WO2023045378A1 - 向用户推荐物品信息的方法、设备、存储介质及程序产品 - Google Patents

向用户推荐物品信息的方法、设备、存储介质及程序产品 Download PDF

Info

Publication number
WO2023045378A1
WO2023045378A1 PCT/CN2022/095341 CN2022095341W WO2023045378A1 WO 2023045378 A1 WO2023045378 A1 WO 2023045378A1 CN 2022095341 W CN2022095341 W CN 2022095341W WO 2023045378 A1 WO2023045378 A1 WO 2023045378A1
Authority
WO
WIPO (PCT)
Prior art keywords
node
nodes
vector
user
preset
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.)
Ceased
Application number
PCT/CN2022/095341
Other languages
English (en)
French (fr)
Inventor
肖智鹏
施文祥
殷威
徐昊
刘欢
李森
张万书
金林荣
裴英瑞
王浩
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.)
Beijing Wodong Tianjun Information Technology Co Ltd
Original Assignee
Beijing Wodong Tianjun Information Technology 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 Beijing Wodong Tianjun Information Technology Co Ltd filed Critical Beijing Wodong Tianjun Information Technology Co Ltd
Publication of WO2023045378A1 publication Critical patent/WO2023045378A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/23—Updating
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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—Recommending goods or services

