Association object identification method and apparatus, electronic device, and storage medium

Abstract

The application discloses a kind of associated object identification method, device, electronic equipment and storage medium, comprising: obtaining target object to be identified;According to the scope of operation information of the target object and the logistics information of the target object, determine the actual business information of the target object;According to the actual business information of the target object, and the actual business information of the first candidate object of preset, from the first candidate object, select the upstream object and / or downstream object associated with the target object from the first candidate object.This method can correct the scope of operation information of the target object by the logistics information of the target object, obtain more accurate actual business information, and identify the associated object according to the more accurate actual business information, and then the upstream object and / or downstream object associated with the target object is more accurate.On the other hand, the data used by the present method is more common scope of operation information and logistics information, so it is more versatile.

Description

Technical Field

[0001] This application relates to the field of information recognition technology, specifically to a method, apparatus, electronic device, and storage medium for identifying associated objects. Background Technology

[0002] Related object identification is a relatively new technological field used to identify relationships between multiple objects. For example, it can be used to recommend potential trading partners to businesses, or to identify potential trading partners during transactions to determine if there are any risks involved.

[0003] However, while current methods for identifying related objects can perform simple identification, their accuracy is low and they are prone to errors. Summary of the Invention

[0004] This application provides a method, apparatus, electronic device, and storage medium for identifying associated objects, aiming to solve the problem of low recognition accuracy in current methods for identifying associated objects.

[0005] Firstly, this application provides a method for identifying associated objects, including:

[0006] Obtain the target object to be identified;

[0007] Based on the target object's business scope information and logistics information, the target object's actual business information is determined;

[0008] Based on the actual operating information of the target object and the actual operating information of the preset first candidate object, select upstream and / or downstream objects associated with the target object from the first candidate object.

[0009] In one possible implementation of this application, determining the actual operating information of the target object based on the target object's business scope information and logistics information includes:

[0010] Obtain multiple second candidate objects that have the same industry information as the target object;

[0011] Extract the first keyword information from the business scope information of the multiple second candidate objects, which appears more than a preset frequency threshold;

[0012] By using a preset association recognition model, the logistics information, business scope information, and first keyword information of the target object are processed to obtain the first target coefficient corresponding to the target object;

[0013] Based on the first target coefficient, the first keyword information is weighted and then fused with the business scope information of the target object to obtain the actual business information of the target object;

[0014] Before selecting upstream and / or downstream objects associated with the target object from the first candidate objects based on the actual operating information of the target object and the actual operating information of the preset first candidate objects, the method further includes:

[0015] Obtain multiple third candidate objects that have the same industry information as the first candidate object;

[0016] Extract second keyword information from the business scope information of the multiple third candidate objects, where the number of occurrences exceeds a preset threshold.

[0017] By using a preset association recognition model, the logistics information, business scope information, and second keyword information of the first candidate object are processed to obtain the second target coefficient corresponding to the first candidate object.

[0018] Based on the second target coefficient, the second keyword information is weighted and then fused with the business scope information of the first candidate to obtain the actual business information of the first candidate.

[0019] In one possible implementation of this application, the preset association recognition model is trained using the following method:

[0020] Obtain target order information with order type "target", and designate the sender object and the recipient object in the same target order information as upstream objects, and the first downstream object associated with the upstream objects, respectively;

[0021] Based on the business scope and logistics information of the upstream object and the business scope and logistics information of the first downstream object, positive sample data is constructed.

[0022] The initial association recognition model is trained using the positive sample data to obtain the preset association recognition model.

[0023] In one possible implementation of this application, the step of training an initial association recognition model using the positive sample data to obtain a preset association recognition model includes:

[0024] For each upstream object, based on the industry information of the corresponding first downstream object, a second downstream object corresponding to each upstream object is selected from a preset sample object. The second downstream object corresponding to each upstream object refers to a sample object that is not in an upstream-downstream relationship with each upstream object.

[0025] Based on the business scope information and mailing information of the upstream object, and the business scope information and mailing information of the second downstream object corresponding to the upstream object, negative sample data is constructed.

[0026] Based on the positive sample data and the negative sample data, the initial association recognition model is trained to obtain the preset association recognition model.

[0027] In one possible implementation of this application, the step of selecting a second downstream object corresponding to each upstream object from a preset sample object based on the industry information of the corresponding first downstream object includes:

[0028] For each upstream object, the industry information of the first downstream object corresponding to each upstream object is statistically analyzed to obtain the downstream industry distribution information corresponding to each upstream object. The industry distribution information includes the industry proportion of multiple different industry information.

[0029] By analyzing the industry information of all first-downstream objects, we can obtain global industry distribution information.

[0030] Based on the industry information of the sample objects, the individual industry proportion of each upstream object is determined from the downstream industry distribution information corresponding to each upstream object, and the global industry proportion of the sample objects is determined from the global industry distribution information.

[0031] Based on the individual industry share of each upstream object and the global industry share of the sample object, a second downstream object corresponding to each upstream object is selected from the sample object.

[0032] In one possible implementation of this application, the step of selecting upstream and / or downstream objects associated with the target object from the first candidate objects based on the actual operating information of the target object and the actual operating information of a preset first candidate object includes:

[0033] Obtain the object identity information of the target object, and the object identity information of a preset first candidate object;

[0034] Based on the actual operating information, object identity information, and logistics information of the target object, as well as the actual operating information, object identity information, and logistics information of the first candidate object, upstream objects and / or downstream objects associated with the target object are selected from the first candidate objects.

[0035] In one possible implementation of this application, the step of selecting upstream and / or downstream objects associated with the target object from the first candidate objects based on the actual operating information, object identity information, and logistics information of the target object, and the actual operating information, object identity information, and logistics information of the first candidate objects, includes:

[0036] The actual operating information, object identity information, and logistics information of the target object, as well as the actual operating information, object identity information, and logistics information of the first candidate object, are fused to obtain fused information.

[0037] By using a preset association relationship recognition model, the fused information is predicted to obtain the association probability between the target object and the first candidate object.

[0038] The first candidate object with an association probability greater than a preset probability threshold is set as the upstream and / or downstream object associated with the target object.

[0039] Secondly, this application provides an associated object identification device, comprising:

[0040] The acquisition unit is used to acquire the target object to be identified.

[0041] The determining unit is used to determine the actual operating information of the target object based on the target object's business scope information and the target object's logistics information;

[0042] The selection unit is used to select an upstream object and / or downstream object associated with the target object from the first candidate object based on the actual operating information of the target object and the actual operating information of the preset first candidate object.

[0043] In one possible implementation of this application, the determining unit is further configured to:

[0044] Obtain multiple second candidate objects that have the same industry information as the target object;

[0045] Extract the first keyword information from the business scope information of the multiple second candidate objects, which appears more than a preset frequency threshold;

[0046] By using a preset association recognition model, the logistics information, business scope information, and first keyword information of the target object are processed to obtain the first target coefficient corresponding to the target object;

[0047] Based on the first target coefficient, the first keyword information is weighted and then fused with the business scope information of the target object to obtain the actual business information of the target object;

[0048] Before selecting upstream and / or downstream objects associated with the target object from the first candidate objects based on the actual operating information of the target object and the actual operating information of the preset first candidate objects, the method further includes:

[0049] Obtain multiple third candidate objects that have the same industry information as the first candidate object;

[0050] Extract second keyword information from the business scope information of the multiple third candidate objects, where the number of occurrences exceeds a preset threshold.

[0051] By using a preset association recognition model, the logistics information, business scope information, and second keyword information of the first candidate object are processed to obtain the second target coefficient corresponding to the first candidate object.

[0052] Based on the second target coefficient, the second keyword information is weighted and then fused with the business scope information of the first candidate to obtain the actual business information of the first candidate.

[0053] In one possible implementation of this application, the determining unit is further configured to:

[0054] Obtain target order information with order type "target", and designate the sender object and the recipient object in the same target order information as upstream objects, and the first downstream object associated with the upstream objects, respectively;

[0055] Based on the business scope and logistics information of the upstream object and the business scope and logistics information of the first downstream object, positive sample data is constructed.

[0056] The initial association recognition model is trained using the positive sample data to obtain the preset association recognition model.

[0057] In one possible implementation of this application, the determining unit is further configured to:

[0058] For each upstream object, based on the industry information of the corresponding first downstream object, a second downstream object corresponding to each upstream object is selected from a preset sample object. The second downstream object corresponding to each upstream object refers to a sample object that is not in an upstream-downstream relationship with each upstream object.

