CN112699305A - Multi-target recommendation method, device, computing equipment and medium - Google Patents

Multi-target recommendation method, device, computing equipment and medium Download PDF

Info

Publication number
CN112699305A
CN112699305A CN202110025577.8A CN202110025577A CN112699305A CN 112699305 A CN112699305 A CN 112699305A CN 202110025577 A CN202110025577 A CN 202110025577A CN 112699305 A CN112699305 A CN 112699305A
Authority
CN
China
Prior art keywords
target
recommended
business
score
recommendation
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202110025577.8A
Other languages
Chinese (zh)
Inventor
李炼淳
任重起
丁长林
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Netease Media Technology Beijing Co Ltd
Original Assignee
Netease Media Technology Beijing Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Netease Media Technology Beijing Co Ltd filed Critical Netease Media Technology Beijing Co Ltd
Priority to CN202110025577.8A priority Critical patent/CN112699305A/en
Publication of CN112699305A publication Critical patent/CN112699305A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The embodiment of the disclosure provides a multi-target recommendation method, a multi-target recommendation device, a computing device and a medium. The method comprises the following steps: extracting input features from user information of a user and object information of an object to be recommended; determining a target score of each business target in a plurality of business targets aiming at an object to be recommended through a multi-target recommendation model based on the input characteristics; weighting the target scores of all the business targets based on the weights corresponding to all the business targets, determining the recommendation scores of the objects to be recommended based on the weighting processing results, and determining the weights according to the expected scores of all the business targets of the objects to be recommended in the current business scene; and recommending the object to be recommended to the user based on the recommendation score of the object to be recommended. According to the technical scheme of the embodiment of the disclosure, a plurality of business targets can be optimized simultaneously, and the income of each business target can be balanced and adjusted.