Definitions

  • the present application relates to data processing technology, and in particular to a method, device, storage medium and program product for recommending item information to users.
  • information can be actively pushed to users, for example, product information can be pushed, and information that users may be interested in can also be pushed.
  • the network topology map can be selected to store the relationship between users and information.
  • the complexity of graphs often makes real-time recommendation a difficult point. Therefore, how to improve the timeliness of personalized recommendation is a technical problem urgently needed to be solved by those skilled in the art.
  • the present application provides a method, device, storage medium and program product for recommending item information to users, so as to solve the problem of poor timeliness of personalized recommendation in the prior art.
  • the first aspect of this application is to provide a method for recommending item information to a user, including:
  • the preset topology graph includes nodes, and the nodes include user nodes , Item nodes, there is an association relationship between user nodes and item nodes connected by edges;
  • the second aspect of this application is to provide a device for personalized recommendation information, including:
  • An acquisition unit configured to acquire a preset topological graph for representing the association relationship between users and items, and update data for updating the preset topological graph; wherein, the preset topological graph includes nodes, and the nodes Including user nodes and item nodes, there is an association relationship between user nodes and item nodes connected by edges;
  • a graph updating unit configured to update edges in the preset topology graph according to the update data to obtain an updated topology graph
  • a vector updating unit configured to obtain a pre-trained vector model corresponding to the preset topology map, and update any node in the topology map according to the updated topology map and the obtained vector model A vector, which is used to characterize the characteristics of the node;
  • the recommending unit is configured to determine similar item nodes of the user node according to the vectors of each node, and recommend item information of the similar item nodes to users corresponding to the user nodes.
  • Another aspect of the present application is to provide an electronic device, comprising:
  • the computer program is stored in the memory and is configured to be executed by the processor to implement the method for recommending item information to the user as described in the first aspect above.
  • Another aspect of the present application is to provide a computer-readable storage medium on which a computer program is stored, and the computer program is executed by a processor to implement the method for recommending item information to a user as described in the first aspect above .
  • Yet another aspect of the present application is to provide a computer program product, including a computer program, which, when executed by a processor, implements the method for recommending item information to a user as described in the first aspect above.
  • the method, device, storage medium, and program product for recommending item information to users include: obtaining a preset topological map used to represent the relationship between the user and the item, and updating data used to update the preset topological map;
  • the preset topology graph includes nodes, the nodes include user nodes and item nodes, and there is an association relationship between user nodes and item nodes connected by edges; update the edges in the preset topology graph according to the update data to obtain the updated topology graph ;
  • the preset topology map and the vector model corresponding to the preset topology map can be set in advance. After the update data exists, the preset topology map can be directly updated based on the update data, and the vector model can be used to update The final topology map is processed to obtain the updated vector of the node. This method can quickly update the node vector based on the updated data, and then recommend information to users more accurately based on the updated node vector.
  • Fig. 1 is a schematic diagram of the architecture of a transductive graph embedding algorithm shown in an exemplary embodiment
  • Fig. 2 is a schematic diagram of an inductive graph embedding algorithm architecture shown in an exemplary embodiment
  • FIG. 3 is a flow chart of a method for personalized recommendation information shown in an exemplary embodiment of the present application
  • FIG. 4 is a schematic diagram of a collection behavior subgraph shown in an exemplary embodiment of the present application.
  • FIG. 5 is a flow chart of a method for personalized recommendation information shown in another exemplary embodiment of the present application.
  • FIG. 6 is a schematic diagram of an initial topology diagram shown in an exemplary embodiment of the present application.
  • FIG. 7 is a schematic diagram of a behavior subgraph shown in an exemplary embodiment of the present application.
  • FIG. 8 is a structural diagram of a device for personalized recommendation information shown in an exemplary embodiment of the present application.
  • FIG. 9 is a structural diagram of a device for personalized recommendation information shown in another exemplary embodiment of the present application.
  • Fig. 10 is a structural diagram of an electronic device shown in an exemplary embodiment of the present application.
  • GCN Graph Convolutional Network, graph convolutional network
  • GRAPHSAGE Graph SAmple and aggreGatE
  • Fig. 1 is a schematic diagram of an architecture of a transductive graph embedding algorithm shown in an exemplary embodiment.
  • the offline data includes data such as user behavior, user attributes, and item attributes. Users and items are modeled as nodes in the graph through graph construction.
  • the user's behavior is used to construct the connection edges between nodes, and the user attributes and item attributes are the characteristics of the nodes in the graph.
  • algorithms such as GCN use the entire adjacency matrix of the graph and graph convolution operations to fuse the information of adjacent nodes to generate a vector representation of the nodes.
  • Fig. 2 is a schematic diagram of an inductive graph embedding algorithm architecture shown in an exemplary embodiment.
  • the full amount of offline data is first collected, which includes user behavior, user attributes, and item attributes. Users and items are modeled as nodes in the graph through graph construction.
  • this application provides a scheme for recommending item information to users.
  • the topology map can be pre-built, and the vector model can also be pre-trained.
  • the topology can be directly updated using the updated data Graph, and process the updated graph based on the vector model to determine the updated vector of the node.
  • This method can quickly update the vector of the node according to the updated data, and then can recommend personalized information to the user according to the updated vector, so that Improve the timeliness of personalized recommendations.
  • Fig. 3 is a flowchart of a method for recommending item information to a user according to an exemplary embodiment of the present application.
  • the method for recommending item information to users includes:
  • Step 301 obtain a preset topology graph used to represent the relationship between users and items, and update data used to update the preset topology graph; wherein, the preset topology graph includes nodes, and the nodes include user nodes and item nodes. There is an association relationship between user nodes and item nodes connected by edges.
  • the method provided in this application can be executed by one or more electronic devices with computing capabilities, and the electronic devices can be used as servers, such as cloud servers, or distributed servers, for example.
  • the server may, for example, provide an information recommendation function, for example, may recommend commodities to the user.
  • the electronic device may be triggered to execute the method of this embodiment when updating data is generated, or the method provided by this embodiment may be executed every preset period of time.
  • the electronic device can obtain a preset topology graph used to represent the relationship between users and items.
  • the preset topology graph includes multiple nodes, and these nodes include user nodes and item nodes.
  • User nodes and item nodes There are edges between nodes, so that user nodes and item nodes are connected through edges.
  • a topology graph including multiple nodes may be constructed in advance based on offline data. For example, if a piece of offline data includes user A accessing information B, node A and node B can be constructed, and an edge between A and B can be constructed to represent the relationship between A and B.
  • the weight value of the edge can also be set according to requirements.
  • the electronic device can also acquire update data, and the update data can include user behavior data. For example, when a user browses a certain product or accesses a certain piece of information, the electronic device can obtain relevant information.
  • the acquired update data is data used to update the preset topology map.
  • Step 302 updating edges in the preset topology graph according to the update data to obtain an updated topology graph.
  • the electronic device may update the acquired preset topology map according to the acquired update data to obtain an updated topology map.
  • the updated topology map may be acquired again as a preset topology map.
  • the electronic device can add edges to the preset topology graph according to the update data to update the topology graph, and can also update the weight value of the edges to update the topology graph. For example, if the update data includes that user A has browsed product C, then an edge between A and C can be constructed, and then the preset topology map can be updated.
  • Step 303 obtain the pre-trained vector model corresponding to the preset topology map, and update the vector of any node in the topology map according to the updated topology map and the obtained vector model, and the vector is used to represent the characteristics of the node.
  • the vector model can also be trained in advance using an initial topology map corresponding to a preset topology map, and the preset topology map can be the same as the initial topology map, or can be obtained by one or more updates based on the initial topology map .
  • the preset topology map is also obtained by processing offline data, the preset topology map has a corresponding relationship with the above-mentioned vector model.
  • the GRAPHSAGE algorithm can be used to train the initial topology graph to obtain a vector model.
  • the behavior subgraph may be collected for each node, and the behavior subgraph includes the node and its neighbor nodes, and also includes the relationship between these nodes.
  • corresponding behavior subgraphs can be collected for nodes P1, P2, and P3, and the electronic device can learn the relationship between adjacent nodes and central nodes in these subgraphs, thereby obtaining a vector model.
  • the electronic device after the electronic device has updated the preset topology map, it can also obtain the vector model corresponding to the preset topology map, and then update the vector model of any node in the topology map according to the updated topology map and the obtained vector model.
  • the vector of the node can represent the feature of the node, specifically, the vector of the node can be generated according to the information of the node itself and the information of the neighbor nodes of the node.
  • the relationship between nodes in the updated topology graph has changed. Therefore, the behavior subgraph of each node can be collected again, or only the behavior subgraph of nodes whose edges have changed with neighboring nodes can be collected.
  • Fig. 4 is a schematic diagram of a collection behavior subgraph shown in an exemplary embodiment of the present application.
  • FIG. 4 there is a topological graph structure as shown in FIG. 41 , in which the behavior subgraph of node 42 is shown as 43 .
  • the updated topology graph is 44
  • the behavior subgraph of node 42 is shown as 45 .
  • the behavior subgraph of another node 46 does not change, therefore, the vector of the node 46 may not be updated, and furthermore, the behavior subgraph of the node 46 does not need to be collected.
  • the updated vector of the node may be determined according to the updated behavior subgraph of the node and the vector model. For example, the information of the node and the adjacent nodes of the node may be input into the vector model, so that the vector model outputs the updated vector of the node.
  • Step 304 Determine similar item nodes of the user node according to the vectors of each node, and recommend item information of similar item nodes to users corresponding to the user node.
  • the electronic device can determine similar nodes among the nodes according to the vectors of each node. For example, when it is necessary to recommend information to user A, the similarity between user A's node and other nodes can be determined, and then similar nodes to user A's node can be obtained.
  • the electronic device may also recommend item information of the item to any node according to similar nodes of any node. For example, if the node of user A has a high similarity with the node of item D, the information of item D can be recommended to user A. For another example, if the node of user A has a high similarity with the node of user E, item information can be recommended to user A based on the information of user E.
  • the method provided in this embodiment is used to recommend item information to a user, and the method is executed by a device configured with the method provided in this embodiment, and the device is usually implemented in the form of hardware and/or software.
  • the method for recommending item information to users includes: acquiring a preset topology map used to characterize the relationship between users and items, and updating data used to update the preset topology map; wherein, the preset topology map
  • the graph includes nodes, nodes include user nodes and item nodes, and there is an association between user nodes and item nodes connected by edges; update the edges in the preset topology graph according to the update data to obtain the updated topology graph; obtain the pre-trained
  • the preset topology map and the vector model corresponding to the preset topology map can be set in advance. After the update data exists, the preset topology map can be directly updated based on the update data, and the vector model can be used to update The final topology map is processed to obtain the updated vector of the node. This method can quickly update the node vector based on the updated data, and then can recommend item information to the user more accurately based on the updated node vector.
  • Fig. 5 is a flowchart of a method for recommending item information to a user according to another exemplary embodiment of the present application.
  • the method for recommending item information to users includes:
  • Step 501 obtain preset offline data, and construct an initial topology map according to the offline data.
  • offline data may be prepared in advance, for example, offline data of the last 7 days may be prepared, and an initial topology map may be constructed based on these offline data.
  • the constructed initial topology graph includes a plurality of nodes, such as user nodes and item nodes, and there are edges between these nodes, and an edge indicates that there is an association relationship between two nodes. For example, there may be an edge between user node A and user node B, indicating that there is an association between user A and user B. For example, there is an edge between user node A and item node C, which indicates the relationship between user A and item C There is a relationship between them.
  • each piece of offline data can be parsed to obtain behavioral data. For example, if user A browses item C, the electronic device can build a node for user A, a node for item C, and build an edge between user A and item C. In this manner, the electronic device can construct an initial topology graph including multiple nodes and relationships between nodes.