[0059] Based on the business scope information and mailing information of the upstream object, and the business scope information and mailing information of the second downstream object corresponding to the upstream object, negative sample data is constructed.

[0060] Based on the positive sample data and the negative sample data, the initial association recognition model is trained to obtain the preset association recognition model.

[0061] In one possible implementation of this application, the determining unit is further configured to:

[0062] For each upstream object, the industry information of the first downstream object corresponding to each upstream object is statistically analyzed to obtain the downstream industry distribution information corresponding to each upstream object. The industry distribution information includes the industry proportion of multiple different industry information.

[0063] By analyzing the industry information of all first-downstream objects, we can obtain global industry distribution information.

[0064] Based on the industry information of the sample objects, the individual industry proportion of each upstream object is determined from the downstream industry distribution information corresponding to each upstream object, and the global industry proportion of the sample objects is determined from the global industry distribution information.

[0065] Based on the individual industry share of each upstream object and the global industry share of the sample object, a second downstream object corresponding to each upstream object is selected from the sample object.

[0066] In one possible implementation of this application, the selection unit is further configured to:

[0067] Obtain the object identity information of the target object, and the object identity information of a preset first candidate object;

[0068] Based on the actual operating information, object identity information, and logistics information of the target object, as well as the actual operating information, object identity information, and logistics information of the first candidate object, upstream objects and / or downstream objects associated with the target object are selected from the first candidate objects.

[0069] In one possible implementation of this application, the selection unit is further configured to:

[0070] The actual operating information, object identity information, and logistics information of the target object, as well as the actual operating information, object identity information, and logistics information of the first candidate object, are fused to obtain fused information.

[0071] By using a preset association relationship recognition model, the fused information is predicted to obtain the association probability between the target object and the first candidate object.

[0072] The first candidate object with an association probability greater than a preset probability threshold is set as the upstream and / or downstream object associated with the target object.

[0073] Thirdly, this application also provides an electronic device, which includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor calls the computer program in the memory, it executes the steps in any of the associated object identification methods provided in this application.

[0074] Fourthly, this application also provides a storage medium storing a computer program, which, when executed by a processor, implements the steps in any of the associated object identification methods provided in this application.

[0075] In summary, the associated object identification method provided in this application includes: obtaining a target object to be identified; determining the actual business information of the target object based on the business scope information and logistics information of the target object; and selecting an upstream object and / or downstream object associated with the target object from the first candidate object based on the actual business information of the target object and the actual business information of a preset first candidate object.

[0076] As can be seen, the associated object identification method provided in this application can, on the one hand, correct the target object's business scope information through the target object's logistics information to obtain more accurate actual business information, and then identify associated objects based on the more accurate actual business information, thereby obtaining more accurate upstream and / or downstream objects associated with the target object. On the other hand, the associated object identification method provided in this application uses relatively common business scope information and logistics information, thus having greater versatility. Attached Figure Description

[0077] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0078] Figure 1 This is a schematic diagram illustrating an application scenario of the associated object identification method provided in the embodiments of this application;

[0079] Figure 2 This is a flowchart illustrating an associated object identification method provided in an embodiment of this application;

[0080] Figure 3 This is a schematic diagram of a process for obtaining actual business information provided in the embodiments of this application;

[0081] Figure 4 This is a schematic diagram of the structure of the preset association relationship recognition model provided in the embodiments of this application;

[0082] Figure 5 This is a flowchart illustrating a process for obtaining a preset association recognition model, as provided in an embodiment of this application.

[0083] Figure 6 This is another flowchart illustrating the process of obtaining a preset association recognition model provided in the embodiments of this application;

[0084] Figure 7 This is another flowchart illustrating the associated object identification method provided in the embodiments of this application;

[0085] Figure 8 This is another structural diagram of the preset association relationship identification model provided in the embodiments of this application;

[0086] Figure 9 This is a schematic diagram of an embodiment of the associated object identification device provided in this application.

[0087] Figure 10 This is a schematic diagram of an embodiment of the electronic device provided in this application. Detailed Implementation

[0088] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0089] In the description of the embodiments of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0090] To enable any person skilled in the art to implement and use this application, the following description is provided. In this description, details are set forth for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be implemented without using these specific details. In other instances, well-known processes will not be described in detail to avoid obscuring the description of the embodiments of this application with unnecessary detail. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in the embodiments of this application.

[0091] This application provides a method, apparatus, electronic device, and storage medium for identifying associated objects. The associated object identification apparatus can be integrated into an electronic device, which may be a server or a terminal, etc.

[0092] The execution subject of the associated object identification method in this application embodiment can be the associated object identification device provided in this application embodiment, or different types of electronic devices such as server equipment, physical host, or user equipment (UE) that integrate the associated object identification device. The associated object identification device can be implemented in hardware or software. The UE can be a terminal device such as a smartphone, tablet computer, laptop computer, handheld computer, desktop computer, or personal digital assistant (PDA).

[0093] The electronic device can operate independently or in a cluster.

[0094] See Figure 1 , Figure 1 This is a schematic diagram of a scenario for the associated object recognition system provided in an embodiment of this application. The associated object recognition system may include an electronic device 101, which integrates an associated object recognition device.

[0095] In addition, such as Figure 1 As shown, the associated object recognition system may also include a memory 102 for storing data, such as text data.

[0096] It should be noted that, Figure 1 The schematic diagram of the associated object identification system shown is merely an example. The associated object identification system and scenario described in this application embodiment are for the purpose of more clearly illustrating the technical solutions of this application embodiment and do not constitute a limitation on the technical solutions provided in this application embodiment. As those skilled in the art will know, with the evolution of associated object identification systems and the emergence of new business scenarios, the technical solutions provided in this invention embodiment are also applicable to similar technical problems.

[0097] The following describes the associated object identification method provided in the embodiments of this application. In the embodiments of this application, an electronic device is used as the execution subject. For the sake of simplicity and ease of description, the execution subject will be omitted in the subsequent method embodiments. The associated object identification method includes: obtaining the target object to be identified; determining the actual business information of the target object based on the business scope information and logistics information of the target object; and selecting upstream and / or downstream objects associated with the target object from the first candidate objects based on the actual business information of the target object and the actual business information of a preset first candidate object.

[0098] Reference Figure 2 , Figure 2 This is a flowchart illustrating an associated object identification method provided in an embodiment of this application. It should be noted that although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here. Specifically, the associated object identification method may include the following steps 201-203, wherein:

[0099] 201. Obtain the target object to be identified.

[0100] The associated object identification method provided in this application can be applied to a variety of scenarios.

[0101] In some embodiments, the associated object identification method provided in this application can be applied to transaction early warning between enterprises in a supply chain. For example, when a courier company receives a shipment request from an upstream enterprise in the supply chain, it can use the associated object identification method provided in this application to identify whether the recipient is a possible downstream enterprise of the upstream enterprise in the supply chain. When the recipient is identified as a non-downstream enterprise corresponding to the upstream enterprise, an early warning is provided to the upstream enterprise.

[0102] In the supply chain, upstream companies can be understood as those located at the upper end of the supply chain, while downstream companies can be understood as those located at the lower end. For example, in the new energy battery supply chain, upstream companies can refer to suppliers of raw materials such as cathode materials, anode materials, electrolytes, separators, solid electrolytes, structural components, and nickel hydroxide, while downstream companies can refer to companies that sell products such as new energy vehicles, electric bicycles, and electric motorcycles.

[0103] In other embodiments, the associated object identification method provided in this application can also be applied to the field of recommending transaction objects. For example, a user can use the associated object identification method provided in this application to query potential transaction objects from multiple preset enterprises. For instance, when an individual merchant owner is used as the user, the user can use the associated object identification method provided in this application to query transaction objects that can serve as procurement channels from multiple preset enterprises.

[0104] The aforementioned preset enterprises can include all enterprises, which are manually collected and stored in a first preset database. This first preset database can be the backend database of the software used by the user during the query, and it stores at least the industry information of all enterprises.

[0105] As can be seen from the above, the target object in step 201 can refer to either an enterprise or an individual, and this application embodiment does not impose any restrictions on this. For ease of understanding, the application scenario of this application embodiment is assumed to be a transaction alert between enterprises in a supply chain, and the target object refers to an enterprise in the supply chain, which can be either an upstream enterprise or a downstream enterprise. For example, when the associated object identification method in this application embodiment is applied to the supply chain of new energy batteries, the target object can refer to either the supplier of raw materials such as positive electrode materials, negative electrode materials, electrolytes, separators, solid electrolytes, structural components, and nickel hydroxide, i.e., upstream enterprises in the supply chain, or the sales enterprise that sells products such as new energy vehicles, electric bicycles, and electric motorcycles powered by new energy batteries, i.e., downstream enterprises in the supply chain.

