. The server 102 sends the audio to be recommended to the terminal, and optionally, may also collect relevant data of interaction behaviors (including behaviors such as clicking, collecting, confirming use, uploading, and the like) fed back by the user to the audio to be recommended, so as to perform iterative update on the first fully connected network and the second fully connected network as a new round of sample data.

It should be noted that the device type of any one of the at least one terminal 101 may include: at least one of a smart phone, a tablet computer, an e-book reader, an MP3(Moving Picture Experts Group Audio Layer III, motion Picture Experts compression standard Audio Layer 3) player, an MP4(Moving Picture Experts Group Audio Layer IV, motion Picture Experts compression standard Audio Layer 4) player, a laptop portable computer, or a desktop computer. For example, the any terminal may be a smartphone, or other hand-held portable electronic device. The following embodiments are illustrated with the terminal comprising a smartphone.

Those skilled in the art will appreciate that the number of terminals described above may be greater or fewer. For example, the number of the terminals may be only one, or several tens or hundreds of the terminals, or more. The number of terminals and the type of the device are not limited in the embodiments of the present disclosure.

Fig. 2 is a flowchart illustrating a method for determining resource information according to an exemplary embodiment, and referring to fig. 2, the method for determining resource information is applied to a computer device, and the computer device is taken as an example for explanation.

In step 201, the server obtains a fusion feature based on the user information of the target account and the resource information of the candidate resource, where the fusion feature is used to represent a feature of the user information and a feature of the resource information.

In step 202, the server obtains a plurality of first behavior parameters based on the fusion feature, where the first behavior parameters are used to indicate a possibility that the target account performs an interaction behavior with the candidate resource.

In step 203, the server obtains a plurality of second behavior parameters based on the fusion feature and the plurality of first behavior parameters, where the second behavior parameters are used to indicate that the one interactive behavior is likely to occur if other interactive behaviors than the one interactive behavior occur.

In step 204, the server makes a resource recommendation for the target account based on the plurality of second behavior parameters.

According to the method provided by the embodiment of the disclosure, the first behavior parameters of each interactive behavior are respectively and independently predicted by obtaining the fusion feature of the features representing the user information and the features of the resource information, the fusion feature and the first behavior parameters of each interactive behavior are comprehensively considered, and the second behavior parameters of each interactive behavior are obtained.

In one possible implementation, the acquiring the fusion characteristics based on the user information of the target account and the resource information of the candidate resource includes:

In one possible embodiment, performing feature extraction on the user information and the resource information through the first fully-connected network to obtain the fusion feature includes:

In one possible embodiment, based on the fused feature, obtaining a plurality of first behavior parameters includes:

and mapping the fusion feature to a plurality of behavior tags to obtain a plurality of first behavior parameters, wherein one behavior tag corresponds to a first behavior parameter of an interaction behavior.

In one possible embodiment, obtaining a plurality of second behavior parameters based on the fused feature and the plurality of first behavior parameters comprises:

acquiring a plurality of behavior characteristics respectively corresponding to the plurality of value intervals based on the plurality of value intervals where the plurality of first behavior parameters are located, wherein the behavior characteristics are used for expressing the parameter characteristics of the first behavior parameters;

splicing the fusion characteristics, a plurality of behavior characteristics, the account identification of the target account and the resource identification of the candidate resource to obtain splicing characteristics;

inputting the splicing characteristic into a second fully-connected network, processing the splicing characteristic through the second fully-connected network to obtain the plurality of second behavior parameters, wherein the second fully-connected network is used for obtaining the second behavior parameters according to the splicing characteristic.

In one possible embodiment, the processing the splicing characteristic through the second fully-connected network to obtain the plurality of second behavior parameters includes:

and recommending the candidate resource to the target account based on the fact that the global parameter is larger than a parameter threshold.

All the above optional technical solutions may be combined arbitrarily to form the optional embodiments of the present disclosure, and are not described herein again.

Fig. 3 is a flowchart illustrating a method for determining resource information according to an exemplary embodiment, where the method for determining resource information is applied to a computer device, and the computer device is taken as a server as an example to describe below, and the embodiment includes the following steps.

In step 301, the server obtains user information of the target account and resource information of the candidate resource.

The candidate resource is a multimedia resource to be recommended to the target account.

The target account is any user account registered in the server, the user corresponding to the target account is one or more users using the target account, and the user can start an application program on the terminal and log in the target account, and then browse, approve and pay attention to each multimedia resource in the application program.

