CN112785391B - Recommendation processing method and device, intelligent equipment and storage medium - Google Patents

Recommendation processing method and device, intelligent equipment and storage medium Download PDF

Info

Publication number
CN112785391B
CN112785391B CN202110144785.XA CN202110144785A CN112785391B CN 112785391 B CN112785391 B CN 112785391B CN 202110144785 A CN202110144785 A CN 202110144785A CN 112785391 B CN112785391 B CN 112785391B
Authority
CN
China
Prior art keywords
recommended
vector
network
network layer
processing
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.)
Active
Application number
CN202110144785.XA
Other languages
Chinese (zh)
Other versions
CN112785391A (en
Inventor
张懿
甘露
吴伟佳
洪庚伟
李羽
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.)
Weimin Insurance Agency Co Ltd
Original Assignee
Weimin Insurance Agency 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 Weimin Insurance Agency Co Ltd filed Critical Weimin Insurance Agency Co Ltd
Priority to CN202110144785.XA priority Critical patent/CN112785391B/en
Publication of CN112785391A publication Critical patent/CN112785391A/en
Application granted granted Critical
Publication of CN112785391B publication Critical patent/CN112785391B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0601—Electronic shopping [e-shopping]
    • G06Q30/0631—Item recommendations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features
    • 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
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G06Q30/0202—Market predictions or forecasting for commercial activities

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Strategic Management (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Data Mining & Analysis (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Business, Economics & Management (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Game Theory and Decision Science (AREA)
  • Human Resources & Organizations (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Operations Research (AREA)
  • Tourism & Hospitality (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Quality & Reliability (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The embodiment of the application discloses a recommendation processing method, a recommendation processing device, intelligent equipment and a storage medium. The method is applied to intelligent equipment, and a pre-trained sequencing model is configured on the intelligent equipment and comprises a first network layer, a second network layer and a third network layer. The method comprises the following steps: and acquiring fusion characteristic data of each element to be recommended in the element set to be recommended. And processing each piece of fusion characteristic data through the first network layer to generate a first characteristic vector and a second characteristic vector corresponding to each element to be recommended. And respectively processing the first characteristic vector and the second characteristic vector corresponding to each element to be recommended through a first sub-network of a target type included in the first network in the second network layer and the second network in the second network layer to obtain a first output vector and a second output vector. And processing the first output vector and the second output vector through a third network layer to obtain predicted values of the elements to be recommended. By adopting the embodiment of the application, the recommendation precision can be improved.

