WO2023173550A1 - 一种跨领域数据推荐方法、装置、计算机设备及介质 - Google Patents

一种跨领域数据推荐方法、装置、计算机设备及介质 Download PDF

Info

Publication number
WO2023173550A1
WO2023173550A1 PCT/CN2022/090364 CN2022090364W WO2023173550A1 WO 2023173550 A1 WO2023173550 A1 WO 2023173550A1 CN 2022090364 W CN2022090364 W CN 2022090364W WO 2023173550 A1 WO2023173550 A1 WO 2023173550A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
domain
source
target
historical
Prior art date
Application number
PCT/CN2022/090364
Other languages
English (en)
French (fr)
Inventor
侯昶宇
Original Assignee
平安科技（深圳）有限公司
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 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2023173550A1 publication Critical patent/WO2023173550A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/335—Filtering based on additional data, e.g. user or group profiles
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
    • G06F16/367—Ontology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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

Definitions

  • This application relates to the field of artificial intelligence technology, and in particular to a cross-domain data recommendation method, device, computer equipment and media.
  • the recommendation system is used for accurate recommendations, that is, to provide users with accurate recommended content and services.
  • a cross-domain data recommendation method includes: obtaining multiple source domain data and multiple target domain data; inputting multiple source domain data and multiple target domain data into a pre-trained cross-domain data recommendation model to make recommendations based on multiple sources.
  • the source domain data determines the data to be recommended from a variety of target domain data; among them, the pre-trained cross-domain data recommendation model is generated based on knowledge graph and user data training, and the knowledge graph is constructed based on a variety of historical source domain data; Output data to be recommended corresponding to data from multiple source fields, and push the data to be recommended to the corresponding client.
  • the following steps are followed to generate a pre-trained cross-domain recommendation model, including: creating a cross-domain data recommendation model; wherein the cross-domain data recommendation model is a twin network, and the twin network includes a source domain sub-model and a target domain sub-model.
  • model construct the topological structure of the source domain based on various historical source domain data to obtain the knowledge graph; input the knowledge graph and user data into the source domain sub-model, and output source domain embedding vectors corresponding to the historical data of various source domains; based on various Historical target domain data constructs training data with different positive and negative ratios; inputs the training data and user data with different positive and negative ratios into the target domain sub-model, and outputs target domain embedding vectors corresponding to various historical target domain data; according to the source domain
  • the embedding vector and the target domain embedding vector generate a pre-trained cross-domain recommendation model.
  • constructing the topological structure of the source domain based on multiple historical source domain data to obtain the knowledge graph includes: obtaining multiple historical source domain data; analyzing the multiple historical source domain data to determine user preferences in the source domain data; based on the user preference data in the source domain, the user preference relationship and product attribute relationship are determined from a variety of historical source domain data; the topological structure of the source domain is constructed based on the user preference relationship and product attribute relationship, and the graph structure of the source domain is generated; The graph structure of the source domain is determined as a knowledge graph.
  • the source domain sub-model includes a graph transformer graph neural network and a first bridge function; the knowledge graph and user data are input into the source domain sub-model, and source domain embedding vectors corresponding to various source domain historical data are output.
  • constructing training data with different positive and negative ratios based on multiple historical target field data includes: obtaining multiple historical target field data; analyzing multiple historical target field data to determine product data in the target field; Construct training data with different positive and negative ratios based on product data in the target field.
  • the target domain sub-model includes a JK-Net network and a second bridging function; the training data and user data with different positive and negative proportions are input into the target domain sub-model, and output corresponding to various historical target domain data
  • the target domain embedding vector includes: obtaining the embedding vector of each product in the training data with different positive and negative ratios; inputting user data into the JK-Net network and outputting the user's own embedding vector in the target domain; embedding each product The vector is spliced with the user's own embedding vector input bridge function in the target domain, and target domain embedding vectors corresponding to various historical target domain data are output.
  • generating a pre-trained cross-domain recommendation model based on the source domain embedding vector and the target domain embedding vector includes: performing similarity calculations based on the source domain embedding vector and the target domain embedding vector to generate a similarity score; The degree score is determined as the model loss value; when the model loss value reaches the preset threshold, a pre-trained cross-domain recommendation model is generated.
  • a cross-domain data recommendation device includes: a data acquisition module, used to obtain multiple source domain data and multiple target domain data; a data input module, used to input multiple source domain data and multiple target domain data into advance
  • data to be recommended is determined from multiple target domain data based on multiple source domain data; among them, the pre-trained cross-domain data recommendation model is generated based on knowledge graph and user data training.
  • the graph is constructed based on multiple historical source domain data; the data push module is used to output data to be recommended corresponding to multiple source domain data, and push the to-be-recommended data to the corresponding client.
  • a computer device which includes a memory and a processor.
  • Computer-readable instructions are stored in the memory. When the computer-readable instructions are executed by the processor, they cause the processor to perform the following steps:
  • the pre-trained cross-domain data recommendation model is generated based on knowledge graph and user data training, and the knowledge graph is constructed based on a variety of historical source domain data;
  • a medium storing computer-readable instructions, wherein when executed by one or more processors, the computer-readable instructions cause one or more processors to perform the following steps:
  • the pre-trained cross-domain data recommendation model is generated based on knowledge graph and user data training, and the knowledge graph is constructed based on a variety of historical source domain data;
  • the cross-domain data recommendation device obtains multiple source domain data and multiple target domain data; inputs multiple source domain data and multiple target domain data into pre-trained cross-domain data
  • the data to be recommended is determined from a variety of target domain data based on a variety of source domain data; among them, the pre-trained cross-domain data recommendation model is generated based on knowledge graph and user data training, and the knowledge graph is based on a variety of Constructed from historical source domain data; output data to be recommended corresponding to multiple source domain data, and push the data to be recommended to the corresponding client.
  • this application uses a knowledge graph to represent the topological structure of the relationship between users and products in different fields, and at the same time combines the impact of user data on the results, a more accurate source field embedding vector can be obtained, making the model more accurate after training. , improving the accuracy of data recommendation.
  • Figure 1 is an implementation environment diagram of the cross-domain data recommendation method provided in one embodiment of the present application.
  • Figure 2 is a schematic diagram of the internal structure of a computer device in one embodiment of the present application.
  • Figure 3 is a schematic diagram of a cross-domain data recommendation method provided in one embodiment of the present application.
  • Figure 4 is a schematic diagram of a cross-domain data recommendation model training method provided in one embodiment of the present application.
  • Figure 5 is a schematic process diagram of the cross-domain data recommendation model training process provided in one embodiment of the present application.
  • Figure 6 is a schematic diagram of a cross-domain data recommendation device provided by an embodiment of the present application.
  • Figure 1 is an implementation environment diagram of the cross-domain data recommendation method provided in one embodiment. As shown in Figure 1, the implementation environment includes a server 110 and a client 120.
  • the server 110 may be a server, which may be an independent server, or may provide cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, and security services. , Content Delivery Network (CDN), and cloud servers for basic cloud computing services such as big data and artificial intelligence platforms, such as server devices that store pre-trained cross-domain data recommendation models.
  • CDN Content Delivery Network
  • the server 110 obtains multiple source domain data and multiple target domain data sent from the client 120, and the server 110 inputs the multiple source domain data and multiple target domain data into pre-training
  • data to be recommended is determined from multiple target domain data based on multiple source domain data.
  • the server 110 outputs the to-be-recommended data corresponding to the multiple source domain data, and pushes the to-be-recommended data to the corresponding client120.
  • the client 120 can be a smartphone, a tablet, a laptop, a desktop computer, etc., but is not limited thereto.
  • the server 110 and the client 120 can be connected through Bluetooth, USB (Universal Serial Bus, Universal Serial Bus) or other communication connection methods, which are not limited in this application.
  • Figure 2 is a schematic diagram of the internal structure of a computer device in one embodiment.
  • the computer device includes a processor, media, memory, and network interfaces connected through a system bus.
  • the medium of the computer device stores an operating system, a database and computer-readable instructions.
  • the database can store a sequence of control information.
  • the processor can implement a cross-domain data recommendation. method.
  • the processor of the computer device is used to provide computing and control capabilities to support the operation of the entire device.
  • Computer readable instructions may be stored in the memory of the computer device. When executed by the processor, the computer readable instructions may cause the processor to perform a cross-domain data recommendation method.
  • the network interface of the computer device is used for communication with the terminal connection.
  • FIG. 2 is only a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer equipment to which the solution of the present application is applied.
  • Specific computer equipment can May include more or fewer parts than shown, or combine certain parts, or have a different arrangement of parts.
  • the medium is a readable storage medium.
  • the cross-domain data recommendation method provided by the embodiment of the present application will be introduced in detail below with reference to Figure 3.
  • This method can be implemented by relying on a computer program and can run on a cross-domain data recommendation device based on the von Neumann system.
  • the computer program can be integrated into an application or run as a stand-alone utility application.
  • AI Artificial Intelligence
  • digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.
  • Basic artificial intelligence technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation/interaction systems, mechatronics and other technologies.
  • Artificial intelligence software technology mainly includes computer vision technology, robotics technology, biometric technology, speech processing technology, natural language processing technology, and machine learning/deep learning.
  • Figure 3 provides a schematic flow chart of a cross-domain data recommendation method for an embodiment of the present application, which is applied to the server.
  • the method according to the embodiment of the present application may include the following steps:
  • multiple source domain data and multiple target domain data are first obtained.
  • the pre-trained cross-domain data recommendation model is generated based on knowledge graph and user data training.
  • the knowledge graph is constructed based on multiple historical source domain data; the pre-trained cross-domain data recommendation model is based on multiple source domain data.