The user corresponding to the target account may be a "new user" of the application, that is, a user who registers the user account in the application for the first time, and the target account may also be an "old user" of the application, that is, a user who has registered the user account in the application for the first time.

Optionally, the user information of the target account may include user attribute information and user behavior information. The user attribute information comprises information which is fully authorized by the user, such as account identification, a nickname of the user, the gender of the user, the age of the user, the occupation of the user or the geographical position of the user. The user behavior information refers to interaction behavior information generated by the target account on recommended resources in a historical recommendation process.

Illustratively, in an audio recommendation scenario, the user behavior information includes at least one of click behavior information, collection behavior information, confirmation usage behavior information, or upload behavior information of a target account on a recommended audio, where the click behavior information indicates whether a user corresponding to the target account clicks the recommended audio, the collection behavior information indicates whether a user corresponding to the target account collects the recommended audio, the confirmation usage behavior information indicates whether a user corresponding to the target account records a short video work using the recommended audio as background music, and the upload behavior information indicates whether a user corresponding to the target account uploads the short video work recorded using the recommended audio as background music.

Exemplarily, in a video recommendation scene, the user behavior information includes click behavior information, praise behavior information, collection behavior information, and attention behavior information of a target account for a recommended video, where the click behavior information and the collection behavior information are similar to those in an audio recommendation scene and are not described herein, and the praise behavior information refers to whether a user corresponding to the target account approves the recommended video, and the attention behavior information refers to whether a user corresponding to the target account pays attention to a publisher of the recommended video after playing.

Exemplarily, in an advertisement delivery scene, the user behavior information includes click behavior information, consumption behavior information and feedback behavior information of a target account for a delivered advertisement, where the click behavior information is similar to an audio recommendation scene and a video recommendation scene and is not described herein, the consumption behavior information refers to whether a user corresponding to the target account purchases a good or service promoted by the delivered advertisement, and the feedback behavior information refers to a feedback behavior generated by the user corresponding to the target account for an original video after the advertisement is played, and includes normal viewing, reduced viewing, exiting from a client, and the like.

In some embodiments, the click behavior information, the collection behavior information, the confirmation use behavior information, the upload behavior information, the praise behavior information, the attention behavior information, the consumption behavior information, and the like in the above exemplary scenario may be represented by a binary value, for example, true represents that the user corresponding to the target account performs a related interaction behavior, and false represents that the user corresponding to the target account does not perform a related interaction behavior.

In some embodiments, if the user corresponding to the target account clicks the recommended resource for multiple times, the click time and the play duration of each time may be collected in the click behavior information, and the total number of clicks of the user corresponding to the target account is counted, at this time, a list or an array may be used to record the click behavior information.

Optionally, the resource information of the candidate resource refers to related information of a resource to be recommended for the target account in the resource recommendation process, the candidate resource may be any multimedia resource in a multimedia library, and the multimedia resource may include at least one of a video resource, an audio resource, a picture resource, a text resource, or a web resource, or the candidate resource may also be a multimedia resource preliminarily screened from the multimedia library by the server after one round of rough ranking.

In some embodiments, the resource information may include at least one of a name, a cover page, a summary, a content tag, or author information of the candidate resource, it being noted that for audio resources, its author corresponds to a singer, for video resources, its author corresponds to a director if it is a movie, a tv show, etc., and its author is typically the video publisher itself (up master) if it is a personal original video.

In the above process, the server may collect user attribute information submitted by the user when registering the target account and subject to repeated authorization of the user, record a behavior log in a history recommendation process on an application program of the terminal, extract user behavior information according to the behavior log, acquire the user attribute information and the user behavior information as user information, further select one or more candidate resources from a multimedia library, take any one of the candidate resources as an example for explanation, and acquire resource information of the candidate resources.

In step 302, the server inputs the user information and the resource information into a first fully-connected network, and performs feature extraction on the user information and the resource information through the first fully-connected network to obtain a fusion feature.

Wherein the fusion characteristics are used for representing the characteristics of the user information and the characteristics of the resource information.

Wherein the first fully connected network is used for extracting the fusion feature, and the first fully connected network comprises at least one fully connected layer.

Optionally, the server performs weighting processing on the user information and the resource information through at least one fully-connected layer in the first fully-connected network, and acquires an output feature of a last fully-connected layer as the fusion feature. It should be noted that, the first fully-connected layer uses the user information and the resource information as input features, then the output feature of each fully-connected layer is used as the input feature of the next fully-connected layer, and so on, and the output feature of the last fully-connected layer is used as a fusion feature.