Description

Recommendation processing method and device, intelligent equipment and storage medium
Technical Field
The application relates to the technical field of artificial intelligence, in particular to a recommendation processing method and device, intelligent equipment and storage medium.
Background
With the rapid development of the internet, the application scenes of the item recommendation method are more and more. The related recommendation provided by the item recommendation method can help the user to better select satisfactory items, and meanwhile, the income of the electronic commerce platform can be promoted.
At present, for a recommendation method under a multi-page recommendation scene, related technologies mostly recommend the user according to elements such as goods and services recently clicked or browsed by the user, for example, the user pays attention to office stationery recently frequently, when the user needs to recommend corresponding elements to be recommended, the office stationery is often ordered before on a page, and the user can conveniently and quickly find the commodity elements such as office stationery. How to better order elements such as goods, services, etc. when recommending these elements becomes a hotspot problem for research.
Disclosure of Invention
The embodiment of the application provides a recommendation processing method and device, intelligent equipment and storage medium, and can be used for better realizing the sorting of elements to be recommended.
The embodiment of the application provides a recommendation processing method, which is applied to intelligent equipment, wherein the intelligent equipment is configured with a pre-trained sequencing model, the sequencing model comprises a first network layer, a second network layer and a third network layer, and the method comprises the following steps:
Acquiring fusion characteristic data of each element to be recommended in an element set to be recommended;
processing the fusion feature data corresponding to each element to be recommended through a first network layer to generate a first feature vector and a second feature vector corresponding to each element to be recommended;
processing a first feature vector corresponding to each element to be recommended through a first sub-network of a target type included in a first network in a second network layer to obtain a first output vector, wherein the first network comprises a plurality of types of first sub-networks, and the target type is matched with the type of the element set to be recommended;
processing a second feature vector corresponding to each element to be recommended through a second network in a second network layer to obtain a second output vector;
and processing the first output vector and the second output vector through a third network layer to obtain a predicted value of each element to be recommended in the element set to be recommended, so as to determine the recommended arrangement sequence of each element to be recommended in the element set to be recommended according to the predicted value.
The embodiment of the application provides a recommendation processing device, which is configured with a pre-trained ranking model, wherein the ranking model comprises a first network layer, a second network layer and a third network layer, and the device comprises:
The fusion characteristic data acquisition module is used for acquiring fusion characteristic data of each element to be recommended in the element set to be recommended;
the fusion characteristic data processing module is used for processing fusion characteristic data corresponding to each element to be recommended through the first network layer to generate a first characteristic vector and a second characteristic vector corresponding to each element to be recommended;
the first output vector determining module is used for processing the first feature vector corresponding to each element to be recommended through a first sub-network of a target type included in a first network in the second network layer to obtain a first output vector, wherein the first network comprises a plurality of first sub-networks of types, and the target type is matched with the type to which the element set to be recommended belongs;
the second output vector determining module is used for processing the second feature vector corresponding to each element to be recommended through a second network in the second network layer to obtain a second output vector;
the predicted value determining module is used for processing the first output vector and the second output vector through a third network layer to obtain predicted values of each element to be recommended in the element set to be recommended, so as to determine the recommended arrangement sequence of each element to be recommended in the element set to be recommended according to the predicted values.
The embodiment of the application provides a smart device, which comprises a processor, a memory and a transceiver, wherein the processor, the memory and the transceiver are mutually connected. The memory is for storing a computer program supporting the smart device to perform the method provided in the first aspect, the computer program comprising program instructions, the processor and transceiver being configured to invoke the program instructions to perform the method provided in the first aspect.
The present embodiments provide a computer readable storage medium storing a computer program comprising program instructions which, when executed by a processor, cause the processor to perform the method provided above.
In the embodiment of the application, two sets of networks are designed to form a recommendation model, first, a first feature vector and a second feature vector corresponding to each element to be recommended are generated through a first network layer aiming at the acquired fusion feature data of the elements to be recommended, then, the first feature vector corresponding to each element to be recommended is processed through a first sub-network which is included in the first network in the second network layer and is related to the type of each element to be recommended, a first output vector is obtained, the second feature vector corresponding to each element to be recommended is processed through a second network in the second network layer, a second output vector is obtained, finally, the first output vector and the second output vector are processed through a third network layer, the predicted value of each element to be recommended in an element set to be recommended is obtained, different first sub-networks are configured for the element set to be recommended in the first network, so that the elements to be recommended are analyzed and processed in a targeted manner based on the type, the elements to be recommended are processed globally, and finally, the ordering of the elements to be recommended is realized through the third network layer, the ordering of the elements to be recommended can be more accurately ordered, the requirements of users are met, and the intelligent service demands of users are met.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic diagram of a system architecture according to an embodiment of the present application;
FIG. 2 is a schematic diagram of a ranking model provided in an embodiment of the present application;
FIG. 3 is a flowchart illustrating a recommendation processing method according to an embodiment of the present disclosure;
FIG. 4 is a schematic diagram of verification results of each ranking model provided in an embodiment of the present application;
FIG. 5 is a schematic diagram of the verification results of the AB test provided in the examples of the present application;
FIG. 6 is a schematic structural diagram of a recommendation processing apparatus according to an embodiment of the present disclosure;
FIG. 7 is another schematic structural diagram of a recommendation processing device according to an embodiment of the present application;
fig. 8 is a schematic structural diagram of an intelligent device according to an embodiment of the present application.
Detailed Description
The following description of the embodiments of the present application will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are only some, but not all, of the embodiments of the present application. All other embodiments, which can be made by one of ordinary skill in the art based on the embodiments herein without making any inventive effort, are intended to be within the scope of the present application.
The recommendation processing method provided by the embodiment of the application can be applied to intelligent equipment, wherein a pre-trained sequencing model is configured on the intelligent equipment, and the sequencing model comprises a first network layer, a second network layer and a third network layer. The smart device includes, but is not limited to, a server, a terminal device, and the like, where the terminal device includes, but is not limited to, a smart phone, a tablet computer, a notebook computer, a desktop computer, and the like. For convenience of description, a smart device will be described as an example. It can be appreciated that the intelligent device configured with the ranking model can be used to execute the recommendation of each element to be recommended included in any one of the online recommendation systems. For example, the online recommendation system may be a commodity recommendation system, a service recommendation system, a song recommendation system, a video recommendation system, and the like, which are specifically determined according to the actual application scenario, and are not limited herein. The service recommendation system may be a system for insurance service recommendation, and the like, and is not limited herein. For convenience of description, the object to be recommended in the recommendation system is called an element to be recommended, or an item to be recommended (item), or the like. The above-mentioned recommended articles are generalized articles, and are the general names of all recommended objects.
Referring to fig. 1, fig. 1 is a schematic diagram of a system architecture according to an embodiment of the present application. As shown in fig. 1, the server 100a, the server 100b, and the terminal device 101 may be connected to each other via a network. The terminal device shown in fig. 1 may include a mobile phone, a tablet computer, a notebook computer, a palm computer, a mobile internet device (mobile internet device, MID), a wearable device (e.g., a smart watch, a smart bracelet, etc.), and the like, which is not limited herein. It should be understood that the server 100a may perform data transmission with the server 100b, and the server 100a and the server 100b may also perform data transmission with the terminal device 101, respectively. The terminal device may be loaded with application programs having various functionalities, and for convenience of description, the application programs having various functionalities loaded on the terminal device may be exemplified by the first application. Alternatively, the first application may be a video application, a music playing application, a game application, a shopping application, or the like, or the first application may be an applet, a web page, or the like, which is not limited herein.
It is understood that the server 100a and the server 100b may be local servers of the first application, remote servers (e.g., cloud servers), or the like, and are not limited herein. The server 100a may be configured to store various types of data (e.g., user characteristic data of a registered user of the first application, element characteristic data of each element to be recommended, etc.) and information related to the first application. The server 100b may be configured to obtain various feature data required for a certain page recommendation scenario from the server 100a, and perform processing according to the feature data to generate a recommendation list, so as to send the recommendation list to a terminal device for display. That is, when the first application in the terminal device transmits a data acquisition request to the server 100b, the server 100b may return a data instance (e.g., a recommendation list as in fig. 1) to the first application in the terminal device based on the received data acquisition request to update the interface presentation instance of the first application in the terminal device.
The method in the embodiment of the application can be applied to intelligent equipment, wherein the intelligent equipment is provided with a pre-trained sequencing model, and the sequencing model comprises a first network layer, a second network layer and a third network layer. Specifically, fusion characteristic data of each element to be recommended in the element set to be recommended is obtained. And processing the fusion characteristic data corresponding to each element to be recommended through the first network layer to generate a first characteristic vector and a second characteristic vector corresponding to each element to be recommended. Processing a first feature vector corresponding to each element to be recommended through a first sub-network of a target type, which is included in a first network in a second network layer, to obtain a first output vector, wherein the first network comprises a plurality of first sub-networks of types, and the target type is matched with the type to which the element set to be recommended belongs. And processing the second feature vector corresponding to each element to be recommended through a second network in the second network layer to obtain a second output vector. And processing the first output vector and the second output vector through a third network layer to obtain a predicted value of each element to be recommended in the element set to be recommended, so as to determine the recommended arrangement sequence of each element to be recommended in the element set to be recommended according to the predicted value. By adopting the embodiment of the application, the recommendation precision can be improved.