  • the initial topology map can make the initial topology map include the latest information of each user, such as the item information recently clicked by the user, so that the timeliness of the initial topology map is better.
  • Fig. 6 is a schematic diagram of an initial topology diagram shown in an exemplary embodiment of the present application.
  • the initial topology graph includes a plurality of nodes, and the nodes are connected by edges, so as to represent that there is an association relationship between two nodes.
  • Step 502 determine the behavior subgraph of each node in the initial topology graph.
  • Step 503 Perform training according to the behavior subgraph to obtain a vector model corresponding to the preset topology graph.
  • the electronic device can train the initial topology graph to obtain a vector model. Specifically, the electronic device can learn the relationship between nodes, and then obtain the relationship between a node and its neighbor nodes.
  • the electronic device can obtain a corresponding behavior subgraph.
  • the behavior subgraph may include a central node and its neighbors. Relationships between these nodes can also be included.
  • Fig. 7 is a schematic diagram of a behavior subgraph shown in an exemplary embodiment of the present application.
  • the initial topology graph can be processed to collect the behavior subgraph of the central node 71 .
  • the behavior subgraph includes multiple adjacent nodes of the central node 71 .
  • the number of adjacent points n can also be set, and each time a behavior subgraph is collected, n adjacent points of the central node can be collected, thereby reducing the size of a single collected behavior subgraph.
  • each time a behavior subgraph is collected one behavior subgraph can be collected for each node, so as to obtain each behavior subgraph of each node.
  • the electronic device can train each collected behavioral subgraph to obtain a vector model corresponding to the preset topology graph.
  • the obtained vector model can learn the relationship between the nodes in each behavior subgraph, and specifically learn the relationship between the adjacent nodes and the central node.
  • steps 502 and 503 may be executed iteratively, so as to sample the behavior subgraph multiple times and perform training, so as to gradually optimize the vector model.
  • Step 504 acquiring a preset topology graph used to characterize the relationship between the user and the item, and update data used to update the preset topology graph.
  • step 504 is similar to that of step 301 .
  • the acquired preset topology map may be the initial topology map constructed in step 501, or may be a topology map obtained after updating the initial topology map.
  • Step 505 obtaining a vocabulary including vocabulary corresponding to each node in the preset topology graph.
  • a vocabulary list may be preset. For example, after the initial topology graph is constructed, the electronic device may determine a vocabulary according to each node in the initial topology graph, so as to obtain a vocabulary including vocabulary corresponding to each node. For example, if the node includes the node of item A, the electronic device may use the information of item A as an entry in the vocabulary.
  • Step 506 according to the vocabulary, filter out the data included in the update data that does not belong to the vocabulary in the vocabulary.
  • the electronic device can perform preliminary filtering on the update data, specifically, it can analyze the subject information included in each piece of update data, such as the operation behavior of subject A on subject B, and determine whether these subjects belong to the vocabulary in the vocabulary, if If not, the corresponding update data is filtered out.
  • the update data is filtered out.
  • Step 507 Update the edges in the preset topology graph according to the remaining update data to obtain an updated topology graph.
  • the subjects included in the remaining updated data after filtering all exist in the vocabulary, therefore, the subjects included in the remaining updated data have corresponding nodes in the preset topology graph, so the preset topology can be updated according to the updated data.
  • the electronic device may determine the relationship between the incremental nodes according to the remaining update data, and update the edges in the preset topology graph according to the relationship between the incremental nodes to obtain the updated topology graph.
  • edges can be added to the preset topology graph according to the updated data, so as to obtain an updated topology graph. For example, if the update data includes detailed information that user B visited item D, the electronic device may add an edge between the node of user B and the node of item D in the preset topology graph.
  • Step 508 obtaining a pre-trained vector model corresponding to a preset topology map.
  • Step 508 is similar to the method of obtaining the vector model in step 302, and will not be repeated here.
  • Step 509 determine the behavior subgraph of any node; wherein, the behavior subgraph includes any node and its corresponding neighbor nodes, and the behavior subgraph is used to represent the behavior of any node and its neighbors relation.
  • the edge relationship between nodes is added in the updated topology graph, therefore, the behavior subgraph of any node can be determined according to the updated topology graph, so as to collect the updated behavior subgraph of any node.
  • the behavior subgraph of each node can be updated, and the behavior subgraph of nodes whose relationship with neighboring nodes has changed can also be updated.
  • the updated behavior subgraph may include an intermediate node, and may also include adjacent nodes of the intermediate node.
  • the updated behavior subgraph of the same node may contain richer information than the behavior subgraph before the update.
  • the order can also be preset, and the neighbor nodes of any node can be determined based on the preset order, and the behavior subgraph of any node can be determined according to the correspondence between any node and its neighbor nodes.
  • the order i can be set, and the electronic device can take any node as the starting point, the i-th node connected to the any node, and the nodes between the any node and the i-th node as the neighbors of the node node.
  • Step 510 according to the behavior subgraph and vector model of any node, determine the update vector of any node.
  • the vector model is obtained by collecting behavioral subgraphs on the basis of the initial topology graph, and then training the behavioral subgraphs. Therefore, adding edges to the initial topology graph based on updated data will not have a great impact on the structure of the vector model itself.
  • the method provided by the present application utilizes the previously trained vector model to process the behavior subgraph, thereby updating the update vector of the central node in each behavior subgraph.
  • the electronic device may also encode the behavior subgraph of any node based on the vector model to obtain the vector code of any node.
  • the vector model includes multiple network structure parameters, and the electronic device can use the attributes of any node and the attributes of the neighbor nodes of any node as input information, and process the input information according to the network structure parameters included in the vector model , get the vector code of any node.
  • Any node can be used as the central node, so that the adjacent nodes of the central node can be aggregated according to the vector model to obtain the vector code of the central node.
  • the electronic device encodes the attribute of any node based on the vector model to obtain the attribute code of any node.
  • the vector model may include an attribute encoding network, and may also include parameters in the attribute encoding network, and then the attribute encoding process of any node may be encoded by using the attribute encoding network and its parameters to obtain an attribute encoding.
  • the electronic device can also use the vector model to encode the attribute code of any node according to the behavior subgraph of any node to obtain the vector code of any node.
  • the vector model can include the topological coding network, and can also include the parameters in the topological coding network, and then can use the topological coding network and its parameters, combined with the behavior subgraph of any node, to encode the vector coding of any node, so that Get the vector encoding of any node.
  • vector codes including node attributes and topology information of the nodes can be obtained.
  • the electronic device can process the vector encoding of any node based on the vector model to obtain the update vector of any node.
  • the vector model may also include parameters of a mapping function, and the mapping function may process the vector encoding of any node to obtain an update vector of the node.
  • the electronic device may determine the mapping function according to the vector model, and the vector model may include mapping function parameters, so that the electronic device may restore the mapping function according to the vector model.
  • the electronic device can determine the update vector of any node according to the mapping function and the vector code of any node. Specifically, the vector code can be used as the input data of the mapping function, and then the vector code can be mapped to the update vector of the node through the mapping function.
  • Step 511 according to the updated vectors of each node, determine the similarity between any user node and other item nodes; determine similar item nodes of the user node according to the similarity between the user node and each item node.
  • the electronic device may determine the similarity between the node and other nodes according to the updated vector of the node and the updated vectors of other nodes.
  • the electronic device can determine the similarity between the vector of user A's node and the vectors of other nodes, and then can determine similar nodes of user A's node.
  • the electronic device may determine the cosine value between the updated vectors of the two nodes as the similarity between the two nodes.
  • Step 512 recommending item information of similar item nodes to users corresponding to the user nodes.
  • the electronic device may recommend the item information of the item to the user represented by any node according to similar nodes of any node.
  • item information may be recommended to user A according to similar nodes of user A's node.
  • the similar node of user A's node is an item node
  • the item information of the item node can be directly recommended to user A.
  • the electronic device may recommend item information to user A according to the information of other user nodes, such as recommending item information of items that other users are interested in to user A.
  • Fig. 8 is a structural diagram of an apparatus for recommending item information to a user according to an exemplary embodiment of the present application.
  • the apparatus 800 for recommending item information to a user includes:
  • the obtaining unit 810 is configured to obtain a preset topological graph used to characterize the relationship between users and items, and update data used to update the preset topological graph; wherein, the preset topological graph includes nodes, the Nodes include user nodes and item nodes, and there is an association between user nodes and item nodes connected by edges;
  • a graph updating unit 820 configured to update edges in the preset topology graph according to the update data to obtain an updated topology graph
  • a vector updating unit 830 configured to obtain a pre-trained vector model corresponding to the preset topology map, and update any node in the topology map according to the updated topology map and the obtained vector model A vector, which is used to characterize the characteristics of the node;
  • the recommendation unit 840 is configured to determine similar item nodes of the user node according to the vectors of each node, and recommend item information of the similar item nodes to users corresponding to the user node.
  • Fig. 9 is a structural diagram of an apparatus for recommending item information to a user according to another exemplary embodiment of the present application.
  • the graph updating unit 820 includes:
  • a vocabulary acquiring module 821 configured to acquire a vocabulary of vocabulary corresponding to each node in the preset topology diagram
  • a filtering module 822 configured to filter out data included in the update data that do not belong to vocabulary in the vocabulary according to the vocabulary
  • An update module 823 configured to update edges in the preset topology graph according to the remaining update data, to obtain an updated topology graph.
  • the update module 823 is specifically configured to:
  • the incremental inter-node relationship is determined according to the remaining update data, and the edges in the preset topology graph are updated according to the incremental inter-node relationship to obtain an updated topology graph.
  • the vector updating unit 830 includes:
  • the subgraph determination module 831 is configured to determine the behavior subgraph of any node according to the updated topology diagram; wherein, the behavior subgraph includes any node and its corresponding neighbor nodes, and the behavior subgraph Used to characterize the relationship between any node and its adjacent points;
  • the vector update module 832 is configured to determine the update vector of any node according to the behavior subgraph of any node and the vector model.
  • the subgraph determination module 831 is specifically configured to determine the neighbor nodes of any node based on the preset order, and determine the behavior of any node according to the correspondence between any node and its neighbor nodes subplot.
  • the vector update module 832 is specifically configured to:
  • the vector encoding of any node is processed based on the vector model to obtain an update vector of any node.
  • the vector update module 832 is specifically configured to:
  • the attribute code of any node is coded according to the behavior subgraph of any node to obtain the vector code of any node.
  • the vector update module 832 is specifically configured to:
  • An update vector of any node is determined according to the mapping function and the vector code of any node.
  • the recommendation unit 840 includes:
  • the similarity determination module 841 is used to determine each similarity between any user node and other item nodes according to the updated vectors of each node;
  • the similar node determination module 842 is configured to determine similar item nodes of the user node according to the similarities between the user node and each item node.
  • the device further includes a preset unit 850, configured to:
  • the vector updating unit 830 is specifically configured to:
  • the preset unit 850 is also used to:
  • Training is performed according to the behavior subgraph to obtain a vector model corresponding to the preset topology graph.
  • Fig. 10 is a structural diagram of an electronic device shown in an exemplary embodiment of the present application.
  • the electronic equipment provided in this embodiment includes:
  • the computer program is stored in the memory 101 and is configured to be executed by the processor 102 to implement any method for recommending item information to a user as described above.
  • This embodiment also provides a computer-readable storage medium on which a computer program is stored,
  • the computer program is executed by the processor to implement any one of the methods for recommending item information to the user as described above.
  • This embodiment also provides a computer program, including program code.
  • the program code executes any method for recommending item information to the user as described above.
  • the aforementioned program can be stored in a computer-readable storage medium.
  • the program executes the steps including the above-mentioned method embodiments; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk and other various media that can store program codes.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • General Engineering & Computer Science (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (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