[0106] For example, when the application scenario of this application embodiment is that the express delivery company provides transaction alerts to enterprises in the supply chain, the electronic device can obtain express delivery orders to be shipped, filter out candidate orders from the order to be shipped where both the sender and recipient are enterprises, and then use the sender in the candidate orders as the target object. For example, the electronic device can obtain express delivery orders to be shipped, and then extract the sender and recipient from the order to be shipped using Named Entity Recognition (NER) technology. Then it can determine whether the obtained sender and recipient are enterprises. If both the sender and recipient are enterprises, the order to be shipped is used as a candidate order, and the obtained sender is used as the target object. During the determination, the sender and recipient can be matched with preset enterprises in a second preset database. If the second preset database contains the matched object, it means that the object is an enterprise. The second preset database can be the back-end management database of the express delivery company, which stores at least the enterprise names and other data of all enterprises obtained through manual collection.

[0107] Understandably, the first preset database must contain the names of all companies. Therefore, the data in the first and second preset databases can be the same or different. For example, if the first preset database contains industry information and company names for all companies, the second preset database can contain both industry information and company names, or it can contain only company names.

[0108] Named entity recognition (NER) is a technique for identifying entities with specific meanings within text. For example, in this embodiment, a knowledge graph model can be used for NER. In some embodiments, the TransE model can be used for NER.

[0109] Among them, the TransE model is a model that treats the relationships in the knowledge graph as a kind of translation vector between entities for entity naming.

[0110] 202. Determine the actual operating information of the target object based on the target object's business scope information and logistics information.

[0111] The business scope information can include information on all the products a company operates. For example, if the target is an upstream company in the supply chain, the target's business scope information can include information on the products the target operates, such as positive electrode materials, negative electrode materials, electrolytes, separators, solid electrolytes, structural components, and nickel hydroxide.

[0112] In some embodiments, electronic devices can obtain information about the business scope of a target entity through a network. For example, an electronic device can search for the name of the target entity using a search engine on the network, and then obtain the target entity's business scope information from web pages that publicly disclose the company's business and product information.

[0113] Logistics information can include a company's shipment information. For example, logistics information can include the types of products a company has historically shipped, the historical shipment quantity for each historical product type, and the average historical shipment interval for each historical product type. Here, "historical shipment product types" refers to the types of shipments a company shipped within a historical period; "historical shipment quantity for each historical product type" refers to the number of shipments for that historical product type within a historical period; and "average historical shipment interval for each historical product type" refers to the average shipment interval for that historical product type within a historical period. The historical period can be set according to the needs of the actual scenario. For example, the three months prior to the generation time of a candidate order can be used as the historical period corresponding to the target object. The description of candidate orders can be found above and will not be elaborated further.

[0114] In some embodiments, the electronic device can read the back-end management database of the courier company to obtain the logistics information of the target object.

[0115] It should be noted that the target entity's business scope information and logistics information are pre-acquired and stored in the database. During step 202, the electronic device can query the corresponding database to obtain the target entity's business scope information and logistics information. The sources of the target entity's business scope information and logistics information can be found above.

[0116] Actual operating information can include information about the products a company actually operates. The difference between operating scope information and actual operating information is that operating scope information, in addition to actual operating information, may also include information about products the company does not actually operate. For example, when a company discloses its operating scope information, it usually discloses information about the products it currently operates, the products it may operate in the future, and products it fabricates for investment attraction. If an electronic device obtains the target company's operating scope information from a webpage that discloses the company's operating product information, the obtained operating scope information will also include information about products it may operate in the future, as well as products fabricated for investment attraction—that is, products not actually operated. If transaction warnings between companies in the supply chain are directly based on the operating product information, misjudgments may occur. Therefore, in this embodiment, more accurate actual operating information can be obtained through the target company's logistics information and operating scope information, and transaction warnings between companies in the supply chain can be made based on the target company's actual operating information.

[0117] In some embodiments, the electronic device can perform word segmentation on the business scope information to obtain the information on each business product disclosed by the enterprise in the business scope information. Then, it matches the information on historical shipment products in the logistics information with the information on each business product to obtain the target product information that is included in both the information on historical shipment products and the information on each business product. The target product information is used as the actual business information of the target object. It is understood that the target product information may include information on one or more products.

[0118] Among them, electronic devices can use open-source models such as word2vec to segment the business scope information to obtain the information on each of the company's publicly disclosed business products in the business scope information.

[0119] The word2vec model is a model that converts words into vector form, including functions such as word segmentation.

[0120] 203. Based on the actual operating information of the target object and the actual operating information of the preset first candidate object, select the upstream object and / or downstream object associated with the target object from the first candidate object.

[0121] The first candidate objects can be composed of all enterprises whose business scope information can be collected. Alternatively, to reduce the computational load when the electronic device performs step 203, historical express order information can be read from the back-end management database of the express company to obtain the sender and receiver objects in the historical express order information, and the sender and receiver objects can be included as enterprises in the first candidate objects. It can be seen that through this method, the first candidate objects only include enterprises that have sending and / or receiving behavior, and enterprises that do not involve the supply chain, such as empty shell companies, can be excluded. Alternatively, all sender objects and / or all receiver objects corresponding to the target object in the candidate orders can be used as the first candidate objects. For example, when the candidate orders include 5 orders: "Sender object A to receiver object B", "Sender object A to receiver object C", "Sender object A to receiver object D", "Sender object E to receiver object A", and "Sender object F to receiver object A", if object A is the target object to be identified, then object B, object C, object D, object E, and object F can be used as the first candidate objects.

[0122] For ease of understanding, unless otherwise stated below, all sender objects and / or all recipient objects corresponding to the target object in the candidate order are considered as the preset first candidate object.

[0123] The method for obtaining the actual business information of the first candidate can be found in step 202, and will not be elaborated further.

[0124] For example, the electronic device can compare the actual operating information of the target object with the actual operating information of each first candidate object for similarity, and select objects from the first candidate objects whose similarity is greater than a preset similarity threshold. The selected objects are then used as upstream and / or downstream objects associated with the target object. Using the example above, if objects B, C, D, E, and F are preset first candidate objects, and object A is the target object to be identified, and among objects B, C, D, E, and F, objects B and E have a similarity greater than the preset similarity threshold between their actual operating information and the actual operating information of object A, then objects B and E are the selected objects associated with the target object. Since object B is the destination of object A, object B can be used as a downstream object of object A. Since object A is the destination of object E, object E can be used as an upstream object of object A.

[0125] It should be noted that the similarity comparison here refers to encoding the actual operating information of the target object and the actual operating information of each first candidate object separately to obtain corresponding semantic features, and then comparing the similarity of these semantic features. For example, electronic devices can use open-source models such as the word2vec model to encode the actual operating information of the target object and the actual operating information of each first candidate object separately to obtain corresponding semantic features. An explanation of the word2vec model can be found above and will not be repeated here.

[0126] The preset similarity threshold can be set according to the actual needs of the scenario. For example, the similarity threshold can be set to 0.9.

[0127] In summary, the associated object identification method provided in this application includes: obtaining a target object to be identified; determining the actual business information of the target object based on the business scope information and logistics information of the target object; and selecting an upstream object and / or downstream object associated with the target object from the first candidate object based on the actual business information of the target object and the actual business information of a preset first candidate object.

[0128] As can be seen, the associated object identification method provided in this application can, on the one hand, correct the target object's business scope information through the target object's logistics information to obtain more accurate actual business information, and then identify associated objects based on the more accurate actual business information, thereby obtaining more accurate upstream and / or downstream objects associated with the target object. On the other hand, the associated object identification method provided in this application uses relatively common business scope information and logistics information, thus having greater versatility.

[0129] In some embodiments, in addition to the method in step 202, product information that frequently appears in the business scope information corresponding to the industry can be statistically obtained from multiple objects whose industry information is the same as that of the target object, and the actual business information of the target object can be determined based on the frequently appearing product information.

[0130] refer to Figure 3 At this point, the step "determine the actual operating information of the target object based on the target object's business scope information and the target object's logistics information" includes:

[0131] 301. Obtain multiple second candidate objects with the same industry information as the target object.