As can be appreciated, please refer to fig. 2, fig. 2 is a schematic structural diagram of a ranking model provided in an embodiment of the present application. As shown in fig. 2, the ranking model in the embodiment of the present application may include a first network layer, a second network layer, and a third network layer. The first network layer comprises an embedded layer network and a splicing and locating layer, and the second network layer comprises a routing network, a first network TSNA and a second network GEN. Wherein the first network TSNA includes m first subnetworks, such as the first subnetwork TSN in FIG. 2 0 …, first subnetwork TSN m-1 . The second network GEN comprises n second sub-networks, such as the second sub-network g in FIG. 2 0 …, second subnetwork g n-1 . Wherein m and n are integers greater than 1. The value of m is the same as the number of page types of the recommended pages included in the multi-page recommendation scene. That is, one of the first networks is used to learn the characteristics of elements to be recommended in one type of recommended page. The third network layer includes an attention unit queue, a splice localization layer, a routing network, and a third network (the third network may also be referred to as a trim network or a trim layer) including m third sub-networks, such as the third sub-network MLP in fig. 2 1 …, third subnetwork MLP m-1 。
The first network layer is mainly used for processing the fusion feature data of each element to be recommended included in the target recommendation page to obtain a first feature vector and a second feature vector. The first network TSNA in the second network layer is mainly used for processing the first feature vector corresponding to each element to be recommended to obtain a first output vector, and the second network GEN in the second network layer is mainly used for processing the second feature vector of each element to be recommended to obtain a second output vector corresponding to each element to be recommended. The third network layer is mainly used for processing the first output vector and the second output vector to obtain a predicted value of the element to be recommended.
It is understood that a first sub-network included in the first network in the second network layer in the present application may be used to learn the characteristics of the elements to be recommended included in one type of recommended page independently. Thus, the number of first sub-networks in the first network is the same as the number of page types of the recommended pages included in the multi-page recommendation scenario, i.e. each type of recommended page has a unique first sub-network that can be used to learn the feature distribution of the elements to be recommended included in the corresponding recommended page. That is, assuming that m types of recommended pages are included in the multi-page recommendation scenario, in the training phase of the ranking model, for one first sub-network in the first network, the one first sub-network may be trained from one sample data corresponding to one type of recommended page. Thus, m sample data corresponding to m types of recommended pages may be trained to obtain m first sub-networks. The sample data corresponding to the recommended page of the type comprises page feature data of the recommended page of the type, user feature data of a sample user and element feature data of sample elements to be recommended, which are included in the recommended page of the type.
Accordingly, in order to learn the correlation characteristics among the recommended pages in the multi-page recommendation scene, so as to improve the learning effect and further improve the recommendation accuracy, the second network of the second network layer can be obtained through training according to sample data corresponding to the recommended pages comprising multiple types in the multi-page recommendation scene. For example, assuming that m types of recommended pages are included in the multi-page recommendation scene, in a training stage of the ranking model, the second network may be obtained by training according to m training data corresponding to the m types of recommended pages. The training data comprises page feature data of a type of recommended page, user feature data of a sample user and element feature data of a sample element to be recommended, wherein the sample page feature data of the type of recommended page comprises element feature data of the sample element to be recommended.
The specific implementation process of each network layer will be described in detail in the following embodiments, and will not be described in detail herein.
The method and apparatus provided in the embodiments of the present application will be described in detail below with reference to fig. 3 to 8, respectively.
Referring to fig. 3, fig. 3 is a flowchart illustrating a recommended processing method according to an embodiment of the present application. The method provided by the embodiment of the application may include the following steps S301 to S305:
S301, acquiring fusion characteristic data of each element to be recommended in the element set to be recommended.
In some possible embodiments, the method provided in the present application may be applied to a smart device, where the smart device includes, but is not limited to, a server, a terminal device, and the like, where the terminal device includes, but is not limited to, a smart phone, a tablet computer, a notebook computer, a desktop computer, and the like. For convenience of description, the following will take the smart device as an example. Specifically, the intelligent device may acquire fusion feature data of each element to be recommended in the element set to be recommended, where the fusion feature data corresponding to one element to be recommended may be generated according to user feature data of a target recommendation user, element feature data of the element to be recommended, and page feature data of a target recommendation page. That is, to generate fusion feature data for each element to be recommended in the set of elements to be recommended, the associated features of the set of elements to be recommended may be first acquired. The association features of the element set to be recommended include: and any one or more of user characteristic data of the target recommendation user, page characteristic data of a target recommendation page corresponding to the element set to be recommended, and element characteristic data of each element to be recommended in at least two elements to be recommended included in the target recommendation page. Furthermore, the associated features of the element set to be recommended are processed based on a first network layer in the ordering model, and fusion feature data corresponding to each element to be recommended can be generated.
For example, for any element to be recommended in the element set to be recommended, the element feature data of the element to be recommended, the user feature data of the target recommendation user and the page feature data of the target recommendation page are spliced, so that fusion feature data corresponding to the element to be recommended can be generated. The splicing mode of the user feature data, the page feature data and the element feature data can be determined according to the actual application scene, and the method is not limited. For example, assuming that the target recommended user is user1, the user feature data corresponding to user1 is feature1, the page feature data corresponding to page parent 1 is feature2, and the element feature data corresponding to page parent 1 is feature3, the fusion feature data ra= [ feature1, feature2, feature3] corresponding to the element item a to be recommended, or the fusion feature data ra= [ feature1, feature3, feature2] corresponding to the element item a to be recommended, or the fusion feature data ra= [ feature2, feature1, feature3] corresponding to the element item a to be recommended, or the fusion feature data ra= [ feature2, feature3, feature1] corresponding to the element item a to be recommended, or the fusion feature data ra= [ feature3, feature1, feature3] corresponding to the element item a to be recommended, or the feature1, feature3, or the like.
The target recommending user is a directivity object of the current element recommendation/article recommendation. For example, if the current item recommendation is directed to the user1, the user1 is the target recommended user, and if the current item recommendation is directed to the user2, the user2 is the target recommended user. It will be appreciated that the target recommendation user is typically a registered user in the recommendation system, where the registered user may be a newly registered user, that is, a user who has not purchased or recorded an order, or may be an old registered user, that is, a user who has purchased or recorded an order, and the like, which is not limited herein. Wherein the user characteristic data of the target recommended user includes at least one of a user gender, a user age, a user browsing history, a user purchase history, and the like.
The recommended page corresponding to the element set to be recommended is a target recommended page. That is, at least two elements to be recommended included in the element set to be recommended in the present application are elements in the recommendation scene of the target recommendation page. For example, assuming that the multi-page recommendation scenario includes a special offer page, a payment completion page, an insurance list page, and a child product page, the target recommendation page may be a special offer page, or a payment completion page, or an insurance list page, or a child product page. The page feature data of the target recommended page can comprise at least one of page identification, user browsing duration, user positioning and the like. The element to be recommended may also be referred to as an item to be recommended, and the element characteristic data of the element to be recommended may be referred to as item characteristic data of the item to be recommended. It should be appreciated that the item characteristic data for the item to be recommended may include at least one of an item identification, an item type, an item price, an item manufacturer, and the like.
For example, assume that the target recommendation page is a policy list page, wherein elements to be recommended included in the policy list page are individual risk categories, wherein risk characteristic data of each risk category may include at least one of a risk name, an overlaid disease category, a price, an underwriting company, a guard, a minimum applied age, a maximum applied age, and the like. The user characteristic data of the target recommended user may include the user's gender, the user's age, the user's annual policy, which policies are being paid, which policies are being invalidated, participating in the health event, whether the user is public number focused, whether the user is a car user, whether the user is a new user, the click behavior in the last 30 days, etc. The page feature data of the policy list page may include page identification, user location, etc.
S302, processing the fusion feature data corresponding to each element to be recommended through a first network layer to generate a first feature vector and a second feature vector corresponding to each element to be recommended.
In some possible embodiments, the first feature vector and the second feature vector corresponding to each element to be recommended may be generated by processing the fused feature data corresponding to each element to be recommended through the first network layer. It is understood that the fused feature data in this application may be feature data in libvm format. That is, for any element to be recommended in the target recommendation page, the fusion feature data corresponding to the element to be recommended may be a multi-hot vector. For example, for gender in the user characteristic data, gender "men" may be converted into a form of (gender_men: 1), and for example, for browsing history in the user characteristic data, assume that the target recommended user browsed product is item a, item b, not browsed The browsing history feature may be constructed as (browsing history_item a:1, browsing history_item b:1, browsing history_item c: 0). Thus, the fusion feature data corresponding to the element to be recommended can be expressed asWherein the superscript j indicates the recommended page, and the subscript i indicates the i-th element to be recommended included in the recommended page j. Therefore, the fusion feature data corresponding to each element to be recommended included in the recommendation page is +.>The fusion profile data can be based on a vector dictionary, i.e. the first preset vector of the present application>Conversion to a dense vector->Specifically, the constructed dense vector may be expressed as: />Where emb represents a vector dictionary, x represents a point multiplication (i.e., multiplication of elements in a vector). Wherein, a dense vector is constructed>Then, for the input of the first sub-network of the target type included in the first network in the second network layer and the input of the second network, the first network layer will add the dense vector to the second sub-network of the target type>And performing different processing to generate a first feature vector and a second feature vector, and further taking the first feature vector as an input of a first sub-network of a target type included in a first network in a second network layer and taking the second feature vector as an input of a second network in the second network layer. Specifically, for the first network layer For the input of a first sub-network of the target type comprised by a network, the first network layer of the present application can be implemented by adding to the dense vector>Summing according to rows to obtain a first eigenvector, which may be denoted +.>For the input of the second network in the second network layer, the first network layer of the present application can be implemented by adding to the dense vector +.>The second feature vector is obtained by splicing after opening according to the row, and can be marked as +.>For example, assume the dense vector +.>Is [ N, M ]]The first network comprises the input variables of the first sub-network of the target type (i.e. the first eigenvector +.>) The dimension is [1, M]The input variable of the second network (i.e. the second eigenvector +.>) Dimension of [1, N x M]Wherein N and M are integers greater than 1.