提供向用户推荐物品信息的方法、设备、存储介质及程序产品，涉及数据处理技术，包括：获取用于表征用户与物品间关联关系的预设拓扑图，以及用于更新预设拓扑图的更新数据；根据更新数据更新预设拓扑图中的边得到更新后的拓扑图；获取预先训练得到的与预设拓扑图对应的向量模型，并根据更新后的拓扑图、获取的向量模型，更新拓扑图中任一节点的向量，向量用于表征节点的特征；根据各节点的向量，确定用户节点的相似物品节点，并向用户节点对应的用户推荐相似物品节点的物品信息。提供的方案中，能够基于更新数据快速更新节点向量，进而可以基于更新后的节点向量更精准的向用户推荐信息。

Description

向用户推荐物品信息的方法、设备、存储介质及程序产品
本申请要求于2021年9月24日提交中国专利局、申请号为202111123077.4、申请名称为“向用户推荐物品信息的方法、设备、存储介质及程序产品”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
本申请涉及数据处理技术，尤其涉及一种向用户推荐物品信息的方法、设备、存储介质及程序产品。
背景技术
为了向用户提供更加丰富的网络信息，可以向用户主动推送信息，比如，可以推送商品信息，还可以推送用户可能感兴趣的资讯等。
为了能够向用户推送其感兴趣的网络信息，现有技术中存在个性化推荐方法，通过将用户与信息进行向量化表示，能够确定各个用户的感兴趣信息，进而向用户推荐。
为了记录用户与信息之间的关系，可以选择网络拓扑图的方式存储用户与信息间的关系。但是，图的复杂性往往使得实时推荐成为一个难点，因此，如何提高个性化推荐的时效性，是本领域技术人员亟需解决的技术问题。
发明内容
本申请提供一种向用户推荐物品信息的方法、设备、存储介质及程序产品，以解决现有技术中个性化推荐时效性差的问题。
本申请的第一个方面是提供一种用于向用户推荐物品信息的方法，包括：
获取用于表征用户与物品间关联关系的预设拓扑图，以及用于更新所述预设拓扑图的更新数据；其中，其中，所述预设拓扑图中包括节点，所述节点包括用户节点、物品节点，通过边连接的用户节点和物品节点之间具有关联关系；
根据所述更新数据更新所述预设拓扑图中的边得到更新后的拓扑图；
获取预先利用与所述预设拓扑图对应的初始拓扑图训练得到的向量模型，并根据更新后的所述拓扑图、获取的所述向量模型，更新所述拓扑图中任一节点的向量，所述向量用于表征节点的特征；
根据各所述节点的向量，确定用户节点的相似物品节点，并向所述用户节点对应的用户推荐所述相似物品节点的物品信息。
本申请的第二个方面是提供一种个性化推荐信息的装置，包括：
获取单元，用于获取用于表征用户与物品间关联关系的预设拓扑图，以及用于更新所述预设拓扑图的更新数据；其中，所述预设拓扑图中包括节点，所述 节点包括用户节点、物品节点，通过边连接的用户节点和物品节点之间具有关联关系；
图更新单元，用于根据所述更新数据更新所述预设拓扑图中的边得到更新后的拓扑图；
向量更新单元，用于获取预先训练得到的与所述预设拓扑图对应的向量模型，并根据更新后的所述拓扑图、获取的所述向量模型，更新所述拓扑图中任一节点的向量，所述向量用于表征节点的特征；
推荐单元，用于根据各所述节点的向量，确定用户节点的相似物品节点，并向所述用户节点对应的用户推荐所述相似物品节点的物品信息。本申请的又一个方面是提供一种电子设备，包括：
存储器；
处理器；以及
计算机程序；
其中，所述计算机程序存储在所述存储器中，并配置为由所述处理器执行以实现如上述第一方面所述的用于向用户推荐物品信息的方法。
本申请的又一个方面是提供一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器执行以实现如上述第一方面所述的用于向用户推荐物品信息的方法。
本申请的又一个方面是提供一种计算机程序产品，包括计算机程序，所述计算机程序在被处理器执行时实现如上述第一方面所述的用于向用户推荐物品信息的方法。
本申请提供的向用户推荐物品信息的方法、设备、存储介质及程序产品的技术效果是：
本申请提供的向用户推荐物品信息的方法、设备、存储介质及程序产品，包括：获取用于表征用户与物品间关联关系的预设拓扑图，以及用于更新预设拓扑图的更新数据；其中，预设拓扑图中包括节点，节点包括用户节点、物品节点，通过边连接的用户节点和物品节点之间具有关联关系；根据更新数据更新预设拓扑图中的边得到更新后的拓扑图；获取预先训练得到的与预设拓扑图对应的向量模型，并根据更新后的拓扑图、获取的向量模型，更新拓扑图中任一节点的向量，向量用于表征节点的特征；根据各节点的向量，确定用户节点的相似物品节点，并向用户节点对应的用户推荐相似物品节点的物品信息。本申请提供的方案中，可以预先设置预设拓扑图，以及与该预设拓扑图对应的向量模型，存在更新数据后，可以直接基于该更新数据更新预设拓扑图，并利用向量模型对更新后的拓扑图进行处理，得到节点更新后的向量。这种方法能够基于更新数据快速更新节点向量，进而可以基于更新后的节点向量更精准的向用 户推荐信息。
附图说明
图1为一示例性实施例示出的直推式图嵌入算法架构示意图；
图2为一示例性实施例示出的归纳式图嵌入算法架构示意图；
图3为本申请一示例性实施例示出的个性化推荐信息的方法的流程图；
图4为本申请一示例性实施例示出的采集行为子图的示意图；
图5为本申请另一示例性实施例示出的个性化推荐信息的方法的流程图；
图6为本申请一示例性实施例示出的初始拓扑图的示意图；
图7为本申请一示例性实施例示出的行为子图的示意图；
图8为本申请一示例性实施例示出的个性化推荐信息的装置的结构图；
图9为本申请另一示例性实施例示出的个性化推荐信息的装置的结构图；
图10为本申请一示例性实施例示出的电子设备的结构图。
具体实施方式
在个性化推荐的方案中，为达到近实时的推荐效果，在全量训练的基础上，需要根据用户新产生的交互行为，及时更新用户的向量结果，以便在进行推荐时结合用户的新近行为，作出分钟级别的效果反馈。
目前存在两类图嵌入方法，一类为直推式图嵌入算法，比如GCN(Graph Convolutional Network，图卷积网络)。另一类为归纳式图嵌入算法，比如GRAPHSAGE(Graph SAmple and aggreGatE)。
图1为一示例性实施例示出的直推式图嵌入算法架构示意图。
如1所示，首先收集全量离线数据，其中离线数据包含用户的行为，用户的属性和物品的属性等数据，通过图构建将用户和物品建模为图中节点。
用户的行为用来构建节点之间的连接边，用户属性和物品属性则是图中节点的特征。构建好全图之后，采用GCN等算法利用图的整个邻接矩阵和图卷积操作融合相邻节点的信息，生成节点的向量表示。
当有新的用户行为产生后，需要基于全量的离线数据以及增量数据，重新构建图，再基于图生产各个节点的向量表示。这种方式更新用户的向量表示速度慢，无法及时基于产生的增量数据快速调整图，因此，不能支持近实时的架构更新。
图2为一示例性实施例示出的归纳式图嵌入算法架构示意图。
如图2所示，首先收集全量离线数据，其中全量数据包含用户的行为，用户的属性和物品的属性等数据，通过图构建将用户和物品建模为图中节点。
以GRAPHSAGE为例，对构建的图进行训练，保存节点之间的映射关系。可基于映射函数对一个节点的邻节点进行处理，从而确定该节点的向量表达。但是归纳式图嵌入算法主要应用于全量结点的离线训练，无法直接基于归纳式图嵌入算法进行个性化的信息推荐。
为了解决上述技术问题，本申请提供一种用于向用户推荐物品信息的方案，本方 案中可以预先构建拓扑图，还可以预先训练得到向量模型，存在更新数据时，可以直接利用更新数据更新拓扑图，并基于向量模型对更新后的图进行处理，确定节点更新后的向量，这种方式能够根据更新数据快速更新节点的向量，进而能够根据更新后的向量向用户个性化的推荐信息，从而提高个性化推荐的时效性。
图3为本申请一示例性实施例示出的用于向用户推荐物品信息的方法的流程图。
如图3所示，本申请提供的用于向用户推荐物品信息的方法，包括：
步骤301，获取用于表征用户与物品间关联关系的预设拓扑图，以及用于更新预设拓扑图的更新数据；其中，预设拓扑图中包括节点，节点包括用户节点、物品节点，通过边连接的用户节点和物品节点之间具有关联关系。
其中，本申请提供的方法可以由具备计算能力的一台或多台电子设备执行，该电子设备例如可以作为服务器，比如可以是云服务器，还可以是分布式服务器等。具体的，该服务器例如可以提供信息推荐功能，例如，可以向用户推荐商品。
进一步的，可以在产生更新数据时，触发电子设备执行本实施例的方法，也可以每隔预设时长执行一次本实施例提供的方法。
实际应用时，电子设备可以获取用于表征用户与物品间关联关系的预设拓扑图，该预设拓扑图中包括多个节点，这些节点中包括用户节点，也包括物品节点，用户节点与物品节点之间具有边，从而通过边连接用户节点和物品节点。
其中，可以预先基于离线数据构建包括多个节点的拓扑图。比如一条离线数据中包括用户A访问了信息B，则可以构建节点A、节点B，还可以构建A与B之间的边，用来表征A和B之间具有关联关系。
具体的，还可以根据需求设置边的权重值。