[0132] Multiple second candidate objects can consist of all enterprises whose business scope information can be collected. For example, the electronic device can query the target industry information of the target object from the first preset database, and read all corresponding industry information from the first preset database as the second candidate objects of the target industry information based on the target industry information. Alternatively, it can read a preset number of corresponding industry information from the first preset database as the second candidate objects of the target industry information based on the target industry information. The preset number can be set according to the needs of the actual scenario.

[0133] The industry information for an object includes the industry corresponding to its business operations. For example, for a company whose business involves supplying raw materials such as cathode materials, anode materials, electrolytes, separators, solid electrolytes, structural components, and nickel hydroxide for new energy batteries, the corresponding industry information could refer to the battery raw material supply industry. For companies selling new energy vehicles, electric bicycles, and electric motorcycles powered by new energy batteries, the corresponding industry information could refer to the new energy vehicle industry. When the object refers to a company, its industry information can be manually collected and then associated with objects in the first preset database. The description of the first preset database can be found above and will not be repeated here.

[0134] Understandably, the difference between industry information and business scope information lies in the fact that industry information refers to the industry in which the object corresponds, and can be the name of the industry. Business scope information refers to the object's products or business operations, and can be the name of the products or the name of the business operations. The same industry information can correspond to multiple business scope information. For example, the business scope information "positive electrode materials, negative electrode materials, electrolytes, separators, solid electrolytes, structural components, and nickel hydroxide for new energy batteries" can correspond to the industry information "battery raw material supply industry," or the business scope information "electrolytes, separators, and solid electrolytes for new energy batteries" can also correspond to the industry information "battery raw material supply industry."

[0135] The difference between logistics information and industry information and business scope information is that the information contained in logistics information may or may not be related to the products / businesses being operated. It may include information unrelated to the products / businesses being operated, such as office supplies, or information related to the products / businesses being operated, such as the products actually being operated. Therefore, it is necessary to use both business scope information and logistics information to accurately obtain the actual business information and filter out information in the logistics information that is unrelated to the products / businesses being operated.

[0136] 302. Extract the first keyword information from the business scope information of the multiple second candidate objects, which appears more than a preset threshold number of times.

[0137] As explained above, the business scope information of an object can include all the product information of the enterprise. Therefore, it can be understood that the first keyword information refers to the product information that appears more than a preset threshold number of times in the business scope information of multiple second candidate objects. This includes common products for enterprises whose business operations are within the target industry. For example, the electronic device can perform word segmentation on the business scope information of each second candidate object and count the number of various product information items obtained after segmentation. The product information with a count greater than the preset threshold number is used as the first keyword information. Open-source models such as the word2vec model can be used to perform word segmentation on the business scope information of each second candidate object. An explanation of the word2vec model can be found above and will not be repeated here.

[0138] For ease of understanding, the following example illustrates the situation: Assume the target industry information mentioned above refers to the battery raw material supply industry, the preset threshold for the number of attempts is 2, and the second candidate objects include: raw material supplier G, raw material supplier H, and raw material supplier I. The business scope information of G, H, and I respectively includes the business product information of "positive electrode material, negative electrode material, electrolyte", "positive electrode material, negative electrode material, separator", and "positive electrode material, negative electrode material, nickel hydroxide". After word segmentation, the various business product information obtained are "positive electrode material", "negative electrode material", "electrolyte", "separator", and "nickel hydroxide". Therefore, the first keyword information obtained after statistics is "positive electrode material" and "negative electrode material", indicating that for companies whose business is in the battery raw material supply industry, their common business products are positive electrode material and negative electrode material.

[0139] The preset number of times threshold can be set according to the needs of the actual scenario, and the preset number of times threshold in the above example should not be used as a limitation on the embodiments of this application.

[0140] 303. By using a preset association recognition model, the logistics information of the target object, the business scope information of the target object, and the first keyword information are processed to obtain the first target coefficient corresponding to the target object.

[0141] The first target coefficient corresponding to the target object is used to integrate the first keyword information corresponding to the target object and the business scope information of the target object to obtain actual business information. The actual business information obtained can more accurately reflect the product information actually operated by the target object.

[0142] The preset relationship identification model is used to identify relationships between objects. In this embodiment, it can be used to identify upstream and downstream relationships between enterprises in a supply chain. (Reference) Figure 4 , Figure 4 The present invention provides a structural model of a relation recognition model 400, which includes:

[0143] Self-attention layer 401 is used to process logistics information, business scope information and keyword information to obtain the target coefficient corresponding to the object. Based on the target coefficient, the keyword information corresponding to the object and the business scope information of the object are fused to obtain the actual business information of the object. The object here can include either the target object or the first candidate object. That is, self-attention layer 401 can be used to obtain the actual business information of the target object or the actual business information of the first candidate object.

[0144] Feature fusion layer 402 is used to fuse the actual operating information of the target object and the first candidate object to obtain fused information. The purpose of feature fusion is to combine two types of information into one to facilitate subsequent prediction. For example, the actual operating information of the target object and the first candidate object can be concatenated to obtain fused information. The fused information includes the actual operating information of both the target object and the first candidate object.

[0145] The identification layer 403 is used to predict the association probability between the target object and the first candidate object based on the fusion information, and select upstream and / or downstream objects associated with the target object from the first candidate objects based on the association probability. For example, the first candidate object with an association probability greater than a preset probability threshold can be used as the upstream and / or downstream objects associated with the target object. The preset probability threshold can be set according to the actual scenario requirements.

[0146] For example, electronic devices can be Figure 4 The self-attention layer 401 of the association recognition model 400 uses formula (1) to determine the first target coefficient of the target object:

[0147]

[0148] Where ω1 refers to the first target coefficient of the target object, and W1, W2, W3, b1, b2, and b3 are all trained parameters. This refers to the semantic features obtained after encoding the logistics information of the target object. This refers to the semantic features obtained after encoding the business scope information of the target object. It refers to the semantic features obtained after encoding the first keyword information corresponding to the target object.

[0149] It is evident that formula (1) considers the logistics information, business scope information, and primary keyword information of the target object when calculating the first target coefficient, thus containing richer information. The first target coefficient can be understood as an attention weight, used to characterize the importance of the primary keyword information.

[0150] It should be noted that in the process of obtaining the actual business information of the first candidate, through... Figure 4When the self-attention layer 401 of the correlation recognition model 400 obtains the first target coefficient of the first candidate object, it can either extract the second keyword information corresponding to the first candidate object based on the industry information of the first candidate object, and obtain the actual business information of the first candidate object based on the second keyword information, or use the first keyword information corresponding to the target object as the second keyword information corresponding to the first candidate object, and obtain the actual business information of the first candidate object based on the second keyword information. If the second keyword information corresponding to the first candidate object is extracted based on the industry information of the first candidate object, then common business product information in the industry corresponding to the first candidate object can be incorporated when calculating the second target coefficient. Therefore, the fused information obtained through the feature fusion layer 402 simultaneously contains common business product information in the industry corresponding to the first candidate object and common business product information in the industry corresponding to the target object, thereby improving the accuracy of recognition. When directly using the first keyword information corresponding to the target object, there is no need to perform extraction operations to obtain the second keyword information corresponding to the first candidate object. Therefore, the computational power requirement of the associated object identification method provided in this application can be reduced. At the same time, when calculating the second target coefficient of the first candidate object, the differences between the common business product information in the industry corresponding to the target object and the business scope information and logistics information of the first candidate object are taken into account, which can also improve the accuracy of identification. The self-attention layer 401 can be set according to actual needs when constructing it.

[0151] Taking the direct use of the first keyword information corresponding to the target object to obtain the second target coefficient of the first candidate object as an example, the electronic device can... Figure 4 The self-attention layer 401 of the association recognition model 400 uses formula (2) to determine the second target coefficient of the first candidate object:

[0152]

[0153] Where ω2 refers to the second target coefficient of the first candidate object, and W1, W2, W3, b1, b2, and b3 are all trained parameters. This refers to the semantic features obtained after encoding the logistics information of the first candidate object. This refers to the semantic features obtained after encoding the business scope information of the first candidate. It refers to the semantic features obtained after encoding the second keyword information of the target object.

[0154] For ease of understanding, unless otherwise stated below, it is assumed that the electronic device directly uses the first keyword information corresponding to the target object to obtain the second target coefficient of the first candidate object. That is, the second keyword information in the following text refers to the first keyword information of the target object.

[0155] 304. Based on the first target coefficient, the first keyword information is weighted and then the weighted first keyword information is merged with the business scope information of the target object to obtain the actual business information of the target object.