Description

Multi-target recommendation method, device, computing equipment and medium
Technical Field
Embodiments of the present disclosure relate to the field of artificial intelligence technologies, and in particular, to a multi-target recommendation method, a multi-target recommendation apparatus, a computing device, and a medium.
Background
This section is intended to provide a background or context to the embodiments of the disclosure recited in the claims. The description herein is not admitted to be prior art by inclusion in this section.
With the development of internet technology, the amount of information is exponentially increasing. Recommendations as a technical means to address information overload and to mine the potential needs of users play an important role in many areas, such as: e-commerce, news information, movie recommendation, etc.
In the related technical scheme, a deep learning-based recommendation ranking model is adopted, historical behavior data of a user is used as a sample to train the model, recalled objects to be recommended, such as commodities and information, are scored and ranked in a prediction stage, N objects to be recommended before scoring are selected to be displayed, and a recommendation process is completed.
Disclosure of Invention
However, the recommendation ranking model in the above technical solution generally only uses the click-through rate as an optimized business objective, and in a real business scenario, it is often necessary to optimize a plurality of business objectives, for example, the stay time index of information, the click-through purchase rate index of e-commerce, and the like.
For this reason, there is a strong need for an improved multi-objective recommendation method that enables optimization of multiple business objectives and balances the gains of the individual business objectives.
In a first aspect of the disclosed embodiments, a multi-target recommendation method is provided, including: extracting input features from user information of a user and object information of an object to be recommended, wherein the input features comprise: user characteristics and object characteristics; determining a target score for each of a plurality of business targets of the object to be recommended through a multi-target recommendation model based on the input features, wherein the multi-target recommendation model is a neural network model for determining the target score of each business target of the object to be recommended; weighting the target score of each business target based on the weight corresponding to each business target, and determining the recommendation score of the object to be recommended based on the result of the weighting, wherein the weight is determined according to the expected score of each business target of the object to be recommended in the current business scene; recommending the object to be recommended to the user based on the recommendation score of the object to be recommended.
In some example embodiments of the present disclosure, the method further comprises: initializing the weight corresponding to each business target; adjusting the weight corresponding to each service target in a stepping mode, and obtaining a group of weights after each adjustment; determining recommendation scores corresponding to each group of weights after the weights are adjusted; determining a difference value between the recommendation score and an expected recommendation score, wherein the expected recommendation score is an expected score of the object to be recommended under the current business scene; and taking at least one group of weights corresponding to the recommendation scores with the difference value smaller than a preset threshold value as the weights of the business targets of the object to be recommended.
In some example embodiments of the present disclosure, the input layers of the multi-target recommendation model include a first network layer and a second network layer, the input features further include cross features, and the extracting the input features from user information of a user and object information of an object to be recommended includes: acquiring original characteristics from user information of a user and object information of an object to be recommended through the first network layer, wherein the original characteristics comprise the user characteristics and the object characteristics; acquiring the cross feature from the user information of the user and the object information of the object to be recommended through the second network layer; and splicing the original features and the cross features to generate the input features.
In some example embodiments of the present disclosure, the first network layer is a multilayer perceptron network, the second network layer is an inner product-based neural network IPNN layer, and the obtaining, by the second network layer, the cross feature from the user information of the user and the object information of the object to be recommended includes: acquiring user characteristics and object characteristics from the user information of the user and the object information of the object to be recommended through the IPNN layer; and carrying out inner product processing on the user characteristics and/or the object characteristics to generate cross characteristics.
In some example embodiments of the present disclosure, the determining, by the multi-objective recommendation model, a target score for each of a plurality of business targets of the object to be recommended includes: determining, by each of the sub-networks, a sub-feature of the input features corresponding to the sub-network; determining the weight of each sub-feature corresponding to the business target through the gating layer corresponding to the business target; weighting each sub-feature based on the weight of the sub-feature to obtain a target feature corresponding to the service target; and determining the target score of each business target based on the target characteristics corresponding to each business target.
In some example embodiments of the present disclosure, the determining, by each of the sub-networks, the sub-feature corresponding to the sub-network in the input feature comprises: extracting high-order cross features corresponding to the sub-networks from the input features through the first cross feature extraction network; extracting second-order cross features corresponding to the sub-networks from the input features through the second cross feature extraction network, wherein the order of the high-order cross features is greater than that of the second-order cross features; and splicing the high-order cross features and the second-order cross features to generate sub-features corresponding to the sub-networks.
In some example embodiments of the present disclosure, the first cross feature extraction network is a multi-layer perceptron network and the second cross feature extraction network is a factorizer network.
In some example embodiments of the present disclosure, the method further comprises: acquiring sample data, wherein the sample data comprises sample characteristics and sample labels, the sample characteristics comprise user characteristics of a sample user and object characteristics of a sample object, and the sample labels comprise actual target scores of all business targets of the sample object; inputting the sample data into the multi-target recommendation model, and determining a predicted target score for each business target of the sample object; determining a loss function corresponding to the business objective based on a difference between the actual objective score and the predicted objective score; and adjusting parameters of the multi-target recommendation model based on the loss function.
In some example embodiments of the present disclosure, the determining a recommendation score of the object to be recommended based on a result of the weighting process includes: and determining the recommendation score of the object to be recommended through a Sigmoid activation function based on the result of the weighting processing.
In some example embodiments of the present disclosure, the multi-objective recommendation model is a neural network model based on soft parameter sharing.
In a second aspect of the embodiments of the present disclosure, there is provided a multi-target recommendation device including: the input feature extraction module is used for extracting input features from user information of a user and object information of an object to be recommended, and the input features comprise: user characteristics and object characteristics; a target score determining module, configured to determine, based on the input feature, a target score for each of a plurality of business targets of the object to be recommended through a multi-target recommendation model, where the multi-target recommendation model is a neural network model that determines the target score for each business target of the object to be recommended; the recommendation score determining module is used for weighting the target score of each business target based on the weight corresponding to each business target and determining the recommendation score of the object to be recommended based on the result of the weighting, wherein the weight is determined according to the expected score of each business target of the object to be recommended in the current business scene; and the recommending module is used for recommending the object to be recommended to the user based on the recommending score of the object to be recommended.
In some example embodiments of the present disclosure, the apparatus further comprises: the weight initialization module initializes the weight corresponding to each business target; the weight adjusting module adjusts the weight corresponding to each business target in a stepping mode, and a group of weights are obtained after each adjustment; the adjustment score determining module is used for determining recommendation scores corresponding to each group of weights after the weights are adjusted; the difference value determining module is used for determining the difference value between the recommendation score and an expected recommendation score, wherein the expected recommendation score is the expected score of the object to be recommended in the current business scene; and the weight selection module is used for taking at least one group of weights corresponding to the recommendation scores with the difference value smaller than a preset threshold value as the weights of all the service targets of the object to be recommended.
In some example embodiments of the present disclosure, the input layers of the multi-objective recommendation model include a first network layer and a second network layer, the input features further include cross features, and the input feature extraction module includes: the original feature extraction unit is used for acquiring original features from user information of a user and object information of an object to be recommended through the first network layer, wherein the original features comprise the user features and the object features; the cross feature extraction unit is used for acquiring the cross feature from the user information of the user and the object information of the object to be recommended through the second network layer; and the first splicing processing unit is used for splicing the original features and the cross features to generate the input features.
In some example embodiments of the present disclosure, the first network layer is a multi-layer perceptron network, the second network layer is an inner product-based neural network IPNN layer, and the cross feature extraction unit is further configured to: acquiring user characteristics and object characteristics from the user information of the user and the object information of the object to be recommended through the IPNN layer; and carrying out inner product processing on the user characteristics and/or the object characteristics to generate cross characteristics.
In some example embodiments of the present disclosure, the hidden layer of the multi-objective recommendation model includes a plurality of sub-networks and a plurality of gated layers, and the objective score determination module includes: a sub-feature determining module, configured to determine, through each of the sub-networks, a sub-feature corresponding to the sub-network in the input features; the weight determining unit is used for determining the weight of each sub-feature corresponding to the business target through the gating layer corresponding to the business target; the weighting processing unit is used for carrying out weighting processing on each sub-feature based on the weight of the sub-feature to obtain a target feature corresponding to the service target; and the score determining unit is used for determining the target score of each business target based on the target characteristics corresponding to each business target.