进一步的，电子设备还可以获取更新数据，该更新数据可以包括用户行为数据。比如用户浏览了某个商品，或者访问了某条信息，则电子设备可以获取到相关信息。
实际应用时，获取的更新数据是用于更新预设拓扑图的数据。
步骤302，根据更新数据更新预设拓扑图中的边得到更新后的拓扑图。
实际应用时，电子设备可以根据获取的更新数据对获取的预设拓扑图进行更新，得到更新后的拓扑图。
在一种实施方式中，更新后的拓扑图可以作为预设拓扑图，被再次获取。
其中，电子设备可以根据更新数据在预设拓扑图中添加边，以更新拓扑图，还可以更新边的权重值，进而更新拓扑图。比如，更新数据中包括用户A浏览了商品C，则可以构建A与C之间的边，进而更新预设拓扑图。
步骤303，获取预先训练得到的与预设拓扑图对应的向量模型，并根据更新后的拓扑图、获取的向量模型，更新拓扑图中任一节点的向量，向量用于表征节点的特征。
具体的，还可以预先利用与预设拓扑图对应的初始拓扑图训练向量模型，预设拓扑图可以与初始拓扑图相同，也可以是在初始拓扑图的基础上进行一次或多次更新得到的。
比如，在基于离线数据构建出拓扑图后，可以基于该拓扑图进行训练，得到向量模型。由于预设拓扑图也是对离线数据进行处理得到的，因此，该预设拓扑图与上述 向量模型具有对应关系。
进一步的，可以采用GRAPHSAGE算法对初始拓扑图进行训练，得到向量模型。具体可以针对每一节点采集其行为子图，该行为子图中包括该节点及其邻节点，还包括这些节点之间的关系。比如针对节点P1、P2、P3都能够采集相应的行为子图，电子设备可以学习这些子图中邻节点与中心节点之间的关系，从而得到向量模型。
实际应用时，电子设备更新完预设拓扑图之后，还可以获取与该预设拓扑图对应的向量模型，进而可以根据更新后的拓扑图、获取的向量模型，更新拓扑图中任一节点的向量。其中，节点的向量可以表征该节点的特征，具体可以根据该节点本身的信息，以及该节点的邻节点的信息生成该节点的向量。
具体的，更新后的拓扑图中节点间的关系发生了变化，因此，可以重新采集各个节点的行为子图，也可以仅采集与邻节点之间发生边变化的节点的行为子图。
图4为本申请一示例性实施例示出的采集行为子图的示意图。
如图4所示，例如存在如图41所示的拓扑图结构，其中的节点42的行为子图为43所示。例如更新后的拓扑图为44，那么节点42的行为子图为45所示。
一种实施方式中，另一节点46的行为子图则不发生变化，因此，可以不更新节点46的向量，进而也不需要采集节点46的行为子图。
进一步的，可以根据节点更新后的行为子图以及向量模型，确定该节点更新后的向量。比如，可以将该节点以及该节点的邻节点信息输入到向量模型中，使得向量模型输出该节点更新后的向量。
步骤304，根据各节点的向量，确定用户节点的相似物品节点，并向用户节点对应的用户推荐相似物品节点的物品信息。
实际应用时，电子设备可以根据各节点的向量，确定节点间的相似节点。比如，在需要向用户A推荐信息时，则可以确定用户A的节点与其他节点之间的相似度，进而得到用户A的节点的相似节点。
其中，电子设备还可以根据任一节点的相似节点，向任一节点推荐物品的物品信息。比如，用户A的节点与物品D的节点相似度高，则可以将物品D的信息推荐给用户A。再比如，用户A的节点与用户E的节点相似度高，则可以根据用户E的信息，向用户A推荐物品信息。
本实施例提供的方法用于向用户推荐物品信息，该方法由设置有本实施例提供的方法的设备执行，该设备通常以硬件和/或软件的方式来实现。
本实施例提供的用于向用户推荐物品信息的方法，包括：获取用于表征用户与物品间关联关系的预设拓扑图，以及用于更新预设拓扑图的更新数据；其中，预设拓扑图中包括节点，节点包括用户节点、物品节点，通过边连接的用户节点和物品节点之间具有关联关系；根据更新数据更新预设拓扑图中的边得到更新后的拓扑图；获取预先训练得到的与预设拓扑图对应的向量模型，并根据更新后的拓扑图、获取的向量模型，更新拓扑图中任一节点的向量，向量用于表征节点的特征；根据各节点的向量，确定用户节点的相似物品节点，并向用户节点对应的用户推荐相似物品节点的物品信息。本申请提供的方法中，可以预先设置预设拓扑图，以及与该预设拓扑图对应的向量模型，存在更新数据后，可以直接基于该更新数据更新预设拓扑图，并利用向量模 型对更新后的拓扑图进行处理，得到节点更新后的向量。这种方法能够基于更新数据快速更新节点向量，进而可以基于更新后的节点向量更精准的向用户推荐物品信息。
图5为本申请另一示例性实施例示出的用于向用户推荐物品信息的方法的流程图。
如图5所示，本申请提供的用于向用户推荐物品信息的方法，包括：
步骤501，获取预设的离线数据，根据离线数据构建初始拓扑图。
其中，可以预先准备离线数据，比如，可以准备最近7天的离线数据，并根据这些离线数据构建初始拓扑图。构建的初始拓扑图中包括多个节点，比如可以包括用户节点，还可以包括物品节点，这些节点之间具有边，通过边表征两个节点之间具有关联关系。比如，用户节点A和用户节点B之间可以具有边，表征用户A和用户B之间具有关联关系，再比如，用户节点A和物品节点C之间具有边，则表征用户A和物品C之间具有关联关系。
具体的，可以解析每条离线数据，从中获取行为数据。比如，用户A浏览了物品C，则电子设备可以构建用户A的节点、物品C的节点，并构建用户A与物品C之间的边。通过这种方式，电子设备能够构建出包括多个节点以及节点之间关系的初始拓扑图。
进一步的，利用最近离线数据构建该初始拓扑图，能够使得初始拓扑图中包括各用户最近的信息，例如最近用户点击的物品信息从而使得初始拓扑图的时效性更好。
图6为本申请一示例性实施例示出的初始拓扑图的示意图。
如图6所示，在初始拓扑图中包括多个节点，通过边连接节点，从而表征两个节点之间具有关联关系。
步骤502，在初始拓扑图中确定每一节点的行为子图。
步骤503，根据行为子图进行训练，得到与预设拓扑图对应的向量模型。
实际应用时，构建出初始拓扑图之后，电子设备可以对该初始拓扑图进行训练，得到向量模型。电子设备具体可以学习节点之间的关系，进而得到一个节点与该节点的邻节点之间的关系。
其中，针对初始拓扑图中的每个节点，电子设备都可以获取相应的行为子图。行为子图中可以包括中心节点，以及该中心节点的邻接点。还可以包括这些节点之间的关系。
图7为本申请一示例性实施例示出的行为子图的示意图。
如图7所示，可以对初始拓扑图进行处理，采集中心节点71的行为子图。该行为子图中包括中心节点71的多个邻节点。
具体的，还可以预先设置阶数k，从而基于k确定每个中心节点的邻接点。比如，若k＝1，则将直接与中心节点连接的节点，确定为该中心节点的邻节点。若k＝2，则将直接与中心节点连接的第一节点，以及直接与第一节点连接的第二节点，均确定为该中心节点的邻节点。
进一步的，还可以设置邻接点数量n，每次采集行为子图，可以采集中心节点的n个邻接点，从而降低单次采集的行为子图的尺寸。
实际应用时，每次采集行为子图时，针对每一节点都可以采集一个行为子图，从 而得到每一节点的每一行为子图。
其中，电子设备可以对采集到的每一行为子图进行训练，得到预设拓扑图对应的向量模型。得到的向量模型能够学习到各个行为子图中的节点间的关系，具体学习到邻接点与中心节点之间的关系。
具体的，可以迭代执行步骤502、503，从而多次采样行为子图，并进行训练，从而逐步的优化向量模型。
步骤504，获取用于表征用户与物品间关联关系的预设拓扑图，以及用于更新预设拓扑图的更新数据。
步骤504与步骤301的实现方式相似。
其中，获取的预设拓扑图可以是步骤501中构建的初始拓扑图，还可以是对初始拓扑图进行更新后得到的拓扑图。
步骤505，获取包括预设拓扑图中各节点对应的词汇的词汇表。
具体的，可以预先设置词汇表。比如，在构建了初始拓扑图之后，电子设备可以根据初始拓扑图中各个节点确定词汇表，从而得到包括各节点对应的词汇的词汇表。比如，节点包括物品A的节点，则电子设备可以将物品A的信息作为词汇表中的一个词条。
步骤506，根据词汇表，过滤掉更新数据中包括的不属于词汇表中的词汇的数据。
进一步的，电子设备可以对更新数据进行初步过滤，具体可以分析每条更新数据中包括的主体信息，比如主体A对主体B的操作行为，并确定这些主体是否是属于词汇表中的词汇，若不是，则过滤掉相应的更新数据。
比如，更新数据中包括主体M，但是词汇表中不包括M对应的词汇，则过滤掉该更新数据。
步骤507，根据剩余的更新数据更新预设拓扑图中的边，得到更新后的拓扑图。
实际应用时，过滤后剩余的更新数据中包括的主体均存在于词汇表中，因此，剩余的更新数据中包括的主体在预设拓扑图中存在相应的节点，因此，可以根据更新数据更新预设拓扑图。
其中，电子设备可以根据剩余的更新数据确定增量节点间关系，并根据增量节点间关系更新预设拓扑图中的边得到更新后的拓扑图。
具体可以根据更新数据在预设拓扑图中添加边，进而得到更新后的拓扑图。比如，更新数据中包括用户B访问了物品D的详细信息，则电子设备可以在预设拓扑图中添加用户B的节点与物品D的节点之间的边。
步骤508，获取预先训练得到的与预设拓扑图对应的向量模型。
步骤508与步骤302中获取向量模型的方式相似，不再赘述。