For example, assume that the fusion profile in this application isThe acquired first preset vectorThen based onConversion formula of dense vector->Can obtain the corresponding dense vector asThus, by +.>According to the row summation, a first eigenvector +.>By>Opening according to the row and then splicing to obtain a second characteristic vector of +. >
S303, processing the first feature vector corresponding to each element to be recommended through a first sub-network of a target type included in a first network in the second network layer to obtain a first output vector.
In some possible embodiments, the first output vector may be obtained by processing, through a first sub-network of a target type included in a first network in the second network layer, a first feature vector corresponding to each element to be recommended. The first network comprises a plurality of types of first sub-networks, and the target type is matched with the type of the element set to be recommended. That is, the first sub-networks of the plurality of types referred to in the embodiments of the present application may be understood as first sub-networks corresponding to the plurality of page types, and one page type corresponds to one first sub-network. For example, assuming that the multi-page recommendation scenario includes 4 recommendation pages, the number of first sub-networks in the first network is 4 (i.e., m=4). The target type may be a type determined from a page identification included in the page feature data of the target recommended page.
It can be appreciated that the first sub-network in the embodiment of the present application may be a multi-layer level (MLP), or may also be a Wide & Deep (a network structure), or may also be a Deep fm (a network structure), which is specifically determined according to an actual application scenario, and is not limited herein. For convenience of description, the first subnetwork is taken as an MLP for illustration in the embodiments of the present application. That is, the first network in the second network layer is a set of page-oriented recommendation scenes MLPs. Wherein, an MLP in the first network is used for learning the characteristic representation of each element to be recommended included in a page recommendation scene. In general, in the embodiment of the present application, the number of MLPs included in the first network is the same as the number of page types of recommended pages included in the multi-page recommendation scenario, that is, each recommended page has an independent MLP. For example, assuming that a recommendation system includes 4 recommendation pages, which are a special sale page, a payment completion page, an insurance list page, and a child product page, respectively, the 4 recommendation pages will correspond to 4 MLPs, one MLP being used to independently learn a feature representation of an element to be recommended included in a corresponding one of the recommendation pages, so as to capture a corresponding data distribution.
In order to realize data distribution capable of respectively learning each page recommendation scene, the embodiment of the application designs a routing network for selective activation, wherein the routing network can determine a target type according to page feature data of a target recommendation page, and further guide first feature vectors corresponding to elements to be recommended in the target recommendation page to MLPs corresponding to the target type for calculation. Specifically, the routing network may guide the first feature vector corresponding to each element to be recommended in the target recommendation page to the MLP corresponding to the target recommendation page for calculation according to the page identifier included in the page feature data of the target recommendation page, so as to obtain the first output vector. Wherein the output of the first network (i.e., the first output vector in the embodiments of the present application) can be calculated by the following formula:
wherein,representing a first output vector, ">Representing the output of the h-th MLP.
S304, processing the second feature vector corresponding to each element to be recommended through a second network in the second network layer to obtain a second output vector.
In some possible embodiments, the second feature vector corresponding to each element to be recommended is processed through a second network in the second network layer to obtain a second output vector. It is understood that the second network in the second network layer may include n second sub-networks, where n is an integer greater than 1. The processing, by the second network in the second network layer, the second feature vector corresponding to each element to be recommended, so as to obtain a second output vector may be understood as: firstly, determining second feature vectors corresponding to elements to be recommended as n vector sets, and further respectively inputting the n vector sets into n second sub-networks to respectively process the corresponding vector sets through each second sub-network to obtain second output vectors. Wherein, determining the second feature vector corresponding to the element to be recommended as n vector sets can be understood as: and splitting the second feature vector corresponding to the element to be recommended to obtain n vector sets. For example, assume that the second feature vector has dimensions of [1, 10 ]N=3. Wherein, if the second feature vector specifically is:then by->Splitting to obtain 3 vector sets of +.> It can be appreciated that the length of each split vector set in the embodiments of the present application may be determined according to the actual application scenario, which is not limited herein.
In some possible embodiments, the foregoing processing, by each second sub-network, the corresponding vector set, to obtain the second output vector may be understood as: obtaining each output vector obtained by processing the corresponding vector set by each second sub-networkWherein, the value range of k is more than or equal to 0 and less than or equal to n-1, and k is an integer. And further, respective output vectors outputted according to respective second sub-networks +.>A second output vector is generated.
Specifically, referring to fig. 2 together, unlike the first sub-network in the second network layer, which is used to independently learn the feature representation of the element to be recommended based on only the first sub-network corresponding to the target type at a time, the second network in the second network layer can be used to learn the global feature representation. Wherein, in order to be able to capture the characteristic representation of different spaces, the embodiment of the application designs a characteristic extraction mode similar to a multi-head attention mechanism for the second network. For the second network, the embodiment of the present application may be designed as a queue formed by a set of second sub-networks, so the output of the second network (i.e. the second output vector in the present application) may be expressed as:
Wherein,representing a second output vector, ">And representing the output of a kth second sub-network, wherein the value range of k is more than or equal to 0 and less than or equal to n-1, and k is an integer. n represents the number of second subnetworks, that is to say the second output vector consists of the output vector of each of the n second subnetworks. The respective second sub-networks included in the second network may also be MLPs, or may be Wide, similar to the first sub-networks included in the first network&Deep (a network structure), or Deep fm (a network structure) may be used, which is specifically determined according to the actual application scenario, and is not limited herein. It is understood that in the second network, each second sub-network is included in the second network +.>Focusing on only the input vector (i.e. the second feature vector in this application +.>) So that features of different spaces can be more fully captured, the specific implementation can be represented by the following formula:
wherein g 0 Focusing on the second feature vectorX in the middle 1 ～x l0 Partial input g 1 Attention to the second eigenvector->X in the middle l0+1 ～x l1 Partial input, and so on, g n-1 Attention to the second eigenvector->Middle->Partial input. The lengths of the eigenvalues in the second eigenvectors focused by different second subnetworks can be adjusted according to the actual application scenario, which is not limited herein.
S305, processing the first output vector and the second output vector through a third network layer to obtain a predicted value of each element to be recommended in the element set to be recommended.
In some possible embodiments, the predicted value of each element to be recommended in the element set to be recommended may be obtained by processing the first output vector and the second output vector through the third network layer. Specifically, the first output vector and the second output vector are processed by the attention unit of the target type in the attention unit queue in the third network layer, so as to obtain a weight vector, wherein the weight vector comprises n weight values. And processing the weight vector, the first output vector and the second output vector through a third network layer to obtain the predicted value of each element to be recommended in the element set to be recommended. The attention unit queue comprises a plurality of attention units of different types, and the target type is matched with the type to which the element set to be recommended belongs. The processing of the first output vector and the second output vector by the attention unit of the target type in the attention unit queue in the third network layer to obtain the weight vector may be understood as: the first output vector and the second output vector are processed through the attention units of the target type in the attention unit queue in the third network layer to obtain a process parameter vector, and then a weight vector is determined according to the process parameter vector, wherein the process parameter vector comprises n process parameters.
In some possible embodiments, the processing, by the third network layer, the weight vector, the first output vector, and the second output vector to obtain the predicted value of each element to be recommended in the element to be recommended set may be understood as: and processing the weight vector, the first output vector and the second output vector through a spliced connection network of the third network layer to obtain a predicted feature vector. Further, the predicted feature vector is processed through a third sub-network of the target type in a third network in the third network layer, so that a predicted value corresponding to each element to be recommended can be determined. It can be appreciated that the third network includes a plurality of third sub-networks of different types, and the target type is matched with the type to which the element set to be recommended belongs.
Specifically, referring to fig. 2 together, the embodiment of the present application assists in element recommendation of the current page recommendation scenario (i.e. the target recommendation page) by designing a set of attention unit queues in the third network layer for extracting useful recommendation knowledge. Similar to the first network, embodiments of the present application may assign an attention unit to each page recommendation scenario for learning a particular extraction rule (i.e., a process parameter vector in the present application). That is, in the implementation of the present application, the first output vector is passed through the attention unit of the target type in the attention unit queue in the second network layer Second output vector +.>Processing to obtain a process parameter vector a j In particular, the process parameter vector a j The method meets the following conditions:
wherein a is j Representing a process parameter vector, process parameter vector a j The n process parameters are respectivelyThat is, the process included in the process parameter vectorThe number of parameters is the same as the number of second subnetworks. />Representing a first output vector, W is the attention weight matrix of the attention unit, +.>Representing the output vector by the respective second subnetwork +.>And (5) forming an output vector of the second network.
Further, embodiments of the present application may utilize a j To calculate and obtain weight vectorWherein the weight vector->The method meets the following conditions:
wherein,is the calculated weight vector,/>Comprises n weight values, that is, < ->The number of weight values in (a) is the same as the number of second sub-networks included in the second network. Thus finally passing throughSpliced connection network pair first output vector in three network layers>Second output vector->Weight vector +.>And processing to obtain the predicted feature vector corresponding to the element to be recommended. Wherein the predictive feature vector satisfies:
wherein,representing a first output vector, " >For the second output vector->Comprising the output vector of the respective second subnetwork, respectively>Representing weight vector +.>Each weight value included in the table, η represents a normalization parameter.
It can be appreciated that in some possible embodiments, the third network layer in the present application further includes a third network, where the third network includes m third sub-networks. It can be understood that, in order to implement the logoss of each page recommendation scene, a routing network for selective activation is also designed in the third layer network in the embodiment of the present application, and the routing network may determine a target type according to the page identifier in the page feature data of the target recommendation page, so as to guide the predicted feature vector corresponding to the element to be recommended in the target recommendation page to the third sub-network corresponding to the target type for calculation. It can be understood that the third sub-network in the embodiment of the present application may be an MLP, or may also be a Wide & Deep (a network structure), or may also be a Deep fm (a network structure), which is specifically determined according to an actual application scenario, and is not limited herein. For convenience of description, the third subnetwork is taken as an MLP for illustration in the embodiments of the present application. Wherein a third sub-network (i.e., an MLP) in the third network is used to calculate a logoss of a page recommendation scene to determine a final output result. As shown in fig. 2, the embodiment of the present application establishes an independent third sub-network for each page recommendation scenario, and in the present design, the embodiment of the present application uses MLP as the fine tuning network, and the final output result (i.e. the predicted value in the present application) can be expressed by the following formula:
Wherein,MLP, < ++corresponding to the page recommendation scene j>And representing the predicted value corresponding to the current page recommendation scene (namely the target recommendation page). In MLP, reLU is selected as an activating unit in the embodiment of the application, and the method has obvious advantages in the aspects of forward gradient transfer, gradient disappearance reduction and the like.
Finally, the embodiment of the application calculates the final loglos by using weighted cross entropy, and increases the l1 rule to control the complexity of the model, wherein the cross entropy function satisfies the following conditions:
wherein,representing cross entropy, β is a super parameter for representing a weight of each page recommendation scene, the value of which is specified by an expert. />Representing a genuine label->Representing the predicted value, λ represents the weight of the l1 regularization term, and Ω represents the parameter l1 regularization of the model.
In the embodiment of the application, fusion characteristic data of each element to be recommended in the element to be recommended set is obtained. And processing the fusion characteristic data corresponding to each element to be recommended through the first network layer to generate a first characteristic vector and a second characteristic vector corresponding to each element to be recommended. Processing a first feature vector corresponding to each element to be recommended through a first sub-network of a target type, which is included in a first network in a second network layer, to obtain a first output vector, wherein the first network comprises a plurality of first sub-networks of types, and the target type is matched with the type to which the element set to be recommended belongs. And processing the second feature vector corresponding to each element to be recommended through a second network in the second network layer to obtain a second output vector. And processing the first output vector and the second output vector through a third network layer to obtain a predicted value of each element to be recommended in the element set to be recommended, so as to determine the recommended arrangement sequence of each element to be recommended in the element set to be recommended according to the predicted value. By adopting the embodiment of the application, the recommendation precision can be improved.
It can be appreciated that, to verify the effectiveness of the ranking model proposed in the present application, the embodiments of the present application verify the model effect using actual data. Wherein. The data set for verification is derived from internet insurance sales data in a real scene, and comprises about 560 ten thousand pieces of data, wherein the data set mainly comprises the following four page recommendation scenes:
special offer (scenario 1): for a recommended scenario for a user with a certain awareness, the number of samples is 390W.
Payment completion page (scenario 2): for the recommended scenario where the product purchasing user has completed, the sample size is 18.3W.
Policy list page (scenario 3): the recommended scenario for the user has not been purchased, and the sample number is 41W.
Child product page (scenario 4): the product recommendation scenario prepared for children/children, sample number 115W.
At the time of testing, the present embodiment uses 80% of the data in the dataset for testing and 20% of the data in the dataset for validation. It can be appreciated that AUC (a parameter for measuring the ranking capability of the model), recall@n (recall) and NDCG (a parameter for measuring the relevance of the ranking) are selected as evaluation indexes for measuring the performance of the model. The other 5 sorting models for comparison with the sorting model proposed in the embodiment of the present application are respectively:
(1) Logistic regression (logistic regression, LR) model: during testing, an LR model is obtained by training the FTRL (a training mode).
(2) Depth and logistic regression (Wide & Deep) model: this structure was used as an intermediate feature interaction layer during testing.
(3) Underlying shared network (bottom shared model, BSM) model: a modified multitasking recommendation network, multitasking sharing the same embedded layer, only doing weighting processing at the last fine tuning layer.
(4) Needle-threading network (CoNet) model: the feature interaction layer adopts a feature interaction method of fixed weight sharing.
(5) Multi-expert-network (MMoE) model: the interaction layer adopts an expert network queue, and the input sample and the control gate determine the input of the fine tuning layer.
When testing, the learning rate is uniformly selected to be 0.002, the weight beta= [0.5,0.1,0.2,0.2], and the regular weight of l1 is 0.1. Each network model is trimmed to an optimal result. Referring to fig. 4, fig. 4 is a schematic diagram of verification results of each ranking model according to an embodiment of the present application. The test results of the ranking model (i.e., SAMN in fig. 4) and the other 5 ranking models (i.e., LR, wide & Deep, BSM, coNet, MMoE in fig. 4) set forth in the present application are shown in fig. 4. As can be seen from fig. 4, in the above-mentioned multi-page recommendation scenario, the recommendation model proposed in the present application exceeds the other 5 ranking models on the training set/test set on the above three indexes (i.e. AUC, recall@n and NDCG), so that the effectiveness of the ranking model proposed in the present application is demonstrated.
Meanwhile, in the embodiment of the present application, an AB test is performed for 7 days on the 4 page recommendation scenes, please refer to fig. 5, and fig. 5 is a schematic diagram of a verification result of the AB test provided in the embodiment of the present application. In this embodiment, the online CR (conversion rate) is normalized, the baseline (tree model) is normalized to 1, and the verification result is shown in fig. 5: in the 4 page recommendation scenes, the CR average lifting value of the 4 page recommendation scenes is as follows, compared with the baseline: 3.98%,1.04%,1.26% and 1.05%. Again, the AB test results demonstrate the effectiveness of the present application.
Referring to fig. 6, fig. 6 is a schematic structural diagram of a recommendation processing apparatus according to an embodiment of the present application. The recommendation processing device is an intelligent device, and a pre-trained sequencing model is configured on the intelligent device and comprises a first network layer, a second network layer and a third network layer. The device of the embodiment of the invention can be arranged in intelligent equipment such as intelligent terminals, servers and the like, the intelligent terminals can be various intelligent mobile phones, tablet personal computers, intelligent wearable equipment, personal computers and the like, the servers mainly refer to some commodity and service recommending platforms and even servers provided by some social application platforms, and the recommending processing device provided by the embodiment of the invention comprises:
The fusion characteristic data acquisition module 61 is configured to acquire fusion characteristic data of each element to be recommended in the element set to be recommended;
the fused feature data processing module 62 is configured to process, through the first network layer, fused feature data corresponding to each element to be recommended, and generate a first feature vector and a second feature vector corresponding to each element to be recommended;
a first output vector determining module 63, configured to process, through a first sub-network of a target type included in a first network in a second network layer, a first feature vector corresponding to each element to be recommended to obtain a first output vector, where the first network includes a plurality of first sub-networks of types, and the target type is matched with a type to which the set of elements to be recommended belongs;
a second output vector determining module 64, configured to process, through a second network in the second network layer, a second feature vector corresponding to each element to be recommended, to obtain a second output vector;
the predicted value determining module 65 is configured to process the first output vector and the second output vector through a third network layer to obtain a predicted value of each element to be recommended in the set of elements to be recommended, so as to determine a recommendation arrangement sequence of each element to be recommended in the set of elements to be recommended according to the predicted value.
Referring to fig. 7, fig. 7 is another schematic structural diagram of the recommendation processing apparatus according to the embodiment of the present application. The device of the embodiment of the invention can be arranged in intelligent equipment such as intelligent terminals, servers and the like, wherein the intelligent terminals can be various intelligent mobile phones, tablet personal computers, intelligent wearable equipment, personal computers and the like, and the servers mainly refer to servers provided by some commodity and service recommending platforms and even some social application platforms, and the device comprises the following structures.
In some possible embodiments, the fused feature data acquisition module 61 includes:
an associated feature obtaining unit 611, configured to obtain an associated feature of the element set to be recommended, where the associated feature includes: any one or more of user characteristic data of a target recommendation user, page characteristic data of a target recommendation page corresponding to a set of elements to be recommended, and article characteristic data of each element to be recommended in at least two elements to be recommended included in the target recommendation page;
and the fusion characteristic data generating unit 612 is configured to generate fusion characteristic data corresponding to each element to be recommended according to the associated characteristic of the set of elements to be recommended through the first network layer.
In some possible embodiments, the dimension of the fused feature data is [1*N ]; the fused feature data processing module 62 includes:
a dense vector determining unit 621, configured to obtain a first preset vector, and generate a dense vector according to the fused feature data and the first preset vector, where a dimension of the first preset vector is [ M, N ], and a dimension of the dense vector is [ N, M ];
the feature vector generating unit 622 is configured to determine a first feature vector and a second feature vector corresponding to each element to be recommended according to the dense vector, where a dimension of the first feature vector is [1, m ], and a dimension of the second feature vector is [1, n×m ].
In some possible embodiments, the second network in the second network layer includes n second sub-networks, where n is an integer greater than 1; the second output vector determination module 64 includes:
a vector set determining unit 641 for determining the second feature vector as n vector sets;
the second output vector determining unit 642 is configured to input the n vector sets into the n second sub-networks respectively, so as to process the corresponding vector sets through the respective second sub-networks respectively, and obtain a second output vector.