In some embodiments, since the user information and the resource information both belong to discrete features, the user information and the resource information may be represented by using multidimensional vectors, that is, the user vector of the user information and the resource vector of the resource information are extracted through the first fully-connected network, and then the user vector and the resource vector are input into the at least one fully-connected layer for weighting, and finally the output feature of the last fully-connected layer is obtained as a fusion feature.

Optionally, the server performs one-hot (one-hot) encoding on the user information and the resource information respectively to obtain a one-hot vector of the user information and a one-hot vector of the resource information, and optionally, the server performs embedding (embedding) processing on the user information and the resource information respectively to obtain an embedding vector of the user information and an embedding vector of the resource information. In the embodiment of the present disclosure, whether the user vector or the resource vector belongs to a one-hot vector or an embedding vector is not specifically limited.

Fig. 4 is a schematic diagram of a fully-connected layer according to an embodiment of the disclosure, please refer to fig. 4, which illustrates that a first fully-connected network includes only one fully-connected layer, as shown in 401, the first fully-connected network can be regarded as including an input layer, a fully-connected layer (also referred to as an hidden layer), and an output layer, and an input feature in the input layer includes [ x [, [ x ] ]₁，x₂，x₃]And 4 neurons are included in the fully connected layer. As shown at 402, x is₁、x₂、x₃(collectively referred to as "x") are input into 4 neurons, and a weight coefficient w is adopted in each neuron^TWeighting the input characteristic x, and outputting z ═ w^Tx + b, where z is an intermediate vector and b is a bias coefficient, and then the intermediate vector z is subjected to nonlinear transformation in an output layer by using an activation function σ to obtain a final output characteristic a of the current neuron, that is, a = σ (z). In the disclosed embodiment, a feature [ x ] is input₁，x₂，x₃]Equivalent to a user vector and a resource vector, eachThe features obtained by fully connecting the output features a of the neurons in the output layer again are equivalent to the fusion features, and are not described herein again.

In step 301-, in the disclosed embodiment, the topmost vector of the first fully-connected network is taken as an example for illustration only as a fusion feature, thus, the network structure with full connection can ensure that the relevant details in the user information and the resource information are not lost, in other embodiments, the server can use other forms of machine learning models to obtain the fusion feature, for example, CNN (Convolutional Neural Networks), LSTM (Long Short-Term Memory ), BLSTM (Bidirectional Long Short-Term Memory, Bidirectional Long Short-Term Memory), etc., and the network type for extracting the fusion feature is not specifically limited in the embodiments of the present disclosure.

In step 303, the server maps the fusion feature to a plurality of behavior tags to obtain a plurality of first behavior parameters, one behavior tag corresponding to a first behavior parameter of an interaction behavior.

The behavior tag is used to identify an interaction behavior, for example, the behavior tag includes a click behavior, a collection behavior, a confirmation usage behavior, an upload behavior, and the like.

The first behavior parameter is used to indicate a possibility that the target account performs an interactive behavior on the candidate resource, and for example, the first behavior parameter includes a Click Rate parameter (CTR), a collection Rate parameter (Like Through Rate, LTR), a confirmation usage Rate parameter (FTR), an Upload Rate parameter (WTR), and the Like.

In step 303, a server obtains a plurality of first behavior parameters based on the fused feature. In some embodiments, the server performs an exponential normalization (sigmoid) process on the fused feature to obtain the first behavior parameters. In one example, the server performs index normalization processing on the fusion characteristics by using a sigmoid function and respectively outputs predicted CTR, LTR, FTR and WTR (4 first behavior parameters).

In step 304, the server obtains a plurality of behavior characteristics corresponding to the plurality of value intervals, respectively, based on the plurality of value intervals in which the plurality of first behavior parameters are located.

Wherein the behavior feature is used for representing parameter characteristics of the first behavior parameter.

In the above process, usually, the first behavior parameters (CTR, LTR, FTR, WTR, etc.) corresponding to each behavior label have a continuous value taking characteristic, optionally, the value taking range of each behavior label is divided into a plurality of value taking intervals at equal intervals, and each value taking interval can be represented by using one multidimensional vector (i.e., behavior characteristics in step 304 described below), so that the first behavior parameters having the continuous value taking characteristic can be represented vectorially. This equidistant division of the continuous features can be visually referred to as a "bucket division process" where one "bucket" is equivalent to one value interval.