[0156] The following is a method for obtaining the actual operating information of the first candidate:

[0157] (A) Obtain multiple third candidates with the same industry information as the first candidate.

[0158] (B) Extract the second keyword information from the business scope information of the multiple third candidate objects, which appears more than a preset threshold number of times.

[0159] (C) By using a preset association recognition model, the logistics information, business scope information and second keyword information of the first candidate object are processed to obtain the second target coefficient corresponding to the first candidate object.

[0160] (D) Based on the second target coefficient, the second keyword information is weighted and then fused with the business scope information of the first candidate to obtain the actual business information of the first candidate.

[0161] The method for obtaining the third candidate object can refer to that for the second candidate object. The explanation and processing method for the second keyword information and the second target coefficient can refer to the explanation and processing method for the first keyword information and the first target coefficient above. The preset association relationship identification model can refer to the association relationship identification model 400.

[0162] For example, electronic devices can be Figure 4 The self-attention layer 401 of the association recognition model 400 performs weighted processing on the first keyword information and merges the weighted first keyword information with the business scope information of the target object to obtain the actual business information of the target object. Simultaneously, electronic devices can also... Figure 4 The self-attention layer 401 of the association recognition model 400 integrates the weighted first keyword information with the business scope information of the first candidate object to obtain the actual business information of the first candidate object.

[0163] After performing step 304, the electronic device can also... Figure 4The feature fusion layer 402 and recognition layer 403 of the relationship identification model 400 select upstream and / or downstream objects associated with the target object from the first candidate objects based on the actual operating information of the target object and the actual operating information of the first candidate objects. For example, electronic devices can... Figure 4 The feature fusion layer 402 of the correlation identification model 400 fuses the actual operating information of the target object and the actual operating information of each first candidate object to obtain the fused information corresponding to each first candidate object, and then... Figure 4 The recognition layer 403 of the correlation recognition model 400 predicts each piece of fused information to obtain the correlation probability corresponding to each first candidate object. Based on the correlation probability of each first candidate object, it selects objects with higher correlation probabilities from the first candidate objects as upstream and / or downstream objects associated with the target object. For details, please refer to the explanation in step 203. The fusion method and the method for obtaining correlation probabilities based on the fused information can be referred to above. For example, the actual operating information of the target object and the actual operating information of each first candidate object can be concatenated to obtain the fused information corresponding to each first candidate object, and then... Figure 4 The recognition layer 403 of the association recognition model 400 predicts each piece of fused information to obtain the association probability corresponding to each first candidate object.

[0164] The pre-defined relationship recognition model can be trained using positive sample data. To quickly obtain positive sample data, historical express delivery orders can be filtered by order type. Based on the filtered order information, positive sample data can be constructed. (Reference) Figure 5 At this point, the pre-set association recognition model is trained using the following method:

[0165] 501. Obtain target order information with the order type being target type, and designate the sender object and the recipient object in the same target order information as upstream objects and the first downstream object associated with the upstream objects, respectively.

[0166] Among them, "order" can refer to a courier order, and "target type" can refer to a business order, that is, an order used to send business documents such as contracts and agreements.

[0167] The reason for using business orders as the target type is that even between two companies that are not in an upstream or downstream relationship in the supply chain, express delivery orders may still be generated. However, only between two companies that are in an upstream or downstream relationship in the supply chain will they send each other business documents such as contracts and agreements. Therefore, the positive sample data constructed based on business orders that send business documents is more accurate, and can be directly filtered according to the order type, while reducing manpower consumption.

[0168] The order type of historical express orders can be carried by the historical express orders and stored in the back-end management database of the express company.

[0169] For example, an electronic device can extract the sender object and the recipient object from the same target order information, treat the sender object as the upstream object, and the recipient object as the first downstream object corresponding to the upstream object. It can be understood that the first downstream object is an object that has an upstream-downstream relationship with the corresponding upstream object in the supply chain.

[0170] It should be noted that each upstream object can have only one or more first downstream objects. For example, if there are multiple target order information sender objects, all of which are object J, and the recipient objects for each target order information are object K, object L, and object M respectively, then if object J is taken as the upstream object, the first downstream objects corresponding to this upstream object include object K, object L, and object M, for a total of 3.

[0171] 502. Construct positive sample data based on the business scope information and logistics information of the upstream object and the business scope information and logistics information of the first downstream object.

[0172] The methods for obtaining the business scope and logistics information of upstream entities, as well as the business scope and logistics information of the first downstream entity, can be referred to the above text and will not be elaborated upon here. The positive sample data obtained through step 502 includes multiple groups of upstream and downstream entities, the business scope and logistics information of the upstream entities in each group, and the business scope and logistics information of the first downstream entity in each group.

[0173] In this step, constructing positive sample data refers to using the information as the input data for the initial association recognition model below. It can be understood that the positive sample data includes the business scope information and logistics information of the upstream object and the business scope information and logistics information of the first downstream object.

[0174] 503. Using the positive sample data, train the initial association recognition model to obtain the preset association recognition model.

[0175] The structure of the initial association recognition model can be referenced from [reference needed]. Figure 4 The specifics will not be elaborated upon.

[0176] To further improve training effectiveness and obtain a high-quality pre-defined association recognition model, electronic devices can simultaneously acquire negative sample data during the initial training of the association recognition model. Negative samples are samples with incorrectly classified labels that are used to reverse-engineer the machine learning model, thereby improving its predictive ability.

[0177] refer to Figure 6 At this point, the step "training the initial association recognition model using the positive sample data to obtain a preset association recognition model" includes:

[0178] 601. For each upstream object, based on the industry information of the corresponding first downstream object, select the second downstream object corresponding to each upstream object from the preset sample objects.

[0179] The preset sample objects can include all enterprises, for example, all enterprises stored in the first preset database mentioned above.

[0180] The explanation and methods for obtaining industry information can be found above, and will not be elaborated upon here.

[0181] The second downstream object of an upstream object refers to an object in the sample objects that is not in an upstream or downstream relationship with the upstream object in the supply chain.

[0182] In some embodiments, the electronic device can select sample objects whose industry information differs from that of the first downstream object, and use them as the second downstream objects corresponding to the upstream object. For example, if an upstream object O has only one corresponding first downstream object, and its corresponding industry is the new energy battery industry, an object whose corresponding industry is not the new energy battery industry can be selected from the sample objects as the second downstream object of upstream object O. The sample consisting of upstream object O and its corresponding second downstream object can then be used as negative sample data to train the initial association relationship recognition model. As another example, if an upstream object J has three corresponding first downstream objects—object K, object L, and object M—and the industries corresponding to objects K, L, and M are the new energy battery industry, the catering industry, and the clothing industry, respectively, an object whose corresponding industry is not the new energy battery industry, the catering industry, or the clothing industry can be selected from the sample objects as the second downstream object of upstream object J. The sample consisting of upstream object J and its corresponding second downstream object can then be used as negative sample data to train the initial association relationship recognition model.

[0183] Similarly, there can be only one or more second downstream objects, and this application embodiment does not impose any restrictions on this.

[0184] However, the negative sample data obtained by the above method is still inaccurate. Therefore, in some other embodiments, for each upstream object, the electronic device can also statistically analyze the distribution of its corresponding first downstream object, as well as the distribution of all first downstream objects in the positive sample data. Combining the distribution of the first downstream objects corresponding to the upstream object and the distribution of all first downstream objects in the positive sample data, a second downstream object corresponding to each upstream object is selected. In this case, the step "For each upstream object, based on the industry information of the corresponding first downstream object, select the second downstream object corresponding to each upstream object from the preset sample objects" includes:

[0185] (1) For each upstream object, the industry information of the first downstream object corresponding to each upstream object is statistically analyzed to obtain the downstream industry distribution information corresponding to each upstream object.

[0186] The industry distribution information includes the industry percentage of multiple different industries.