  • a variety of target field data determines the mathematical model for recommending the data to be recommended.
  • a cross-domain data recommendation model is first created; wherein the cross-domain data recommendation model is a twin network, and the twin network includes a source domain sub-model and a target domain sub-model.
  • the server can input the multiple source domain data and multiple target domain data into the pre-trained cross-domain data.
  • the data is processed in the recommendation model.
  • S103 Output data to be recommended corresponding to data in multiple source fields, and push the data to be recommended to the corresponding client.
  • step S102 after processing according to the pre-trained cross-domain data recommendation model in step S102, data to be recommended corresponding to multiple source domain data can be output, and finally the data to be recommended can be pushed to the corresponding client. Make a presentation.
  • the cross-domain data recommendation device acquires multiple source domain data and multiple target domain data; the multiple source domain data and the multiple target domain data are input into the pre-trained cross-domain data recommendation model to predict the data based on the cross-domain data recommendation model.
  • Multiple source domain data determines the data to be recommended from multiple target domain data; among them, the pre-trained cross-domain data recommendation model is generated based on knowledge graph and user data training, and the knowledge graph is constructed based on multiple historical source domain data ; Output data to be recommended corresponding to data from multiple source fields, and push the data to be recommended to the corresponding client.
  • this application uses a knowledge graph to represent the topological structure of the relationship between users and products in different fields, and at the same time combines the impact of user data on the results, a more accurate source field embedding vector can be obtained, making the model more accurate after training. , improving the accuracy of data recommendation.
  • Figure 4 provides a schematic flow chart for generating a pre-trained cross-domain recommendation model according to an embodiment of the present application.
  • the method according to the embodiment of the present application may include the following steps:
  • the cross-domain data recommendation model is a twin network, which includes a source domain sub-model and a target domain sub-model;
  • the model adopts the structure of twin network, which can map inputs from different fields into a new and identical vector space.
  • twin network In engineering practice, the embedding content required each time is greatly reduced, which effectively improves the model efficiency and achieves greater success. Online use of scale data.
  • it is more convenient to recommend cold starts for content in multiple fields at the same time.
  • multiple historical source domain data are first obtained, and then multiple historical source domain data are analyzed to determine user preference data in the source domain, and then based on the user preference data in the source domain, multiple historical source domain data are obtained.
  • the user preference relationship and product attribute relationship are determined from the historical source domain data.
  • the topological structure of the source domain is constructed based on the user preference relationship and product attribute relationship, and the graph structure of the source domain is generated.
  • the graph structure of the source domain is determined as a knowledge graph. .
  • the source domain sub-model includes graph transformer graph neural network and first bridge function.
  • the knowledge graph is first input into the graph transformer graph neural network, and the embedding vector of user preferences in the source field is output. Then the user data is input into the graph transformer graph neural network, and the embedding vector of the user's own data is output. Finally, the embedding vector of the user's own data is output.
  • the embedding vectors of user preferences in the source domain and the embedding vectors of the user's own data are input into the bridge function for vector splicing, and source domain embedding vectors corresponding to various source domain historical data are output.
  • multiple historical target field data are first obtained, then multiple historical target field data are analyzed to determine product data in the target field, and finally training data with different positive and negative ratios are constructed based on the product data in the target field.
  • S205 input the training data and user data with different positive and negative ratios into the target domain sub-model, and output target domain embedding vectors corresponding to various historical target domain data;
  • the target domain sub-model includes the JK-Net network and the second bridge function.
  • the embedding vector of each product in the training data with different positive and negative ratios is first obtained, then the user data is input into the JK-Net network, the user's own embedding vector in the target field is output, and finally each product is The product's embedding vector is spliced with the user's own embedding vector in the target domain input bridge function, and target domain embedding vectors corresponding to various historical target domain data are output.
  • JK-Net network is a deep GNN architecture network.
  • JK-Net aggregates different fields through adaptive learning of nodes at different positions, thereby improving the representation of nodes.
  • S206 Generate a pre-trained cross-domain recommendation model based on the source domain embedding vector and the target domain embedding vector.
  • Figure 5 is a schematic block diagram of the cross-domain recommendation model training process.
  • user preferences are determined based on multiple historical source domain data, and a graph structure of the origin domain is constructed based on the relationship between user preferences and product attributes.
  • a graph structure of the origin domain is constructed based on the relationship between user preferences and product attributes.
  • use the graph transformer graph neural network structure to obtain the graph embedding representation of the subgraph composed of the nodes that the user prefers in the source domain, and use this as the embedding representation of the user's preferences in the source domain; according to the user's For preferences in the source domain, the graph network is also used to obtain an embedded representation of the user's own data.
  • the final embedding representation of the source domain is obtained.
  • the embedding representation obtained in this way is equivalent to a personalized preference migration function for different users. Instead of migrating the preferences of all users through a unified function, user information is more effectively utilized.
  • the JK-Net method can be used to solve the problem that the number of layers of the traditional graph network cannot be too deep, and obtain the embedded representation of each product in the target field.
  • the graph network can be used to obtain the user's position in the target field. After obtaining its own embedded representation, it then obtains the user’s final embedded representation in the target domain through a bridge network. Finally, through the embedding representation of the source domain and the embedding representation of the target domain, the final score is obtained to determine whether the user will be interested in the content of the target domain.
  • the model as a whole adopts a twin network structure.
  • the target domain and source domain use similar network structures respectively.
  • the same loss function is used for optimization during the training process.
  • Parameters are shared in the functions that bridge user data and domain product data.
  • the model targets multiple target fields, it can adopt a structure similar to triple network.
  • it can construct training data with different positive and negative ratios for different target fields, so that the model can calculate recommendation results in multiple target fields at the same time. .
  • this application effectively utilizes the topological structure formed by the relationship between users and products in different fields, and at the same time combines the impact of the user's own information on the results to achieve a more accurate embedding representation of the source domain and target domain, thus Improved the recommendation accuracy of the model.
  • the cross-domain data recommendation device acquires multiple source domain data and multiple target domain data; the multiple source domain data and the multiple target domain data are input into the pre-trained cross-domain data recommendation model to predict the data based on the cross-domain data recommendation model.
  • Multiple source domain data determines the data to be recommended from multiple target domain data; among them, the pre-trained cross-domain data recommendation model is generated based on knowledge graph and user data training, and the knowledge graph is constructed based on multiple historical source domain data ; Output data to be recommended corresponding to data from multiple source fields, and push the data to be recommended to the corresponding client.
  • this application uses a knowledge graph to represent the topological structure of the relationship between users and products in different fields, and at the same time combines the impact of user data on the results, a more accurate source field embedding vector can be obtained, making the model more accurate after training. , improving the accuracy of data recommendation.
  • FIG. 6 shows a schematic structural diagram of a cross-domain data recommendation device provided by an exemplary embodiment of the present application, applied to a server.
  • the cross-domain data recommendation device can be implemented as all or part of the device through software, hardware, or a combination of both.
  • the device 1 includes a data acquisition module 10 , a data input module 20 , and a data push module 30 .
  • the data acquisition module 10 is used to acquire multiple source domain data and multiple target domain data;
  • the data input module 20 is used to input multiple source domain data and multiple target domain data into the pre-trained cross-domain data recommendation model, so as to determine the data to be recommended from the multiple target domain data based on the multiple source domain data; wherein ,
  • the pre-trained cross-domain data recommendation model is generated based on knowledge graph and user data training.
  • the knowledge graph is constructed based on multiple historical source domain data;
  • the data push module 30 is used to output data to be recommended corresponding to multiple source domain data, and push the data to be recommended to the corresponding client.
  • the high-voltage rear identification device provided in the above embodiments performs the high-voltage rear identification method
  • only the division of the above-mentioned functional modules is used as an example.
  • the above-mentioned functions can be allocated from different modules as needed.
  • the functional modules are completed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above.
  • the high-pressure rear identification device and the high-voltage rear identification method embodiment provided in the above embodiments belong to the same concept. Please refer to the method embodiment for details of the implementation process, which will not be described again here.
  • the cross-domain data recommendation device acquires multiple source domain data and multiple target domain data; the multiple source domain data and the multiple target domain data are input into the pre-trained cross-domain data recommendation model to predict the data based on the cross-domain data recommendation model.
  • Multiple source domain data determines the data to be recommended from multiple target domain data; among them, the pre-trained cross-domain data recommendation model is generated based on knowledge graph and user data training, and the knowledge graph is constructed based on multiple historical source domain data ; Output data to be recommended corresponding to data from multiple source fields, and push the data to be recommended to the corresponding client.
  • this application uses a knowledge graph to represent the topological structure of the relationship between users and products in different fields, and at the same time combines the impact of user data on the results, a more accurate source field embedding vector can be obtained, making the model more accurate after training. , improving the accuracy of data recommendation.
  • a computer device in one embodiment, includes a memory, a processor, and a computer program stored in the memory and executable on the processor.
  • the processor executes the computer program, the following steps are implemented: Obtaining multiple source fields data and multiple target domain data; input multiple source domain data and multiple target domain data into the pre-trained cross-domain data recommendation model to determine the data to be recommended from multiple target domain data based on multiple source domain data; Among them, the pre-trained cross-domain data recommendation model is generated based on the knowledge graph and user data training.