Further, for each predicted first behavior parameter, the server only needs to judge into which value interval the first behavior parameter specifically falls under the behavior label, and the parameter characteristic of the first behavior parameter can be represented by adopting the behavior characteristic corresponding to the value interval in which the first behavior parameter falls.

In the step 303-304, the first behavior parameter is estimated by the server based on the fusion characteristic, and then the behavior characteristic is determined according to the value range in which the first behavior parameter falls, so that the expression capability of the behavior characteristic can be improved, and the accuracy of the resource recommendation process is higher.

In step 305, the server concatenates the fusion feature, the behavior features, the account id of the target account, and the resource id of the candidate resource to obtain a concatenation feature.

In the above process, the server obtains the fusion feature in the step 302, obtains the plurality of behavior features in the step 304, then extracts the account identifier from the user information in the step 301, extracts the resource identifier from the resource information, and then concatenates the fusion feature, the plurality of behavior features, the account identifier, and the resource identifier to obtain a concatenation feature.

Optionally, the account identifier (user id, uid for short) is the target account itself, or a user identifier allocated by the server to each user, and optionally, the resource identifier is a resource identifier allocated by the server to each multimedia resource, and in an audio recommendation scenario, the resource identifier is also referred to as music id, mid for short.

In some embodiments, if the sizes of the fusion feature, the behavior features, the account identifier, and the resource identifier are not consistent, the server adjusts the sizes of the fusion feature, the behavior features, the account identifier, and the resource identifier to be consistent through a padding (padding) operation, and then directly connects the fusion feature, the behavior features, the account identifier, and the resource identifier in length, so as to obtain the splicing feature, that is, implement a splicing (concat) operation.

In one example, assuming that the fusion feature is a 256 × 128-dimensional feature vector, the 4 behavior features are 4 256 × 8-dimensional feature vectors, and both uid and pid are 256 × 32-dimensional feature vectors, the four can be pieced together to form a 256 × 256 (128+4 × 8+2 × 32) or 256 × 224-dimensional piecing feature.

In other embodiments, the server may also predict the corresponding second behavior parameter by processing the fusion feature, the behavior features, the account identifier, and the resource identifier in an element addition manner, a bilinear fusion manner, and the like, and inputting the processed feature into the second fully-connected network.

In step 306, the server inputs the splicing characteristic into a second fully-connected network, and processes the splicing characteristic through the second fully-connected network to obtain a plurality of second behavior parameters.

Wherein the second fully connected network is configured to obtain a second behavior parameter according to the splicing characteristic, and the second fully connected network includes at least one fully connected layer.

The second behavior parameter is used to indicate a possibility of an interactive behavior occurring when other interactive behaviors than the interactive behavior occur, for example, the second behavior parameter includes a Click Rate parameter (CTR), a collection Rate parameter (Like Through Rate, LTR), a confirmation usage Rate parameter (FTR), an Upload Rate parameter (WTR), and the Like.

Optionally, the server performs weighting processing on the splicing feature through at least one fully-connected layer in the second fully-connected network, and maps the output feature of the last fully-connected layer to the plurality of behavior tags to obtain the plurality of second behavior parameters. It should be noted that, the first fully-connected layer uses the splicing characteristic as an input characteristic, then the output characteristic of each fully-connected layer is used as an input characteristic of the next fully-connected layer, and so on, the output characteristic of the last fully-connected layer is subjected to sigmoid processing to be mapped to a plurality of behavior tags, so as to obtain a plurality of second behavior parameters.

The internal structure of the second fully connected network is similar to that of the first fully connected network, and is not described herein.

In the above step 304-.

In other embodiments, the server may use other forms of machine learning models to obtain the second behavior parameter, for example, CNN (Convolutional Neural Networks), LSTM (Long Short-Term Memory ), BLSTM (Bidirectional Long Short-Term Memory, Bidirectional Long Short-Term Memory), etc., and the disclosed embodiments do not specifically limit the type of network for predicting the second behavior parameter.

It should be noted here that the types of the second behavior parameters and the first behavior parameters need to be kept the same, that is, the second behavior parameters and the first behavior parameters are consistent in number and also correspond to the same behavior tags, in this embodiment of the present disclosure, the first behavior parameters are behavior parameters preliminarily predicted by the first fully-connected network, and the second behavior parameters are behavior parameters precisely predicted again by the second fully-connected network by comprehensively considering a topmost vector (fusion feature) of the first fully-connected network, behavior features, account id, and resource id of each first behavior parameter, and fully considering potential links between different behavior parameters corresponding to different behavior tags, so as to greatly improve accuracy of resource recommendation.