[0187] Among them, the downstream industry distribution information of the upstream object includes the industry proportion of each industry in the first downstream object corresponding to the upstream object, which can be understood as the popularity of enterprises in different industries as the first downstream object. The industry proportion in the downstream industry distribution information can be calculated from the number of first downstream objects corresponding to each industry. For ease of understanding, let's continue to use the upstream object J in the above text as an example. If the first downstream objects corresponding to the upstream object J include objects K, L and M, a total of 3, and the industries corresponding to objects K, L and M are the new energy battery industry, the catering industry and the clothing industry, respectively, then the downstream industry distribution information of the upstream object J includes the industry proportions corresponding to the new energy battery industry, the catering industry and the clothing industry, respectively. For example, the industry proportions corresponding to the new energy battery industry, the catering industry and the clothing industry can be calculated by formulas (3)-(5):

[0188] B1 J =I1 J / I all J Formula (3)

[0189] B2 J =I2 J / I all J Formula (4)

[0190] B3 J =I3 J / I all J Formula (5)

[0191] Among them, B1J I1 represents the industry share of the new energy battery industry. J Let I be the number of first downstream objects corresponding to the new energy battery industry among the first downstream objects of upstream object J. all J This represents the total number of the first downstream objects of upstream object J. (B2) J I2 represents the industry share of the catering industry. J This represents the number of first downstream objects corresponding to the catering industry within the first downstream object of upstream object J. (B3) J I3 represents the industry share of the apparel industry. J Let J be the number of first downstream objects in the apparel industry among the first downstream objects of upstream object J.

[0192] Due to I1 J I2 J and I3 J Both are 1, therefore B1 J B2 J and B3 J Both are 1 / 3, meaning that for upstream object J, its corresponding downstream industry distribution information includes: B1 J B2 J and B3 J The information is 1 / 3 for each of the three categories: K, L, and M. The number of objects in the clothing industry, catering industry, and clothing industry, respectively, accounts for 1 / 3 of the total number of objects.

[0193] By analyzing the downstream industry distribution information corresponding to each upstream object, electronic devices can determine the popularity of companies in different industries as the first downstream object of that upstream object. The greater the popularity, the greater the likelihood that a company in the corresponding industry will be the first downstream object of that upstream object; the less popular, the less likely that a company in the corresponding industry will be the first downstream object of that upstream object.

[0194] (2) Statistically analyze the industry information of all first downstream objects to obtain global industry distribution information.

[0195] The global industry distribution information can include the proportion of the number of first downstream objects corresponding to different industries to the total number of first downstream objects, as described in step (1). The global industry distribution information also includes multiple industry proportions. For example, if the total number of first downstream objects is 200, and the number of first downstream objects corresponding to the new energy battery industry is 100, the number of first downstream objects corresponding to the catering industry is 50, and the number of first downstream objects corresponding to the clothing industry is 50, then the industry proportion of the new energy battery industry is 0.5, and the industry proportions of the catering industry and the clothing industry are both 0.25. That is, the global industry distribution information includes three proportions: 0.5, 0.25, and 0.25.

[0196] The purpose of obtaining global industry distribution information is to determine the popularity of companies in each industry as primary downstream targets from a global industry perspective. The greater the popularity, the greater the likelihood that companies in the corresponding industry will be primary downstream targets; the less popular, the less likely that companies in the corresponding industry will be primary downstream targets.

[0197] It can be seen that through steps (1)-(2), the electronic device can obtain the probability that different industries are the first downstream objects for each upstream object individual, and the probability that different industries are the first downstream objects when considering the global situation without considering the differences of upstream objects. Then, by combining the probability that different industries are the first downstream objects for individuals and the probability that different industries are the first downstream objects for the global situation, the accuracy of constructing negative sample data can be improved.

[0198] (3) Based on the industry information of the sample object, determine the individual industry proportion of each upstream object from the downstream industry distribution information corresponding to each upstream object, and determine the global industry proportion corresponding to the industry information of the sample object from the global industry distribution information.

[0199] The individual industry proportion corresponding to the upstream object refers to the industry proportion of the sample object in the downstream industry distribution information corresponding to the upstream object. When the sample object refers to all enterprises, the individual industry proportion corresponding to the upstream object refers to the industry proportion of all enterprises in the downstream industry distribution information corresponding to the upstream object. In this case, since all enterprises include the upstream object, if the second downstream object corresponding to the upstream object includes the upstream object itself, it should be removed. For ease of understanding, let's continue with the example in step (1). Suppose that for the upstream object J, in addition to itself, the sample object contains two enterprises, a and b. If the industry of sample object a is the new energy battery industry, then for sample object a, the individual industry proportion corresponding to the upstream object J is 1 / 3. This means that for sample object a, the industry "new energy battery industry" accounts for 1 / 3 of the downstream enterprises of the upstream object J. That is, 1 / 3 of the downstream enterprises of the upstream object J correspond to the industry "new energy battery industry". If the industry of sample object b is the catering industry, then for sample object b, the proportion of individual industries corresponding to upstream object J is also 1 / 3. This means that for sample object b, the proportion of the industry "catering industry" among the downstream enterprises of upstream object J is 1 / 3, that is, 1 / 3 of the downstream enterprises of upstream object J are in the "catering industry".

[0200] The global industry percentage corresponding to the sample object refers to the industry percentage of the sample object in the global industry distribution information. For ease of understanding, let's continue with the example in step (2). If the industry of sample object c is the new energy battery industry, then the global industry percentage corresponding to sample object c is 0.5. If the industry of sample object d is the catering industry, then the global industry percentage corresponding to sample object d is 0.25.

[0201] The difference between individual industry proportion and global industry proportion lies in the fact that global industry proportion does not consider the individual differences between each upstream object and is used to characterize the probability that the industry of a sample object is also the industry of a downstream object. The higher the global industry proportion, the higher the probability that the industry of the sample object is also the industry of a downstream object. On the other hand, individual industry proportion takes into account the individual differences between each upstream object and calculates the proportion of the sample object's industry information in the downstream industry corresponding to that upstream object.

[0202] (4) Based on the individual industry percentage corresponding to each upstream object and the global industry percentage corresponding to the sample object, select the second downstream object corresponding to each upstream object from the sample object.

[0203] For example, the probability that a sample object is the corresponding second downstream object for each upstream object can be calculated using formula (6), and the second downstream object corresponding to each upstream object can be selected from the sample objects based on the obtained probability.

[0204] P i j =(1-B i j )T j Formula (6)

[0205] Among them, P i j B is the probability that the j-th object in the sample is the second downstream object corresponding to the i-th upstream object. i j For the j-th object in the sample, T represents the individual industry percentage corresponding to the i-th upstream object. j It represents the global industry percentage of the j-th object in the sample.

[0206] It can be seen that, through equation (6), B i j The smaller the value of T, the lower the probability that, for the i-th upstream object, a company in the industry of the j-th object in the sample objects will be the first downstream object of that upstream object, and the lower the probability that the j-th object in the sample objects will be the second downstream object corresponding to the i-th upstream object. From a global perspective, T... j The larger the value, the higher the probability that the enterprise in the industry of the j-th object in the sample is the first downstream object, and the higher the probability that the j-th object in the sample is the second downstream object corresponding to the i-th upstream object, thus allowing for the selection of reasonable second downstream objects.

[0207] In other embodiments, the industry information of each object in the sample can be further statistically analyzed to obtain another global industry distribution information for each industry. This second global industry distribution information is then incorporated when calculating the probability that the sample object is the corresponding second downstream object for each upstream object. The calculation method for this second global industry distribution information can refer to step (2), and will not be elaborated further.

[0208] For example, the probability that a sample object is the corresponding second downstream object for each upstream object can be calculated using equation (7).

[0209] P i j =(1-B i j )T j H j Formula (7)

[0210] Among them, Pi j B is the probability that the j-th object in the sample is the second downstream object corresponding to the i-th upstream object. i j For the j-th object in the sample, T represents the individual industry percentage corresponding to the i-th upstream object. j H represents the global industry percentage of the j-th object in the sample. j It is another global industry distribution information of the j-th object in the sample objects.

[0211] 602. Based on the business scope information and mailing information of the upstream object, and the business scope information and mailing information of the second downstream object corresponding to the upstream object, construct negative sample data.

[0212] The methods for obtaining the business scope and logistics information of upstream entities, as well as the business scope and logistics information of the second downstream entity, can be referred to the above text and will not be elaborated further. The negative sample data obtained through step 603 contains multiple groups of upstream and downstream entities, including the business scope and logistics information of the upstream entities in each group, as well as the business scope and logistics information of the second downstream entities in each group.

[0213] In other embodiments, after obtaining the business scope information and mailing information of each upstream object, as well as the business scope information and mailing information of the second downstream object corresponding to each upstream object, and constructing the initial sample data, negative sample data for training can be further filtered from the obtained initial sample data. For example, the initial sample data can be filtered using the spy technology in the positive and unlabeled algorithm to obtain negative sample data for training.

[0214] Among them, the spy algorithm is an algorithm that randomly selects a portion of positive samples from the positive sample data as spy samples, hides their labels, and then mixes the selected spy samples into the initial sample data. By leveraging the high consistency between the behavior of the spy samples and the positive samples in the initial sample data, it filters out highly credible negative sample data from the initial sample data.