In some possible embodiments, the predictor determination module 65 includes:
a weight vector determining subunit 651, configured to process, by using an attention unit of a target type in an attention unit queue in the third network layer, the first output vector and the second output vector to obtain a weight vector, where the weight vector includes n weight values;
a predicted value determining unit 652, configured to process, through the third network layer, the weight vector, the first output vector, and the second output vector, to obtain a predicted value of each element to be recommended in the set of elements to be recommended;
the attention unit queue comprises a plurality of attention units in different types, and the target type is matched with the type of the element set to be recommended.
In some possible implementations, the weight vector determination subunit 651 includes:
a process parameter vector determining subunit 6511, configured to process, by using an attention unit of a target type in an attention unit queue in the third network layer, the first output vector and the second output vector to obtain a process parameter vector, where the process parameter vector includes n process parameters;
The weight vector generation subunit 6512 is configured to determine a weight vector according to the process parameter vector, where the weight vector includes n weight values.
In some possible embodiments, the predictor determining unit 652 includes:
a prediction feature vector determining subunit 6521, configured to process, through a Concatenation network of the third network layer, the weight vector, the first output vector, and the second output vector, to obtain a prediction feature vector;
a predicted value determining subunit 6522, configured to process the predicted feature vector through a third sub-network of a target type in a third network in the third network layer, and determine a predicted value corresponding to each element to be recommended;
the third network comprises a plurality of third sub-networks with different types, and the target type is matched with the type of the element set to be recommended.
In some possible embodiments, the fused feature data of each element to be recommended in the set of elements to be recommended includes: and the target recommends page feature data of the page, and the target type is determined according to the page feature data.
Based on the same inventive concept, the principle and beneficial effects of the solution to the problem of the recommended processing device provided in the embodiments of the present application are similar to those of the message processing method in the embodiments of the present application, and may refer to the principle and beneficial effects of implementation of the method, and the relationships between the steps executed by each relevant module may refer to the descriptions of the relevant content in the foregoing embodiments, which are not repeated herein for brevity.
Referring to fig. 8, fig. 8 is a schematic structural diagram of an intelligent device according to an embodiment of the present application. The intelligent device is provided with a pre-trained sequencing model, and the sequencing model comprises a first network layer, a second network layer and a third network layer. As shown in fig. 8, the smart device in this embodiment may include: one or more processors 801, memory 802, and a transceiver 803. The processor 801, the memory 802, and the transceiver 803 are connected through a bus 804.
The processor 801 (or central processing unit (Central Processing Unit, CPU)) is a computing core and a control core of the data processing apparatus, which can parse various instructions within the terminal apparatus and process various data of the terminal apparatus, for example: the CPU can be used for analyzing a startup and shutdown instruction sent by a user to the terminal equipment and controlling the terminal equipment to perform startup and shutdown operation; and the following steps: the CPU can transmit various kinds of interaction data between the internal structures of the terminal device, and so on. The processor 801 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (programmable logic device, PLD), or the like. The PLD may be a field-programmable gate array (FPGA), general-purpose array logic (generic array logic, GAL), or the like.
The transceiver 803 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI, mobile communication interface, etc.), and the control of the processor 801 may be used to transmit and receive data; the transceiver 803 may also be used for transmission and interaction of data inside the terminal device.
The Memory 802 (Memory) is a Memory device in the terminal device for storing programs and data. It will be appreciated that the memory 802 herein may include both built-in memory of the terminal device and extended memory supported by the terminal device. Memory 802 provides storage space that stores the operating system of the terminal device, which may include, but is not limited to: android systems, iOS systems, windows Phone systems, etc., which are not limiting in this application. The memory 802 may include volatile memory (RAM), such as random-access memory (RAM); the memory 802 may also include a non-volatile memory (non-volatile memory), such as a flash memory (flash memory), a Solid State Drive (SSD), etc.; the memory 803 may also include a combination of the above types of memory.
The memory 802 is used for storing a computer program comprising program instructions, and the processor 801 and the transceiver 803 are used for executing the program instructions stored in the memory 802, performing the following operations:
Acquiring fusion characteristic data of each element to be recommended in an element set to be recommended;
processing the fusion feature data corresponding to each element to be recommended through a first network layer to generate a first feature vector and a second feature vector corresponding to each element to be recommended;
processing a first feature vector corresponding to each element to be recommended through a first sub-network of a target type included in a first network in a second network layer to obtain a first output vector, wherein the first network comprises a plurality of types of first sub-networks, and the target type is matched with the type of the element set to be recommended;
processing a second feature vector corresponding to each element to be recommended through a second network in a second network layer to obtain a second output vector;
and processing the first output vector and the second output vector through a third network layer to obtain a predicted value of each element to be recommended in the element set to be recommended, so as to determine the recommended arrangement sequence of each element to be recommended in the element set to be recommended according to the predicted value.
In some possible embodiments, the processor 801 is configured to:
acquiring the association characteristics of the element set to be recommended, wherein the association characteristics comprise: any one or more of user characteristic data of a target recommendation user, page characteristic data of a target recommendation page corresponding to a set of elements to be recommended, and article characteristic data of each element to be recommended in at least two elements to be recommended included in the target recommendation page;
And generating fusion characteristic data corresponding to each element to be recommended according to the associated characteristics of the element set to be recommended through a first network layer.
In some possible embodiments, the dimension of the fused feature data is [1*N ]; the processor 801 is configured to:
acquiring a first preset vector, and generating a dense vector according to the fusion characteristic data and the first preset vector, wherein the dimension of the first preset vector is [ M, N ], and the dimension of the dense vector is [ N, M ];
and determining a first characteristic vector and a second characteristic vector corresponding to each element to be recommended according to the dense vector, wherein the dimension of the first characteristic vector is [1, M ], and the dimension of the second characteristic vector is [1, N x M ].
In some possible embodiments, the second network in the second network layer includes n second sub-networks, where n is an integer greater than 1; the processor 801 is configured to:
determining the second feature vector as a set of n vectors;
and respectively inputting the n vector sets into the n second sub-networks to respectively process the corresponding vector sets through each second sub-network to obtain a second output vector.
In some possible embodiments, the processor 801 is configured to:
Processing the first output vector and the second output vector through an attention unit of a target type in an attention unit queue in the third network layer to obtain a weight vector, wherein the weight vector comprises n weight values;
processing the weight vector, the first output vector and the second output vector through the third network layer to obtain a predicted value of each element to be recommended in the element to be recommended set;
the attention unit queue comprises a plurality of attention units in different types, and the target type is matched with the type of the element set to be recommended.
In some possible embodiments, the processor 801 is configured to:
processing the first output vector and the second output vector through an attention unit of a target type in an attention unit queue in the third network layer to obtain a process parameter vector, wherein the process parameter vector comprises n process parameters;
and determining a weight vector according to the process parameter vector, wherein the weight vector comprises n weight values.
In some possible embodiments, the processor 801 is configured to:
Processing the weight vector, the first output vector and the second output vector through a spliced connection network of the third network layer to obtain a predicted feature vector;
processing the prediction feature vector through a third sub-network of a target type in a third network in the third network layer, and determining a prediction value corresponding to each element to be recommended;
the third network comprises a plurality of third sub-networks with different types, and the target type is matched with the type of the element set to be recommended.
In some possible embodiments, the fused feature data of each element to be recommended in the set of elements to be recommended includes: and the target recommends page feature data of the page, and the target type is determined according to the page feature data.
It should be appreciated that in some possible embodiments, the processor 801 described above may be a central processing unit (central processing unit, CPU), which may also be other general purpose processors, digital signal processors (digital signal processor, DSP), application specific integrated circuits (application specific integrated circuit, ASIC), off-the-shelf programmable gate arrays (field programmable gate array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or the like. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The memory 802 may include read only memory and random access memory, and provides instructions and data to the processor 801. A portion of memory 802 may also include non-volatile random access memory. For example, the memory 802 may also store information of device type.
Based on the same inventive concept, the principle and beneficial effects of the solution to the problem of the intelligent device provided in the embodiments of the present application are similar to those of the message processing method in the embodiments of the present application, and may refer to the principle and beneficial effects of implementation of the method, and the relationship between each step executed by the processor 801 may refer to the description of the related content in the foregoing embodiments, which is not repeated herein for brevity.
Embodiments of the present application also provide a computer readable storage medium having one or more instructions stored therein, where the one or more instructions are adapted to be loaded by a processor and execute the recommended processing method described in the method embodiments above.
The present application also provides a computer program product containing instructions, which when run on a computer, cause the computer to perform the recommended processing method described in the above method embodiment.
Embodiments of the present application also provide a computer program product or computer program comprising computer instructions stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions so that the computer device performs the method of the recommended processing described above.
The steps in the method of the embodiment of the application can be sequentially adjusted, combined and deleted according to actual needs.
The modules in the device of the embodiment of the application can be combined, divided and deleted according to actual needs.
Those of ordinary skill in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be implemented by a program to instruct related hardware, the program may be stored in a computer readable storage medium, and the readable storage medium may include: flash disk, read-Only Memory (ROM), random-access Memory (Random Access Memory, RAM), magnetic or optical disk, and the like.
The above disclosure is illustrative of a preferred embodiment of the present application and, of course, should not be taken as limiting the scope of the invention, and those skilled in the art will recognize that all or part of the above embodiments can be practiced with modification within the spirit and scope of the appended claims.