步骤509，根据更新后的拓扑图，确定任一节点的行为子图；其中，行为子图中包括任一节点及其对应的邻节点，行为子图用于表征任一节点及其邻接点的关系。
进一步的，更新后的拓扑图中添加了节点间的边关系，因此，可以根据更新后的拓扑图，确定任一节点的行为子图，从而采集任一节点更新后的行为子图。
实际应用时，可以更新每个节点的行为子图，还可以更新与邻节点关系改变的节点的行为子图。
其中，更新后的行为子图中可以包括中间节点，还可以包括该中间节点的邻节点。同一节点更新后的行为子图可能比更新前的行为子图中包括的信息更丰富。
具体的，还可以预先设置阶数，并基于预设阶数确定任一节点的邻节点，并根据任一节点及其邻节点的对应关系，确定任一节点的行为子图。比如，可以设置阶数i，电子设备可以将以任一节点为起点，与该任一节点连接的第i个节点，以及该任一节点与第i个节点之间的节点作为该节点的邻节点。
步骤510，根据任一节点的行为子图、向量模型，确定任一节点的更新向量。
具体的，向量模型是在初始拓扑图基础上采集行为子图，再对行为子图进行训练得到的。因此，基于更新数据在初始拓扑图中添加边后，对向量模型本身的结构不会影响很大。基于此，本申请提供的方法利用此前训练完毕的向量模型，对行为子图进行处理，从而更新每一行为子图中中心节点的更新向量。
进一步的，电子设备还可以基于向量模型对任一节点的行为子图进行编码处理，得到任一节点的向量编码。向量模型中包括多个网络结构的参数，电子设备可以将任一节点的属性，以及该任一节点的邻节点的属性作为输入信息，并根据向量模型中包括的网络结构参数对输入信息进行处理，得到任一节点的向量编码。可以将任一节点作为中心节点，从而可以根据向量模型对该中心节点的邻节点进行聚合，以得到中心节点的向量编码。
实际应用时，电子设备基于向量模型对任一节点的属性进行编码，得到任一节点的属性编码。向量模型中可以包括属性编码网络，还可以包括该属性编码网络中的参数，进而可以利用属性编码网络及其参数，对任一节点的属性进行编码处理，得到属性编码。
电子设备还可以利用向量模型，根据任一节点的行为子图对任一节点的属性编码进行编码，得到任一节点的向量编码。向量模型中可以包括拓扑编码网络，还可以包括该拓扑编码网络中的参数，进而可以利用拓扑编码网络及其参数，结合任一节点的行为子图，对任一节点的向量编码进行编码，从而得到任一节点的向量编码。
通过这种方式，能够得到即包括节点属性，以及该节点拓扑结构信息的向量编码。
实际应用时，电子设备可以基于向量模型对任一节点的向量编码进行处理，得到任一节点的更新向量。其中，向量模型中还可以包括映射函数的参数，映射函数可以对任一节点的向量编码进行处理，得到该节点的更新向量。
其中，电子设备可以根据向量模型确定映射函数，向量模型中可以包括映射函数参数，使得电子设备可以根据向量模型还原出映射函数。电子设备可以根据映射函数、任一节点的向量编码，确定任一节点的更新向量。具体可以将向量编码作为映射函数的输入数据，进而通过映射函数能够将向量编码映射为该节点的更新向量。
步骤511，根据更新后的各节点的向量，确定任一用户节点与其他各物品节点之间的各个相似度；根据用户节点与各物品节点间的各个相似度，确定用户节点的相似物品节点。
具体的，在需要为与一个节点对应的用户推荐信息后，电子设备可以根据该节点更新后的向量，以及其他节点更新后的向量，确定该节点与其他各节点之间的相似度。
比如，需要向用户A推荐信息，则电子设备可以确定用户A节点的向量与其他各 节点的向量之间的相似度，进而可以确定出用户A节点的相似节点。
进一步的，电子设备可以确定两个节点更新后的向量间的余弦值，将其作为这两个节点的相似度。
步骤512，向用户节点对应的用户推荐相似物品节点的物品信息。
电子设备可以根据任一节点的相似节点，向该任一节点代表的用户推荐物品的物品信息。
其中，比如可以根据用户A节点的相似节点向用户A推荐物品信息。比如，用户A节点的相似节点是物品节点，则可以直接向用户A推荐该物品节点的物品信息。再比如，用户A节点的相似节点是其他用户节点，则电子设备可以根据其他用户节点的信息，向用户A推荐物品信息，比如向用户A推荐该其他用户感兴趣的物品的物品信息。
图8为本申请一示例性实施例示出的用于向用户推荐物品信息的装置的结构图。
如图8所示，本实施例提供的用于向用户推荐物品信息的装置800，包括：
获取单元810，用于获取用于表征用户与物品间关联关系的预设拓扑图，以及用于更新所述预设拓扑图的更新数据；其中，所述预设拓扑图中包括节点，所述节点包括用户节点、物品节点，通过边连接的用户节点和物品节点之间具有关联关系；
图更新单元820，用于根据所述更新数据更新所述预设拓扑图中的边得到更新后的拓扑图；
向量更新单元830，用于获取预先训练得到的与所述预设拓扑图对应的向量模型，并根据更新后的所述拓扑图、获取的所述向量模型，更新所述拓扑图中任一节点的向量，所述向量用于表征节点的特征；
推荐单元840，用于根据各所述节点的向量，确定用户节点的相似物品节点，并向用户节点对应的用户推荐所述相似物品节点的物品信息。
本申请提供的个性化推荐信息的装置的实现方式、原理、效果与图3所示施例类似，不再赘述。
图9为本申请另一示例性实施例示出的用于向用户推荐物品信息的装置的结构图。
如图9所示，本申请提供的用于向用户推荐物品信息的装置900，所述图更新单元820，包括：
词表获取模块821，用于获取与所述预设拓扑图中各节点对应的词汇的词汇表；
过滤模块822，用于根据所述词汇表，过滤掉所述更新数据中包括的不属于所述词汇表中的词汇的数据；
更新模块823，用于根据剩余的更新数据更新所述预设拓扑图中的边，得到更新后的拓扑图。
在一种可选的实施方式中，所述更新模块823具体用于：
根据剩余的更新数据确定增量节点间关系，并根据增量节点间关系更新所述预设拓扑图中的边得到更新后的拓扑图。
在一种可选的实施方式中，所述向量更新单元830包括：
子图确定模块831，用于根据更新后的所述拓扑图，确定任一节点的行为子图；其中，所述行为子图中包括任一节点及其对应的邻节点，所述行为子图用于表征所述任一节点及其邻接点的关系；
向量更新模块832，用于根据任一节点的行为子图、所述向量模型，确定任一节点的更新向量。
在一种可选的实施方式中，子图确定模块831具体用于基于预设阶数确定任一节点的邻节点，并根据任一节点及其邻节点的对应关系，确定任一节点的行为子图。
在一种可选的实施方式中，所述向量更新模块832具体用于：
基于所述向量模型对任一节点的行为子图进行编码处理，得到任一节点的向量编码；
基于所述向量模型对所述任一节点的向量编码进行处理，得到任一节点的更新向量。
在一种可选的实施方式中，所述向量更新模块832具体用于：
基于所述向量模型对任一节点的属性进行编码，得到任一节点的属性编码；
利用所述向量模型，根据任一节点的行为子图对所述任一节点的属性编码进行编码，得到任一节点的向量编码。
在一种可选的实施方式中，所述向量更新模块832具体用于：
根据所述向量模型确定映射函数；
根据所述映射函数、所述任一节点的向量编码，确定任一节点的更新向量。
在一种可选的实施方式中，所述推荐单元840包括：
相似度确定模块841，用于根据更新后的各节点的向量，确定任一用户节点与其他各物品节点之间的各个相似度；
相似节点确定模块842，用于根据所述用户节点与各物品节点间的各个相似度，确定所述用户节点的相似物品节点。
在一种可选的实施方式中，所述装置还包括预设单元850，用于：
获取预设的离线数据，根据所述离线数据构建初始拓扑图；
相应的，所述向量更新单元830具体用于：
获取所述初始拓扑图。
在一种可选的实施方式中，所述预设单元850还用于：
在所述初始拓扑图中确定每一节点的行为子图；
根据所述行为子图进行训练，得到与所述预设拓扑图对应的向量模型。
图10为本申请一示例性实施例示出的电子设备的结构图。
如图10所示，本实施例提供的电子设备包括：
存储器101；
处理器102；以及
计算机程序；
其中，所述计算机程序存储在所述存储器101中，并配置为由所述处理器102执行以实现如上所述的任一种用于向用户推荐物品信息的方法。
本实施例还提供一种计算机可读存储介质，其上存储有计算机程序，
所述计算机程序被处理器执行以实现如上所述的任一种用于向用户推荐物品信息的方法。
本实施例还提供一种计算机程序，包括程序代码，当计算机运行所述计算机程序时，所述程序代码执行如上所述的任一种用于向用户推荐物品信息的方法。
本领域普通技术人员可以理解：实现上述各方法实施例的全部或部分步骤可以通过程序指令相关的硬件来完成。前述的程序可以存储于一计算机可读取存储介质中。该程序在执行时，执行包括上述各方法实施例的步骤；而前述的存储介质包括：ROM、RAM、磁碟或者光盘等各种可以存储程序代码的介质。
最后应说明的是：以上各实施例仅用以说明本申请的技术方案，而非对其限制；尽管参照前述各实施例对本申请进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分或者全部技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本申请各实施例技术方案的范围。