[0215] 603. Based on the positive sample data and the negative sample data, train the initial association relationship recognition model to obtain the preset association relationship recognition model.

[0216] In some embodiments, to further improve the accuracy of associated object identification, during step 203, the electronic device may also simultaneously acquire the object identity information of the object, and combine the object identity information, logistics information, and actual operating information of each object to select upstream and / or downstream objects associated with the target object from the first candidate objects. (See reference) Figure 7 At this point, the step "Based on the actual operating information of the target object and the actual operating information of the preset first candidate object, select the upstream object and / or downstream object associated with the target object from the first candidate object" includes:

[0217] 701. Obtain the object identity information of the target object and the object identity information of the preset first candidate object.

[0218] Object identity information is used to describe the identity of an object. For example, when the object refers to a company, the object identity information may include the company's legal representative, company address, etc., to describe the company's identity. This object identity information can be manually collected and stored in the first preset database mentioned above. When the electronic device executes step 701, it reads the object identity information from the first preset database.

[0219] 702. Based on the actual operating information, object identity information and logistics information of the target object, and the actual operating information, object identity information and logistics information of the first candidate object, select the upstream object and / or downstream object associated with the target object from the first candidate object.

[0220] For example, electronic devices can be Figure 8 The relationship identification model 800 in the middle selects upstream and / or downstream objects associated with the target object from the first candidate objects based on the actual operating information, object identity information, and logistics information of each object. The relationship identification model 800 includes:

[0221] Self-attention layer 801 is used to process logistics information, business scope information and first keyword information to obtain the target coefficient corresponding to the object. Based on the target coefficient, the keyword information corresponding to the object and the business scope information of the object are fused to obtain the actual business information of the object. The object here can include either the target object or the first candidate object. That is, self-attention layer 401 can be used to obtain the actual business information of the target object or the actual business information of the first candidate object.

[0222] Feature fusion layer 802 is used to fuse the actual operating information of the target object and the first candidate object to obtain fused information. The purpose of feature fusion is to combine two types of information into one to facilitate subsequent prediction. For example, the actual operating information of the target object and the first candidate object can be concatenated to obtain fused information. The fused information includes the actual operating information, object identity information, and logistics information of the target object, as well as the actual operating information, object identity information, and logistics information of the first candidate object.

[0223] The identification layer 803 is used to determine the association probability between the target object and the first candidate object based on the fusion information, and to select upstream and / or downstream objects associated with the target object from the first candidate objects based on the association probability.

[0224] At this point, the step "selecting upstream and / or downstream objects associated with the target object from the first candidate objects based on the actual operating information, object identity information, and logistics information of the target object, and the actual operating information, object identity information, and logistics information of the first candidate object" may include:

[0225] (i) The actual operating information, object identity information and logistics information of the target object and the actual operating information, object identity information and logistics information of the first candidate object are fused to obtain fused information.

[0226] The fusion process can be the same as described above, that is, fusion can be performed by splicing. In step 701, the fused information simultaneously includes the actual operating information, object identity information, and logistics information of the target object, as well as the actual operating information, object identity information, and logistics information of the first candidate object, with the aim of integrating the information for subsequent prediction.

[0227] (ii) The fused information is predicted using a preset association recognition model to obtain the association probability between the target object and the first candidate object.

[0228] The preset association recognition model can refer to Association Relationship Recognition Model 800.

[0229] (iii) Set the first candidate object with an association probability greater than a preset probability threshold as the upstream and / or downstream object associated with the target object.

[0230] The preset probability threshold is used to evaluate the magnitude of the association probability and can be set according to the needs of the actual scenario.

[0231] For example, electronic devices can implement steps (i) and (iii) through the recognition layer 803 of the association recognition model 800, which will not be described in detail.

[0232] To better implement the associated object identification method in the embodiments of this application, based on the associated object identification method, the embodiments of this application also provide an associated object identification device, such as... Figure 9 The diagram shown is a schematic representation of an embodiment of the associated object identification device in this application. The associated object identification device 900 includes:

[0233] Acquisition unit 901 is used to acquire the target object to be identified;

[0234] The determining unit 902 is used to determine the actual operating information of the target object based on the target object's business scope information and the target object's logistics information;

[0235] Selection unit 903 is used to select upstream and / or downstream objects associated with the target object from the first candidate objects based on the actual operating information of the target object and the actual operating information of the preset first candidate objects.

[0236] In one possible implementation of this application, the determining unit 902 is further configured to:

[0237] Obtain multiple second candidate objects that have the same industry information as the target object;

[0238] Extract the first keyword information from the business scope information of the multiple second candidate objects, which appears more than a preset frequency threshold;

[0239] By using a preset association recognition model, the logistics information, business scope information, and first keyword information of the target object are processed to obtain the first target coefficient corresponding to the target object;

[0240] Based on the first target coefficient, the first keyword information is weighted and then fused with the business scope information of the target object to obtain the actual business information of the target object;

[0241] Before selecting upstream and / or downstream objects associated with the target object from the first candidate objects based on the actual operating information of the target object and the actual operating information of the preset first candidate objects, the method further includes:

[0242] Obtain multiple third candidate objects that have the same industry information as the first candidate object;

[0243] Extract second keyword information from the business scope information of the multiple third candidate objects, where the number of occurrences exceeds a preset threshold.

[0244] By using a preset association recognition model, the logistics information, business scope information, and second keyword information of the first candidate object are processed to obtain the second target coefficient corresponding to the first candidate object.

[0245] Based on the second target coefficient, the second keyword information is weighted and then fused with the business scope information of the first candidate to obtain the actual business information of the first candidate.

[0246] In one possible implementation of this application, the determining unit 902 is further configured to:

[0247] Obtain target order information with order type "target", and designate the sender object and the recipient object in the same target order information as upstream objects, and the first downstream object associated with the upstream objects, respectively;

[0248] Based on the business scope and logistics information of the upstream object and the business scope and logistics information of the first downstream object, positive sample data is constructed.

[0249] The initial association recognition model is trained using the positive sample data to obtain the preset association recognition model.

[0250] In one possible implementation of this application, the determining unit 902 is further configured to:

[0251] For each upstream object, based on the industry information of the corresponding first downstream object, a second downstream object corresponding to each upstream object is selected from a preset sample object. The second downstream object corresponding to each upstream object refers to a sample object that is not in an upstream-downstream relationship with each upstream object.

[0252] Based on the business scope information and mailing information of the upstream object, and the business scope information and mailing information of the second downstream object corresponding to the upstream object, negative sample data is constructed.

[0253] Based on the positive sample data and the negative sample data, the initial association recognition model is trained to obtain the preset association recognition model.

[0254] In one possible implementation of this application, the determining unit 902 is further configured to:

[0255] For each upstream object, the industry information of the first downstream object corresponding to each upstream object is statistically analyzed to obtain the downstream industry distribution information corresponding to each upstream object. The industry distribution information includes the industry proportion of multiple different industry information.

[0256] By analyzing the industry information of all first-downstream objects, we can obtain global industry distribution information.

[0257] Based on the industry information of the sample objects, the individual industry proportion of each upstream object is determined from the downstream industry distribution information corresponding to each upstream object, and the global industry proportion of the sample objects is determined from the global industry distribution information.

[0258] Based on the individual industry share of each upstream object and the global industry share of the sample object, a second downstream object corresponding to each upstream object is selected from the sample object.

[0259] In one possible implementation of this application, the selection unit 903 is further configured to:

[0260] Obtain the object identity information of the target object, and the object identity information of a preset first candidate object;

[0261] Based on the actual operating information, object identity information, and logistics information of the target object, as well as the actual operating information, object identity information, and logistics information of the first candidate object, upstream objects and / or downstream objects associated with the target object are selected from the first candidate objects.

[0262] In one possible implementation of this application, the selection unit 903 is further configured to:

[0263] The actual operating information, object identity information, and logistics information of the target object, as well as the actual operating information, object identity information, and logistics information of the first candidate object, are fused to obtain fused information.

[0264] By using a preset association relationship recognition model, the fused information is predicted to obtain the association probability between the target object and the first candidate object.

[0265] The first candidate object with an association probability greater than a preset probability threshold is set as the upstream and / or downstream object associated with the target object.

[0266] In practice, each of the above units can be implemented as an independent entity or can be arbitrarily combined to be implemented as the same or several entities. For the specific implementation of each of the above units, please refer to the previous method embodiments, which will not be repeated here.