Fig. 5 is a schematic diagram of a method for determining resource information according to an embodiment of the present disclosure, please refer to fig. 5, where a first fully-connected network 501 and a second fully-connected network 502 may be collectively referred to as an "integrated network (stacking network)", where the first fully-connected network 501 includes 4 fully-connected layers L1-L4, a layer L1 outputs a 512-dimensional feature, a layer L2 outputs a 256-dimensional feature, a layer L3 outputs a 128-dimensional feature, a layer L4 outputs a 128-dimensional feature (using the feature as a fusion feature), based on the fusion feature, 4 behavior features CTR-L, LTR-L, FTR-L, WTR-L, a fusion feature of a layer L4, 4 behavior features CTR-L, LTR-L, FTR-L, WTR-L, account id, and a resource identifier mid are input into the second fully-connected network 502, in the second fully-connected network 502, 3 fully-connected layers L1 '-L3', L1 'output a 128-dimensional signature, L2' output a 64-dimensional signature, and L3 'output a 64-dimensional signature, and then final 4 second behavior parameters CTR, LTR, FTR, WTR are predicted based on the 64-dimensional signature of the L3' layer.

In step 307, the server makes a resource recommendation for the target account based on the plurality of second behavior parameters.

Optionally, the server determines a target behavior parameter from the plurality of second behavior parameters according to the optimization demand information, and recommends the candidate resource to the target account based on that the target behavior parameter is greater than a parameter threshold. The optimization demand information can be flexibly changed according to different scenes, for example, in an audio recommendation scene, the FTR parameter is emphasized, in a video recommendation, the LTR parameter is emphasized, and in an advertisement putting scene, the CTR parameter is emphasized.

Optionally, the server performs weighted summation on the plurality of second behavior parameters to obtain a global parameter, and recommends the candidate resource to the target account based on the fact that the global parameter is greater than a parameter threshold. Therefore, the second behavior parameters under the behavior labels of various dimensions can be comprehensively considered, the overall resource recommendation accuracy can be improved, and the user experience is improved.

Optionally, the server repeatedly performs an operation of obtaining a plurality of second behavior parameters on each candidate resource, so as to obtain a plurality of second behavior parameters of each of the plurality of candidate resources laterally, then may sort any one of the second behavior parameters (or global parameters) in a descending order, and recommend the candidate resource ranked in the top N digits to the target account, where N is an integer greater than or equal to 1.

Fig. 6 is a schematic flowchart of a resource recommendation process according to an embodiment of the present disclosure, please refer to fig. 6, in the resource recommendation process, a user interacts with a target application on a terminal to generate an interaction behavior, a server collects user attribute information and user behavior information as user information, and combines resource information of each multimedia resource in a multimedia library, so as to train a first fully connected network and a second fully connected network in an offline model training phase, and then, through model transmission, the first fully connected network and the second fully connected network trained offline can be put into an online model evaluation phase, when a real-time recommendation is performed, the terminal sends a resource recommendation request, the server returns each second behavior parameter predicted by the second fully connected network, and then recalls and sorts, and sends the multimedia resource meeting a recommendation condition to the terminal, a closed loop on the service can be achieved. In an exemplary audio recommendation scene, by applying the method for determining resource information according to the embodiment of the disclosure, the effect of improving the uploading amount of full-platform audio is up to more than 1%.

Fig. 7 is a block diagram illustrating a logical structure of a resource information determination apparatus according to an exemplary embodiment. Referring to fig. 7, the apparatus includes a first obtaining unit 701, a second obtaining unit 702, a third obtaining unit 703, and a resource recommending unit 704.

A first obtaining unit 701 configured to perform, based on user information of a target account and resource information of a candidate resource, obtaining a fusion feature, where the fusion feature is used to represent a feature of the user information and a feature of the resource information;

a second obtaining unit 702, configured to perform obtaining, based on the fusion feature, a plurality of first behavior parameters, where the first behavior parameters are used to indicate a possibility that the target account performs an interaction behavior on the candidate resource;

a third obtaining unit 703 configured to perform obtaining, based on the fusion feature and a plurality of the first behavior parameters, a plurality of second behavior parameters, where the second behavior parameters are used to indicate a possibility of occurrence of the one interactive behavior in a case where other interactive behaviors than the one interactive behavior occur;

a resource recommending unit 704 configured to perform resource recommendation on the target account based on the plurality of second behavior parameters.