In some example embodiments of the present disclosure, the sub-network comprises a first cross feature extraction network and a second cross feature extraction network, the sub-feature determination module is further configured to: extracting high-order cross features corresponding to the sub-networks from the input features through the first cross feature extraction network; extracting second-order cross features corresponding to the sub-networks from the input features through the second cross feature extraction network, wherein the order of the high-order cross features is greater than that of the second-order cross features; and splicing the high-order cross features and the second-order cross features to generate sub-features corresponding to the sub-networks.
In some example embodiments of the present disclosure, the first cross feature extraction network is a multi-layer perceptron network and the second cross feature extraction network is a factorizer network.
In some example embodiments of the present disclosure, the apparatus further comprises: the sample acquisition module is used for acquiring sample data, wherein the sample data comprises sample characteristics and sample labels, the sample characteristics comprise user characteristics of a sample user and object characteristics of a sample object, and the sample labels comprise actual target scores of all business targets of the sample object; the predicted score determining module is used for inputting the sample data into the multi-target recommendation model and determining predicted target scores of all business targets of the sample object; a loss function determining module, configured to determine a loss function corresponding to the business objective based on a difference between the actual objective score and the predicted objective score; and the parameter adjusting module is used for adjusting the parameters of the multi-target recommendation model based on the loss function.
In some example embodiments of the present disclosure, the recommendation score determination module is further to: and determining the recommendation score of the object to be recommended through a Sigmoid activation function based on the result of the weighting processing.
In some example embodiments of the present disclosure, the multi-objective recommendation model is a neural network model based on soft parameter sharing.
In a third aspect of embodiments of the present disclosure, there is provided a computing device comprising: a processor and a memory, the memory storing executable instructions, the processor being configured to invoke the memory-stored executable instructions to perform the method of any of the first aspects described above.
In a fourth aspect of embodiments of the present disclosure, there is provided a medium having stored thereon a program which, when executed by a processor, implements the method as described in any one of the above first aspects.
According to the technical scheme of the embodiment of the disclosure, on one hand, the target scores of all the business targets of the object to be recommended are determined through the multi-target recommendation model, and the multiple business targets of the object to be recommended can be optimized simultaneously; on the other hand, the weight of each business target is determined according to the expected score of each business target of the object to be recommended in the current business scene, and the weight of each business target can be adjusted according to business needs, so that the profit of each business target can be balanced and adjusted.
Drawings
The above and other objects, features and advantages of exemplary embodiments of the present disclosure will become readily apparent from the following detailed description read in conjunction with the accompanying drawings. Several embodiments of the present disclosure are illustrated by way of example, and not by way of limitation, in the figures of the accompanying drawings and in which:
FIG. 1 schematically illustrates a block diagram of an application scenario of a multi-objective recommendation method according to an embodiment of the present disclosure;
FIG. 2 schematically illustrates a flow diagram of a multi-objective recommendation method, in accordance with some embodiments of the present disclosure;
FIG. 3 schematically illustrates a block diagram of a recommendation system applying a multi-objective recommendation method according to some embodiments of the present disclosure;
FIG. 4 schematically illustrates a flow diagram of weight adjustment, according to some embodiments of the present disclosure;
FIG. 5 schematically illustrates a schematic diagram of a computer-readable storage medium, according to some embodiments of the present disclosure;
FIG. 6 schematically illustrates a block diagram of a multi-target recommendation device, in accordance with some embodiments of the present disclosure;
FIG. 7 schematically illustrates a block diagram of a computing device, in accordance with some embodiments of the present disclosure.
In the drawings, the same or corresponding reference numerals indicate the same or corresponding parts.
Detailed Description
The principles and spirit of the present disclosure will be described below with reference to a number of exemplary embodiments. It is understood that these embodiments are given solely for the purpose of enabling those skilled in the art to better understand and to practice the present disclosure, and are not intended to limit the scope of the present disclosure in any way. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
As will be appreciated by one skilled in the art, embodiments of the present disclosure may be embodied as a system, apparatus, device, method, or computer program product. Accordingly, the present disclosure may be embodied in the form of: entirely hardware, entirely software (including firmware, resident software, micro-code, etc.), or a combination of hardware and software.
According to the embodiment of the disclosure, a multi-target recommendation method, a multi-target recommendation device, a computing device and a medium are provided.
In this context, it is to be understood that the terms referred to:
the object to be recommended is as follows: representing an object recommended to a user in a current business scene, for example, in a shopping business scene, an object to be recommended is a commodity; under the scene of news information, the object to be recommended is information.
And (4) service target: the service target used for determining whether to recommend the object to be recommended to the user may be, for example, a click rate index, a stay time index, a conversion rate index, or the like.
Sub-network layer: the neural network layer is used for characterizing at least one type of characteristics of input characteristics in a current business scene, for example, the sub-network layer provided with the multi-objective recommendation model comprises three sub-networks, wherein a first sub-network can be used for characterizing characteristics corresponding to a user browsing operation, a second sub-network can be used for characterizing characteristics corresponding to a purchasing operation, and a third sub-network can be used for characterizing characteristics corresponding to a collecting operation.
A gate control layer: for determining the weight of the output characteristics of each sub-network corresponding to the traffic objective.
The multi-target recommendation model comprises the following steps: the model is used for determining the target scores of the business targets of the objects to be recommended, and the Multi-target recommendation model can be a neural network model based on soft parameter sharing, such as an MMOE (Multi-gate Mixed Experts) model.
IPNN (Inner Product-based Neural Network ): the method is based on the idea of multiplication and accumulation, and utilizes multiplication operation to embody a DNN (deep neural Networks) network structure of feature intersection, namely, each feature is subjected to inner product to obtain intersection features.
FM (Factorization Machines): the factorization machine is a machine learning algorithm based on matrix decomposition, and the specific principle is as follows:
the general linear model in machine learning is shown as the following formula (1),
y＝w1x1+w2x2+…+wnxn (1)
a disadvantage of linear models is that the interrelationships between features cannot be captured. To solve this problem, polynomial regression is introduced as the following formula (2):
Figure BDA0002890199200000081
in the formula (2), since the weight w between every two features needs to be learned, n (n-1)/2 parameters need to be learned, n is the number of features, and the value of n is generally large, which causes the model to be particularly complex. Because data in reality is often sparse, the model cannot learn parameters. Inspired by matrix decomposition, to solve the above problem, a factorizer was introduced.
The factorization machine FM decomposes a sparse matrix into a product of two low-order matrices, namely w, based on the idea of matrix decompositioni,j＝vivjThe above formula (2) can be evolved into the following formula (3):
Figure BDA0002890199200000082
in the above equation (2), the w parameter cannot be learned in the case of sparse data, while the w parameter in equation (3) can be approximated by multiplying the hidden vectors of i and j, so that the problem that the parameter cannot be learned in the sparse data is solved.
Moreover, any number of elements in the drawings are by way of example and not by way of limitation, and any nomenclature is used solely for differentiation and not by way of limitation.
The principles and spirit of the present disclosure are explained in detail below with reference to several representative embodiments of the present disclosure.
Summary of The Invention
The inventor finds that, in order to solve the problem of simultaneous optimization of multiple targets, in one technical scheme, a shared underlying multitask model is adopted to determine target scores of various business targets of an object to be recommended. However, in this solution, conflicts often occur between optimized business objectives, and it is difficult to balance the gains of the business objectives.
Based on the above, the basic idea of the present disclosure is: determining the target score of each business target of the object to be recommended through a multi-target recommendation model, carrying out weighting processing on the target score of each business target based on the weight corresponding to each business target, determining the weight corresponding to each business target according to the expected score of each business target of the object to be recommended in the current business scene, determining the recommendation score of the object to be recommended based on the weighting processing result, and recommending the object to be recommended to a user based on the recommendation score. According to the technical scheme of the embodiment of the disclosure, on one hand, the target scores of all the business targets of the object to be recommended are determined through the multi-target recommendation model, and the multiple business targets of the object to be recommended can be optimized simultaneously; on the other hand, the weight of each business target is determined according to the expected score of each business target of the object to be recommended in the current business scene, and the weight of each business target can be adjusted according to business needs, so that the profit of each business target can be balanced and adjusted.
Having described the general principles of the present disclosure, various non-limiting embodiments of the present disclosure are described in detail below.
Application scene overview
It should be noted that the following application scenarios are merely illustrated to facilitate understanding of the spirit and principles of the present disclosure, and embodiments of the present disclosure are not limited in this respect. Rather, embodiments of the present disclosure may be applied to any scenario where applicable.
Fig. 1 schematically illustrates a block diagram of an application scenario of a multi-target recommendation method according to an embodiment of the present disclosure.