[0267] Since the associated object identification device can execute the steps in the associated object identification method in any embodiment, it can achieve the beneficial effects that the associated object identification method in any embodiment of this application can achieve, as detailed in the preceding description, and will not be repeated here.

[0268] Furthermore, in order to better implement the associated object identification method in the embodiments of this application, the associated object identification method...

[0269] Based on this, embodiments of this application also provide an electronic device, see below. Figure 10 , Figure 10 This illustration shows a structural schematic diagram of an electronic device according to an embodiment of this application. Specifically, the electronic device provided in this embodiment includes a processor 1001. The processor 1001 is used to execute a computer program stored in a memory 1002 to implement each step of the associated object identification method in any embodiment; or, the processor 1001 is used to execute a computer program stored in a memory 1002 to implement, for example... Figure 9 The functions of each unit in the corresponding embodiment.

[0270] For example, a computer program may be divided into one or more modules / units, one or more of which are stored in memory 1002 and executed by processor 1001 to complete the embodiments of this application. One or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in a computer device.

[0271] The electronic device may include, but is not limited to, processor 1001 and memory 1002. Those skilled in the art will understand that the illustrations are merely examples of an electronic device and do not constitute a limitation on the electronic device. It may include more or fewer components than illustrated, or combine certain components, or use different components.

[0272] Processor 1001 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting various parts of the electronic device through various interfaces and lines.

[0273] The memory 1002 can be used to store computer programs and / or modules. The processor 1001 implements various functions of the computer device by running or executing the computer programs and / or modules stored in the memory 1002 and by calling the data stored in the memory 1002. The memory 1002 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device (such as audio data, video data, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0274] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the associated object identification device, electronic device and its corresponding unit described above can be referred to the description of the associated object identification method in any embodiment, and will not be repeated here.

[0275] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be accomplished by instructions, or by instructions controlling related hardware. These instructions can be stored in a storage medium and loaded and executed by a processor.

[0276] Therefore, embodiments of this application provide a storage medium storing a computer program. When the computer program is executed by a processor, it performs the steps in the associated object identification method in any embodiment of this application. For specific operations, please refer to the description of the associated object identification method in any embodiment, which will not be repeated here.

[0277] The storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0278] Since the instructions stored in the storage medium can execute the steps in the associated object identification method in any embodiment of this application, the beneficial effects that the associated object identification method in any embodiment of this application can achieve can be realized, as detailed in the preceding description, and will not be repeated here.

[0279] The foregoing has provided a detailed description of an associated object identification method, apparatus, storage medium, and electronic device provided in the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for identifying associated objects, characterized in that, include: Obtain the target object to be identified; Based on the target object's business scope information and logistics information, the target object's actual business information is determined; Based on the actual operating information of the target object and the actual operating information of the preset first candidate object, select the upstream object and / or downstream object associated with the target object from the first candidate object; The step of determining the actual operating information of the target object based on its business scope information and logistics information includes: Obtain multiple second candidate objects that have the same industry information as the target object; Extract the first keyword information from the business scope information of the multiple second candidate objects, which appears more than a preset frequency threshold; By using a preset association recognition model, the logistics information, business scope information, and first keyword information of the target object are processed to obtain the first target coefficient corresponding to the target object; Based on the first target coefficient, the first keyword information is weighted and then fused with the business scope information of the target object to obtain the actual business information of the target object; Before selecting upstream and / or downstream objects associated with the target object from the first candidate objects based on the actual operating information of the target object and the actual operating information of the preset first candidate objects, the method further includes: Obtain multiple third candidate objects that have the same industry information as the first candidate object; Extract second keyword information from the business scope information of the multiple third candidate objects, where the number of occurrences exceeds a preset threshold. By using a preset association recognition model, the logistics information, business scope information, and second keyword information of the first candidate object are processed to obtain the second target coefficient corresponding to the first candidate object. Based on the second target coefficient, the second keyword information is weighted and then fused with the business scope information of the first candidate to obtain the actual business information of the first candidate.

2. The method for identifying associated objects according to claim 1, characterized in that, The preset association recognition model is trained using the following method: Obtain target order information with order type "target", and designate the sender object and the recipient object in the same target order information as upstream objects, and the first downstream object associated with the upstream objects, respectively; Based on the business scope and logistics information of the upstream object and the business scope and logistics information of the first downstream object, positive sample data is constructed. The initial association recognition model is trained using the positive sample data to obtain the preset association recognition model.

3. The method for identifying associated objects according to claim 2, characterized in that, The step of training the initial association recognition model using the positive sample data to obtain the preset association recognition model includes: For each upstream object, based on the industry information of the corresponding first downstream object, a second downstream object corresponding to each upstream object is selected from a preset sample object. The second downstream object corresponding to each upstream object refers to a sample object that is not in an upstream-downstream relationship with each upstream object. Based on the business scope information and mailing information of the upstream object, and the business scope information and mailing information of the second downstream object corresponding to the upstream object, negative sample data is constructed. Based on the positive sample data and the negative sample data, the initial association recognition model is trained to obtain the preset association recognition model.

4. The method for identifying associated objects according to claim 3, characterized in that, For each upstream object, selecting a second downstream object corresponding to each upstream object from a preset sample object based on the industry information of the corresponding first downstream object includes: For each upstream object, the industry information of the first downstream object corresponding to each upstream object is statistically analyzed to obtain the downstream industry distribution information corresponding to each upstream object. The industry distribution information includes the industry proportion of multiple different industry information. By statistically analyzing the industry information of all first downstream objects, global industry distribution information is obtained, wherein the global industry distribution information includes the industry proportions corresponding to different industry information. Based on the industry information of the sample objects, the individual industry proportion of each upstream object is determined from the downstream industry distribution information corresponding to each upstream object, and the global industry proportion of the sample objects is determined from the global industry distribution information. Based on the individual industry share of each upstream object and the global industry share of the sample object, a second downstream object corresponding to each upstream object is selected from the sample object.

5. The method for identifying associated objects according to any one of claims 1-4, characterized in that, The step of selecting upstream and / or downstream objects associated with the target object from the first candidate objects based on the actual operating information of the target object and the actual operating information of the preset first candidate objects includes: Obtain the object identity information of the target object, and the object identity information of a preset first candidate object; Based on the actual operating information, object identity information, and logistics information of the target object, as well as the actual operating information, object identity information, and logistics information of the first candidate object, upstream objects and / or downstream objects associated with the target object are selected from the first candidate objects.

6. The method for identifying associated objects according to claim 5, characterized in that, The step of selecting upstream and / or downstream objects associated with the target object from the first candidate objects based on the actual operating information, object identity information, and logistics information of the target object, and the actual operating information, object identity information, and logistics information of the first candidate objects, includes: The actual operating information, object identity information, and logistics information of the target object, as well as the actual operating information, object identity information, and logistics information of the first candidate object, are fused to obtain fused information. By using a preset association relationship recognition model, the fused information is predicted to obtain the association probability between the target object and the first candidate object. The first candidate object with an association probability greater than a preset probability threshold is set as the upstream and / or downstream object associated with the target object.

7. A device for identifying associated objects, characterized in that, include: The acquisition unit is used to acquire the target object to be identified. The determining unit is used to determine the actual operating information of the target object based on the target object's business scope information and the target object's logistics information; The selection unit is used to select an upstream object and / or downstream object associated with the target object from the first candidate object based on the actual operating information of the target object and the actual operating information of the preset first candidate object. The determining unit is further configured to acquire multiple second candidate objects with the same industry information as the target object; and extract first keyword information from the business scope information of the multiple second candidate objects, which appears more than a preset frequency threshold. By using a preset association recognition model, the logistics information, business scope information, and first keyword information of the target object are processed to obtain the first target coefficient corresponding to the target object; Based on the first target coefficient, the first keyword information is weighted and then fused with the business scope information of the target object to obtain the actual business information of the target object; The determining unit is further configured to acquire multiple third candidate objects with the same industry information as the first candidate object; and extract second keyword information from the business scope information of the multiple third candidate objects, where the number of occurrences is greater than a preset threshold. By using a preset association recognition model, the logistics information, business scope information, and second keyword information of the first candidate object are processed to obtain the second target coefficient corresponding to the first candidate object. Based on the second target coefficient, the second keyword information is weighted and then fused with the business scope information of the first candidate to obtain the actual business information of the first candidate.

8. An electronic device, characterized in that, The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the associated object identification method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the steps of the associated object identification method according to any one of claims 1 to 6.

Images