  • the knowledge graph is constructed based on a variety of historical source domain data; the data to be recommended corresponding to the multiple source domain data is output, and the data to be recommended is output The data is pushed to the corresponding client.
  • the processor when it generates a pre-trained cross-domain recommendation model, it specifically performs the following operations: creates a cross-domain data recommendation model; wherein the cross-domain data recommendation model is a twin network, and the twin network includes a source domain sub-model and Target domain sub-model; construct the topological structure of the source domain based on multiple historical source domain data to obtain the knowledge graph; input the knowledge graph and user data into the source domain sub-model, and output source domain embedding vectors corresponding to various source domain historical data; Construct training data with different positive and negative ratios based on various historical target field data; input the training data and user data with different positive and negative ratios into the target field sub-model, and output the target field embedding vectors corresponding to various historical target field data; Generate a pre-trained cross-domain recommendation model based on the source domain embedding vector and the target domain embedding vector.
  • the processor constructs the topological structure of the source domain based on multiple historical source domain data, and when obtaining the knowledge graph, specifically performs the following operations: obtains multiple historical source domain data; analyzes the multiple historical source domain data, and determines User preference data in the source field; based on the user preference data in the source field, the user preference relationship and product attribute relationship are determined from a variety of historical source field data; the topology structure of the source field is constructed based on the user preference relationship and product attribute relationship, and the source field is generated.
  • the graph structure of the domain determine the graph structure of the source domain as a knowledge graph.
  • the processor when the processor inputs the knowledge graph and user data into the source domain sub-model and outputs source domain embedding vectors corresponding to various source domain historical data, it specifically performs the following operations: input the knowledge graph into the graph transformer graph neural In the network, the embedding vector of user preferences in the source domain is output; the user data is input into the graph transformer graph neural network, and the embedding vector of the user's own data is output; the embedding vector of user preferences in the source domain and the embedding vector of the user's own data are input into the bridge function Perform vector splicing and output source domain embedding vectors corresponding to historical data in multiple source domains.
  • the processor when the processor constructs training data with different positive and negative ratios based on multiple historical target domain data, the processor specifically performs the following operations: obtains multiple historical target domain data; analyzes multiple historical target domain data, and determines Product data in the target field; construct training data with different positive and negative proportions based on the product data in the target field.
  • the processor performs the following operations when inputting training data and user data with different positive and negative proportions into the target domain sub-model and outputting target domain embedding vectors corresponding to multiple historical target domain data: obtaining different The embedding vector of each product in the training data in positive and negative proportions; input the user data into the JK-Net network and output the user's own embedding vector in the target field; compare the embedding vector of each product with the user's own embedding in the target field The vectors are input into the bridge function for splicing, and target domain embedding vectors corresponding to various historical target domain data are output.
  • the processor when the processor generates a pre-trained cross-domain recommendation model based on the source domain embedding vector and the target domain embedding vector, the processor specifically performs the following operations: performs similarity calculation based on the source domain embedding vector and the target domain embedding vector, and generates Similarity score; determine the similarity score as the model loss value; when the model loss value reaches the preset threshold, a pre-trained cross-domain recommendation model is generated.
  • the cross-domain data recommendation device acquires multiple source domain data and multiple target domain data; the multiple source domain data and the multiple target domain data are input into the pre-trained cross-domain data recommendation model to predict the data based on the cross-domain data recommendation model.
  • Multiple source domain data determines the data to be recommended from multiple target domain data; among them, the pre-trained cross-domain data recommendation model is generated based on knowledge graph and user data training, and the knowledge graph is constructed based on multiple historical source domain data ; Output data to be recommended corresponding to data from multiple source fields, and push the data to be recommended to the corresponding client.
  • this application uses a knowledge graph to represent the topological structure of the relationship between users and products in different fields, and at the same time combines the impact of user data on the results, a more accurate source field embedding vector can be obtained, making the model more accurate after training. , improving the accuracy of data recommendation.
  • a medium storing computer-readable instructions When executed by one or more processors, the computer-readable instructions cause the one or more processors to perform the following steps: obtain multiple source fields. data and multiple target domain data; input multiple source domain data and multiple target domain data into the pre-trained cross-domain data recommendation model to determine the data to be recommended from multiple target domain data based on multiple source domain data; Among them, the pre-trained cross-domain data recommendation model is generated based on the knowledge graph and user data training. The knowledge graph is constructed based on a variety of historical source domain data; the data to be recommended corresponding to the multiple source domain data is output, and the data to be recommended is output The data is pushed to the corresponding client.
  • the medium storing computer-readable instructions may be non-volatile or volatile.
  • the processor when it generates a pre-trained cross-domain recommendation model, it specifically performs the following operations: creates a cross-domain data recommendation model; wherein the cross-domain data recommendation model is a twin network, and the twin network includes a source domain sub-model and Target domain sub-model; construct the topological structure of the source domain based on multiple historical source domain data to obtain the knowledge graph; input the knowledge graph and user data into the source domain sub-model, and output source domain embedding vectors corresponding to various source domain historical data; Construct training data with different positive and negative ratios based on various historical target field data; input the training data and user data with different positive and negative ratios into the target field sub-model, and output the target field embedding vectors corresponding to various historical target field data; Generate a pre-trained cross-domain recommendation model based on the source domain embedding vector and the target domain embedding vector.
  • the processor constructs the topological structure of the source domain based on multiple historical source domain data, and when obtaining the knowledge graph, specifically performs the following operations: obtains multiple historical source domain data; analyzes the multiple historical source domain data, and determines User preference data in the source field; based on the user preference data in the source field, the user preference relationship and product attribute relationship are determined from a variety of historical source field data; the topology structure of the source field is constructed based on the user preference relationship and product attribute relationship, and the source field is generated.
  • the graph structure of the domain determine the graph structure of the source domain as a knowledge graph.
  • the processor when the processor inputs the knowledge graph and user data into the source domain sub-model and outputs source domain embedding vectors corresponding to various source domain historical data, it specifically performs the following operations: input the knowledge graph into the graph transformer graph neural In the network, the embedding vector of user preferences in the source domain is output; the user data is input into the graph transformer graph neural network, and the embedding vector of the user's own data is output; the embedding vector of user preferences in the source domain and the embedding vector of the user's own data are input into the bridge function Perform vector splicing and output source domain embedding vectors corresponding to historical data in multiple source domains.
  • the processor when the processor constructs training data with different positive and negative ratios based on multiple historical target domain data, the processor specifically performs the following operations: obtains multiple historical target domain data; analyzes multiple historical target domain data, and determines Product data in the target field; construct training data with different positive and negative proportions based on the product data in the target field.
  • the processor performs the following operations when inputting training data and user data with different positive and negative proportions into the target domain sub-model and outputting target domain embedding vectors corresponding to multiple historical target domain data: obtaining different The embedding vector of each product in the training data in positive and negative proportions; input the user data into the JK-Net network and output the user's own embedding vector in the target field; compare the embedding vector of each product with the user's own embedding in the target field The vectors are input into the bridge function for splicing, and target domain embedding vectors corresponding to various historical target domain data are output.
  • the processor when the processor generates a pre-trained cross-domain recommendation model based on the source domain embedding vector and the target domain embedding vector, the processor specifically performs the following operations: performs similarity calculation based on the source domain embedding vector and the target domain embedding vector, and generates Similarity score; determine the similarity score as the model loss value; when the model loss value reaches the preset threshold, a pre-trained cross-domain recommendation model is generated.
  • the cross-domain data recommendation device acquires multiple source domain data and multiple target domain data; the multiple source domain data and the multiple target domain data are input into the pre-trained cross-domain data recommendation model to predict the data based on the cross-domain data recommendation model.
  • Multiple source domain data determines the data to be recommended from multiple target domain data; among them, the pre-trained cross-domain data recommendation model is generated based on knowledge graph and user data training, and the knowledge graph is constructed based on multiple historical source domain data ; Output data to be recommended corresponding to data from multiple source fields, and push the data to be recommended to the corresponding client.
  • this application uses a knowledge graph to represent the topological structure of the relationship between users and products in different fields, and at the same time combines the impact of user data on the results, a more accurate source field embedding vector can be obtained, making the model more accurate after training. , improving the accuracy of data recommendation.
  • the computer program can be stored in a computer-readable medium. When the program is executed When doing so, it may include the processes of the above method embodiments.
  • the aforementioned media can be non-volatile media such as magnetic disks, optical disks, read-only memory (Read-Only Memory, ROM), or random access memory (Random Access Memory, RAM), etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Databases & Information Systems (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Business, Economics & Management (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • Animal Behavior & Ethology (AREA)
  • General Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Evolutionary Computation (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • General Business, Economics & Management (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种跨领域数据推荐方法、装置、计算机设备及介质，方法包括：获取多种源领域数据和多种目标领域数据（S101）；将所述多种源领域数据和所述多种目标领域数据输入预先训练的跨领域数据推荐模型中，以根据所述多种源领域数据从所述多种目标领域数据确定出待推荐数据；其中，所述预先训练的跨领域数据推荐模型是基于知识图谱和用户数据训练生成的，所述知识图谱是根据多种历史源领域数据构建的（S102）；输出所述多种源领域数据对应的待推荐数据，并将所述待推荐数据推送至相应客户端（S103）。通过知识图谱来表征用户和不同领域产品之间关系所构成的拓扑结构，同时结合了用户数据对结果的影响，从而得到更加精准的源领域嵌入向量，使得模型训练后的精度更高，提升了数据推荐的准确性。