Referring to fig. 1, the application scenario may include: at least one client terminal 110 and a server terminal 120, wherein the client terminal 110 is installed with a plurality of applications, such as a shopping application, a video application, a news application, etc. The client terminal 110 and the server terminal 120 communicate with each other through a network 130. Taking a shopping application scenario as an example, a user opens a shopping application program installed on a client 110 with a business target of click rate and purchase rate, inputs a name of a commodity to be searched in a search box, a server 120 determines a target score aiming at the click rate and the purchase rate of the commodity to be recommended through a multi-target recommendation model according to user information of the user and commodity information of the commodity to be recommended, performs weighting processing on the target score according to weights corresponding to the click rate and the purchase rate to obtain a recommendation score of the commodity to be recommended, and recommends the commodity for the user based on the recommendation score.
The client 110 may be a mobile phone, a tablet computer, a desktop computer, a portable notebook computer, or a vehicle-mounted terminal. The server side 120 may be a physical server including independent hosts, or a virtual server carried by a host cluster, or a cloud server. The Network 130 may be a wired Network or a wireless Network, for example, the Network 130 may be a PSTN (public switched Telephone Network) or the internet.
Exemplary method
In combination with the above application scenarios, a multi-target recommendation method according to an exemplary embodiment of the present disclosure is described below with reference to fig. 2. It should be noted that the above application scenarios are merely illustrative for the convenience of understanding the spirit and principles of the present disclosure, and embodiments of the present disclosure are not limited in this respect. Rather, embodiments of the present disclosure may be applied to any scenario where applicable.
Referring to fig. 2, in step S210, an input feature is extracted from user information of a user and object information of an object to be recommended, the input feature including: user characteristics and object characteristics.
In an example embodiment, taking a shopping scenario as an example, the user information of the user may include user basic information and user historical behavior information, the user basic information may include information such as age, gender, and occupation of the user, and the user historical behavior information may include user historical purchased commodity information, collected commodity information, comment commodity information, and the like. The object to be recommended is a commodity, and the object information of the object to be recommended may include information such as a commodity name, a commodity model, a commodity size, and a commodity color.
Further, in an example embodiment, the input features may be extracted from the user information of the user and the commodity information of the commodity through a feature extraction neural network, for example, the input features may be extracted from the user information of the user and the commodity information of the commodity through a multi-layer perceptron neural network. The input features may include user features and merchandise features.
In step S220, based on the input features, a target score for each of a plurality of business targets of the object to be recommended is determined through a multi-target recommendation model, which is a neural network model that determines the target scores for each business target of the object to be recommended.
In an example embodiment, the multi-objective recommendation model may be a neural network model based on soft parameter sharing, for example, the multi-objective recommendation model may be an MMOE model. The service target is a service target in a current service scene, and taking the service scene as a shopping scene as an example, the service target can be a click rate, a click purchase rate, a page dwell time and the like of a commodity.
Further, the input features extracted in step S210 are input into the multi-target recommendation model, and target scores for each business target of the to-be-recommended goods are determined by the multi-target recommendation model. In an example embodiment, the multi-target recommendation model includes a plurality of sub-networks and a plurality of gating layers, and determining, by the multi-target recommendation model, a target score for each of a plurality of business targets of the object to be recommended includes: inputting the input features into the sub-network; determining sub-features corresponding to the sub-networks in the input features through each sub-network; determining the weight of each sub-feature corresponding to the business target through a gating layer corresponding to the business target; weighting each sub-feature based on the weight of the sub-feature to obtain a target feature corresponding to the service target; and determining the target score of each business target based on the target characteristics corresponding to each business target.
The sub-networks are neural network layers for representing at least one type of characteristics of input characteristics in a current business scene, for example, the sub-network layer provided with the multi-objective recommendation model comprises three sub-networks, wherein a first sub-network is used for representing characteristics corresponding to user browsing operations, a second sub-network is used for representing characteristics corresponding to purchasing operations, and a third sub-network is used for representing characteristics corresponding to collecting operations. The gating layer corresponds to a business target in a current business scene and is used for determining the weight of the output characteristics of each sub-network corresponding to the business target and carrying out weighting processing on the output characteristics of each expert network to obtain the target characteristics corresponding to the business target.
Further, the target characteristics corresponding to the business target can be input into the target score prediction network, and the target score corresponding to the business target is determined. In an example embodiment, the target score prediction network may be a tower network in an MMOE model.
In step S230, the target scores of the business targets are weighted based on the weights corresponding to the business targets, and the recommendation score of the object to be recommended is determined based on the result of the weighting, where the weights are determined according to the expected scores of the business targets of the object to be recommended in the current business scenario.
In an example embodiment, the expected score of each business target in the current business scenario may be preset, and the weight corresponding to each business target is determined according to the expected score of each business target in the current business scenario, for example, for a shopping scenario, since the click purchase rate is important, that is, the expected score of the click purchase rate is higher, the weight corresponding to the click purchase rate may be set higher; for news information, the weight of the page dwell time can be set higher because the page dwell time is more important, i.e., the expected score of the page dwell time is higher.
Further, the target score of each business target in the multiple business targets is weighted based on the weight corresponding to each business target, and the recommendation score of the object to be recommended is determined. For example, the recommendation score of the object to be recommended may be determined by Sigmoid activation function based on the result of the weighting processing
In step S240, the object to be recommended is recommended to the user based on the recommendation score of the object to be recommended.
In an example embodiment, the commodities to be recommended are sorted based on the recommendation scores of the commodities to be recommended, and the commodities to be recommended with the recommendation scores larger than a preset threshold are selected and recommended to the user.
According to the technical solution in the example embodiment of fig. 2, on one hand, the target scores of the business targets of the object to be recommended are determined by the multi-target recommendation model, and a plurality of business targets of the object to be recommended can be optimized simultaneously; on the other hand, the weight of each business target is determined according to the expected score of each business target of the object to be recommended in the current business scene, and the weight of each business target can be adjusted according to business needs, so that the profit of each business target can be balanced and adjusted.
In addition, before the multi-objective recommendation model is applied, the multi-objective recommendation model needs to be trained. Therefore, in an example embodiment, the multi-target recommendation method further comprises: acquiring sample data, wherein the sample data comprises sample characteristics and sample labels, the sample characteristics comprise user characteristics of sample users and object characteristics of sample objects, and the sample labels comprise actual target scores of all business targets of the sample objects; inputting sample data into a multi-target recommendation model, and determining a predicted target score for each business target of a sample object; determining a loss function corresponding to the business target based on a difference value between the actual target score and the predicted target score; and adjusting parameters of the multi-target recommendation model based on a loss function.
It should be noted that different loss functions may be set for different business targets, for example, the business targets may be divided into a classification business target and a regression business target, for example, it is determined whether a commodity link is clicked as the classification business target, the page dwell time is the regression business target, a cross entropy loss function is adopted for the classification business target, and a square loss function is adopted for the regression business target.
FIG. 3 schematically illustrates a block diagram of a recommendation system applying a multi-objective recommendation method according to some embodiments of the present disclosure.
Referring to fig. 3, the user log module 305 is configured to obtain historical behavior data of the user, which may include purchased goods data, collected goods data, browsed goods data, commented goods data, and the like, taking a shopping scenario as an example.
The sample module 310 is configured to generate sample data according to the historical behavior data of the user, where the sample data may include user information of the user, object information of the sample object, and a sample tag. Taking a business target as a click rate target as an example, a sample label of a user click commodity page link after commodity exposure is a positive sample, and a sample label of an exposure non-click is a negative sample.
The input feature extraction module 315 includes a multi-layer perceptron 312 and an IPNN (inner product-based Neural Network) 314. The multilayer perceptron 312 is configured to obtain original features from user information of a user and object information of an object to be recommended, where the original features include user features and object features; the IPNN 314 is configured to obtain cross features from user information of a user and object information of an object to be recommended. The input feature extraction module 315 is configured to perform stitching processing on the original features extracted by the multilayer perceptron 312 and the cross features extracted by the IPNN 314, so as to generate input features.
In an example embodiment, important features in the sample features, such as user features, commodity category features, and the like, may be used to generate cross-features using the network structure of the IPNN. For example, let eiAn embedded vector representing the ith feature, ejAn embedded vector representing the jth feature, then e is generated based on the inner productiAnd ejCan be expressed as the following formula (4):
Figure BDA0002890199200000131
wherein lpIn order to be a cross-feature,
Figure BDA0002890199200000132
as a weight matrix, p ═ pi,j}，i＝1…N，j＝1…N，pi,j＝<ei,ej>。
Further, letThe original characteristic of the sample data is lzThen the input feature extraction module 315 performs the following equation (5) on the original feature lzAnd cross feature lpAnd performing splicing processing to obtain an input characteristic z:
z＝concat([lz,lp]) (5)
by extracting the cross features of the sample data and splicing the original features and the cross features of the sample data, richer model features of the sample data can be obtained, and the prediction result of the multi-target recommendation model obtained by training is more accurate.