Claims (9)

1. A recommendation processing method, wherein the method is applied to an intelligent device, and the intelligent device is configured with a pre-trained ranking model, the ranking model includes a first network layer, a second network layer and a third network layer, and the method includes:
Acquiring fusion characteristic data of each element to be recommended in an element set to be recommended;
processing the fusion feature data corresponding to each element to be recommended through a first network layer to generate a first feature vector and a second feature vector corresponding to each element to be recommended;
processing a first feature vector corresponding to each element to be recommended through a first sub-network of a target type, which is included in a first network in a second network layer, to obtain a first output vector, wherein the first network comprises a plurality of types of first sub-networks, and the target type is matched with the type to which the element set to be recommended belongs;
processing a second feature vector corresponding to each element to be recommended through a second network in a second network layer to obtain a second output vector;
processing the first output vector and the second output vector through a third network layer to obtain a predicted value of each element to be recommended in the element set to be recommended, so as to determine a recommendation arrangement sequence of each element to be recommended in the element set to be recommended according to the predicted value;
wherein the dimension of the fusion characteristic data is [1 ]N]The method comprises the steps of carrying out a first treatment on the surface of the The processing, by the first network layer, the fused feature data corresponding to each element to be recommended, and generating a first feature vector and a second feature vector corresponding to each element to be recommended, including: acquiring a first preset vector, and generating a dense vector according to the fusion characteristic data and the first preset vector, wherein the dimension of the first preset vector is [ M, N ] ]The dimension of the dense vector is [ N, M ]]The method comprises the steps of carrying out a first treatment on the surface of the Determining a first feature vector and a second feature vector corresponding to each element to be recommended according to the dense vector, wherein the dimension of the first feature vector is [1, M]The second feature vector has dimensions of [1, N ]>M]；
The processing, by the first sub-network of the target type included in the first network in the second network layer, the first feature vector corresponding to each element to be recommended to obtain a first output vector includes: inputting the first feature vector corresponding to each element to be recommended into a first sub-network of the target type for calculation to obtain the first output vector;
wherein the second network in the second network layer comprises n second sub-networks, n is an integer greater than 1; the processing, by a second network in the second network layer, the second feature vector corresponding to each element to be recommended to obtain a second output vector includes: determining the second feature vector as a set of n vectors; and respectively inputting the n vector sets into the n second sub-networks to respectively process the corresponding vector sets through each second sub-network to obtain the second output vector.
2. The method of claim 1, wherein the obtaining the fused feature data of each element to be recommended in the set of elements to be recommended comprises:
acquiring the association characteristics of the element set to be recommended, wherein the association characteristics comprise: any one or more of user characteristic data of a target recommendation user, page characteristic data of a target recommendation page corresponding to a set of elements to be recommended, and article characteristic data of each element to be recommended in at least two elements to be recommended included in the target recommendation page;
and generating fusion characteristic data corresponding to each element to be recommended according to the associated characteristics of the element set to be recommended through a first network layer.
3. The method of claim 1, wherein the processing, by the third network layer, the first output vector and the second output vector to obtain the predicted value of each element to be recommended in the set of elements to be recommended includes:
processing the first output vector and the second output vector through an attention unit of a target type in an attention unit queue in the third network layer to obtain a weight vector, wherein the weight vector comprises n weight values;
Processing the weight vector, the first output vector and the second output vector through the third network layer to obtain a predicted value of each element to be recommended in the element to be recommended set;
the attention unit queue comprises a plurality of attention units in different types, and the target type is matched with the type of the element set to be recommended.
4. A method according to claim 3, wherein said processing the first output vector and the second output vector by the attention unit of the target type in the attention unit queue in the third network layer to obtain a weight vector comprises:
processing the first output vector and the second output vector through an attention unit of a target type in an attention unit queue in the third network layer to obtain a process parameter vector, wherein the process parameter vector comprises n process parameters;
and determining a weight vector according to the process parameter vector, wherein the weight vector comprises n weight values.
5. The method of claim 3, wherein the processing, by the third network layer, the weight vector, the first output vector, and the second output vector to obtain the predicted value of each element to be recommended in the set of elements to be recommended includes:
Processing the weight vector, the first output vector and the second output vector through a spliced connection network of the third network layer to obtain a predicted feature vector;
processing the prediction feature vector through a third sub-network of a target type in a third network in the third network layer, and determining a prediction value corresponding to each element to be recommended;
the third network comprises a plurality of third sub-networks with different types, and the target type is matched with the type of the element set to be recommended.
6. The method of any one of claims 1-5, wherein the fused feature data for each element to be recommended in the set of elements to be recommended comprises: and the target type is determined according to the page feature data of the target recommended page corresponding to the element set to be recommended.
7. A recommendation processing apparatus, wherein the apparatus is a network device, and a pre-trained ranking model is configured on the apparatus, the ranking model includes a first network layer, a second network layer, and a third network layer, and the apparatus includes:
the fusion characteristic data acquisition module is used for acquiring fusion characteristic data of each element to be recommended in the element set to be recommended;
The fusion characteristic data processing module is used for processing fusion characteristic data corresponding to each element to be recommended through the first network layer to generate a first characteristic vector and a second characteristic vector corresponding to each element to be recommended;
the first output vector determining module is used for processing the first feature vector corresponding to each element to be recommended through a first sub-network of a target type, which is included in a first network in the second network layer, to obtain a first output vector, wherein the first network comprises a plurality of first sub-networks of types, and the target type is matched with the type to which the element set to be recommended belongs;
the second output vector determining module is used for processing the second feature vector corresponding to each element to be recommended through a second network in the second network layer to obtain a second output vector;
the predicted value determining module is used for processing the first output vector and the second output vector through a third network layer to obtain a predicted value of each element to be recommended in the element set to be recommended, so as to determine the recommended arrangement sequence of each element to be recommended in the element set to be recommended according to the predicted value;
wherein the dimension of the fusion characteristic data is [1 ] N]The method comprises the steps of carrying out a first treatment on the surface of the The fusion characteristic data processing module comprises: a dense vector determining unit, configured to obtain a first preset vector, generate a dense vector according to the fused feature data and the first preset vector, and determine a first preset direction of the first dense vectorThe dimension of the quantity is [ M, N]The dimension of the dense vector is [ N, M ]]The method comprises the steps of carrying out a first treatment on the surface of the A feature vector generation unit, configured to determine a first feature vector and a second feature vector corresponding to each element to be recommended according to the dense vector, where a dimension of the first feature vector is [1, M]The second feature vector has dimensions of [1, N ]>M]；
The first output vector determining module is specifically configured to: inputting the first feature vector corresponding to each element to be recommended into a first sub-network of the target type for calculation to obtain the first output vector;
wherein the second network in the second network layer comprises n second sub-networks, n is an integer greater than 1; the second output vector determination module includes: a vector set determining unit configured to determine the second feature vector as n vector sets; and the second output vector determining unit is used for respectively inputting the n vector sets into the n second sub-networks so as to respectively process the corresponding vector sets through each second sub-network to obtain the second output vector.
8. An intelligent device, comprising a processor, a memory and a transceiver, wherein the processor, the memory and the transceiver are connected with each other;
the memory is for storing a computer program comprising program instructions, the processor and the transceiver being configured to invoke the program instructions to perform the method of any of claims 1-6.
9. A computer readable storage medium, characterized in that the computer readable storage medium stores a computer program comprising program instructions which, when executed by a processor, cause the processor to perform the method of any of claims 1-6.
CN202110144785.XA 2021-02-02 2021-02-02 Recommendation processing method and device, intelligent equipment and storage medium Active CN112785391B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110144785.XA CN112785391B (en) 2021-02-02 2021-02-02 Recommendation processing method and device, intelligent equipment and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110144785.XA CN112785391B (en) 2021-02-02 2021-02-02 Recommendation processing method and device, intelligent equipment and storage medium