Description

一种跨领域数据推荐方法、装置、计算机设备及介质
优先权申明
本申请要求于2022年3月14日提交中国专利局、申请号为202210248145.8，发明名称为“一种跨领域数据推荐方法、装置、计算机设备及介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及人工智能技术领域，特别涉及一种跨领域数据推荐方法、装置、计算机设备及介质。
背景技术
近年来，互联网技术的日益发展和普及给用户带来了大量的信息，满足了用户对信息的需求。但随着信息呈指数级增长，使得用户难以从海量的数据中筛选出自己真正想要的信息。在这种情况下，推荐系统应运而生，推荐系统用于精准推荐，即向用户提供精准的推荐内容和服务。
在推荐系统领域，数据稀疏性和冷启动是仍然存在且颇具挑战性的问题，研究者们提出了很多种解决方案。近年来，出现了一种新的研究趋势，即跨领域推荐，旨在缓解数据稀疏性和冷启动对推荐系统性能的影响。现实中，在不同推荐领域都会遇到这些问题。例如，某在线购物网站拥有不止一个商品领域，如书籍、美妆、电子产品、影片等。同一个用户的在不同领域的喜好或许是相似的，因而将用户在某个领域的偏好特征迁移到目标域来提高目标域的推荐性能是一个不错的选择。
传统的冷启动推荐有两种方法，一是通过设计决策策略解决冷启动问题；二是利用辅助信息帮助冷启动(如用户属性、项目属性等)。但由于个体差异，不同领域的偏好是复杂的，发明人意识到现有的方法都不能很好的使用各类信息，用公用的偏好桥也不能准确的抓获复杂多样的关系，从而使得模型训练后的精度较低，降低了数据推荐的准确性。
发明内容
基于此，有必要针对数据推荐的准确性低的问题，提供一种跨领域数据推荐方法、装置、计算机设备及介质。
一种跨领域数据推荐方法，方法包括：获取多种源领域数据和多种目标领域数据；将多种源领域数据和多种目标领域数据输入预先训练的跨领域数据推荐模型中，以根据多种源领域数据从多种目标领域数据确定出待推荐数据；其中，预先训练的跨领域数据推荐模型是基于知识图谱和用户数据训练生成的，知识图谱是根据多种历史源领域数据构建的；输出多种源领域数据对应的待推荐数据，并将待推荐数据推送至相应客户端。
在其中一个实施例中，按照以下步骤生成预先训练的跨领域推荐模型，包括：创建跨领域数据推荐模型；其中，跨领域数据推荐模型为孪生网络，孪生网络包括源领域子模型和目标领域子模型；根据多种历史源领域数据构建源领 域的拓扑结构，得到知识图谱；将知识图谱和用户数据输入源领域子模型中，输出多种源领域历史数据对应的源领域嵌入向量；根据多种历史目标领域数据构建不同正负比例的训练的数据；将不同正负比例的训练的数据与用户数据输入目标领域子模型中，输出多种历史目标领域数据对应的目标领域嵌入向量；根据源领域嵌入向量与目标领域嵌入向量生成预先训练的跨领域推荐模型。
在其中一个实施例中，根据多种历史源领域数据构建源领域的拓扑结构，得到知识图谱，包括：获取多种历史源领域数据；分析多种历史源领域数据，确定出源领域的用户喜好数据；基于源领域的用户喜好数据从多种历史源领域数据中确定出用户喜好关系和产品属性关系；根据用户喜好关系和产品属性关系构建源领域的拓扑结构，生成源领域的图结构；将源领域的图结构确定为知识图谱。
在其中一个实施例中，源领域子模型包括graph transformer图神经网络和第一桥接函数；将知识图谱和用户数据输入源领域子模型中，输出多种源领域历史数据对应的源领域嵌入向量，包括：将知识图谱输入graph transformer图神经网络中，输出源领域用户喜好的嵌入向量；将用户数据输入graph transformer图神经网络中，输出用户自身数据的嵌入向量；将源领域用户喜好的嵌入向量与用户自身数据的嵌入向量输入桥接函数中进行向量拼接，输出多种源领域历史数据对应的源领域嵌入向量。
在其中一个实施例中，根据多种历史目标领域数据构建不同正负比例的训练的数据，包括：获取多种历史目标领域数据；分析多种历史目标领域数据，确定出目标领域的产品数据；根据目标领域的产品数据构建不同正负比例的训练的数据。
在其中一个实施例中，目标领域子模型包括JK-Net网络和第二桥接函数；将不同正负比例的训练的数据与用户数据输入目标领域子模型中，输出多种历史目标领域数据对应的目标领域嵌入向量，包括：获取不同正负比例的训练的数据中每个产品的嵌入向量；将用户数据输入JK-Net网络中，输出用户在目标领域自身的嵌入向量；将每个产品的嵌入向量与用户在目标领域自身的嵌入向量输入桥接函数中进行拼接，输出多种历史目标领域数据对应的目标领域嵌入向量。
在其中一个实施例中，根据源领域嵌入向量与目标领域嵌入向量生成预先训练的跨领域推荐模型，包括：根据源领域嵌入向量与目标领域嵌入向量进行相似度计算，生成相似度分数；将相似度分数确定为模型损失值；当模型损失值到达预设阈值时，生成预先训练的跨领域推荐模型。
一种跨领域数据推荐装置，装置包括：数据获取模块，用于获取多种源领域数据和多种目标领域数据；数据输入模块，用于将多种源领域数据和多种目标领域数据输入预先训练的跨领域数据推荐模型中，以根据多种源领域数据从多种目标领域数据确定出待推荐数据；其中，预先训练的跨领域数据推荐模型是基于知识图谱和用户数据训练生成的，知识图谱是根据多种历史源领域数据构建的；数据推送模块，用于输出多种源领域数据对应的待推荐数据，并将待推荐数据推送至相应客户端。
一种计算机设备，其中，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述计算机可读指令被所述处理器执行时，使得所述处理器执行如下步骤的指令：
获取多种源领域数据和多种目标领域数据；
将所述多种源领域数据和所述多种目标领域数据输入预先训练的跨领域数据推荐模型中，以根据所述多种源领域数据从所述多种目标领域数据确定出待推荐数据；其中，所述预先训练的跨领域数据推荐模型是基于知识图谱和用户数据训练生成的，所述知识图谱是根据多种历史源领域数据构建的；
输出所述多种源领域数据对应的待推荐数据，并将所述待推荐数据推送至相应客户端。
一种存储有计算机可读指令的介质，其中，所述计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器执行如下步骤的指令：
获取多种源领域数据和多种目标领域数据；
将所述多种源领域数据和所述多种目标领域数据输入预先训练的跨领域数据推荐模型中，以根据所述多种源领域数据从所述多种目标领域数据确定出待推荐数据；其中，所述预先训练的跨领域数据推荐模型是基于知识图谱和用户数据训练生成的，所述知识图谱是根据多种历史源领域数据构建的；
输出所述多种源领域数据对应的待推荐数据，并将所述待推荐数据推送至相应客户端。
上述跨领域数据推荐方法、装置、设备和介质，跨领域数据推荐装置获取多种源领域数据和多种目标领域数据；将多种源领域数据和多种目标领域数据输入预先训练的跨领域数据推荐模型中，以根据多种源领域数据从多种目标领域数据确定出待推荐数据；其中，预先训练的跨领域数据推荐模型是基于知识图谱和用户数据训练生成的，知识图谱是根据多种历史源领域数据构建的；输出多种源领域数据对应的待推荐数据，并将待推荐数据推送至相应客户端。由于本申请通过知识图谱来表征用户和不同领域产品之间关系所构成的拓扑结构，同时结合了用户数据对结果的影响，从而得到更加精准的源领域嵌入向量，使得模型训练后的精度更高，提升了数据推荐的准确性。
附图说明
此处的附图被并入说明书中并构成本说明书的一部分，示出了符合本申请的实施例，并与说明书一起用于解释本申请的原理。
图1为本申请一个实施例中提供的跨领域数据推荐方法的实施环境图；
图2为本申请一个实施例中计算机设备的内部结构示意图；
图3为本申请一个实施例中提供的跨领域数据推荐方法的方法示意图；
图4为本申请一个实施例中提供的跨领域数据推荐模型训练方法的方法示意图；
图5为本申请一个实施例中提供的跨领域数据推荐模型训练过程的过程示意图；
图6是本申请实施例提供的一种跨领域数据推荐装置的装置示意图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处所描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
可以理解，本申请所使用的术语“第一”、“第二”等可在本文中用于描 述各种元件，但这些元件不受这些术语限制。这些术语仅用于将第一个元件与另一个元件区分。
图1为一个实施例中提供的跨领域数据推荐方法的实施环境图，如图1所示，在该实施环境中，包括服务端110以及客户端120。
服务端110可以为服务器，该服务器具体可以是独立的服务器，也可以是提供云服务、云数据库、云计算、云函数、云存储、网络服务、云通信、中间件服务、域名服务、安全服务、内容分发网络(Content Delivery Network，CDN)、以及大数据和人工智能平台等基础云计算服务的云服务器，例如为保存预先训练的跨领域数据推荐模型的服务器设备。当需要进行跨领域数据推荐时，服务端110获取来自客户端120的发送的多种源领域数据和多种目标领域数据，服务端110将多种源领域数据和多种目标领域数据输入预先训练的跨领域数据推荐模型中，以根据多种源领域数据从多种目标领域数据确定出待推荐数据，服务端110输出多种源领域数据对应的待推荐数据，并将待推荐数据推送至相应客户端120。
需要说明的是，客户端120可为智能手机、平板电脑、笔记本电脑、台式计算机等，但并不局限于此。服务端110以及客户端120可以通过蓝牙、USB(Universal Serial Bus，通用串行总线)或者其他通讯连接方式进行连接，本申请在此不做限制。
图2为一个实施例中计算机设备的内部结构示意图。如图2所示，该计算机设备包括通过系统总线连接的处理器、介质、存储器和网络接口。其中，该计算机设备的介质存储有操作系统、数据库和计算机可读指令，数据库中可存储有控件信息序列，该计算机可读指令被处理器执行时，可使得处理器实现一种跨领域数据推荐方法。该计算机设备的处理器用于提供计算和控制能力，支撑整个设备的运行。该计算机设备的存储器中可存储有计算机可读指令，该计算机可读指令被处理器执行时，可使得处理器执行一种跨领域数据推荐方法。该计算机设备的网络接口用于与终端连接通信。本领域技术人员可以理解，图2中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定，具体的计算机设备可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。其中，介质为一种可读的存储介质。