With continued reference to FIG. 3, the sub-network layer includes n sub-networks 320, each sub-network 320 including a multi-tier perceptron 322 and a factorizer 324. The multilayer perceptron 322 is configured to extract a high-order cross feature corresponding to the sub-network from the input features; the factoring machine 324 is used to extract second order cross features in the input features corresponding to the sub-network 320. The sub-network layer is used for performing splicing processing on the high-order cross features extracted by the multi-layer perceptron 322 and the second-order cross features extracted by the factorization machine 324 to generate sub-features corresponding to the sub-networks 320.
For example, let the vector output by the multi-level perceptron 322 be ydnnThe vector output by the factorizer 324 is yFMWill y isdnnAnd yFMThe splicing process is performed according to the following formula, and the result after the splicing process is used as a sub-feature output by the sub-network 320, which is shown in the following formula (6):
y＝concat([yFM,ydnn]) (6)
wherein y is the sub-feature after splicing treatment, yFMVector, y, output by the factorizer 324dnnIs the vector output by the multi-layered perceptron 322.
By combining the high-order feature extraction network and the low-order feature extraction network, the multi-target recommendation model can learn the cross information of the high-order features and the low-order features more effectively, and the model prediction result is more accurate.
Next, the gating layer 325 is used to determine the weight of the sub-feature output by each sub-network 320 corresponding to the business target; and carrying out weighting processing on each sub-feature based on the weight of the sub-feature to obtain a target feature corresponding to the service target. For example, if the shared bottom layer of the multi-target recommendation model is divided into n different sub-networks, namely, expert networks, each sub-network is fused with the kth gating layer, and the fusion result is represented by the following formula (7):
Figure BDA0002890199200000141
wherein f isk(x) As a result of merging the sub-network with the kth gating layer, fi(x) Sub-feature output for ith sub-network, gk(x)iWeights corresponding to the sub-features of the ith sub-network output for the kth gating layer.
Further, the kth gate layer may be represented by the following formula (8):
gk(x)＝soft max(Wgkx) (8)
wherein, WgkFor the trained weight matrix, x is the input of the gating layer.
In an example embodiment, the traffic target network layer may include a plurality of traffic target networks, the input of which is the target feature output by the gating layer 325, and the output is the target score of the traffic target. In fig. 3, the traffic target network layer includes a traffic target 1 network 326, a traffic target 2 network 328, and a traffic target 3 network 330, and the output of the traffic target k network can be represented by the following formula (9):
yk＝hk(fk(x)) (9)
wherein h iskRepresenting the kth traffic target network, i.e. the tower network, ykRepresenting the output of the kth traffic destination network.
By adding the gating layer into the multi-target recommendation model, different sample data can be diversified to use the sharing layer, namely the sub-network layer, so that the influence caused by the correlation difference of different targets can be weakened, and meanwhile, all the service targets can learn the information among each other by using a transfer learning method.
Continuing with fig. 3, the score fusion network 335 performs fusion processing on the target scores output by the respective service target networks. For example, let the output results of k service target networks be logit respectivelyiWhere i is 1, 2, …, k, the recommendation score for the output of the score fusion network 335 is the following formula (10):
Figure BDA0002890199200000151
wherein,
Figure BDA0002890199200000152
for the weight corresponding to each business objective, Output is the final recommendation score Output by the score fusion network 335.
Figure BDA0002890199200000153
The larger the value of (d), the larger the influence of the business objective on the recommendation score of the entire output, and the higher the corresponding on-line index.
The offline testing module 340 is configured to, under the offline testing condition, adjust the weight corresponding to each service target in a step-by-step manner according to the expected score of each service target in the current service scenario, and select a weight parameter with a better offline result to be online. For example, the offline testing module 340 may adjust the weight corresponding to each service target in an equidistant stepping manner, or may adjust the weight corresponding to each service target in a non-equidistant stepping manner.
The sorting module 345 is configured to sort the objects to be recommended according to the recommendation scores of the objects to be recommended, and select the object to be recommended whose recommendation score is greater than a predetermined threshold value to recommend to the user.
Fig. 4 schematically illustrates a flow diagram of weight adjustment, according to some embodiments of the present disclosure.
Referring to fig. 4, in step S410, weights corresponding to the respective traffic targets are initialized.
In an example embodiment, the weight corresponding to each business objective may be initialized to 0, or may be initialized to other appropriate empirical values. For example, taking a shopping scenario as an example, the business objectives may include: click rate goal, purchase rate goal, plus shopping cart rate goal, the corresponding set of weights is {0, 0, 0 }.
In step S420, the weights corresponding to the service targets are adjusted in a step-by-step manner, and a set of weights is obtained after each adjustment.
In the example embodiment, the weight corresponding to each service target may be adjusted in an equidistant stepping manner, or may be adjusted in a non-equidistant stepping manner. Or the weight can be adjusted by adopting an equidistant stepping mode for one service target, and the weight can be adjusted by adopting a non-equidistant stepping mode for the other service target.
In step S430, recommendation scores corresponding to each set of weights after the weight adjustment are determined.
In an example embodiment, a set of weights corresponding to a plurality of business targets is obtained after the weights are adjusted, the target scores of the business targets are weighted based on the weights corresponding to the business targets, and recommendation scores corresponding to the set of weights are determined based on the weighting structure.
In step S440, a difference between the recommendation score and the expected recommendation score is determined, and the expected recommendation score is an expected score of the object to be recommended in the current business scenario.
In an example embodiment, an expected score, i.e. an expected recommendation score, of an object to be recommended may be set for a current business scenario, and a difference between a recommendation score corresponding to each set of weights and the expected recommendation score may be determined.
In step S450, at least one set of weights corresponding to the recommendation scores whose difference is smaller than the predetermined threshold is used as the weight of each business target of the object to be recommended.
In an example embodiment, at least one group of weights corresponding to the recommendation scores with the difference values smaller than a predetermined threshold value is determined, the at least one group of weights is used as the weight of each business target of the objects to be recommended, and the predetermined threshold value can be determined according to the number of the objects to be recommended. For example, if there are a plurality of recommendation scores whose difference values are smaller than a predetermined threshold, a group of weights with the smallest difference value may be selected as the weights of the business targets of the object to be recommended, or a plurality of groups of weights corresponding to the plurality of recommendation scores may be respectively used as the weights of the business targets of the object to be recommended.
According to the technical solution in the example embodiment of fig. 4, the weight corresponding to each business target is adjusted in a step-by-step manner, and the weight of each business target can be adjusted in real time according to the need of the business target, so that the profit of each business target can be dynamically and balancedly adjusted.
Exemplary Medium
Having described the methods of the exemplary embodiments of the present disclosure, the media of the exemplary embodiments of the present disclosure are described next.
In some possible embodiments, various aspects of the present disclosure may also be implemented as a medium having stored thereon program code for implementing steps in the multi-target recommendation method according to various exemplary embodiments of the present disclosure described in the above-mentioned "exemplary methods" section of this specification when the program code is executed by a processor of a device.
In some possible embodiments, the program code is executable by a processor of the device to perform the following steps: extracting input features from user information of a user and object information of an object to be recommended, wherein the input features comprise: user characteristics and object characteristics; determining a target score for each of a plurality of business targets of the object to be recommended through a multi-target recommendation model based on the input features, wherein the multi-target recommendation model is a neural network model for determining the target score of each business target of the object to be recommended; weighting the target score of each business target based on the weight corresponding to each business target, and determining the recommendation score of the object to be recommended based on the result of the weighting, wherein the weight is determined according to the expected score of each business target of the object to be recommended in the current business scene; recommending the object to be recommended to the user based on the recommendation score of the object to be recommended.
Referring to fig. 5, a program product 500 for implementing the above-described data processing method according to an embodiment of the present disclosure is described, which may employ a portable compact disc read only memory and include program code, and may be run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto.
It should be noted that: the above-mentioned medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example but not limited to: an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a random access memory, a read only memory, an erasable programmable read only memory, an optical fiber, a portable compact disk read only memory, an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
A readable signal medium may include a propagated data signal with readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take a variety of forms, including, but not limited to: an electromagnetic signal, an optical signal, or any suitable combination of the foregoing. A readable signal medium may also be any readable medium that is not a readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.
Program code embodied on a readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, optical fiber cable, radio frequency signals, etc., or any suitable combination of the foregoing.
Program code for carrying out operations for the present disclosure may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user computing device, partly on a remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device over any kind of network, including a local area network or a wide area network, or may be connected to an external computing device (e.g., over the internet using an internet service provider).
Exemplary devices
Having described the media of the exemplary embodiments of the present disclosure, the multi-target recommendation device of the exemplary embodiments of the present disclosure will next be described with reference to fig. 6.