Claims (15)

  1. 一种用于向用户推荐物品信息的方法，其特征在于，包括：
    获取用于表征用户与物品间关联关系的预设拓扑图，以及用于更新所述预设拓扑图的更新数据；其中，所述预设拓扑图中包括节点，所述节点包括用户节点、物品节点，通过边连接的用户节点和物品节点之间具有关联关系；
    根据所述更新数据更新所述预设拓扑图中的边得到更新后的拓扑图；
    获取预先训练得到的与所述预设拓扑图对应的向量模型，并根据更新后的所述拓扑图、获取的所述向量模型，更新所述拓扑图中任一节点的向量，所述向量用于表征节点的特征；
    根据各所述节点的向量，确定用户节点的相似物品节点，并向所述用户节点对应的用户推荐所述相似物品节点的物品信息。
  2. 根据权利要求1所述的方法，其特征在于，所述根据所述更新数据更新所述预设拓扑图中的边得到更新后的拓扑图，包括：
    获取与所述预设拓扑图中各节点对应的词汇的词汇表；
    根据所述词汇表，过滤掉所述更新数据中包括的不属于所述词汇表中的词汇的数据；
    根据剩余的更新数据更新所述预设拓扑图中的边，得到更新后的拓扑图。
  3. 根据权利要求2所述的方法，其特征在于，所述根据剩余的更新数据更新所述预设拓扑图中的边，得到更新后的拓扑图，包括：
    根据剩余的更新数据确定增量节点间关系，并根据增量节点间关系更新所述预设拓扑图中的边得到更新后的拓扑图。
  4. 根据权利要求1所述的方法，其特征在于，所述根据更新后的所述拓扑图、获取的所述向量模型，更新所述拓扑图中任一节点的向量，包括：
    根据更新后的所述拓扑图，确定任一节点的行为子图；其中，所述行为子图中包括任一节点及其对应的邻节点，所述行为子图用于表征所述任一节点及其邻接点的关系；
    根据任一节点的行为子图、所述向量模型，确定任一节点的更新向量。
  5. 根据权利要求4所述的方法，其特征在于，所述根据更新后的所述拓扑图，确定任一节点对应的行为子图，包括：
    基于预设阶数确定任一节点的邻节点，并根据任一节点及其邻节点的对应关系，确定任一节点的行为子图。
  6. 根据权利要求4所述的方法，其特征在于，所述根据任一节点的行为子图、所述向量模型，确定任一节点的更新向量，包括：
    基于所述向量模型对任一节点的行为子图进行编码处理，得到任一节点的向量编码；
    基于所述向量模型对所述任一节点的向量编码进行处理，得到任一节点的更新向量。
  7. 根据权利要求6所述的方法，其特征在于，所述基于所述向量模型对任一节点 的行为子图进行编码处理，得到任一节点的向量编码，包括：
    基于所述向量模型对任一节点的属性进行编码，得到任一节点的属性编码；
    利用所述向量模型，根据任一节点的行为子图对所述任一节点的属性编码进行编码，得到任一节点的向量编码。
  8. 根据权利要求6所述的方法，其特征在于，所述基于所述向量模型对所述任一节点的向量编码进行处理，得到任一节点的更新向量，包括：
    根据所述向量模型确定映射函数；
    根据所述映射函数、所述任一节点的向量编码，确定任一节点的更新向量。
  9. 根据权利要求1所述的方法，其特征在于，所述根据更新后的各节点的向量，确定用户节点的相似物品节点，包括：
    根据更新后的各节点的向量，确定任一用户节点与其他各物品节点之间的各个相似度；
    根据所述用户节点与各物品节点间的各个相似度，确定所述用户节点的相似物品节点。
  10. 根据权利要求1-9任一项所述的方法，其特征在于，所述预设拓扑图包括初始拓扑图；
    所述方法还包括：
    获取预设的离线数据，根据所述离线数据构建初始拓扑图；
    相应的，所述获取预设拓扑图包括：获取所述初始拓扑图。
  11. 根据权利要求10所述的方法，其特征在于，还包括：
    在所述初始拓扑图中确定每一节点的行为子图；
    根据所述行为子图进行训练，得到与所述预设拓扑图对应的向量模型。
  12. 一种用于向用户推荐物品信息的装置，其特征在于，包括：
    获取单元，用于获取用于表征用户与物品间关联关系的预设拓扑图，以及用于更新所述预设拓扑图的更新数据；其中，所述预设拓扑图中包括节点，所述节点包括用户节点、物品节点，通过边连接的用户节点和物品节点之间具有关联关系；
    图更新单元，用于根据所述更新数据更新所述预设拓扑图中的边得到更新后的拓扑图；
    向量更新单元，用于获取预先训练得到的与所述预设拓扑图对应的向量模型，并根据更新后的所述拓扑图、获取的所述向量模型，更新所述拓扑图中任一节点的向量，所述向量用于表征节点的特征；
    推荐单元，用于根据各所述节点的向量，确定用户节点的相似物品节点，并向所述用户节点对应的用户推荐所述相似物品节点的物品信息。
  13. 一种电子设备，其特征在于，包括：
    存储器；
    处理器；以及
    计算机程序；
    其中，所述计算机程序存储在所述存储器中，并配置为由所述处理器执行以实现如权利要求1-11任一种所述的方法。
  14. 一种计算机可读存储介质，其特征在于，其上存储有计算机程序，
    所述计算机程序被处理器执行以实现如权利要求1-11任一种所述的方法。
  15. 一种计算机程序产品，包括计算机程序，所述计算机程序在被处理器执行时实现根据权利要求1-11中任一项所述的方法。
PCT/CN2022/095341 2021-09-24 2022-05-26 向用户推荐物品信息的方法、设备、存储介质及程序产品 Ceased WO2023045378A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111123077.4A CN113886686B (zh) 2021-09-24 2021-09-24 向用户推荐物品信息的方法、设备、存储介质及程序产品
CN202111123077.4 2021-09-24