According to the device provided by the embodiment of the disclosure, the first behavior parameters of each interactive behavior are respectively and independently predicted by acquiring the fusion feature of the features representing the user information and the features of the resource information, the fusion feature and the first behavior parameters of each interactive behavior are comprehensively considered, and the second behavior parameters of each interactive behavior are acquired.

In a possible implementation, based on the apparatus composition of fig. 7, the first obtaining unit 701 includes:

and the extraction subunit is configured to perform inputting the user information and the resource information into a first fully-connected network, and performing feature extraction on the user information and the resource information through the first fully-connected network to obtain the fusion feature, wherein the first fully-connected network is used for extracting the fusion feature.

In one possible embodiment, the extraction subunit is configured to perform:

In one possible implementation, the second obtaining unit 702 is configured to perform:

In a possible implementation manner, based on the apparatus composition of fig. 7, the third obtaining unit 703 includes:

the acquiring subunit is configured to execute a plurality of value intervals based on the plurality of first behavior parameters, and acquire a plurality of behavior characteristics corresponding to the plurality of value intervals respectively, where the behavior characteristics are used to represent parameter characteristics of the first behavior parameters;

the splicing subunit is configured to splice the fusion feature, the plurality of behavior features, the account identifier of the target account and the resource identifier of the candidate resource to obtain a splicing feature;

and the processing subunit is configured to perform inputting the splicing characteristic into a second fully-connected network, and process the splicing characteristic through the second fully-connected network to obtain the plurality of second behavior parameters, where the second fully-connected network is configured to obtain the second behavior parameters according to the splicing characteristic.

In one possible embodiment, the processing subunit is configured to perform:

In one possible embodiment, the resource recommendation unit 704 is configured to perform:

With regard to the apparatus in the above-described embodiment, the specific manner in which each unit performs the operation has been described in detail in the embodiment of the determination method regarding the resource information, and will not be elaborated here.

Fig. 8 is a schematic structural diagram of a computer device 800 according to an embodiment of the present disclosure, where the computer device 800 may generate a relatively large difference due to different configurations or performances, and may include one or more processors (CPUs) 801 and one or more memories 802, where the memory 802 stores at least one program code, and the at least one program code is loaded and executed by the processors 801 to implement the method for determining resource information according to the embodiments. Certainly, the computer device 800 may further have a wired or wireless network interface, a keyboard, an input/output interface, and other components to facilitate input and output, and the computer device 800 may further include other components for implementing the device functions, which are not described herein again.

In an exemplary embodiment, there is also provided a storage medium, such as a memory, including at least one program code, which is executable by a processor in a terminal to perform the method for determining resource information in the above embodiments. Alternatively, the storage medium may be a non-transitory computer-readable storage medium, for example, the non-transitory computer-readable storage medium may include a ROM (Read-Only Memory), a RAM (Random-Access Memory), a CD-ROM (Compact Disc Read-Only Memory), a magnetic tape, a floppy disk, an optical data storage device, and the like.

In an exemplary embodiment, a computer program product is also provided, which includes one or more program codes executable by a processor of a computer device to perform the method for determining resource information provided by the above embodiments.

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 disclosure is intended to cover any variations, 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 will be understood that the present disclosure is not limited to the precise arrangements described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A method for determining resource information, comprising:

2. The method for determining resource information according to claim 1, wherein the acquiring of the fusion feature based on the user information of the target account and the resource information of the candidate resource includes:

3. The method according to claim 2, wherein the performing feature extraction on the user information and the resource information through the first fully-connected network to obtain the fusion feature comprises:

4. The method according to claim 1, wherein the obtaining a plurality of first behavior parameters based on the fusion feature comprises:

5. The method according to claim 1, wherein the obtaining a plurality of second behavior parameters based on the fusion feature and the plurality of first behavior parameters comprises:

6. The method according to claim 5, wherein the processing the splicing feature through the second fully-connected network to obtain the second behavior parameters comprises:

7. The method for determining resource information according to claim 1, wherein the recommending resources to the target account based on the second behavior parameters includes:

8. An apparatus for determining resource information, comprising:

9. A computer device, comprising:

one or more processors;

wherein the one or more processors are configured to execute the program code to implement the method of determining resource information of any one of claims 1 to 7.

10. A storage medium, wherein at least one program code in the storage medium, when executed by one or more processors of a computer device, enables the computer device to perform the method of determining resource information of any one of claims 1 to 7.