Referring to fig. 6, the multi-target recommendation apparatus 600 includes: an input feature extraction module 610, configured to extract input features from user information of a user and object information of an object to be recommended, where the input features include: user characteristics and object characteristics; a target score determining module 615, configured to determine, based on the input feature, a target score for each of a plurality of business targets of the object to be recommended through a multi-target recommendation model, where the multi-target recommendation model is a neural network model that determines the target score for each business target of the object to be recommended; a recommendation score determining module 620, configured to perform weighting processing on the target score of each service target based on a weight corresponding to each service target, and determine a recommendation score of the object to be recommended based on a result of the weighting processing, where the weight is determined according to an expected score of each service target of the object to be recommended in a current service scenario; and a recommending module 625, configured to recommend the object to be recommended to the user based on the recommendation score of the object to be recommended.
In some example embodiments of the present disclosure, the apparatus 600 further comprises: a weight initialization module, configured to initialize the weight corresponding to each service target; the weight adjusting module is used for adjusting the weight corresponding to each business target in a stepping mode, and a group of weights are obtained after each adjustment; the adjustment score determining module is used for determining recommendation scores corresponding to each group of weights after the weights are adjusted; a difference value determining module, configured to determine a difference value between the recommendation score and an expected recommendation score, where the expected recommendation score is an expected score of the object to be recommended in the current service scenario; and the weight selection module is used for taking at least one group of weights corresponding to the recommendation scores with the difference value smaller than a preset threshold value as the weights of the business targets of the object to be recommended.
In some example embodiments of the present disclosure, the input layers of the multi-objective recommendation model include a first network layer and a second network layer, the input features further include cross features, and the input feature extraction module 610 includes: the original feature extraction unit is used for acquiring original features from user information of a user and object information of an object to be recommended through the first network layer, wherein the original features comprise the user features and the object features; the cross feature extraction unit is used for acquiring the cross feature from the user information of the user and the object information of the object to be recommended through the second network layer; and the first splicing processing unit is used for splicing the original features and the cross features to generate the input features.
In some example embodiments of the present disclosure, the first network layer is a multi-layer perceptron network, the second network layer is an inner product-based neural network IPNN layer, and the cross feature extraction unit is further configured to: acquiring user characteristics and object characteristics from the user information of the user and the object information of the object to be recommended through the IPNN layer; and carrying out inner product processing on the user characteristics and/or the object characteristics to generate cross characteristics.
In some example embodiments of the present disclosure, the hidden layer of the multi-target recommendation model includes a plurality of sub-networks and a plurality of gated layers, and the target score determining module 615 includes: a sub-feature determining module, configured to determine, through each of the sub-networks, a sub-feature corresponding to the sub-network in the input features; the weight determining unit is used for determining the weight of each sub-feature corresponding to the business target through the gating layer corresponding to the business target; the weighting processing unit is used for carrying out weighting processing on each sub-feature based on the weight of the sub-feature to obtain a target feature corresponding to the service target; and the score determining unit is used for determining the target score of each business target based on the target characteristics corresponding to each business target.
In some example embodiments of the present disclosure, the sub-network comprises a first cross feature extraction network and a second cross feature extraction network, the sub-feature determination module is further configured to: extracting high-order cross features corresponding to the sub-networks from the input features through the first cross feature extraction network; extracting second-order cross features corresponding to the sub-networks from the input features through the second cross feature extraction network, wherein the order of the high-order cross features is greater than that of the second-order cross features; and splicing the high-order cross features and the second-order cross features to generate sub-features corresponding to the sub-networks.
In some example embodiments of the present disclosure, the first cross feature extraction network is a multi-layer perceptron network and the second cross feature extraction network is a factorizer network.
In some example embodiments of the present disclosure, the apparatus 600 further comprises: the sample acquisition module is used for acquiring sample data, wherein the sample data comprises sample characteristics and sample labels, the sample characteristics comprise user characteristics of a sample user and object characteristics of a sample object, and the sample labels comprise actual target scores of all business targets of the sample object; the predicted score determining module is used for inputting the sample data into the multi-target recommendation model and determining predicted target scores of all business targets of the sample object; a loss function determining module, configured to determine a loss function corresponding to the business objective based on a difference between the actual objective score and the predicted objective score; and the parameter adjusting module is used for adjusting the parameters of the multi-target recommendation model based on the loss function.
In some example embodiments of the present disclosure, the recommendation score determination module 620 is further configured to: and determining the recommendation score of the object to be recommended through a Sigmoid activation function based on the result of the weighting processing.
In some example embodiments of the present disclosure, the multi-objective recommendation model is a neural network model based on soft parameter sharing.
Exemplary computing device
Having described the methods, media, and apparatus of the exemplary embodiments of the present disclosure, a computing device in accordance with another exemplary embodiment of the present disclosure is described next.
As will be appreciated by one skilled in the art, aspects of the present disclosure may be embodied as a system, method or program product. Accordingly, various aspects of the present disclosure may be embodied in the form of: an entirely hardware embodiment, an entirely software embodiment (including firmware, microcode, etc.) or an embodiment combining hardware and software aspects that may all generally be referred to herein as a "circuit," module "or" system.
In some possible embodiments, a computing device according to embodiments of the present disclosure may include at least one processor, and at least one memory. Wherein the memory stores program code that, when executed by the processor, causes the processor to perform the steps of the multi-objective recommendation method according to various exemplary embodiments of the present disclosure described in the "exemplary methods" section above in this specification. For example, the processor may perform the steps as shown in fig. 2: step S210, extracting input features from the user information of the user and the object information of the object to be recommended, wherein the input features comprise: user characteristics and object characteristics; step S220, based on the input characteristics, determining a target score of each business target in a plurality of business targets of the object to be recommended through a multi-target recommendation model, wherein the multi-target recommendation model is a neural network model for determining the target score of each business target of the object to be recommended; step S230, weighting the target scores of the business targets based on the weights corresponding to the business targets, and determining the recommendation scores of the objects to be recommended based on the weighting results, wherein the weights are determined according to the expected scores of the business targets of the objects to be recommended in the current business scene; recommending the object to be recommended to a user based on the recommendation score of the object to be recommended. As another example, the processor may also perform the steps as shown in fig. 4.
An electronic device 700 according to an example embodiment of the present disclosure is described below with reference to fig. 7. The electronic device 700 shown in fig. 7 is only an example and should not bring any limitations to the functionality and scope of use of the embodiments of the present disclosure.
As shown in fig. 7, electronic device 700 is embodied in the form of a general purpose computing device. The components of the electronic device 700 may include, but are not limited to: the at least one processing unit 710, the at least one memory unit 720, and a bus 730 that couples various system components including the memory unit 720 and the processing unit 710.
Bus 730 represents one or more of any of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor, or a local bus using any of a variety of bus architectures.
The storage unit 720 may include a readable medium in the form of a volatile Memory, such as a RAM (Random Access Memory) 721 and/or a cache Memory 722, and may further include a ROM (Read-Only Memory) 723.
The storage unit 720 may also include a program/utility 725 having a set (at least one) of program modules 724, such program modules 724 including, but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which, or some combination thereof, may comprise an implementation of a network environment.
The electronic device 700 may also communicate with one or more external devices 740 (e.g., keyboard, pointing device, bluetooth device, etc.), with one or more devices that enable a user to interact with the electronic device 700, and/or with any devices (e.g., router, modem, etc.) that enable the electronic device 700 to communicate with one or more other computing devices. Such communication may occur via an input/output (I/O) interface 750. Also, the electronic device 700 may communicate with one or more networks (e.g., a local area network, a wide area network, and/or a public network, such as the Internet) via the network adapter 760. As shown, the network adapter 760 communicates with the other modules of the electronic device 700 via the bus 730. It should be appreciated that although not shown in the figures, other hardware and/or software modules may be used in conjunction with the electronic device 700, including but not limited to: microcode, device drivers, Redundant processing units, external disk drive Arrays, RAID (Redundant array of independent disks) systems, tape drives, and data backup storage systems, among others.
It should be noted that although several units or sub-units of the multi-target recommendation device are mentioned in the above detailed description, such division is merely exemplary and not mandatory. Indeed, the features and functionality of two or more modules or units described above may be embodied in one module or unit, according to embodiments of the present disclosure. Conversely, the features and functions of one module or unit described above may be further divided into embodiments by a plurality of modules or units.
Further, while the operations of the disclosed methods are depicted in the drawings in a particular order, this does not require or imply that these operations must be performed in this particular order, or that all of the illustrated operations must be performed, to achieve desirable results. Additionally or alternatively, certain steps may be omitted, multiple steps combined into one step execution, and/or one step broken down into multiple step executions.
While the spirit and principles of the present disclosure have been described with reference to several particular embodiments, it is to be understood that the present disclosure is not limited to the particular embodiments disclosed, nor is the division of aspects, which is for convenience only as the features in such aspects may not be combined to benefit. The disclosure is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.