Publications (2)

Publication Number Publication Date
CN112785391A CN112785391A (en) 2021-05-11
CN112785391B true CN112785391B (en) 2024-02-09

Family

ID=75760536

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110144785.XA Active CN112785391B (en) 2021-02-02 2021-02-02 Recommendation processing method and device, intelligent equipment and storage medium

Country Status (1)

Country Link
CN (1) CN112785391B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113407851B (en) * 2021-07-15 2024-05-03 北京百度网讯科技有限公司 Method, device, equipment and medium for determining recommended information based on double-tower model

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109299396A (en) * 2018-11-28 2019-02-01 东北师范大学 Merge the convolutional neural networks collaborative filtering recommending method and system of attention model
WO2020108483A1 (en) * 2018-11-28 2020-06-04 腾讯科技（深圳）有限公司 Model training method, machine translation method, computer device and storage medium
CN111782968A (en) * 2020-07-02 2020-10-16 北京字节跳动网络技术有限公司 Content recommendation method and device, readable medium and electronic equipment
CN111881342A (en) * 2020-06-23 2020-11-03 北京工业大学 Recommendation method based on graph twin network
WO2020228514A1 (en) * 2019-05-13 2020-11-19 腾讯科技（深圳）有限公司 Content recommendation method and apparatus, and device and storage medium
CN112102015A (en) * 2020-11-17 2020-12-18 腾讯科技（深圳）有限公司 Article recommendation method, meta-network processing method, device, storage medium and equipment
CN112163165A (en) * 2020-10-21 2021-01-01 腾讯科技（深圳）有限公司 Information recommendation method, device, equipment and computer readable storage medium
CN112232546A (en) * 2020-09-09 2021-01-15 北京三快在线科技有限公司 Recommendation probability estimation method and device, electronic equipment and storage medium

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109299396A (en) * 2018-11-28 2019-02-01 东北师范大学 Merge the convolutional neural networks collaborative filtering recommending method and system of attention model
WO2020108483A1 (en) * 2018-11-28 2020-06-04 腾讯科技（深圳）有限公司 Model training method, machine translation method, computer device and storage medium
WO2020228514A1 (en) * 2019-05-13 2020-11-19 腾讯科技（深圳）有限公司 Content recommendation method and apparatus, and device and storage medium
CN111881342A (en) * 2020-06-23 2020-11-03 北京工业大学 Recommendation method based on graph twin network
CN111782968A (en) * 2020-07-02 2020-10-16 北京字节跳动网络技术有限公司 Content recommendation method and device, readable medium and electronic equipment
CN112232546A (en) * 2020-09-09 2021-01-15 北京三快在线科技有限公司 Recommendation probability estimation method and device, electronic equipment and storage medium
CN112163165A (en) * 2020-10-21 2021-01-01 腾讯科技（深圳）有限公司 Information recommendation method, device, equipment and computer readable storage medium
CN112102015A (en) * 2020-11-17 2020-12-18 腾讯科技（深圳）有限公司 Article recommendation method, meta-network processing method, device, storage medium and equipment

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
Xiaoling Xia ; Le Li.''Based_on_Multiple_Attention_and_User_Preferences_for_Recommendation".《IEEE》.2020,全文. *
基于MLP改进型深度神经网络学习资源推荐算法;樊海玮;史双;张博敏;张艳萍;蔺琪;孙欢;;计算机应用研究(09);全文 *
基于全局与局部相融合的方面注意力推荐模型;张天龙;韩立新;;中国科技论文(11);全文 *

Also Published As

Publication number Publication date
CN112785391A (en) 2021-05-11

Similar Documents

Publication Publication Date Title
CN112000819B (en) Multimedia resource recommendation method and device, electronic equipment and storage medium
CN105247507B (en) Method, system and storage medium for the influence power score for determining brand
CN109872242B (en) Information pushing method and device
CN110008397B (en) Recommendation model training method and device
EP4181026A1 (en) Recommendation model training method and apparatus, recommendation method and apparatus, and computer-readable medium
CN110008973B (en) Model training method, method and device for determining target user based on model
CN112308650B (en) Recommendation reason generation method, device, equipment and storage medium
CN109117442B (en) Application recommendation method and device
CN110532469B (en) Information recommendation method, device, equipment and storage medium
CN111159563A (en) Method, device and equipment for determining user interest point information and storage medium
US20220358175A1 (en) Method and system of personalized blending for content recommendation
CN116108267A (en) Recommendation method and related equipment
CN112785391B (en) Recommendation processing method and device, intelligent equipment and storage medium
CN117391824B (en) Method and device for recommending articles based on large language model and search engine
CN112036987B (en) Method and device for determining recommended commodity
CN113836390A (en) Resource recommendation method and device, computer equipment and storage medium
CN115730143A (en) Recommendation system, method, terminal and medium based on task alignment meta learning and augmentation graph
CN111507471A (en) Model training method, device, equipment and storage medium
CN116777529B (en) Object recommendation method, device, equipment, storage medium and program product
CN114048826B (en) Recommendation model training method, device, equipment and medium
WO2024131762A1 (en) Recommendation method and related device
CN117217852B (en) Behavior recognition-based purchase willingness prediction method and device
CN110502698B (en) Information recommendation method, device, equipment and storage medium
CN111861635B (en) Friend recommendation method, device and equipment for commodity sharing
CN117473157A (en) Sequence recommendation method and device based on personalized prompt learning

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
GR01 Patent grant
GR01 Patent grant