下面将结合附图3，对本申请实施例提供的跨领域数据推荐方法进行详细介绍。该方法可依赖于计算机程序实现，可运行于基于冯诺依曼体系的跨领域数据推荐装置上。该计算机程序可集成在应用中，也可作为独立的工具类应用运行。
本申请实施例可以基于人工智能技术对相关的数据进行获取和处理。其中，人工智能(Artificial Intelligence，AI)是利用数字计算机或者数字计算机控制的机器模拟、延伸和扩展人的智能，感知环境、获取知识并使用知识获得最佳结果的理论、方法、技术及应用系统。
人工智能基础技术一般包括如传感器、专用人工智能芯片、云计算、分布式存储、大数据处理技术、操作/交互系统、机电一体化等技术。人工智能软件技术主要包括计算机视觉技术、机器人技术、生物识别技术、语音处理技术、自然语言处理技术以及机器学习/深度学习等几大方向。
请参见图3，为本申请实施例提供了一种跨领域数据推荐方法的流程示意图，应用于服务端。如图3所示，本申请实施例的方法可以包括以下步骤：
S101，获取多种源领域数据和多种目标领域数据；
在一种可能的实现方式中，在进行跨领域数据推荐时，首先获取多种源领域数据和多种目标领域数据。
S102，将多种源领域数据和多种目标领域数据输入预先训练的跨领域数据推荐模型中，以根据多种源领域数据从多种目标领域数据确定出待推荐数据；
其中，预先训练的跨领域数据推荐模型是基于知识图谱和用户数据训练生成的，知识图谱是根据多种历史源领域数据构建的；预先训练的跨领域数据推荐模型是根据多种源领域数据从多种目标领域数据确定出待推荐数据进行推荐的数学模型。
在本申请实施例中，在生成预先训练的跨领域数据推荐模型时，首先创建跨领域数据推荐模型；其中，跨领域数据推荐模型为孪生网络，孪生网络包括源领域子模型和目标领域子模型，然后根据多种历史源领域数据构建源领域的拓扑结构，得到知识图谱，再将知识图谱和用户数据输入源领域子模型中，输出多种源领域历史数据对应的源领域嵌入向量，其次根据多种历史目标领域数据构建不同正负比例的训练的数据，再将不同正负比例的训练的数据与用户数据输入目标领域子模型中，输出多种历史目标领域数据对应的目标领域嵌入向量，最后根据源领域嵌入向量与目标领域嵌入向量生成预先训练的跨领域推荐模型。
在一种可能的实现方式中，服务端在根据步骤S101获取到多种源领域数据和多种目标领域数据后，可将多种源领域数据和多种目标领域数据输入到预先训练的跨领域数据推荐模型中进行处理。
S103，输出多种源领域数据对应的待推荐数据，并将待推荐数据推送至相应客户端。
在一种可能的实现方式中，在根据步骤S102中的预先训练的跨领域数据推荐模型进行处理后，可输出多种源领域数据对应的待推荐数据，最后将待推荐数据推送至相应客户端进行展示。
在本申请实施例中，跨领域数据推荐装置获取多种源领域数据和多种目标领域数据；将多种源领域数据和多种目标领域数据输入预先训练的跨领域数据推荐模型中，以根据多种源领域数据从多种目标领域数据确定出待推荐数据；其中，预先训练的跨领域数据推荐模型是基于知识图谱和用户数据训练生成的，知识图谱是根据多种历史源领域数据构建的；输出多种源领域数据对应的待推荐数据，并将待推荐数据推送至相应客户端。由于本申请通过知识图谱来表征用户和不同领域产品之间关系所构成的拓扑结构，同时结合了用户数据对结果的影响，从而得到更加精准的源领域嵌入向量，使得模型训练后的精度更高，提升了数据推荐的准确性。
请参见图4，为本申请实施例提供了一种生成预先训练的跨领域推荐模型的流程示意图。如图4所示，本申请实施例的方法可以包括以下步骤：
S201，创建跨领域数据推荐模型；
其中，跨领域数据推荐模型为孪生网络，孪生网络包括源领域子模型和目标领域子模型；
通常，模型采用孪生网络的结构，可以将不同领域的输入映射到了新的相同的向量空间中，并且在工程实践中每次需要embedding的内容被大大减少，有效的提升了模型效率，实现更大规模数据的线上使用。同时，基于孪生网络自身的高扩展性，可以更加方便的同时针对多个领域的内容进行推荐冷启动。
S202，根据多种历史源领域数据构建源领域的拓扑结构，得到知识图谱；
在本申请实施例中，在生成知识图谱时，首先获取多种历史源领域数据，再分析多种历史源领域数据，确定出源领域的用户喜好数据，然后基于源领域的用户喜好数据从多种历史源领域数据中确定出用户喜好关系和产品属性关系，其次根据用户喜好关系和产品属性关系构建源领域的拓扑结构，生成源领域的图结构，最后将源领域的图结构确定为知识图谱。
S203，将知识图谱和用户数据输入源领域子模型中，输出多种源领域历史数据对应的源领域嵌入向量；
其中，源领域子模型包括graph transformer图神经网络和第一桥接函数。
在本申请实施例中，首先将知识图谱输入graph transformer图神经网络中，输出源领域用户喜好的嵌入向量，然后将用户数据输入graph transformer图神经网络中，输出用户自身数据的嵌入向量，最后将源领域用户喜好的嵌入向量与用户自身数据的嵌入向量输入桥接函数中进行向量拼接，输出多种源领域历史数据对应的源领域嵌入向量。
S204，根据多种历史目标领域数据构建不同正负比例的训练的数据；
在本申请实施例中，首先获取多种历史目标领域数据，然后分析多种历史目标领域数据，确定出目标领域的产品数据，最后根据目标领域的产品数据构建不同正负比例的训练的数据。
S205，将不同正负比例的训练的数据与用户数据输入目标领域子模型中，输出多种历史目标领域数据对应的目标领域嵌入向量；
其中，目标领域子模型包括JK-Net网络和第二桥接函数。
在本申请实施例中，首先获取不同正负比例的训练的数据中每个产品的嵌入向量，然后将用户数据输入JK-Net网络中，输出用户在目标领域自身的嵌入向量，最后将每个产品的嵌入向量与用户在目标领域自身的嵌入向量输入桥接函数中进行拼接，输出多种历史目标领域数据对应的目标领域嵌入向量。
具体的，JK-Net网络是一种深层GNN架构的网络，JK-Net通过自适应学习处在不同位置的节点聚合不同领域，从而可以改善节点的表示形式。
S206，根据源领域嵌入向量与目标领域嵌入向量生成预先训练的跨领域推荐模型。
在本申请实施例中，在生成预先训练的跨领域推荐模型时，首先根据源领域嵌入向量与目标领域嵌入向量进行相似度计算，生成相似度分数，然后将相似度分数确定为模型损失值，最后当模型损失值到达预设阈值时，生成预先训练的跨领域推荐模型。
例如图5所示，图5是跨领域推荐模型训练过程的过程示意框图，首先，根据多种历史源领域数据确定出用户喜好，根据用户喜好和产品属性的关系，构建起源领域的图结构。以此图结构为基础，使用graph transformer图神经网络结构，获取用户在源领域使用喜好的节点所构成的子图的图嵌入表示，并以此作为用户在源领域喜好的嵌入表示；根据用户的在源领域的喜好，同样使用图网络获取用户自身数据的嵌入表示。
之后通过一个桥接网络(如一个简单线性层，但本方法采用一个LSTM结构，将用户自身的嵌入表示和用户在源领域的喜好通过LSTM结构合并成一个向量)，得到最终源领域的嵌入表示,这样得到的嵌入表示相当于是针对不同用户的个性化偏好迁移函数，不再是通过一个统一的函数对所有用户的偏好进行迁移，更有效的利用了用户信息。
同样，在目标领域，使用JK-Net的方式可以解决传统图网络层数不能太深的问题，获取目标领域每个产品的嵌入表示，仿照在源领域的方式，使用图网络得到用户在目标领域自身的嵌入表示后，再通过一个桥接网络得到用户在目标领域的最终嵌入表示。最终，通过源领域的嵌入表示和目标领域的嵌入表示，得到最终的分数，确定用户是否会对目标领域内容感兴趣。
模型整体采用孪生网络结构，目标域和源领域分别使用类似的网络结构，在训练过程中使用相同的损失函数进行优化，在用户数据和领域产品数据桥接的函数共享参数。同时，模型在针对多个目标领域时，可以采用类似triple network的结构，在训练时可以针对不同目标领域构建不同正负比例的训练的数据，让模型可以同时计算在多个目标领域的推荐结果。
需要说明的是，本申请有效的利用了用户、不同领域商品之间关系所构成的拓扑结构，同时结合了用户自身信息对结果的影响，实现了更加精准的源领域、目标域嵌入表示，从而提升了模型的推荐精度。
在本申请实施例中，跨领域数据推荐装置获取多种源领域数据和多种目标领域数据；将多种源领域数据和多种目标领域数据输入预先训练的跨领域数据推荐模型中，以根据多种源领域数据从多种目标领域数据确定出待推荐数据；其中，预先训练的跨领域数据推荐模型是基于知识图谱和用户数据训练生成的，知识图谱是根据多种历史源领域数据构建的；输出多种源领域数据对应的待推荐数据，并将待推荐数据推送至相应客户端。由于本申请通过知识图谱来表征用户和不同领域产品之间关系所构成的拓扑结构，同时结合了用户数据对结果的影响，从而得到更加精准的源领域嵌入向量，使得模型训练后的精度更高，提升了数据推荐的准确性。
下述为本申请装置实施例，可以用于执行本申请方法实施例。对于本申请装置实施例中未披露的细节，请参照本申请方法实施例。
请参见图6，其示出了本申请一个示例性实施例提供的跨领域数据推荐装置的结构示意图，应用于服务器。该跨领域数据推荐装置可以通过软件、硬件或者两者的结合实现成为设备的全部或一部分。该装置1包括数据获取模块10、数据输入模块20、数据推送模块30。
数据获取模块10，用于获取多种源领域数据和多种目标领域数据；
数据输入模块20，用于将多种源领域数据和多种目标领域数据输入预先训练的跨领域数据推荐模型中，以根据多种源领域数据从多种目标领域数据确定出待推荐数据；其中，预先训练的跨领域数据推荐模型是基于知识图谱和用户数据训练生成的，知识图谱是根据多种历史源领域数据构建的；
数据推送模块30，用于输出多种源领域数据对应的待推荐数据，并将待推荐数据推送至相应客户端。