Claims (10)

1. A multi-objective recommendation method, comprising:
extracting input features from user information of a user and object information of an object to be recommended, wherein the input features comprise: user characteristics and object characteristics;
determining a target score for each of a plurality of business targets of the object to be recommended through a multi-target recommendation model based on the input features, wherein the multi-target recommendation model is a neural network model for determining the target score of each business target of the object to be recommended;
weighting the target score of each business target based on the weight corresponding to each business target, and determining the recommendation score of the object to be recommended based on the result of the weighting, wherein the weight is determined according to the expected score of each business target of the object to be recommended in the current business scene;
recommending the object to be recommended to the user based on the recommendation score of the object to be recommended.
2. The method of claim 1, further comprising:
initializing the weight corresponding to each business target;
adjusting the weight corresponding to each service target in a stepping mode, and obtaining a group of weights after each adjustment;
determining recommendation scores corresponding to each group of weights after the weights are adjusted;
determining a difference value between the recommendation score and an expected recommendation score, wherein the expected recommendation score is an expected score of the object to be recommended under the current business scene;
and taking at least one group of weights corresponding to the recommendation scores with the difference value smaller than a preset threshold value as the weights of the business targets of the object to be recommended.
3. The method of claim 1, wherein the input layers of the multi-objective recommendation model comprise a first network layer and a second network layer, the input features further comprise cross features, and the extracting the input features from the user information of the user and the object information of the object to be recommended comprises:
acquiring original characteristics from user information of a user and object information of an object to be recommended through the first network layer, wherein the original characteristics comprise the user characteristics and the object characteristics;
acquiring the cross feature from the user information of the user and the object information of the object to be recommended through the second network layer;
and splicing the original features and the cross features to generate the input features.
4. The method according to claim 3, wherein the first network layer is a multi-layer perceptron network, the second network layer is an inner product-based neural network IPNN layer, and the obtaining of the cross feature from the user information of the user and the object information of the object to be recommended through the second network layer comprises:
acquiring user characteristics and object characteristics from the user information of the user and the object information of the object to be recommended through the IPNN layer;
and carrying out inner product processing on the user characteristics and/or the object characteristics to generate cross characteristics.
5. The method of claim 1, wherein the hidden layer of the multi-objective recommendation model comprises a plurality of sub-networks and a plurality of gated layers, and the determining, by the multi-objective recommendation model, the target score for each of a plurality of business targets of the object to be recommended comprises:
determining, by each of the sub-networks, a sub-feature of the input features corresponding to the sub-network;
determining the weight of each sub-feature corresponding to the business target through the gating layer corresponding to the business target;
weighting each sub-feature based on the weight of the sub-feature to obtain a target feature corresponding to the service target;
and determining the target score of each business target based on the target characteristics corresponding to each business target.
6. The method of claim 5, wherein the sub-networks comprise a first cross feature extraction network and a second cross feature extraction network, and wherein determining, by each of the sub-networks, the sub-features of the input feature corresponding to the sub-network comprises:
extracting high-order cross features corresponding to the sub-networks from the input features through the first cross feature extraction network;
extracting second-order cross features corresponding to the sub-networks from the input features through the second cross feature extraction network, wherein the order of the high-order cross features is greater than that of the second-order cross features;
and splicing the high-order cross features and the second-order cross features to generate sub-features corresponding to the sub-networks.
7. The method of claim 6, wherein the first cross feature extraction network is a multi-tier perceptron network and the second cross feature extraction network is a factorizer network.
8. A multi-objective recommendation device, comprising:
the input feature extraction module is used for extracting input features from user information of a user and object information of an object to be recommended, and the input features comprise: user characteristics and object characteristics;
a target score determining module, configured to determine, based on the input feature, a target score for each of a plurality of business targets of the object to be recommended through a multi-target recommendation model, where the multi-target recommendation model is a neural network model that determines the target score for each business target of the object to be recommended;
the recommendation score determining module is used for weighting the target score of each business target based on the weight corresponding to each business target and determining the recommendation score of the object to be recommended based on the result of the weighting, wherein the weight is determined according to the expected score of each business target of the object to be recommended in the current business scene;
and the recommending module is used for recommending the object to be recommended to the user based on the recommending score of the object to be recommended.
9. A computing device, comprising: a processor and a memory, the memory storing executable instructions, the processor to invoke the memory-stored executable instructions to perform the method of any of claims 1 to 7.
10. A medium having stored thereon a program which, when executed by a processor, carries out the method of any one of claims 1 to 7.
CN202110025577.8A 2021-01-08 2021-01-08 Multi-target recommendation method, device, computing equipment and medium Pending CN112699305A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110025577.8A CN112699305A (en) 2021-01-08 2021-01-08 Multi-target recommendation method, device, computing equipment and medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110025577.8A CN112699305A (en) 2021-01-08 2021-01-08 Multi-target recommendation method, device, computing equipment and medium