Publications (1)

Publication Number Publication Date
WO2023045378A1 true WO2023045378A1 (zh) 2023-03-30

Family

ID=79006606

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/095341 Ceased WO2023045378A1 (zh) 2021-09-24 2022-05-26 向用户推荐物品信息的方法、设备、存储介质及程序产品

Country Status (2)

Country Link
CN (1) CN113886686B (zh)
WO (1) WO2023045378A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113886686B (zh) * 2021-09-24 2025-08-22 北京沃东天骏信息技术有限公司 向用户推荐物品信息的方法、设备、存储介质及程序产品
CN115526474A (zh) * 2022-09-20 2022-12-27 中国平安财产保险股份有限公司 派工方法、装置、计算机设备及存储介质

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140136549A1 (en) * 2012-11-14 2014-05-15 Homer Tlc, Inc. System and method for automatic product matching
CN110472087A (zh) * 2019-08-07 2019-11-19 腾讯科技（深圳）有限公司 一种表情图像推荐方法、装置、设备及介质
US20200210833A1 (en) * 2018-12-28 2020-07-02 Visa International Service Association Method, System, and Computer Program Product for Determining Relationships of Entities Associated with Interactions
CN111460232A (zh) * 2020-03-17 2020-07-28 中国平安人寿保险股份有限公司 功能模块的查找方法、装置、终端和计算机可读存储介质
US20200250734A1 (en) * 2019-02-01 2020-08-06 Target Brands, Inc. Item recommendations using convolutions on weighted graphs
CN112416579A (zh) * 2020-11-10 2021-02-26 电子科技大学 一种时间敏感的多方数据融合激励方法
KR20210053823A (ko) * 2020-05-27 2021-05-12 베이징 바이두 넷컴 사이언스 앤 테크놀로지 코., 엘티디. 멀티미디어 리소스 추천 방법, 장치, 전자 기기 및 저장 매체
CN113886686A (zh) * 2021-09-24 2022-01-04 北京沃东天骏信息技术有限公司 向用户推荐物品信息的方法、设备、存储介质及程序产品

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109242633B (zh) * 2018-09-20 2022-04-08 创新先进技术有限公司 一种基于二部图网络的商品推送方法和装置
CN111651671B (zh) * 2020-05-27 2023-11-21 腾讯科技（深圳）有限公司 用户对象推荐方法、装置、计算机设备和存储介质
CN112115377B (zh) * 2020-09-11 2022-05-27 安徽农业大学 一种基于社交关系的图神经网络链路预测推荐方法
CN113077319B (zh) * 2021-04-19 2024-07-19 北京沃东天骏信息技术有限公司 针对微详情页面的动态推荐方法及装置
CN113220866B (zh) * 2021-04-28 2023-01-06 西安电子科技大学 基于用户社交网络和产品相似度的推荐方法

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140136549A1 (en) * 2012-11-14 2014-05-15 Homer Tlc, Inc. System and method for automatic product matching
US20200210833A1 (en) * 2018-12-28 2020-07-02 Visa International Service Association Method, System, and Computer Program Product for Determining Relationships of Entities Associated with Interactions
US20200250734A1 (en) * 2019-02-01 2020-08-06 Target Brands, Inc. Item recommendations using convolutions on weighted graphs
CN110472087A (zh) * 2019-08-07 2019-11-19 腾讯科技（深圳）有限公司 一种表情图像推荐方法、装置、设备及介质
CN111460232A (zh) * 2020-03-17 2020-07-28 中国平安人寿保险股份有限公司 功能模块的查找方法、装置、终端和计算机可读存储介质
KR20210053823A (ko) * 2020-05-27 2021-05-12 베이징 바이두 넷컴 사이언스 앤 테크놀로지 코., 엘티디. 멀티미디어 리소스 추천 방법, 장치, 전자 기기 및 저장 매체
CN112416579A (zh) * 2020-11-10 2021-02-26 电子科技大学 一种时间敏感的多方数据融合激励方法
CN113886686A (zh) * 2021-09-24 2022-01-04 北京沃东天骏信息技术有限公司 向用户推荐物品信息的方法、设备、存储介质及程序产品

Also Published As

Publication number Publication date
CN113886686B (zh) 2025-08-22
CN113886686A (zh) 2022-01-04

Similar Documents

Publication Publication Date Title
CN105706078B (zh) 实体集合的自动定义
CN111723292B (zh) 基于图神经网络的推荐方法、系统、电子设备及存储介质
CN110704411A (zh) 适用于艺术领域的知识图谱搭建方法及装置、电子设备
CN113254630B (zh) 一种面向全球综合观测成果的领域知识图谱推荐方法
US9594756B2 (en) Automated ranking of contributors to a knowledge base
US10268749B1 (en) Clustering sparse high dimensional data using sketches
CN117216405A (zh) 基于知识空间表示的个性化学习资源推荐方法
CN113744023B (zh) 一种基于图卷积网络的双通道协同过滤推荐方法
CN112446739B (zh) 一种基于分解机和图神经网络的点击率预测方法及系统
CN105488211A (zh) 基于特征分析的用户群确定方法
CN110196849B (zh) 基于大数据治理技术实现用户画像构建处理的系统及其方法
CN110069713B (zh) 一种基于用户上下文感知的个性化推荐方法
US12596888B2 (en) Contextualization of generative language models based on entity resource identifiers
WO2024099037A1 (zh) 数据处理、实体链接方法、装置和计算机设备
WO2023045378A1 (zh) 向用户推荐物品信息的方法、设备、存储介质及程序产品
CN120494960A (zh) 基于知识图谱推理的金融业务合规风险识别方法
CN120780915A (zh) 一种教育资源序列推荐方法及相关装置
WO2020147259A1 (zh) 一种用户画像方法、装置、可读存储介质及终端设备
CN115495471B (zh) 基于用户特征进行hql执行语句预测的方法及其相关设备
US11675766B1 (en) Scalable hierarchical clustering
CN114443671A (zh) 推荐模型的更新方法、装置、计算机设备和存储介质
CN114648121A (zh) 一种数据处理方法、装置、电子设备和存储介质
CN109828965B (zh) 一种数据处理的方法及电子设备
CN111966889A (zh) 一种图嵌入向量的生成方法以及推荐网络模型的生成方法
CN116467466B (zh) 基于知识图谱的编码推荐方法、装置、设备及介质

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: 22871439

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 050724)

122 Ep: pct application non-entry in european phase

Ref document number: 22871439

Country of ref document: EP

Kind code of ref document: A1