需要说明的是，上述实施例提供的高压后部识别装置在执行高压后部识别方法时，仅以上述各功能模块的划分进行举例说明，实际应用中，可以根据需要而将上述功能分配由不同的功能模块完成，即将设备的内部结构划分成不同 的功能模块，以完成以上描述的全部或者部分功能。另外，上述实施例提供的高压后部识别装置与高压后部识别方法实施例属于同一构思，其体现实现过程详见方法实施例，这里不再赘述。
上述本申请实施例序号仅仅为了描述，不代表实施例的优劣。
在本申请实施例中，跨领域数据推荐装置获取多种源领域数据和多种目标领域数据；将多种源领域数据和多种目标领域数据输入预先训练的跨领域数据推荐模型中，以根据多种源领域数据从多种目标领域数据确定出待推荐数据；其中，预先训练的跨领域数据推荐模型是基于知识图谱和用户数据训练生成的，知识图谱是根据多种历史源领域数据构建的；输出多种源领域数据对应的待推荐数据，并将待推荐数据推送至相应客户端。由于本申请通过知识图谱来表征用户和不同领域产品之间关系所构成的拓扑结构，同时结合了用户数据对结果的影响，从而得到更加精准的源领域嵌入向量，使得模型训练后的精度更高，提升了数据推荐的准确性。
在一个实施例中，提出了一种计算机设备，设备包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，处理器执行计算机程序时实现以下步骤：获取多种源领域数据和多种目标领域数据；将多种源领域数据和多种目标领域数据输入预先训练的跨领域数据推荐模型中，以根据多种源领域数据从多种目标领域数据确定出待推荐数据；其中，预先训练的跨领域数据推荐模型是基于知识图谱和用户数据训练生成的，知识图谱是根据多种历史源领域数据构建的；输出多种源领域数据对应的待推荐数据，并将待推荐数据推送至相应客户端。
在一个实施例中，处理器执行生成预先训练的跨领域推荐模型时，具体执行以下操作：创建跨领域数据推荐模型；其中，跨领域数据推荐模型为孪生网络，孪生网络包括源领域子模型和目标领域子模型；根据多种历史源领域数据构建源领域的拓扑结构，得到知识图谱；将知识图谱和用户数据输入源领域子模型中，输出多种源领域历史数据对应的源领域嵌入向量；根据多种历史目标领域数据构建不同正负比例的训练的数据；将不同正负比例的训练的数据与用户数据输入目标领域子模型中，输出多种历史目标领域数据对应的目标领域嵌入向量；根据源领域嵌入向量与目标领域嵌入向量生成预先训练的跨领域推荐模型。
在一个实施例中，处理器执行根据多种历史源领域数据构建源领域的拓扑结构，得到知识图谱时，具体执行以下操作：获取多种历史源领域数据；分析多种历史源领域数据，确定出源领域的用户喜好数据；基于源领域的用户喜好数据从多种历史源领域数据中确定出用户喜好关系和产品属性关系；根据用户喜好关系和产品属性关系构建源领域的拓扑结构，生成源领域的图结构；将源领域的图结构确定为知识图谱。
在一个实施例中，处理器执行将知识图谱和用户数据输入源领域子模型中，输出多种源领域历史数据对应的源领域嵌入向量时，具体执行以下操作：将知识图谱输入graph transformer图神经网络中，输出源领域用户喜好的嵌入向量；将用户数据输入graph transformer图神经网络中，输出用户自身数据的嵌入向量；将源领域用户喜好的嵌入向量与用户自身数据的嵌入向量输入桥接函数中进行向量拼接，输出多种源领域历史数据对应的源领域嵌入向量。
在一个实施例中，处理器执行根据多种历史目标领域数据构建不同正负比 例的训练的数据时，具体执行以下操作：获取多种历史目标领域数据；分析多种历史目标领域数据，确定出目标领域的产品数据；根据目标领域的产品数据构建不同正负比例的训练的数据。
在一个实施例中，处理器执行将不同正负比例的训练的数据与用户数据输入目标领域子模型中，输出多种历史目标领域数据对应的目标领域嵌入向量时，具体执行以下操作：获取不同正负比例的训练的数据中每个产品的嵌入向量；将用户数据输入JK-Net网络中，输出用户在目标领域自身的嵌入向量；将每个产品的嵌入向量与用户在目标领域自身的嵌入向量输入桥接函数中进行拼接，输出多种历史目标领域数据对应的目标领域嵌入向量。
在一个实施例中，处理器执行根据源领域嵌入向量与目标领域嵌入向量生成预先训练的跨领域推荐模型时，具体执行以下操作：根据源领域嵌入向量与目标领域嵌入向量进行相似度计算，生成相似度分数；将相似度分数确定为模型损失值；当模型损失值到达预设阈值时，生成预先训练的跨领域推荐模型。
在本申请实施例中，跨领域数据推荐装置获取多种源领域数据和多种目标领域数据；将多种源领域数据和多种目标领域数据输入预先训练的跨领域数据推荐模型中，以根据多种源领域数据从多种目标领域数据确定出待推荐数据；其中，预先训练的跨领域数据推荐模型是基于知识图谱和用户数据训练生成的，知识图谱是根据多种历史源领域数据构建的；输出多种源领域数据对应的待推荐数据，并将待推荐数据推送至相应客户端。由于本申请通过知识图谱来表征用户和不同领域产品之间关系所构成的拓扑结构，同时结合了用户数据对结果的影响，从而得到更加精准的源领域嵌入向量，使得模型训练后的精度更高，提升了数据推荐的准确性。
在一个实施例中，提出了一种存储有计算机可读指令的介质，该计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器执行以下步骤：获取多种源领域数据和多种目标领域数据；将多种源领域数据和多种目标领域数据输入预先训练的跨领域数据推荐模型中，以根据多种源领域数据从多种目标领域数据确定出待推荐数据；其中，预先训练的跨领域数据推荐模型是基于知识图谱和用户数据训练生成的，知识图谱是根据多种历史源领域数据构建的；输出多种源领域数据对应的待推荐数据，并将待推荐数据推送至相应客户端。所述存储有计算机可读指令的介质可以是非易失性，也可以是易失性。
在一个实施例中，处理器执行生成预先训练的跨领域推荐模型时，具体执行以下操作：创建跨领域数据推荐模型；其中，跨领域数据推荐模型为孪生网络，孪生网络包括源领域子模型和目标领域子模型；根据多种历史源领域数据构建源领域的拓扑结构，得到知识图谱；将知识图谱和用户数据输入源领域子模型中，输出多种源领域历史数据对应的源领域嵌入向量；根据多种历史目标领域数据构建不同正负比例的训练的数据；将不同正负比例的训练的数据与用户数据输入目标领域子模型中，输出多种历史目标领域数据对应的目标领域嵌入向量；根据源领域嵌入向量与目标领域嵌入向量生成预先训练的跨领域推荐模型。
在一个实施例中，处理器执行根据多种历史源领域数据构建源领域的拓扑结构，得到知识图谱时，具体执行以下操作：获取多种历史源领域数据；分析多种历史源领域数据，确定出源领域的用户喜好数据；基于源领域的用户喜好数据从多种历史源领域数据中确定出用户喜好关系和产品属性关系；根据用户 喜好关系和产品属性关系构建源领域的拓扑结构，生成源领域的图结构；将源领域的图结构确定为知识图谱。
在一个实施例中，处理器执行将知识图谱和用户数据输入源领域子模型中，输出多种源领域历史数据对应的源领域嵌入向量时，具体执行以下操作：将知识图谱输入graph transformer图神经网络中，输出源领域用户喜好的嵌入向量；将用户数据输入graph transformer图神经网络中，输出用户自身数据的嵌入向量；将源领域用户喜好的嵌入向量与用户自身数据的嵌入向量输入桥接函数中进行向量拼接，输出多种源领域历史数据对应的源领域嵌入向量。
在一个实施例中，处理器执行根据多种历史目标领域数据构建不同正负比例的训练的数据时，具体执行以下操作：获取多种历史目标领域数据；分析多种历史目标领域数据，确定出目标领域的产品数据；根据目标领域的产品数据构建不同正负比例的训练的数据。
在一个实施例中，处理器执行将不同正负比例的训练的数据与用户数据输入目标领域子模型中，输出多种历史目标领域数据对应的目标领域嵌入向量时，具体执行以下操作：获取不同正负比例的训练的数据中每个产品的嵌入向量；将用户数据输入JK-Net网络中，输出用户在目标领域自身的嵌入向量；将每个产品的嵌入向量与用户在目标领域自身的嵌入向量输入桥接函数中进行拼接，输出多种历史目标领域数据对应的目标领域嵌入向量。
在一个实施例中，处理器执行根据源领域嵌入向量与目标领域嵌入向量生成预先训练的跨领域推荐模型时，具体执行以下操作：根据源领域嵌入向量与目标领域嵌入向量进行相似度计算，生成相似度分数；将相似度分数确定为模型损失值；当模型损失值到达预设阈值时，生成预先训练的跨领域推荐模型。
在本申请实施例中，跨领域数据推荐装置获取多种源领域数据和多种目标领域数据；将多种源领域数据和多种目标领域数据输入预先训练的跨领域数据推荐模型中，以根据多种源领域数据从多种目标领域数据确定出待推荐数据；其中，预先训练的跨领域数据推荐模型是基于知识图谱和用户数据训练生成的，知识图谱是根据多种历史源领域数据构建的；输出多种源领域数据对应的待推荐数据，并将待推荐数据推送至相应客户端。由于本申请通过知识图谱来表征用户和不同领域产品之间关系所构成的拓扑结构，同时结合了用户数据对结果的影响，从而得到更加精准的源领域嵌入向量，使得模型训练后的精度更高，提升了数据推荐的准确性。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，该计算机程序可存储于一计算机可读取介质中，该程序在执行时，可包括如上述各方法的实施例的流程。其中，前述的介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory，ROM)等非易失性介质，或随机存储记忆体(Random Access Memory，RAM)等。
以上实施例的各技术特征可以进行任意的组合，为使描述简洁，未对上述实施例中的各个技术特征所有可能的组合都进行描述，然而，只要这些技术特征的组合不存在矛盾，都应当认为是本说明书记载的范围。
以上实施例仅表达了本申请的几种实施方式，其描述较为具体和详细，但并不能因此而理解为对本申请专利范围的限制。应当指出的是，对于本领域的普通技术人员来说，在不脱离本申请构思的前提下，还可以做出若干变形和改进，这些都属于本申请的保护范围。因此，本申请专利的保护范围应以所附权利要求为准。

Claims (20)

  1. 一种跨领域数据推荐方法，其中，所述方法包括：
    获取多种源领域数据和多种目标领域数据；
    将所述多种源领域数据和所述多种目标领域数据输入预先训练的跨领域数据推荐模型中，以根据所述多种源领域数据从所述多种目标领域数据确定出待推荐数据；其中，所述预先训练的跨领域数据推荐模型是基于知识图谱和用户数据训练生成的，所述知识图谱是根据多种历史源领域数据构建的；
    输出所述多种源领域数据对应的待推荐数据，并将所述待推荐数据推送至相应客户端。
  2. 根据权利要求1所述的方法，其中，按照以下步骤生成预先训练的跨领域推荐模型，包括：
    创建跨领域数据推荐模型；其中，所述跨领域数据推荐模型为孪生网络，所述孪生网络包括源领域子模型和目标领域子模型；
    根据多种历史源领域数据构建源领域的拓扑结构，得到知识图谱；
    将所述知识图谱和用户数据输入所述源领域子模型中，输出所述多种源领域历史数据对应的源领域嵌入向量；
    根据多种历史目标领域数据构建不同正负比例的训练的数据；
    将所述不同正负比例的训练的数据与所述用户数据输入所述目标领域子模型中，输出所述多种历史目标领域数据对应的目标领域嵌入向量；
    根据所述源领域嵌入向量与所述目标领域嵌入向量生成预先训练的跨领域推荐模型。
  3. 根据权利要求2所述的方法，其中，所述根据多种历史源领域数据构建源领域的拓扑结构，得到知识图谱，包括：
    获取多种历史源领域数据；
    分析所述多种历史源领域数据，确定出源领域的用户喜好数据；
    基于所述源领域的用户喜好数据从所述多种历史源领域数据中确定出用户喜好关系和产品属性关系；
    根据所述用户喜好关系和产品属性关系构建源领域的拓扑结构，生成源领域的图结构；
    将所述源领域的图结构确定为知识图谱。
  4. 根据权利要求2所述的方法，其中，所述源领域子模型包括graph transformer图神经网络和第一桥接函数；
    所述将所述知识图谱和用户数据输入所述源领域子模型中，输出所述多种源领域历史数据对应的源领域嵌入向量，包括：
    将所述知识图谱输入所述graph transformer图神经网络中，输出源领域用户喜好的嵌入向量；
    将所述用户数据输入所述graph transformer图神经网络中，输出用户自身数据的嵌入向量；
    将所述源领域用户喜好的嵌入向量与所述用户自身数据的嵌入向量输入所述桥接函数中进行向量拼接，输出所述多种源领域历史数据对应的源领域嵌入向量。
  5. 根据权利要求2所述的方法，其中，所述根据多种历史目标领域数据构建不同正负比例的训练的数据，包括：
    获取多种历史目标领域数据；
    分析所述多种历史目标领域数据，确定出目标领域的产品数据；
    根据所述目标领域的产品数据构建不同正负比例的训练的数据。
  6. 根据权利要求2所述的方法，其中，所述目标领域子模型包括JK-Net网络和第二桥接函数；
    所述将所述不同正负比例的训练的数据与所述用户数据输入所述目标领域子模型中，输出所述多种历史目标领域数据对应的目标领域嵌入向量，包括：
    获取所述不同正负比例的训练的数据中每个产品的嵌入向量；
    将所述用户数据输入所述JK-Net网络中，输出用户在目标领域自身的嵌入向量；
    将所述每个产品的嵌入向量与所述用户在目标领域自身的嵌入向量输入所述桥接函数中进行拼接，输出所述多种历史目标领域数据对应的目标领域嵌入向量。
  7. 根据权利要求2所述的方法，其中，所述根据所述源领域嵌入向量与所述目标领域嵌入向量生成预先训练的跨领域推荐模型，包括：
    根据所述源领域嵌入向量与所述目标领域嵌入向量进行相似度计算，生成相似度分数；
    将所述相似度分数确定为模型损失值；
    当所述模型损失值到达预设阈值时，生成预先训练的跨领域推荐模型。
  8. 一种跨领域数据推荐装置，其中，所述装置包括：
    数据获取模块，用于获取多种源领域数据和多种目标领域数据；
    数据输入模块，用于将所述多种源领域数据和所述多种目标领域数据输入预先训练的跨领域数据推荐模型中，以根据所述多种源领域数据从所述多种目标领域数据确定出待推荐数据；其中，所述预先训练的跨领域数据推荐模型是基于知识图谱和用户数据训练生成的，所述知识图谱是根据多种历史源领域数据构建的；
    数据推送模块，用于输出所述多种源领域数据对应的待推荐数据，并将所述待推荐数据推送至相应客户端。
  9. 一种计算机设备，其中，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述计算机可读指令被所述处理器执行时，使得所述处理器执行如下步骤的指令：
    获取多种源领域数据和多种目标领域数据；
    将所述多种源领域数据和所述多种目标领域数据输入预先训练的跨领域数据推荐模型中，以根据所述多种源领域数据从所述多种目标领域数据确定出待推荐数据；其中，所述预先训练的跨领域数据推荐模型是基于知识图谱和用户数据训练生成的，所述知识图谱是根据多种历史源领域数据构建的；
    输出所述多种源领域数据对应的待推荐数据，并将所述待推荐数据推送至相应客户端。
  10. 根据权利要求9所述的计算机设备，其中，按照以下步骤生成预先训练的跨领域推荐模型，包括：
    创建跨领域数据推荐模型；其中，所述跨领域数据推荐模型为孪生网络，所述孪生网络包括源领域子模型和目标领域子模型；
    根据多种历史源领域数据构建源领域的拓扑结构，得到知识图谱；
    将所述知识图谱和用户数据输入所述源领域子模型中，输出所述多种源领域历史数据对应的源领域嵌入向量；
    根据多种历史目标领域数据构建不同正负比例的训练的数据；
    将所述不同正负比例的训练的数据与所述用户数据输入所述目标领域子模型中，输出所述多种历史目标领域数据对应的目标领域嵌入向量；
    根据所述源领域嵌入向量与所述目标领域嵌入向量生成预先训练的跨领域推荐模型。
  11. 根据权利要求10所述的计算机设备，其中，所述根据多种历史源领域数据构建源领域的拓扑结构，得到知识图谱，包括：
    获取多种历史源领域数据；
    分析所述多种历史源领域数据，确定出源领域的用户喜好数据；
    基于所述源领域的用户喜好数据从所述多种历史源领域数据中确定出用户喜好关系和产品属性关系；
    根据所述用户喜好关系和产品属性关系构建源领域的拓扑结构，生成源领域的图结构；
    将所述源领域的图结构确定为知识图谱。
  12. 根据权利要求10所述的计算机设备，其中，所述源领域子模型包括graph transformer图神经网络和第一桥接函数；
    所述将所述知识图谱和用户数据输入所述源领域子模型中，输出所述多种源领域历史数据对应的源领域嵌入向量，包括：
    将所述知识图谱输入所述graph transformer图神经网络中，输出源领域用户喜好的嵌入向量；
    将所述用户数据输入所述graph transformer图神经网络中，输出用户自身数据的嵌入向量；
    将所述源领域用户喜好的嵌入向量与所述用户自身数据的嵌入向量输入所述桥接函数中进行向量拼接，输出所述多种源领域历史数据对应的源领域嵌入向量。
  13. 根据权利要求10所述的计算机设备，其中，所述根据多种历史目标领域数据构建不同正负比例的训练的数据，包括：
    获取多种历史目标领域数据；
    分析所述多种历史目标领域数据，确定出目标领域的产品数据；
    根据所述目标领域的产品数据构建不同正负比例的训练的数据。
  14. 根据权利要求10所述的计算机设备，其中，所述目标领域子模型包括JK-Net网络和第二桥接函数；
    所述将所述不同正负比例的训练的数据与所述用户数据输入所述目标领域子模型中，输出所述多种历史目标领域数据对应的目标领域嵌入向量，包括：
    获取所述不同正负比例的训练的数据中每个产品的嵌入向量；
    将所述用户数据输入所述JK-Net网络中，输出用户在目标领域自身的嵌入向量；
    将所述每个产品的嵌入向量与所述用户在目标领域自身的嵌入向量输入所述桥接函数中进行拼接，输出所述多种历史目标领域数据对应的目标领域嵌入向量。
  15. 一种存储有计算机可读指令的介质，其中，所述计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器执行如下步骤的指令：
    获取多种源领域数据和多种目标领域数据；
    将所述多种源领域数据和所述多种目标领域数据输入预先训练的跨领域数据推荐模型中，以根据所述多种源领域数据从所述多种目标领域数据确定出待 推荐数据；其中，所述预先训练的跨领域数据推荐模型是基于知识图谱和用户数据训练生成的，所述知识图谱是根据多种历史源领域数据构建的；
    输出所述多种源领域数据对应的待推荐数据，并将所述待推荐数据推送至相应客户端。
  16. 根据权利要求15所述的介质，其中，按照以下步骤生成预先训练的跨领域推荐模型，包括：
    创建跨领域数据推荐模型；其中，所述跨领域数据推荐模型为孪生网络，所述孪生网络包括源领域子模型和目标领域子模型；
    根据多种历史源领域数据构建源领域的拓扑结构，得到知识图谱；
    将所述知识图谱和用户数据输入所述源领域子模型中，输出所述多种源领域历史数据对应的源领域嵌入向量；
    根据多种历史目标领域数据构建不同正负比例的训练的数据；
    将所述不同正负比例的训练的数据与所述用户数据输入所述目标领域子模型中，输出所述多种历史目标领域数据对应的目标领域嵌入向量；
    根据所述源领域嵌入向量与所述目标领域嵌入向量生成预先训练的跨领域推荐模型。
  17. 根据权利要求16所述的介质，其中，所述根据多种历史源领域数据构建源领域的拓扑结构，得到知识图谱，包括：
    获取多种历史源领域数据；
    分析所述多种历史源领域数据，确定出源领域的用户喜好数据；
    基于所述源领域的用户喜好数据从所述多种历史源领域数据中确定出用户喜好关系和产品属性关系；
    根据所述用户喜好关系和产品属性关系构建源领域的拓扑结构，生成源领域的图结构；
    将所述源领域的图结构确定为知识图谱。
  18. 根据权利要求16所述的介质，其中，所述源领域子模型包括graph transformer图神经网络和第一桥接函数；
    所述将所述知识图谱和用户数据输入所述源领域子模型中，输出所述多种源领域历史数据对应的源领域嵌入向量，包括：
    将所述知识图谱输入所述graph transformer图神经网络中，输出源领域用户喜好的嵌入向量；
    将所述用户数据输入所述graph transformer图神经网络中，输出用户自身数据的嵌入向量；
    将所述源领域用户喜好的嵌入向量与所述用户自身数据的嵌入向量输入所述桥接函数中进行向量拼接，输出所述多种源领域历史数据对应的源领域嵌入向量。
  19. 根据权利要求16所述的介质，其中，所述根据多种历史目标领域数据构建不同正负比例的训练的数据，包括：
    获取多种历史目标领域数据；
    分析所述多种历史目标领域数据，确定出目标领域的产品数据；
    根据所述目标领域的产品数据构建不同正负比例的训练的数据。
  20. 根据权利要求16所述的介质，其中，所述目标领域子模型包括JK-Net网络和第二桥接函数；
    所述将所述不同正负比例的训练的数据与所述用户数据输入所述目标领域子模型中，输出所述多种历史目标领域数据对应的目标领域嵌入向量，包括：
    获取所述不同正负比例的训练的数据中每个产品的嵌入向量；
    将所述用户数据输入所述JK-Net网络中，输出用户在目标领域自身的嵌入向量；
    将所述每个产品的嵌入向量与所述用户在目标领域自身的嵌入向量输入所述桥接函数中进行拼接，输出所述多种历史目标领域数据对应的目标领域嵌入向量。
PCT/CN2022/090364 2022-03-14 2022-04-29 一种跨领域数据推荐方法、装置、计算机设备及介质 WO2023173550A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210248145.8 2022-03-14
CN202210248145.8A CN114661887A (zh) 2022-03-14 2022-03-14 一种跨领域数据推荐方法、装置、计算机设备及介质

Publications (1)

Publication Number Publication Date
WO2023173550A1 true WO2023173550A1 (zh) 2023-09-21

Family

ID=82028672

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/090364 WO2023173550A1 (zh) 2022-03-14 2022-04-29 一种跨领域数据推荐方法、装置、计算机设备及介质

Country Status (2)

Country Link
CN (1) CN114661887A (zh)
WO (1) WO2023173550A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116629983B (zh) * 2023-07-24 2023-09-22 成都晓多科技有限公司 基于用户偏好的跨领域商品推荐方法及系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200210881A1 (en) * 2018-12-27 2020-07-02 General Electric Company Cross-domain featuring engineering
CN112084383A (zh) * 2020-09-07 2020-12-15 中国平安财产保险股份有限公司 基于知识图谱的信息推荐方法、装置、设备及存储介质
CN112699310A (zh) * 2020-12-30 2021-04-23 航天信息股份有限公司 基于深度神经网络的冷启动跨域混合推荐的方法及系统
WO2022041979A1 (zh) * 2020-08-28 2022-03-03 腾讯科技（深圳）有限公司 一种信息推荐模型的训练方法和相关装置

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200210881A1 (en) * 2018-12-27 2020-07-02 General Electric Company Cross-domain featuring engineering
WO2022041979A1 (zh) * 2020-08-28 2022-03-03 腾讯科技（深圳）有限公司 一种信息推荐模型的训练方法和相关装置
CN112084383A (zh) * 2020-09-07 2020-12-15 中国平安财产保险股份有限公司 基于知识图谱的信息推荐方法、装置、设备及存储介质
CN112699310A (zh) * 2020-12-30 2021-04-23 航天信息股份有限公司 基于深度神经网络的冷启动跨域混合推荐的方法及系统

Also Published As

Publication number Publication date
CN114661887A (zh) 2022-06-24

Similar Documents

Publication Publication Date Title
US20220222920A1 (en) Content processing method and apparatus, computer device, and storage medium
US11416268B2 (en) Aggregate features for machine learning
JP7017640B2 (ja) データ拡張方策の学習
US20230259784A1 (en) Regularized neural network architecture search
US20180246952A1 (en) Method and device for generating online question paths from existing question banks using a knowledge graph
WO2023065859A1 (zh) 物品推荐方法、装置及存储介质
US11086857B1 (en) Method and system for semantic search with a data management system
JP2022050379A (ja) 意味検索方法、装置、電子機器、記憶媒体およびコンピュータプログラム
WO2022140900A1 (zh) 个人知识图谱构建方法、装置及相关设备
US11443202B2 (en) Real-time on the fly generation of feature-based label embeddings via machine learning
WO2023045184A1 (zh) 一种文本类别识别方法、装置、计算机设备及介质
CN114329029B (zh) 对象检索方法、装置、设备及计算机存储介质
CN113239176B (zh) 语义匹配模型训练方法、装置、设备及存储介质
CN113821527A (zh) 哈希码的生成方法、装置、计算机设备及存储介质
WO2023173550A1 (zh) 一种跨领域数据推荐方法、装置、计算机设备及介质
US11651041B2 (en) Method and system for storing a plurality of documents
CN111291563A (zh) 词向量对齐方法和词向量对齐模型训练方法
CN117009621A (zh) 信息搜索方法、装置、电子设备、存储介质及程序产品
WO2019144046A1 (en) Distributed high performance computing using distributed average consensus
CN111897943A (zh) 会话记录搜索方法、装置、电子设备及存储介质
KR102389555B1 (ko) 가중 트리플 지식 그래프를 생성하는 장치, 방법 및 컴퓨터 프로그램
CN115510203B (zh) 问题答案确定方法、装置、设备、存储介质及程序产品
US20230110458A1 (en) Method and system for matching investors with companies
He et al. Neural Ordinary Differential Equations with Envolutionary Weights
CN117093874A (zh) 文本生成方法、装置、计算机设备、介质和程序产品

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22931577

Country of ref document: EP

Kind code of ref document: A1