Publications (1)

Publication Number Publication Date
CN112699305A true CN112699305A (en) 2021-04-23

Family

ID=75513544

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110025577.8A Pending CN112699305A (en) 2021-01-08 2021-01-08 Multi-target recommendation method, device, computing equipment and medium

Country Status (1)

Country Link
CN (1) CN112699305A (en)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113516522A (en) * 2021-09-14 2021-10-19 腾讯科技（深圳）有限公司 Media resource recommendation method, and training method and device of multi-target fusion model
CN113626719A (en) * 2021-10-12 2021-11-09 腾讯科技（深圳）有限公司 Information recommendation method, device, equipment, storage medium and computer program product
CN113672807A (en) * 2021-08-05 2021-11-19 杭州网易云音乐科技有限公司 Recommendation method, device, medium, device and computing equipment
CN113742590A (en) * 2021-09-07 2021-12-03 北京沃东天骏信息技术有限公司 Recommendation method and device, storage medium and electronic equipment
CN113742578A (en) * 2021-08-13 2021-12-03 杭州网易云音乐科技有限公司 Data recommendation method and device, electronic equipment and storage medium
CN113837808A (en) * 2021-09-27 2021-12-24 北京有竹居网络技术有限公司 Promotion information pushing method, device, equipment, medium and product
CN113919893A (en) * 2021-12-14 2022-01-11 腾讯科技（深圳）有限公司 Information pushing method and device, electronic equipment and readable medium
CN114430504A (en) * 2022-01-28 2022-05-03 腾讯科技（深圳）有限公司 Recommendation method and related device for media content
CN114462584A (en) * 2022-04-11 2022-05-10 北京达佳互联信息技术有限公司 Recommendation model training method, recommendation device, server and medium

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170262948A1 (en) * 2016-03-08 2017-09-14 International Business Machines Corporation Determination of targeted food recommendation
CN109408731A (en) * 2018-12-27 2019-03-01 网易（杭州）网络有限公司 A kind of multiple target recommended method, multiple target recommended models generation method and device
CN111291266A (en) * 2020-02-13 2020-06-16 腾讯科技（北京）有限公司 Artificial intelligence based recommendation method and device, electronic equipment and storage medium
CN111523044A (en) * 2020-07-06 2020-08-11 南京梦饷网络科技有限公司 Method, computing device, and computer storage medium for recommending target objects
CN111538907A (en) * 2020-06-05 2020-08-14 支付宝(杭州)信息技术有限公司 Object recommendation method, system and device

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170262948A1 (en) * 2016-03-08 2017-09-14 International Business Machines Corporation Determination of targeted food recommendation
CN109408731A (en) * 2018-12-27 2019-03-01 网易（杭州）网络有限公司 A kind of multiple target recommended method, multiple target recommended models generation method and device
CN111291266A (en) * 2020-02-13 2020-06-16 腾讯科技（北京）有限公司 Artificial intelligence based recommendation method and device, electronic equipment and storage medium
CN111538907A (en) * 2020-06-05 2020-08-14 支付宝(杭州)信息技术有限公司 Object recommendation method, system and device
CN111523044A (en) * 2020-07-06 2020-08-11 南京梦饷网络科技有限公司 Method, computing device, and computer storage medium for recommending target objects

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
JIAQI MA ET AL.: "Modeling Task Relationships in Multi-task learning with Multi-gate Mixture-of-Experts", 《RESEARCH TRACK》, 23 August 2018 (2018-08-23), pages 1930 - 1938 *

Cited By (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113672807A (en) * 2021-08-05 2021-11-19 杭州网易云音乐科技有限公司 Recommendation method, device, medium, device and computing equipment
CN113672807B (en) * 2021-08-05 2024-03-05 杭州网易云音乐科技有限公司 Recommendation method, recommendation device, recommendation medium, recommendation device and computing equipment
CN113742578A (en) * 2021-08-13 2021-12-03 杭州网易云音乐科技有限公司 Data recommendation method and device, electronic equipment and storage medium
CN113742590A (en) * 2021-09-07 2021-12-03 北京沃东天骏信息技术有限公司 Recommendation method and device, storage medium and electronic equipment
WO2023040494A1 (en) * 2021-09-14 2023-03-23 腾讯科技（深圳）有限公司 Resource recommendation method, and multi-target fusion model training method and apparatus
CN113516522A (en) * 2021-09-14 2021-10-19 腾讯科技（深圳）有限公司 Media resource recommendation method, and training method and device of multi-target fusion model
CN113837808A (en) * 2021-09-27 2021-12-24 北京有竹居网络技术有限公司 Promotion information pushing method, device, equipment, medium and product
CN113837808B (en) * 2021-09-27 2024-02-20 北京有竹居网络技术有限公司 Promotion information pushing method, device, equipment, medium and product
CN113626719A (en) * 2021-10-12 2021-11-09 腾讯科技（深圳）有限公司 Information recommendation method, device, equipment, storage medium and computer program product
CN113626719B (en) * 2021-10-12 2022-02-08 腾讯科技（深圳）有限公司 Information recommendation method, device, equipment, storage medium and computer program product
WO2023061087A1 (en) * 2021-10-12 2023-04-20 腾讯科技（深圳）有限公司 Information recommendation method and apparatus, and electronic device, computer-readable storage medium and computer program product
CN113919893B (en) * 2021-12-14 2022-03-15 腾讯科技（深圳）有限公司 Information pushing method and device, electronic equipment and readable medium
CN113919893A (en) * 2021-12-14 2022-01-11 腾讯科技（深圳）有限公司 Information pushing method and device, electronic equipment and readable medium
CN114430504B (en) * 2022-01-28 2023-03-10 腾讯科技（深圳）有限公司 Recommendation method and related device for media content
CN114430504A (en) * 2022-01-28 2022-05-03 腾讯科技（深圳）有限公司 Recommendation method and related device for media content
CN114462584B (en) * 2022-04-11 2022-07-22 北京达佳互联信息技术有限公司 Recommendation model training method, recommendation device, server and medium
CN114462584A (en) * 2022-04-11 2022-05-10 北京达佳互联信息技术有限公司 Recommendation model training method, recommendation device, server and medium

Similar Documents

Publication Publication Date Title
CN112699305A (en) Multi-target recommendation method, device, computing equipment and medium
US20220198289A1 (en) Recommendation model training method, selection probability prediction method, and apparatus
EP3862893A1 (en) Recommendation model training method, recommendation method, device, and computer-readable medium
WO2023040494A9 (en) Resource recommendation method, and multi-target fusion model training method and apparatus
CN110969516A (en) Commodity recommendation method and device
CN109918662B (en) Electronic resource label determination method, device and readable medium
CN111046286A (en) Object recommendation method and device and computer storage medium
CN111966914A (en) Content recommendation method and device based on artificial intelligence and computer equipment
CN113392359A (en) Multi-target prediction method, device, equipment and storage medium
CN105761102B (en) Method and device for predicting commodity purchasing behavior of user
CN111950593A (en) Method and device for recommending model training
CN114117216A (en) Recommendation probability prediction method and device, computer storage medium and electronic equipment
CN114463091A (en) Information push model training and information push method, device, equipment and medium
WO2024002167A1 (en) Operation prediction method and related apparatus
WO2023185925A1 (en) Data processing method and related apparatus
CN111461757B (en) Information processing method and device, computer storage medium and electronic equipment
CN115186192A (en) Information processing method, device, storage medium and equipment
CN117217284A (en) Data processing method and device
CN117009650A (en) Recommendation method and device
WO2023050143A1 (en) Recommendation model training method and apparatus
Chen et al. Controllable Multi-Objective Re-ranking with Policy Hypernetworks
WO2024012360A1 (en) Data processing method and related apparatus
CN113744009A (en) Target object output method and device, computer readable medium and electronic equipment
EP4398128A1 (en) Recommendation method and related device
CN116204709A (en) Data processing method and related device

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination