Entity disambiguation method and device

Abstract

The invention provides an entity disambiguation method and device, and the method comprises the steps: selecting a candidate entity node pair comprising a first node and a second node from a plurality of entity nodes of an enterprise knowledge graph; when the first key information of the first node is not matched with the second key information of the second node, determining a first probability that the first node and the second node are the same entity node based on the multiple pieces of auxiliary information; predicting a second probability that the first node and the second node are the same entity node based on a graph neural network corresponding to the enterprise knowledge graph; and based on the first probability and the second probability, determining a target probability that the first node and the second node are the same entity node, and when the target probability is greater than a probability threshold, disambiguating the first node and the second node. Therefore, the accuracy of entity disambiguation can be improved.

Description

Technical Field

[0001] The present application relates to the field of Internet technology, and in particular to an entity disambiguation method and device. Background Art

[0002] In related technologies, if one wants to disambiguate entity nodes in a knowledge graph, most of them perform similarity matching based on basic attributes such as name and contact information. This makes it difficult to fully utilize the high-dimensional complex associations in the knowledge graph, resulting in low accuracy in entity disambiguation. Summary of the invention

[0003] The embodiments of the present application provide an entity disambiguation method and device, which can improve the accuracy of entity disambiguation.

[0004] The technical solution of the embodiment of the present application is implemented as follows: The present application embodiment provides an entity disambiguation method, the method comprising: Acquire an enterprise knowledge graph, wherein the enterprise knowledge graph includes a plurality of entity nodes, wherein the plurality of entity nodes include at least one of enterprise personnel and enterprises, and each entity node has corresponding key information; Selecting a candidate entity node pair from the enterprise knowledge graph, wherein the candidate entity node pair includes a first node and a second node; When the first key information of the first node does not match the second key information of the second node, respectively obtain multiple auxiliary information of the first node and the second node, and determine a first probability that the first node and the second node are the same entity node based on the multiple auxiliary information; Obtaining a graph neural network corresponding to the enterprise knowledge graph, and based on the graph neural network, predicting a second probability that the first node and the second node are the same entity node; Based on the first probability and the second probability, a target probability that the first node and the second node are the same entity node is determined, and when the target probability reaches a probability threshold, the first node and the second node are disambiguated.

[0005] The present application embodiment provides an entity disambiguation device, the device comprising: An acquisition module, used to acquire an enterprise knowledge graph, wherein the enterprise knowledge graph includes a plurality of entity nodes, wherein the plurality of entity nodes include at least one of enterprise personnel and enterprises, and each entity node has corresponding key information; A selection module, configured to select a candidate entity node pair from the enterprise knowledge graph, wherein the candidate entity node pair includes a first node and a second node; a first determination module, configured to respectively obtain a plurality of auxiliary information of the first node and the second node when the first key information of the first node does not match the second key information of the second node, and determine a first probability that the first node and the second node are the same entity node based on the plurality of auxiliary information; A prediction module, used to obtain a graph neural network corresponding to the enterprise knowledge graph, and based on the graph neural network, predict a second probability that the first node and the second node are the same entity node; The second determination module is used to determine the target probability that the first node and the second node are the same entity node based on the first probability and the second probability, and to disambiguate the first node and the second node when the target probability reaches a probability threshold.

[0006] In the above scheme, the key information includes the identifier of the enterprise personnel corresponding to the entity node, the parent company of the enterprise to which the entity node belongs, the enterprise name of the enterprise to which the entity node belongs, the beneficiary of the enterprise to which the entity node belongs, the control relationship of the enterprise to which the entity node belongs, and at least two of the actual controller of the enterprise to which the entity node belongs; the device also includes a matching module, which is used to match the identifier of the enterprise personnel corresponding to the first node with the identifier of the enterprise personnel corresponding to the second node when the key information includes the identifier of the enterprise personnel corresponding to the entity node to obtain a first matching result; when the key information includes the parent company of the enterprise to which the entity node belongs, the parent company of the enterprise to which the first node belongs with the parent company of the enterprise to which the second node belongs to obtain a second matching result; when the key information includes the enterprise name of the enterprise to which the entity node belongs, the matching module is used to match the parent company of the enterprise personnel corresponding to the first node with the parent company of the enterprise personnel corresponding to the second node to obtain a second matching result. The enterprise name of the enterprise to which the first node belongs is matched with the enterprise name of the enterprise to which the second node belongs to obtain a third matching result; when the key information includes the beneficiary of the enterprise to which the entity node belongs, the beneficiary of the enterprise to which the first node belongs is matched with the beneficiary of the enterprise to which the second node belongs to obtain a fourth matching result; when the key information includes the control relationship of the enterprise to which the entity node belongs, the control relationship of the enterprise to which the first node belongs is matched with the control relationship of the enterprise to which the second node belongs to obtain a fifth matching result; when the key information includes the actual controller of the enterprise to which the entity node belongs, the actual controller of the enterprise to which the first node belongs is matched with the actual controller of the enterprise to which the second node belongs to obtain a sixth matching result; when the matching results obtained based on the key information matching of the first node and the second node both fail to match, it is determined that the first key information does not match the second key information.

[0007] In the above scheme, the first determination module is also used to obtain, for each of the auxiliary information, the target similarity between the auxiliary information of the first node and the auxiliary information of the second node; obtain a reference enterprise knowledge graph including multiple reference entity nodes; for each of the auxiliary information, based on the reference enterprise knowledge graph, when two reference entity nodes belong to the same entity node, determine a first reference probability that the similarity of the two reference entity nodes reaches the target similarity, and when the two reference entity nodes do not belong to the same entity node, determine a second reference probability that the similarity reaches the target similarity; based on the first reference probabilities corresponding to multiple of the auxiliary information and the second reference probabilities corresponding to multiple of the auxiliary information, determine a first probability that the first node and the second node are the same entity node.

[0008] In the above scheme, the first determination module is also used to input the auxiliary information of the first node and the auxiliary information of the second node into the probability prediction model for each of the auxiliary information to obtain the intermediate probability that the first node and the second node are the same entity node; and fuse the intermediate probabilities corresponding to each of the auxiliary information to obtain the first probability that the first node and the second node are the same entity node.

[0009] In the above scheme, the prediction module is also used to obtain the attribute information of the first node and the attribute information of each first neighbor node of the first node based on the graph neural network; perform feature extraction on the attribute information of the first node to obtain the attribute characteristics of the first node, and perform feature extraction on the attribute information of each first neighbor node to obtain the attribute characteristics of each first neighbor node; determine the correlation value between each first neighbor node and the first node based on the attribute characteristics of the first node and the attribute characteristics of each first neighbor node; determine the first feature vector of the first node based on each of the correlation values ​​and the attribute characteristics of each first neighbor node, and obtain the second feature vector of the second node; based on the first feature vector and the second feature vector, predict the second probability that the first node and the second node are the same entity node.

[0010] In the above scheme, the prediction module is also used to perform a dot product on the first eigenvector and the second eigenvector to obtain a dot product result, and multiply the modulus of the first eigenvector and the modulus of the second eigenvector to obtain a product; determine a ratio of the dot product result to the product, and use the ratio as the second probability.

[0011] In the above scheme, the prediction module is also used to perform the following processing for each of the first neighbor nodes: splicing the attribute characteristics of the first node with the attribute characteristics of the first neighbor node to obtain a first splicing feature, and based on the first splicing feature, determining the correlation between the first neighbor node and the first node; normalizing the correlation to obtain a correlation value, and the correlation value is used to indicate the importance of the corresponding first neighbor node to the first node.

[0012] In the above scheme, the prediction module is also used to obtain a first time point, the first time point being the time point at which the second neighbor node of the first node or the connection edge of the first node most recently changed; obtain a second time point, the second time point being the time point at which the third neighbor node of the first neighbor node or the connection edge of the first neighbor node most recently changed; determine the time feature based on the first time point and the second time point; obtain the relationship type of the connection relationship between the first node and the first neighbor node, and perform feature extraction on the relationship type to obtain a relationship type feature; splice the first splicing feature, the time feature and the relationship type feature to obtain a second splicing feature; and determine the correlation between the first neighbor node and the first node based on the second splicing feature.

[0013] In the above scheme, the selection module is also used to obtain the name of each entity node in the enterprise knowledge graph; based on the name of each entity node, the names of any two entity nodes in the enterprise knowledge graph are matched for similarity, and two entity nodes whose similarity is greater than or equal to the similarity threshold are selected as the candidate entity node pair.

[0014] An embodiment of the present application provides an electronic device, including: A memory for storing executable instructions; The processor is used to implement the entity disambiguation method provided in the embodiment of the present application when executing the executable instructions stored in the memory.

[0015] An embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, the processor will execute the entity disambiguation method provided by the embodiment of the present application.

[0016] An embodiment of the present application provides a computer program product, which includes a computer program or computer executable instructions. A processor executes the computer program or computer executable instructions so that the electronic device executes the entity disambiguation method provided by the embodiment of the present application.

[0017] The embodiments of the present application have the following beneficial effects: After determining the candidate entity node pair including the first node and the second node in the enterprise knowledge graph, by matching the key information of the first node with the key information of the second node, if the key information of the first node and the second node do not match, based on multiple auxiliary information of the first node and the second node, determine the first probability that the first node and the second node are the same entity node, and based on the graph neural network corresponding to the enterprise knowledge graph, predict the second probability that the first node and the second node are the same entity node, and then determine the target probability based on the first probability and the second probability, so that when the target probability reaches the probability threshold, the first node and the second node are disambiguated. In this way, compared with the method of determining whether it is the same entity node by similarity matching based only on basic attributes such as the name and contact information of the node, combining key information, auxiliary information and graph neural network to determine whether the first node and the second node are the same entity node, enriches the judgment dimension of determining whether the two nodes are the same entity node, improves the accuracy of determining whether the two nodes are the same entity node, and thus improves the accuracy of entity disambiguation. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a schematic diagram of the architecture of an entity disambiguation system 100 provided in an embodiment of the present application; Figure 2 is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application; Figure 3 is a flowchart of an entity disambiguation method provided in an embodiment of the present application; Figure 4 is a schematic diagram of a first process for determining a first probability provided in an embodiment of the present application; Figure 5 is a schematic diagram of a second process for determining a first probability provided in an embodiment of the present application; Figure 6 It is a schematic diagram of the process of implementing the prediction of the second probability in this application; Figure 7 This is a schematic diagram of a first node and a first neighbor node provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0020] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0021] In the following description, the terms "first\second\third" involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0023] Before further describing the embodiments of the present application in detail, the nouns and terms involved in the embodiments of the present application are explained. The nouns and terms involved in the embodiments of the present application are subject to the following interpretations.

[0024] 1) Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. Artificial intelligence technology is an interdisciplinary subject that covers a wide range of fields, including both hardware-level and software-level technologies. Basic AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics and other technologies.

[0025] 2) Client, also known as the user end, refers to the program corresponding to the server that provides local services to users. Except for some applications that can only run locally, it is generally installed on an ordinary client and needs to cooperate with the server to run, that is, it requires corresponding servers and service programs in the network to provide corresponding services. In this way, specific communication connections need to be established on the client and server sides to ensure the normal operation of the application.

[0026] 3) Entity disambiguation, also known as entity reference disambiguation or named entity disambiguation, aims to merge entity references (such as names of people, places, organizations, etc.) that refer to the same entity in the text.

[0027] 4) Graph Neural Networks (GNN) is a type of artificial neural network used to process data that can be represented as graphs. Compared with traditional two-dimensional convolutional neural networks that act on two-dimensional images, graph neural networks expand their objects of action to graph data that can represent three-dimensional grid forms. The key design element of graph neural networks is the use of paired message passing so that graph nodes can be iteratively updated by exchanging information with their neighbors.

[0028] 5) Knowledge graph is a structured semantic knowledge base consisting of nodes and edges connecting two nodes, which organizes and represents entities or concepts and the relationships between them in the form of a graph; in a knowledge graph, nodes usually represent entities or concepts, while edges represent various relationships between nodes (such as "belongs to", "located in", "creates", etc.).

[0029] See also Figure 1 , Figure 1 It is a schematic diagram of the architecture of the entity disambiguation system 100 provided in an embodiment of the present application. To realize the application scenario of entity disambiguation, a terminal (terminal 400 is shown as an example) is connected to a server 200 via a network 300. The network 300 may be a wide area network or a local area network, or a combination of the two. The terminal 400 is used for users to use a client 401, which is displayed on a display interface (display interface 401-1 is shown as an example). The terminal 400 and the server 200 are connected to each other via a wired or wireless network.

[0030] The terminal 400 is used to send the enterprise knowledge graph to the server 200; Server 200 is used to obtain an enterprise knowledge graph, which includes multiple entity nodes, and the multiple entity nodes include at least one of enterprise personnel and enterprises, and each entity node has corresponding key information; select candidate entity node pairs from the enterprise knowledge graph, and the candidate entity node pairs include a first node and a second node; when the first key information of the first node does not match the second key information of the second node, respectively obtain multiple auxiliary information of the first node and the second node, and determine a first probability that the first node and the second node are the same entity node based on the multiple auxiliary information; obtain a graph neural network corresponding to the enterprise knowledge graph, and predict a second probability that the first node and the second node are the same entity node based on the graph neural network; determine a target probability that the first node and the second node are the same entity node based on the first probability and the second probability, and when the target probability reaches a probability threshold, disambiguate the first node and the second node.

[0031] In some embodiments, the server 200 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content distribution networks (CDN, Content Deliver Network), and big data and artificial intelligence platforms. The terminal 400 may be a smart phone, a tablet computer, a laptop computer, a desktop computer, a set-top box, an intelligent voice interaction device, a smart home appliance, a vehicle-mounted terminal, an aircraft, and a mobile device (e.g., a mobile phone, a portable music player, a personal digital assistant, a dedicated messaging device, a portable gaming device, a smart speaker, and a smart watch), etc., but is not limited thereto. The terminal device and the server may be directly or indirectly connected via wired or wireless communication, which is not limited in the embodiments of the present application.

[0032] See also Figure 2 , Figure 2 is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. In practical applications, the electronic device can be Figure 1 The server 200 or terminal 400 shown, see Figure 2 , Figure 2 The electronic device shown includes: at least one processor 410, a memory 450, at least one network interface 420 and a user interface 430. The various components in the electronic device are coupled together via a bus system 440. It is understood that the bus system 440 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 440 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, the bus system 440 is not shown in FIG. Figure 2 Various buses are labeled as bus system 440 .

[0033] The processor 410 may be an integrated circuit chip having signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., wherein the general-purpose processor may be a microprocessor or any conventional processor, etc.

[0034] The user interface 430 includes one or more output devices 431 that enable presentation of media content, including one or more speakers and / or one or more visual display screens. The user interface 430 also includes one or more input devices 432, including user interface components that facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons and controls.

[0035] The memory 450 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state memory, hard disk drives, optical disk drives, etc. The memory 450 may optionally include one or more storage devices that are physically remote from the processor 410.

[0036] The memory 450 includes a volatile memory or a non-volatile memory, and may also include both volatile and non-volatile memories. The non-volatile memory may be a read-only memory (ROM), and the volatile memory may be a random access memory (RAM). The memory 450 described in the embodiments of the present application is intended to include any suitable type of memory.

[0037] In some embodiments, memory 450 can store data to support various operations, examples of which include programs, modules, and data structures, or a subset or superset thereof, as exemplarily described below.

[0038] Operating system 451, including system programs for processing various basic system services and performing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks; A network communication module 452, used to reach other electronic devices via one or more (wired or wireless) network interfaces 420, exemplary network interfaces 420 include: Bluetooth, Wireless Compatibility Certification (WiFi), and Universal Serial Bus (USB), etc.; a presentation module 453 for enabling presentation of information via one or more output devices 431 (e.g., display screen, speaker, etc.) associated with the user interface 430 (e.g., a user interface for operating peripherals and displaying content and information); The input processing module 454 is used to detect one or more user inputs or interactions from one of the one or more input devices 432 and translate the detected inputs or interactions.

[0039] In some embodiments, the device provided in the embodiments of the present application can be implemented in software. Figure 2 The entity disambiguation device 455 stored in the memory 450 is shown, which can be software in the form of a program and a plug-in, etc., and includes the following software modules: an acquisition module 4551, a selection module 4552, a first determination module 4553, a prediction module 4554, and a second determination module 4555. These modules are logical, and therefore can be arbitrarily combined or further split according to the functions implemented. The functions of each module will be described below.

[0040] In other embodiments, the device provided in the embodiments of the present application may be implemented in hardware. As an example, the entity disambiguation device provided in the embodiments of the present application may be a processor in the form of a hardware decoding processor, which is programmed to execute the entity disambiguation method provided in the embodiments of the present application. For example, the processor in the form of a hardware decoding processor may adopt one or more application specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field programmable gate arrays (FPGAs), or other electronic components.

[0041] In some embodiments, the terminal or server can implement the entity disambiguation method provided in the embodiments of the present application by running a computer program. For example, the computer program can be a native program or software module in the operating system; it can be a native application (APP, Application), that is, a program that needs to be installed in the operating system to run, such as an instant messaging APP, a web browser APP; it can also be a small program, that is, a program that only needs to be downloaded to a browser environment to run; it can also be a small program that can be embedded in any APP. In short, the above-mentioned computer program can be an application, module or plug-in in any form.

[0042] Based on the above description of the entity disambiguation system and electronic device provided in the embodiment of the present application, the entity disambiguation method provided in the embodiment of the present application is described below. In actual implementation, the entity disambiguation method provided in the embodiment of the present application can be implemented by a terminal or a server alone, or by a terminal and a server in collaboration, so that Figure 1 The server 200 in the embodiment of the present application alone executes the entity disambiguation method provided in the embodiment of the present application as an example for explanation. Figure 3 , Figure 3 is a flow chart of the entity disambiguation method provided in the embodiment of the present application. Figure 3 The steps shown are explained.

[0043] Step 101, the server obtains the enterprise knowledge graph, the enterprise knowledge graph includes multiple entity nodes, the multiple entity nodes include at least one of enterprise personnel and enterprises, and each entity node has corresponding key information.

[0044] In actual implementation, enterprise multi-source data (such as equity, bidding and other information) is obtained from the database or the Internet, and data cleaning is performed on the enterprise multi-source data, such as deduplication, field unification, address / telephone / name formatting and outlier processing, to obtain enterprise standard data; based on the enterprise standard data, an enterprise knowledge graph is constructed. Specifically, the enterprise standard data is parsed to obtain the relationship between enterprises and natural persons (such as enterprise personnel) and between enterprises and enterprises, natural persons and natural persons, and between enterprises and natural persons in the enterprise standard data, and the enterprises and natural persons are mapped as entity nodes, and the relationship is mapped as an edge to obtain the enterprise knowledge graph; Among them, the enterprise knowledge graph is a directed multi-relationship graph G=(V, E, R), where V is the set of enterprise and predecessor nodes, E is the relationship edge used to indicate investment relationship, holding relationship, employment relationship, cooperation relationship or contact information, and R is a set of relationship types (such as investment, holding, same address, same phone number, etc.).

[0045] It should be noted that the key information can be used to determine the identity uniqueness of the corresponding entity node, wherein the key information includes at least two of the identification of the enterprise personnel corresponding to the entity node, the parent company of the enterprise to which the entity node belongs, the enterprise name of the enterprise to which the entity node belongs, the beneficiaries of the enterprise to which the entity node belongs, the control relationship of the enterprise to which the entity node belongs, and the actual controller of the enterprise to which the entity node belongs; wherein the control relationship is used to indicate other entity nodes controlled (such as holding or managing) by the corresponding entity node, and / or other entity nodes that control the corresponding entity node; Among them, multiple entity nodes include at least one of enterprise personnel and enterprises, which means that for each entity node, it can indicate the enterprise or the enterprise personnel, but because the name of the enterprise is unique, in the subsequent disambiguation process, the object of disambiguation is the entity node indicating the enterprise personnel.

[0046] Step 102, select candidate entity node pairs from the enterprise knowledge graph, where the candidate entity node pairs include a first node and a second node.

[0047] In actual implementation, the nodes in the enterprise knowledge graph are used to indicate entities, and entity node pairs are also entity pairs, which are used to indicate two entities. The process of selecting candidate entity nodes from the enterprise knowledge graph can be to obtain the name of each entity node in the enterprise knowledge graph; based on the name of each entity node, perform similarity matching on the names of any two entity nodes in the enterprise knowledge graph, and select two entity nodes whose similarity is greater than or equal to the similarity threshold as candidate entity node pairs.

[0048] It should be noted that the name of the entity node refers to the name of the enterprise or the name of the enterprise personnel. At the same time, the similarity threshold can be pre-set, for example, it can be 90% or 100%, etc., which is not limited in the embodiments of the present application; wherein, the candidate entity node pair can be one pair or more pairs, which is also not limited in the embodiments of the present application.

[0049] In some embodiments, after determining the candidate entity nodes, for each candidate entity node, it can be determined whether the two entity nodes included in the candidate entity nodes are the same entity node. Specifically, the first key information of the first node and the second key information of the second entity node are obtained, and the first key information and the second key information are matched. As described above, the key information includes at least two of the identification of the enterprise personnel corresponding to the entity node, the parent company of the enterprise to which the entity node belongs, the enterprise name of the enterprise to which the entity node belongs, the beneficiary of the enterprise to which the entity node belongs, the control relationship of the enterprise to which the entity node belongs, and the actual controller of the enterprise to which the entity node belongs; Based on this, when the key information includes the identifier of the enterprise personnel corresponding to the entity node, the identifier of the enterprise personnel corresponding to the first node is matched with the identifier of the enterprise personnel corresponding to the second node to obtain a first matching result; when the key information includes the parent company of the enterprise to which the entity node belongs, the parent company of the enterprise to which the first node belongs is matched with the parent company of the enterprise to which the second node belongs to obtain a second matching result; when the key information includes the enterprise name of the enterprise to which the entity node belongs, the enterprise name of the enterprise to which the first node belongs is matched with the enterprise name of the enterprise to which the second node belongs to obtain a third matching result; when the key information includes the beneficiary of the enterprise to which the entity node belongs, the beneficiary of the enterprise to which the first node belongs is matched with the beneficiary of the enterprise to which the second node belongs to obtain a fourth matching result; when the key information includes the control relationship of the enterprise to which the entity node belongs, the control relationship of the enterprise to which the first node belongs is matched with the control relationship of the enterprise to which the second node belongs to obtain a fifth matching result; when the key information includes the actual controller of the enterprise to which the entity node belongs, the actual controller of the enterprise to which the first node belongs is matched with the actual controller of the enterprise to which the second node belongs to obtain a sixth matching result; when the matching results obtained based on the key information matching of the first node and the second node both fail to match, it is determined that the first key information does not match the second key information.

[0050] It should be noted that matching the identification of the enterprise personnel corresponding to the first node with the identification of the enterprise personnel corresponding to the second node means determining whether the identification of the enterprise personnel corresponding to the first node is consistent with the identification of the enterprise personnel corresponding to the second node, that is, the first matching result is used to indicate whether the identification of the enterprise personnel corresponding to the first node is consistent with the identification of the enterprise personnel corresponding to the second node. If they are consistent, it means that the first matching result is matched successfully, and if they are inconsistent, it means that the first matching result fails to match; Matching the parent company of the enterprise to which the first node belongs with the parent company of the enterprise to which the second node belongs refers to determining whether the parent company of the enterprise to which the first node belongs is consistent with the parent company of the enterprise to which the second node belongs, that is, the second matching result is used to indicate whether the parent company of the enterprise to which the first node belongs is consistent with the parent company of the enterprise to which the second node belongs. If they are consistent, it means that the second matching result is matched successfully, and if they are inconsistent, it means that the second matching result is matched unsuccessfully; Matching the enterprise name of the enterprise to which the first node belongs with the enterprise name of the enterprise to which the second node belongs refers to determining whether the enterprise name of the enterprise to which the first node belongs is consistent with the enterprise name of the enterprise to which the second node belongs, that is, the third matching result is used to indicate whether the enterprise name of the enterprise to which the first node belongs is consistent with the enterprise name of the enterprise to which the second node belongs. If they are consistent, it means that the third matching result is matched successfully, and if they are inconsistent, it means that the third matching result is matched unsuccessfully; Matching the beneficiary of the enterprise to which the first node belongs with the beneficiary of the enterprise to which the second node belongs refers to determining whether the beneficiary of the enterprise to which the first node belongs is consistent with the beneficiary of the enterprise to which the second node belongs, that is, the fourth matching result is used to indicate whether the beneficiary of the enterprise to which the first node belongs is consistent with the beneficiary of the enterprise to which the second node belongs. If they are consistent, it means that the fourth matching result is matched successfully, and if they are inconsistent, it means that the fourth matching result is matched unsuccessfully; Matching the control relationship of the enterprise to which the first node belongs with the control relationship of the enterprise to which the second node belongs refers to determining whether there is a control relationship between the control relationship of the enterprise to which the first node belongs and the enterprise to which the second node belongs, for example, whether the enterprise to which the first node belongs controls or manages the enterprise to which the second node belongs, or whether the enterprise to which the first node belongs is controlled or managed by the enterprise to which the second node belongs, that is, the fifth matching result is used to indicate whether there is a control relationship between the enterprise to which the first node belongs and the enterprise to which the second node belongs. If there is a control relationship, it means that the fifth matching result is matched successfully, and if there is no control relationship, it means that the fifth matching result is matched unsuccessfully; Matching the actual controller of the enterprise to which the first node belongs with the actual controller of the enterprise to which the second node belongs refers to determining whether the actual controller of the enterprise to which the first node belongs is consistent with the actual controller of the enterprise to which the second node belongs, that is, the fourth matching result is used to indicate whether the actual controller of the enterprise to which the first node belongs is consistent with the actual controller of the enterprise to which the second node belongs. If they are consistent, it means that the sixth matching result is matched successfully, and if they are inconsistent, it means that the sixth matching result fails to match.

[0051] Then, when all matching results based on matching the first key information with the second key information fail to match, it is determined that the first key information does not match the second key information; when there is a matching result based on matching the first key information with the second key information that matches successfully, it is determined that the first key information matches the second key information.

[0052] It should be noted that, for the enterprise to which the entity node belongs, if the entity node is an enterprise, then the enterprise to which the entity node belongs refers to the enterprise indicated by the entity node; if the entity node is an enterprise employee, then the enterprise to which the entity node belongs refers to the enterprise to which the enterprise employee corresponding to the entity node belongs; among them, for the control relationship of the enterprise to which the entity node belongs, if the entity node is an enterprise, then the control relationship of the enterprise to which the entity node belongs refers to the control relationship of the enterprise indicated by the entity node; if the entity node is an enterprise employee, then the control relationship of the enterprise to which the entity node belongs refers to the control relationship of the enterprise to which the enterprise employee indicated by the entity node belongs, or the control relationship of the enterprise employee indicated by the entity node, such as holding or managing other entity nodes.

[0053] Step 103, when the first key information of the first node does not match the second key information of the second node, obtain multiple auxiliary information of the first node and the second node respectively, and determine a first probability that the first node and the second node are the same entity node based on the multiple auxiliary information.

[0054] In actual implementation, when the first key information of the first node matches the second key information of the second node, it is determined that the first node and the second node are the same entity node, thereby disambiguating the first node and the second node; when the first key information of the first node does not match the second key information of the second node, multiple auxiliary information of the first node and the second node are obtained respectively; wherein the auxiliary information may be the address information, industry information, registration time, registered capital, number of joint bids, contact person, etc. of the entity node.

[0055] In actual implementation, there are multiple ways to determine the first probability that the first node and the second node are the same entity node based on multiple auxiliary information. Next, the process of determining the first probability that the first node and the second node are the same entity node based on multiple auxiliary information is explained based on different methods.

[0056] In some embodiments, see Figure 4 , Figure 4 is a schematic diagram of a first process for determining a first probability provided in an embodiment of the present application, based on Figure 4 The process of determining a first probability that the first node and the second node are the same entity node based on the plurality of auxiliary information is implemented by the following steps.

[0057] Step 1031a: for each auxiliary information, obtain the target similarity between the auxiliary information of the first node and the auxiliary information of the second node.

[0058] In actual implementation, for each auxiliary information, the process of obtaining the target similarity between the auxiliary information of the first node and the auxiliary information of the second node can be, for each auxiliary information, matching the auxiliary information of the first node with the auxiliary information of the second node to obtain the target similarity between the auxiliary information of the first node and the auxiliary information of the second node.

[0059] Step 1032a, obtaining a reference enterprise knowledge graph including multiple reference entity nodes.

[0060] It should be noted that the reference enterprise knowledge graph here is a labeled enterprise knowledge graph, that is, the entity nodes that are the same entity nodes have been marked in the enterprise knowledge graph, which is equivalent to which entity nodes have been marked as the same entity nodes in the enterprise knowledge graph.

[0061] Step 1033a, for each auxiliary information, based on the reference enterprise knowledge graph, when the two reference entity nodes belong to the same entity node, determine the first reference probability that the similarity of the two reference entity nodes reaches the target similarity, and when the two reference entity nodes do not belong to the same entity node, determine the second reference probability that the similarity reaches the target similarity.

[0062] It should be noted that, based on the reference enterprise knowledge graph, when two reference entity nodes belong to the same entity node, determining the first reference probability that the similarity of the two reference entity nodes reaches the target similarity means that, in the reference enterprise knowledge graph, for two reference entity nodes belonging to the same entity node, determining the first reference probability that the similarity of the corresponding auxiliary information of the two entity nodes reaches the target similarity, that is, obtaining all reference entity node pairs belonging to the same entity node in the reference enterprise knowledge graph; obtaining the number of reference entity node pairs whose similarity of corresponding auxiliary information reaches the target similarity among all reference entity node pairs belonging to the same entity node, and taking the ratio of the number to the total number of all reference entity node pairs belonging to the same entity node as the first reference probability corresponding to the corresponding auxiliary information, that is, determining the first reference probability that the similarity of the two reference entity nodes reaches the target similarity when the two reference entity nodes belong to the same entity node; Accordingly, based on the reference enterprise knowledge graph, when two reference entity nodes do not belong to the same entity node, determining the first reference probability that the similarity of the two reference entity nodes reaches the target similarity means, in the reference enterprise knowledge graph, for two reference entity nodes that do not belong to the same entity node, determining the first reference probability that the similarity of the corresponding auxiliary information of the two entity nodes reaches the target similarity, that is, obtaining all reference entity node pairs that do not belong to the same entity node in the reference enterprise knowledge graph; obtaining the number of reference entity node pairs whose corresponding auxiliary information similarity reaches the target similarity among all reference entity node pairs that do not belong to the same entity node, and taking the ratio of the number to the total number of all reference entity node pairs that do not belong to the same entity node as the second reference probability corresponding to the corresponding auxiliary information, that is, the above-mentioned second reference probability that the similarity of the two reference entity nodes reaches the target similarity when the two reference entity nodes do not belong to the same entity node.

[0063] Step 1034a: Determine a first probability that the first node and the second node are the same entity node based on the first reference probabilities corresponding to the plurality of auxiliary information and the second reference probabilities corresponding to the plurality of auxiliary information.

[0064] It should be noted that, since each auxiliary information corresponds to a target similarity in the foregoing text, each auxiliary information corresponds to a first reference probability and a second reference probability. Based on this, based on the first reference probabilities corresponding to multiple auxiliary information and the second reference probabilities corresponding to multiple auxiliary information, the process of determining the first probability that the first node and the second node are the same entity node can be to perform a continuous product of the multiple first reference probabilities to obtain a third probability, and to perform a continuous product of the multiple second reference probabilities to obtain a fourth probability, that is: ...Formula (1); ...Formula (2); in, Refers to auxiliary information, refers to the i-th auxiliary information, x refers to the target similarity, refers to the target similarity corresponding to the i-th auxiliary information, n refers to the number of auxiliary information, S refers to the case of belonging to the same entity node, and Diff refers to the case of not belonging to the same entity node. Refers to the first reference probability corresponding to the i-th auxiliary information, Refers to the second reference probability corresponding to the i-th auxiliary information, refers to the third probability, Refers to the fourth probability.

[0065] Then, for any two entity nodes in the enterprise knowledge graph, the marginal probability that can achieve the target similarity corresponding to the auxiliary information is determined, and based on the reference enterprise knowledge graph, the fifth probability indicating the proportion of reference entity node pairs belonging to the same entity node and the sixth probability indicating the proportion of reference entity node pairs belonging to non-identical entity nodes are determined; based on the third probability, the fourth probability, the fifth probability, the sixth probability and the marginal probability, the first probability that the first node and the second node are the same entity node is determined.

[0066] It should be noted that the marginal probability is also the total probability. For any two entity nodes in the enterprise knowledge graph, the process of determining the marginal probability of reaching the similarity corresponding to the auxiliary information is to obtain the entity node pairs in the enterprise knowledge graph, and then obtain the number of entity node pairs in which at least one auxiliary information in the entity node pairs can reach the corresponding target similarity, and take the ratio of this number to the entity node pairs in the enterprise knowledge graph as the marginal probability; The process of determining the fifth probability indicating the proportion of reference entity node pairs belonging to the same entity node based on the reference enterprise knowledge graph may be to obtain reference entity node pairs in the reference enterprise knowledge graph, then obtain the number of reference entity node pairs belonging to the same entity node in the reference entity node pairs, and use the ratio of the number to the reference entity node pairs in the reference enterprise knowledge graph as the fifth probability indicating the proportion of reference entity node pairs belonging to the same entity node; Accordingly, based on the reference enterprise knowledge graph, the process of determining the sixth probability indicating the proportion of reference entity node pairs that belong to non-identical entity nodes may be to obtain reference entity node pairs in the reference enterprise knowledge graph, then obtain the number of reference entity node pairs that do not belong to the same entity node in the reference entity node pairs, and use the ratio of the number to the reference entity node pairs in the reference enterprise knowledge graph as the sixth probability indicating the proportion of reference entity node pairs that belong to non-identical entity nodes; In actual implementation, based on the third probability, the fourth probability, the fifth probability, the sixth probability and the marginal probability, the process of determining the first probability that the first node and the second node are the same entity node may be: taking the product of the third probability and the fifth probability as the first product, and taking the product of the fourth probability and the sixth probability as the second product; obtaining the ratio of the first product and the marginal probability as the first ratio, and summing the first product and the second product to obtain a first summation result; obtaining the ratio of the first product and the first summation result as the second ratio, and subtracting the first ratio from the second ratio to obtain the first probability, that is: ...Formula (3); in, is the first probability, is the third probability, is the fourth probability, is the marginal probability, is the fifth probability, The sixth probability.

[0067] In other embodiments, see Figure 5 , Figure 5 is a schematic diagram of a second process for determining a first probability provided in an embodiment of the present application, based on Figure 5 The process of determining the first probability that the first node and the second node are the same entity node based on the plurality of auxiliary information is implemented by the following steps.

[0068] Step 1031b, for each piece of auxiliary information, input the auxiliary information of the first node and the auxiliary information of the second node into the probability prediction model to obtain the intermediate probability that the first node and the second node are the same entity node.

[0069] In actual implementation, the auxiliary information of the first node and the auxiliary information of the second node are input into the probability prediction model, and the intermediate probability that the first node and the second node are the same entity node is obtained, which is the intermediate probability that the first node and the second node indicated by each auxiliary information are the same entity node, wherein the intermediate probability here can be the similarity between the auxiliary information of the first node and the auxiliary information of the second node, that is, the target similarity mentioned above.

[0070] Step 1032b: fuse the intermediate probabilities corresponding to each auxiliary information to obtain a first probability that the first node and the second node are the same entity node.

[0071] In actual implementation, there are many ways to fuse the intermediate probabilities corresponding to each auxiliary information to obtain the first probability that the first node and the second node are the same entity node. Next, taking two of the methods as examples, the process of fusing the intermediate probabilities corresponding to each auxiliary information to obtain the first probability that the first node and the second node are the same entity node is explained.

[0072] In some embodiments, the process of fusing the intermediate probabilities corresponding to each auxiliary information to obtain the first probability that the first node and the second node are the same entity node can be to average the intermediate probabilities corresponding to each auxiliary information to obtain the first probability that the first node and the second node are the same entity node; or, to obtain the weight of the intermediate probability corresponding to each auxiliary information, and based on the weight, to perform weighted summation of the intermediate probabilities corresponding to each auxiliary information to obtain the first probability that the first node and the second node are the same entity node.

[0073] In other embodiments, the process of fusing the intermediate probabilities corresponding to each auxiliary information to obtain the first probability that the first node and the second node are the same entity node may be: obtaining a first parameter value and multiple second parameter values, where the second parameter values ​​correspond to the auxiliary information one by one; multiplying the intermediate probability with the corresponding second parameter value to obtain a third product, and summing the multiple third products and the first parameter value to obtain a second summation result; and determining the first probability based on the second summation result, that is: ...Formula (4); in, is the first probability, is the intermediate probability corresponding to the ith auxiliary information, that is, the target similarity corresponding to the ith auxiliary information, is the first parameter value, is the second parameter value corresponding to the i-th auxiliary information, and n is the number of auxiliary information.

[0074] It should be noted that to It may be pre-set or obtained through training, and this is not limited in the embodiments of the present application.

[0075] In some embodiments, the above two methods can also be combined to determine the first probability. For example, after obtaining the corresponding first probabilities based on the above two methods, the two first probabilities are averaged or weighted summed to obtain the final first probability. This is not limited to the embodiments of the present application.

[0076] Step 104: obtain a graph neural network corresponding to the enterprise knowledge graph, and based on the graph neural network, predict a second probability that the first node and the second node are the same entity node.

[0077] In actual implementation, the graph neural network corresponding to the enterprise knowledge graph is obtained, and the enterprise knowledge graph is used as the input of the graph neural network to predict the second probability that the first node and the second node are the same entity node; wherein, see Figure 6 , Figure 6 This is a flow chart of the implementation of the second probability prediction in this application, based on Figure 6 , based on the graph neural network, the process of predicting the second probability that the first node and the second node are the same entity node can be achieved through the following steps.

[0078] Step 1041, based on the graph neural network, obtain the attribute information of the first node and the attribute information of each first neighbor node of the first node.

[0079] It should be noted that for attribute information, if the corresponding entity node indicates an enterprise, the attribute information may be industry type, registered capital, establishment time or change time, address information, business status (such as continued existence, cancellation or revocation), enterprise scale (such as number of employees, number of branches), etc.; if the corresponding entity node indicates enterprise personnel, the attribute information refers to identity identification (such as ID card, passport), role attributes (such as whether legal representative, senior executive, shareholder), historical employment record (such as length of service in the company, companies where the employee has worked), other public information (such as number of investments participated in, geographical characteristics), etc.

[0080] Step 1042: extract features from the attribute information of the first node to obtain the attribute features of the first node, and extract features from the attribute information of each first neighboring node to obtain the attribute features of each first neighboring node.

[0081] In actual implementation, after obtaining the attribute information of the entity node, feature extraction is performed on the attribute information of the entity node to obtain the attribute features of the entity node, that is, to convert the attribute information of the entity node into a numerical feature vector, and obtain the attribute features of each entity node, that is, the feature vector ,in, , where F is the feature dimension. For example, if the entity node is an enterprise, the attribute information includes industry type, registered capital, establishment time, and address information. Then the industry type is discretized to obtain the industry type code; the registered capital is logarithmically or bucketed to obtain the numerical feature (Registered capital); extract the time difference between the establishment time and the current time, and normalize the time difference to obtain the time feature; encode the address by geographic coordinates to obtain the address code, and then splice the above results to obtain the attribute features of the corresponding entity node, that is, [Industry type code, (registered capital), time characteristics, address code].

[0082] Among them, for the above example, when the entity node is an enterprise, the edge between the entity node and the neighbor node represents the relationship between enterprises or between people and enterprises, including investment, holding, appointment, upstream and downstream supply, cooperative bidding, etc.; at the same time, different relationship types and strengths can also be digitized or embedded as relationship type features. Here, the relevant processing of relationship feature characteristics will be explained below, and the embodiments of this application will not be repeated here.

[0083] Step 1043: Determine the correlation value between each first neighbor node and the first node based on the attribute characteristics of the first node and the attribute characteristics of each first neighbor node.

[0084] It should be noted that the first neighbor node is a node that is directly connected to the first node, and the number of the first node is at least one. For example, see Figure 7 , Figure 7 is a schematic diagram of a first node and a first neighbor node provided in an embodiment of the present application, based on Figure 7 , the entity node v indicated by 701 is the first node, and the remaining four entity nodes u1, u2, u3 and u4 are neighbor nodes of the entity node v, that is, the first neighbor nodes.

[0085] In actual implementation, the process of determining the correlation value between each first neighbor node and the first node based on the attribute feature of the first node and the attribute feature of each first neighbor node may be to perform the following processing for each first neighbor node: splicing the attribute feature of the first node with the attribute feature of the first neighbor node to obtain a first splicing feature, and determining the correlation between the first neighbor node and the first node based on the first splicing feature, that is: ...Formula (5); in, refers to the correlation between the first node v and the first neighbor node u, is a weight vector of size 2F, is the attribute feature of the first node v, is the attribute feature of the first neighbor node u, It refers to the concatenated feature obtained by concatenating the attribute feature of the first node v and the attribute feature of the first neighbor node u. is a non-linear activation function.

[0086] Then, the correlation is normalized to obtain the correlation value, namely: ...Formula (6); in, It refers to the correlation value between the first node v and the first neighbor node u, that is, the weight of the first neighbor node u to the first node v. exp indicates an exponential function with the natural constant e as the base. Indicates a domain consisting of all first neighbor nodes u that are directly connected to the first node v.

[0087] It should be noted that the correlation value is used to indicate the importance of the corresponding first neighbor node to the first node, that is, the weight. For example, if the first neighbor node u is the ultimate controlling shareholder node of the first node v, and the investment ratio between the two is very high, then Relatively large, that is is relatively large; if the first neighbor node u indicates an insignificant neighbor node, then Relatively small, that is Relatively small.

[0088] In practical applications, enterprise relationships and enterprise personnel are not fixed. For example, the holding relationship between enterprises or the position of enterprise personnel may change. Therefore, the graph neural network is also dynamic. Based on this, the correlation between the first neighbor node and the first node can also be determined in combination with the time feature and the relationship type feature. Specifically, based on the first splicing feature, the process of determining the correlation between the first neighbor node and the first node can be, obtaining a first time point, the first time point is the time point when the second neighbor node of the first node or the connection edge of the first node is most recently changed; obtaining a second time point, the second time point is the time point when the third neighbor node of the first neighbor node or the connection edge of the first neighbor node is most recently changed; determining the time feature based on the first time point and the second time point; obtaining the relationship type of the connection relationship between the first node and the first neighbor node, and extracting features of the relationship type to obtain the relationship type feature; splicing the first splicing feature, the time feature and the relationship type feature to obtain the second splicing feature; determining the correlation between the first neighbor node and the first node based on the second splicing feature, that is: ...Formula (7); in, Indicates the relationship type feature of the edge between the first node v and the first neighbor node u, Indicates the first time point, Indicates the second time point, is the time characteristic function, is the time feature, refers to the correlation between the first node v and the first neighbor node u, is a weight vector of size 2F, is the attribute feature of the first node v, is the attribute feature of the first neighbor node u, It refers to the concatenated features obtained by concatenating the attribute features of the first node v, the attribute features of the first neighbor node u, the time features, and the relationship type features. is a non-linear activation function.

[0089] It should be noted that the second neighbor node includes the corresponding first neighbor node, that is, the second neighbor node indicates all neighbor nodes of the first node, and the third neighbor node indicates all neighbor nodes of the first neighbor node.

[0090] It should be noted that the semantics of different businesses can be distinguished based on relationship features, and based on time features, for example, when considering the time point when the equity relationship between v and u is generated or changed, the time decay or time weight is calculated by comparing the current time, thereby highlighting that recent changes are more valuable for reference than long-term stable relationships.

[0091] In this way, by introducing relationship type characteristics and time characteristics, the focus can be flexibly adjusted for different business scenarios. For example, when judging whether two natural persons with the same name are the same person, if their joint controlling relationship path has recently undergone changes such as equity transfer and capital increase, then the changed relationship may obtain a higher weight, that is, a greater relevance, thereby strengthening the focus on recent dynamic information, making the judgment of whether two natural persons with the same name are the same person more accurate.

[0092] Step 1044: Based on the correlation values ​​and the attribute characteristics of each first neighboring node, determine the first feature vector of the first node, and obtain the second feature vector of the second node.

[0093] In actual implementation, the process of determining the first feature vector of the first node based on each correlation value and the attribute characteristics of each first neighbor node may be, for each first neighbor node, performing product processing on the correlation value and the attribute characteristics of the corresponding first neighbor node to obtain the product result of the first neighbor node; accumulating and summing the product results of each first neighbor node to obtain the sum result; and determining the first feature vector of the first node based on the sum result, that is: ...Formula (8); in, refers to the first eigenvector, refers to the correlation value between the first node v and the first neighbor node u, Indicates a domain consisting of all first neighbor nodes u that have a direct connection relationship with the first node v, is the attribute feature of the first neighbor node, is a non-linear activation function.

[0094] In some embodiments, when determining When determining the first feature vector of the first node based on the first neighbor node, a different attention mechanism can be used, that is, based on the different weight vectors a described above, to determine multiple correlation values, and then determine the first feature vector of the first node based on the multiple correlation values. Based on this, the process of determining the first feature vector of the first node based on each correlation value and the attribute characteristics of each first neighbor node can be, with reference to formulas (5) and (6), to determine multiple correlation values ​​based on multiple preset weight vectors; for each correlation value, perform product processing on the correlation value and the attribute characteristics of each first neighbor node respectively to obtain a product result, and perform cumulative summation on the product results corresponding to each first neighbor node to obtain a summation result corresponding to the correlation value; based on the summation result, determine the intermediate feature vector corresponding to the correlation value; fuse the intermediate feature vectors corresponding to each correlation value, that is, fuse multiple intermediate feature vectors to obtain the first feature vector of the first node; The process of fusing multiple intermediate feature vectors to obtain the first feature vector of the first node may be to perform weighted concatenation on the multiple intermediate feature vectors to obtain the first feature vector of the first node, that is: ...Formula (9); in, It refers to one of the multiple correlation values ​​determined by different attention mechanisms between the first node v and the first neighbor node u. Indicates a domain consisting of all first neighbor nodes u that have a direct connection relationship with the first node v, is the attribute feature of the first neighbor node, is a nonlinear activation function, and Refers to the intermediate feature vector for the Kth correlation value; wherein the weight of each intermediate feature vector in weighted concatenation is pre-set; Alternatively, the process of fusing multiple intermediate feature vectors to obtain the first feature vector of the first node may be to average the multiple intermediate feature vectors to obtain the first feature vector of the first node, that is: ...Formula (10); in, It refers to one of the multiple correlation values ​​determined by different attention mechanisms between the first node v and the first neighbor node u. Indicates a domain consisting of all first neighbor nodes u that have a direct connection relationship with the first node v, is the attribute feature of the first neighbor node, is a nonlinear activation function, and refers to the intermediate eigenvector for the Kth correlation value.

[0095] In this way, determining the first feature vector of the first node based on multiple correlation values ​​determined by using different attention mechanisms can enhance the robustness of the first feature vector.

[0096] In some embodiments, after determining the attribute features of the entity node, the attribute features may be linearly transformed to obtain the target attribute features, namely: ...Formula (11); in, is the target attribute feature of the first node v, W is a pre-set parameter matrix, is the attribute feature of the first node v.

[0097] Then, all the above operations are performed based on the target attribute features, that is, all the attribute features involved in the above process are replaced with the target attribute features, so that the process of determining the correlation between entity nodes can be simplified; wherein, for the process of determining the first feature vector of the first node based on multiple weight vectors, the number of parameter matrices when determining the target attribute features is also multiple, and at the same time corresponds one-to-one to the weight vectors, that is, the number of determined target attribute features is also multiple, so as to determine the first feature vector of the first node based on multiple attribute features and corresponding weight vectors.

[0098] In actual implementation, the process of obtaining the second eigenvector of the second node is the same as the process of obtaining the first eigenvector of the first node, and is not described in detail in the embodiment of the present application.

[0099] Step 1045 , predicting a second probability that the first node and the second node are the same entity node based on the first eigenvector and the second eigenvector.

[0100] In actual implementation, based on the first eigenvector and the second eigenvector, the process of predicting the second probability that the first node and the second node are the same entity node may be to perform a dot product on the first eigenvector and the second eigenvector to obtain a dot product result, and to multiply the modulus of the first eigenvector and the modulus of the second eigenvector to obtain a product; determine the ratio of the dot product result to the product, and use the ratio as the second probability, that is: ...Formula (12); in, is the first node, is the second node, Indicates the second probability, indicates the first eigenvector of the first node, Indicates the second eigenvector of the second node.

[0101] Step 105 , based on the first probability and the second probability, determine the target probability that the first node and the second node are the same entity node, and when the target probability reaches a probability threshold, disambiguate the first node and the second node.

[0102] In actual implementation, based on the first probability and the second probability, the process of determining the target probability that the first node and the second node are the same entity node can be, for example, weighted summing the first probability and the second probability to obtain the target probability that the first node and the second node are the same entity node; or, obtaining the degree of matching between the first key information and the second key information, as well as the difference characteristics between the first node and the second node in other information, and based on the first probability, the second probability, the degree of matching and the difference in characteristics, performing a probability prediction on whether the first node and the second node are the same entity node to obtain the target probability.

[0103] It should be noted that the process of performing probability prediction on the first node and the second node being the same entity node based on the first probability, the second probability, the degree of matching and the difference feature to obtain the target probability can be achieved through a trained probability prediction model, or the first probability, the second probability, the degree of matching and the difference feature can be directly input into the binary classification model to obtain a binary classification result indicating that the first node and the second node are the same entity node.

[0104] Among them, for the difference characteristics of the first node and the second node in other information, the other information is information other than key information and auxiliary information. Specifically, the difference between the first node and the second node in other information is obtained, and features are extracted from the difference to obtain the difference characteristics; for example, the other information may be the contact information of the node, etc.

[0105] In actual implementation, the probability threshold is pre-set, for example, the probability threshold may be 0.9, so that when the target probability is greater than or equal to 0.9, the first node and the second node are determined to be the same entity node, and the first node and the second node are disambiguated.

[0106] In some embodiments, when the target probability is less than the probability threshold and greater than or equal to the second probability threshold, the first node and the second node may be sent to a reviewer so that the reviewer manually reviews the first node and the second node to determine whether the first node and the second node are the same entity node; here, the second probability threshold may also be pre-set, and the second probability threshold may be less than the probability threshold, for example, it may be 0.7; In some embodiments, when the target probability is less than a third probability threshold, it is determined that the first node and the second node are not the same entity node; wherein the third probability threshold may also be pre-set, the third probability threshold may be the same as the second probability threshold, or may be less than the second probability threshold, and the embodiments of the present application do not limit this.

[0107] In some embodiments, feedback information can also be collected periodically to indicate whether the disambiguation result is correct; based on the feedback information, the models described above, such as the probability prediction model when determining the second probability, the graph neural network or the probability prediction model or the binary classification model when determining the target probability, are updated to continuously adapt to data evolution. In this way, parameters are regularly updated through online learning to always maintain an agile response to the latest data, thereby ensuring accuracy in entity disambiguation.

[0108] As an example, the entity disambiguation method provided in the embodiment of the present application is specifically described.

[0109] In actual implementation, multi-source information of Company A and its subsidiary Company B is collected from public reports, executive appointment and removal announcements, company telephone / address registrations, investment and financing information platforms and other channels. The multi-source information is deduplicated and formatted in a unified manner, that is, repeated entries of the same ID number are excluded and all name coding formats are unified (processing traditional and simplified Chinese, full-width and half-width spaces, common aliases, etc.). If there is a manual review interface, when a manual check finds that the ID number of a certain name registered in a certain data source is obviously incorrect or missing, it is corrected or supplemented at this stage. In this way, if two records about "Zhang San" are obtained, that is, Record 1: Zhang San appears in the list of existing shareholders of Company A. "Zhang San", record 2: The name of the chairman of the board of directors of enterprise B (a subsidiary of enterprise A) is also "Zhang San". After deduplication, format unification, correction or supplementation of multi-source data of the enterprise, "Zhang San (shareholder)" and "Zhang San (chairman)" already have comparable structured fields in the database, for example: Zhang San (shareholder): name = "Zhang San", role = "shareholder", company = "Company A", registered address = "address X", time information = "investment time", etc., Zhang San (chairman): name = "Zhang San", role = "chairman", company = "Company B", registered address = "address Y", time information = "start time of employment", etc.; Then, based on the multi-source information of the enterprise, the enterprise knowledge graph is constructed. For example, "Company A" and "Company B" are taken as two enterprise nodes, "Zhang San (shareholder)" and "Zhang San (chairman)" are taken as natural person nodes or individual nodes respectively, and an "investment / holding" or "subsidiary" relationship edge is added to the graph to reflect that Company B is a subsidiary of Company A, that is, the control relationship of Company A→Company B; at the same time, additional attribute fields are stored for each node and edge, such as timestamp, address, shareholding ratio, etc. Next, based on the key information, a preliminary judgment is made on "Zhang San (shareholder)" and "Zhang San (chairman)". Specifically, it is checked whether the encrypted ID number, passport number or other unique ID is completely consistent. If they are consistent, they are directly determined to be the same person, or it is judged whether the parent company or actual controller of the affiliated enterprise is consistent. If they are consistent, they are directly determined to be the same person; if they are inconsistent, auxiliary information is collected, such as the registered address of the shareholder of Company A and the registered address of the chairman of Company B, the time when the shareholder "Zhang San" of Company A invested and the time when the chairman "Zhang San" of Company B served, the contact information of "Zhang San (shareholder)" and "Zhang San (chairman)", etc., so as to judge the similarity between "Zhang San (shareholder)" and "Zhang San (chairman)" for each auxiliary information, such as address similarity, time overlap, contact information correlation, etc.; finally, multiple similarities are formed into a vector, and based on the vector, the probability that "Zhang San (shareholder)" and "Zhang San (chairman)" are the same person is calculated, which is the first probability mentioned above; Then, the enterprise knowledge graph is used as input and input into the graph neural network. Based on the graph neural network, the node features of each node are determined (for example, each natural person node contains its role type, industry background, time information, i.e., investment time or tenure time, etc.; the enterprise node contains registered capital, industry code, geographic location, etc.) and the edge features of the node (such as the equity ratio of the "subsidiary", the appointment time of the "board of directors", etc.); then, the attention mechanism is used to determine the feature vector of "Zhang San (shareholder)" and the feature vector of "Zhang San (chairman)" based on the node features and edge features. For example, the attention mechanism is used to determine the key neighbors of "Zhang San (shareholder)", such as other shareholders of Company A, and the key neighbors of "Zhang San (chairman)", such as other directors and executives of Company B, and the feature vectors of "Zhang San (shareholder)" and "Zhang San (chairman)" are determined by combining the features of these important neighbors. Based on the feature vectors of "Zhang San (shareholder)" and "Zhang San (chairman)", the similarity between "Zhang San (shareholder)" and "Zhang San (chairman)" is determined, which is equivalent to the probability that they are the same person, that is, the second probability mentioned above; Finally, the first probability and the second probability are combined to determine the final probability that "Zhang San (shareholder)" and "Zhang San (chairman)" are the same person. When the final probability is greater than a set threshold (such as 0.8), "Zhang San (shareholder)" and "Zhang San (chairman)" will be automatically merged; if the final probability falls in the middle range (such as 0.5~0.8), it will be submitted to the manual review interface for further confirmation by business personnel.

[0110] Applying the above-mentioned embodiment of the present application, after determining the candidate entity node pair including the first node and the second node in the enterprise knowledge graph, by matching the key information of the first node with the key information of the second node, if the key information of the first node and the second node do not match, based on multiple auxiliary information of the first node and the second node, determine the first probability that the first node and the second node are the same entity node, and based on the graph neural network corresponding to the enterprise knowledge graph, predict the second probability that the first node and the second node are the same entity node, and then determine the target probability based on the first probability and the second probability, so that when the target probability reaches the probability threshold, the first node and the second node are disambiguated. In this way, compared with the method of determining whether it is the same entity node by similarity matching based only on basic attributes such as the name and contact information of the node, combining key information, auxiliary information and graph neural network to determine whether the first node and the second node are the same entity node, enriches the judgment dimension of determining whether the two nodes are the same entity node, improves the accuracy of determining whether the two nodes are the same entity node, and thus improves the accuracy of entity disambiguation.

[0111] The following is a description of an exemplary structure of the entity disambiguation device 455 provided in the embodiment of the present application implemented as a software module. In some embodiments, for example, Figure 2 As shown, the software modules stored in the entity disambiguation device 455 of the memory 450 may include: An acquisition module 4551 is used to acquire an enterprise knowledge graph, wherein the enterprise knowledge graph includes a plurality of entity nodes, wherein the plurality of entity nodes include at least one of enterprise personnel and enterprises, and each entity node has corresponding key information; A selection module 4552, configured to select a candidate entity node pair from the enterprise knowledge graph, wherein the candidate entity node pair includes a first node and a second node; a first determination module 4553, configured to, when the first key information of the first node does not match the second key information of the second node, respectively obtain a plurality of auxiliary information of the first node and the second node, and determine a first probability that the first node and the second node are the same entity node based on the plurality of auxiliary information; A prediction module 4554 is used to obtain a graph neural network corresponding to the enterprise knowledge graph, and based on the graph neural network, predict a second probability that the first node and the second node are the same entity node; The second determination module 4555 is used to determine the target probability that the first node and the second node are the same entity node based on the first probability and the second probability, and to disambiguate the first node and the second node when the target probability reaches a probability threshold.

[0112] In some embodiments, the key information includes at least two of the identifier of the enterprise personnel corresponding to the entity node, the parent company of the enterprise to which the entity node belongs, the enterprise name of the enterprise to which the entity node belongs, the beneficiary of the enterprise to which the entity node belongs, the control relationship of the enterprise to which the entity node belongs, and the actual controller of the enterprise to which the entity node belongs; the device also includes a matching module, which is used to match the identifier of the enterprise personnel corresponding to the first node with the identifier of the enterprise personnel corresponding to the second node when the key information includes the identifier of the enterprise personnel corresponding to the entity node to obtain a first matching result; when the key information includes the parent company of the enterprise to which the entity node belongs, the parent company of the enterprise to which the first node belongs with the parent company of the enterprise to which the second node belongs to obtain a second matching result; when the key information includes the enterprise name of the enterprise to which the entity node belongs, the matching module is used to match the parent company of the enterprise personnel corresponding to the first node with the parent company of the enterprise personnel corresponding to the second node to obtain a second matching result. The enterprise name of the enterprise to which the first node belongs is matched with the enterprise name of the enterprise to which the second node belongs to obtain a third matching result; when the key information includes the beneficiary of the enterprise to which the entity node belongs, the beneficiary of the enterprise to which the first node belongs is matched with the beneficiary of the enterprise to which the second node belongs to obtain a fourth matching result; when the key information includes the control relationship of the enterprise to which the entity node belongs, the control relationship of the enterprise to which the first node belongs is matched with the control relationship of the enterprise to which the second node belongs to obtain a fifth matching result; when the key information includes the actual controller of the enterprise to which the entity node belongs, the actual controller of the enterprise to which the first node belongs is matched with the actual controller of the enterprise to which the second node belongs to obtain a sixth matching result; when the matching results obtained based on the key information matching of the first node and the second node both fail to match, it is determined that the first key information does not match the second key information.

[0113] In some embodiments, the first determination module 4553 is also used to obtain, for each of the auxiliary information, a target similarity between the auxiliary information of the first node and the auxiliary information of the second node; obtain a reference enterprise knowledge graph including multiple reference entity nodes; for each of the auxiliary information, based on the reference enterprise knowledge graph, when two reference entity nodes belong to the same entity node, determine a first reference probability that the similarity of the two reference entity nodes reaches the target similarity, and when the two reference entity nodes do not belong to the same entity node, determine a second reference probability that the similarity reaches the target similarity; based on the first reference probabilities corresponding to multiple of the auxiliary information and the second reference probabilities corresponding to multiple of the auxiliary information, determine a first probability that the first node and the second node are the same entity node.

[0114] In some embodiments, the first determination module 4553 is also used to input the auxiliary information of the first node and the auxiliary information of the second node into a probability prediction model for each of the auxiliary information to obtain an intermediate probability that the first node and the second node are the same entity node; and fuse the intermediate probabilities corresponding to each of the auxiliary information to obtain a first probability that the first node and the second node are the same entity node.

[0115] In some embodiments, the prediction module 4554 is also used to obtain the attribute information of the first node and the attribute information of each first neighbor node of the first node based on the graph neural network; perform feature extraction on the attribute information of the first node to obtain the attribute characteristics of the first node, and perform feature extraction on the attribute information of each first neighbor node to obtain the attribute characteristics of each first neighbor node; determine the correlation value of each first neighbor node with the first node based on the attribute characteristics of the first node and the attribute characteristics of each first neighbor node; determine the first feature vector of the first node based on each of the correlation values ​​and the attribute characteristics of each first neighbor node, and obtain the second feature vector of the second node; based on the first feature vector and the second feature vector, predict the second probability that the first node and the second node are the same entity node.

[0116] In some embodiments, the prediction module 4554 is further used to perform a dot product on the first eigenvector and the second eigenvector to obtain a dot product result, and multiply the modulus of the first eigenvector and the modulus of the second eigenvector to obtain a product; determine a ratio of the dot product result to the product, and use the ratio as the second probability.

[0117] In some embodiments, the prediction module 4554 is also used to perform the following processing for each of the first neighbor nodes: splicing the attribute characteristics of the first node with the attribute characteristics of the first neighbor node to obtain a first splicing feature, and based on the first splicing feature, determining the correlation between the first neighbor node and the first node; normalizing the correlation to obtain a correlation value, and the correlation value is used to indicate the importance of the corresponding first neighbor node to the first node.

[0118] In some embodiments, the prediction module 4554 is also used to obtain a first time point, the first time point being the time point at which the second neighbor node of the first node or the connection edge of the first node most recently changed; obtain a second time point, the second time point being the time point at which the third neighbor node of the first neighbor node or the connection edge of the first neighbor node most recently changed; determine a time feature based on the first time point and the second time point; obtain a relationship type of the connection relationship between the first node and the first neighbor node, and perform feature extraction on the relationship type to obtain a relationship type feature; splice the first splicing feature, the time feature and the relationship type feature to obtain a second splicing feature; and determine the correlation between the first neighbor node and the first node based on the second splicing feature.

[0119] In some embodiments, the selection module 4552 is also used to obtain the name of each of the entity nodes in the enterprise knowledge graph; based on the name of each of the entity nodes, perform similarity matching on the names of any two entity nodes in the enterprise knowledge graph, and select two entity nodes whose similarity is greater than or equal to a similarity threshold as the candidate entity node pair.

[0120] The present application embodiment provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the electronic device executes the entity disambiguation method described in the present application embodiment, for example, Figure 3 The entity disambiguation method shown.

[0121] The embodiment of the present application provides a computer-readable storage medium storing executable instructions, wherein the executable instructions are stored. When the executable instructions are executed by a processor, the processor will execute the entity disambiguation method provided by the embodiment of the present application, for example, Figure 3 The entity disambiguation method shown.

[0122] In some embodiments, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a flash memory, a magnetic surface memory, an optical disk, or a CD-ROM; or it may be various devices including one or any combination of the above memories.

[0123] In some embodiments, computer executable instructions may be in the form of a program, software, software module, script or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine or other unit suitable for use in a computing environment.

[0124] As an example, computer-executable instructions may, but need not, correspond to a file in a file system, may be stored as part of a file storing other programs or data, such as in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files storing one or more modules, subroutines, or code portions).

[0125] As an example, computer executable instructions may be deployed to be executed on one electronic device, or on multiple electronic devices located at one site, or on multiple electronic devices distributed at multiple sites and interconnected by a communication network.

[0126] In summary, the embodiments of the present application have the following beneficial effects: (1) Traditional simple rules need to be frequently updated when facing new data types (such as new industries, new investment forms). In this application, the probabilistic prediction model and graph neural network are embedded to adaptively learn feature distribution, without the need for large amounts of manual intervention.

[0127] (2) Compared with the method of matching only user name similarity or single attribute, the present application can extract features from multiple dimensions based on multiple auxiliary features, thereby improving the accuracy of entity disambiguation.

[0128] (3) Regularly update parameters through online learning to always remain agile and responsive to the latest data, thereby ensuring accuracy in entity disambiguation.

[0129] It should be noted that in the embodiments of the present application, it involves obtaining relevant data including enterprise multi-source data. When the embodiments of the present application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0130] The above is only an embodiment of the present application and is not intended to limit the protection scope of the present application. Any modifications, equivalent substitutions and improvements made within the spirit and scope of the present application are included in the protection scope of the present application.

Claims

1. An entity disambiguation method, characterized in that: The method comprises: Acquire an enterprise knowledge graph, wherein the enterprise knowledge graph includes a plurality of entity nodes, wherein the plurality of entity nodes include at least one of enterprise personnel and enterprises, and each entity node has corresponding key information; Selecting a candidate entity node pair from the enterprise knowledge graph, wherein the candidate entity node pair includes a first node and a second node; When the first key information of the first node does not match the second key information of the second node, respectively obtain multiple auxiliary information of the first node and the second node, and determine a first probability that the first node and the second node are the same entity node based on the multiple auxiliary information; Obtaining a graph neural network corresponding to the enterprise knowledge graph, and based on the graph neural network, predicting a second probability that the first node and the second node are the same entity node; Based on the first probability and the second probability, a target probability that the first node and the second node are the same entity node is determined, and when the target probability reaches a probability threshold, the first node and the second node are disambiguated.

2. The method according to claim 1, characterized in that The key information includes at least two of the identification of the enterprise personnel corresponding to the entity node, the parent company of the enterprise to which the entity node belongs, the enterprise name of the enterprise to which the entity node belongs, the beneficiary of the enterprise to which the entity node belongs, the control relationship of the enterprise to which the entity node belongs, and the actual controller of the enterprise to which the entity node belongs; Before respectively acquiring a plurality of auxiliary information of the first node and the second node, the method further includes: When the key information includes an identifier of an enterprise employee corresponding to the entity node, matching the identifier of the enterprise employee corresponding to the first node with the identifier of the enterprise employee corresponding to the second node to obtain a first matching result; When the key information includes the parent company of the enterprise to which the entity node belongs, matching the parent company of the enterprise to which the first node belongs with the parent company of the enterprise to which the second node belongs to obtain a second matching result; When the key information includes the enterprise name of the enterprise to which the entity node belongs, the enterprise name of the enterprise to which the first node belongs is matched with the enterprise name of the enterprise to which the second node belongs to obtain a third matching result; When the key information includes a beneficiary of the enterprise to which the entity node belongs, matching the beneficiary of the enterprise to which the first node belongs with the beneficiary of the enterprise to which the second node belongs to obtain a fourth matching result; When the key information includes the control relationship of the enterprise to which the physical node belongs, the control relationship of the enterprise to which the first node belongs is matched with the control relationship of the enterprise to which the second node belongs to obtain a fifth matching result; When the key information includes the actual controller of the enterprise to which the entity node belongs, the actual controller of the enterprise to which the first node belongs is matched with the actual controller of the enterprise to which the second node belongs to obtain a sixth matching result; When the matching results obtained based on the key information matching of the first node and the second node both fail to match, it is determined that the first key information does not match the second key information.

3. The method according to claim 1, characterized in that The determining, based on the plurality of auxiliary information, a first probability that the first node and the second node are the same entity node includes: For each piece of the auxiliary information, obtaining a target similarity between the auxiliary information of the first node and the auxiliary information of the second node; Obtaining a reference enterprise knowledge graph including a plurality of reference entity nodes; For each of the auxiliary information, based on the reference enterprise knowledge graph, when two reference entity nodes belong to the same entity node, determining a first reference probability that the similarity of the two reference entity nodes reaches the target similarity, and when the two reference entity nodes do not belong to the same entity node, determining a second reference probability that the similarity reaches the target similarity; Based on the first reference probabilities corresponding to the plurality of auxiliary information and the second reference probabilities corresponding to the plurality of auxiliary information, a first probability that the first node and the second node are the same entity node is determined.

4. The method according to claim 1, characterized in that The determining, based on the plurality of auxiliary information, a first probability that the first node and the second node are the same entity node includes: For each of the auxiliary information, input the auxiliary information of the first node and the auxiliary information of the second node into a probability prediction model to obtain an intermediate probability that the first node and the second node are the same entity node; The intermediate probabilities corresponding to each of the auxiliary information are fused to obtain a first probability that the first node and the second node are the same entity node.

5. The method according to claim 1, characterized in that The predicting, based on the graph neural network, a second probability that the first node and the second node are the same entity node includes: Based on the graph neural network, acquiring attribute information of the first node and attribute information of each first neighbor node of the first node; Performing feature extraction on the attribute information of the first node to obtain the attribute feature of the first node, and performing feature extraction on the attribute information of each of the first neighboring nodes to obtain the attribute feature of each of the first neighboring nodes; Determine, based on the attribute feature of the first node and the attribute feature of each of the first neighboring nodes, a correlation value between each of the first neighboring nodes and the first node; Based on each of the correlation values ​​and the attribute characteristics of each of the first neighboring nodes, determine a first feature vector of the first node, and obtain a second feature vector of the second node; Based on the first feature vector and the second feature vector, a second probability that the first node and the second node are the same entity node is predicted.

6. The method according to claim 5, characterized in that The predicting, based on the first feature vector and the second feature vector, a second probability that the first node and the second node are the same entity node includes: Performing a dot product of the first eigenvector and the second eigenvector to obtain a dot product result, and multiplying the modulus of the first eigenvector and the modulus of the second eigenvector to obtain a product; A ratio of the dot product result to the product is determined, and the ratio is used as the second probability.

7. The method according to claim 5, characterized in that The determining, based on the attribute feature of the first node and the attribute feature of each of the first neighboring nodes, a correlation value between each of the first neighboring nodes and the first node includes: The following processing is performed for each of the first neighbor nodes: splicing the attribute feature of the first node with the attribute feature of the first neighboring node to obtain a first splicing feature, and determining the correlation between the first neighboring node and the first node based on the first splicing feature; The correlation is normalized to obtain a correlation value, where the correlation value is used to indicate the importance of the corresponding first neighbor node to the first node.

8. The method according to claim 7, characterized in that The determining, based on the first splicing feature, the correlation between the first neighbor node and the first node includes: Obtaining a first time point, where the first time point is a time point at which a second neighbor node of the first node or a connection edge of the first node is most recently changed; Obtaining a second time point, where the second time point is a time point at which a third neighbor node of the first neighbor node or a connection edge of the first neighbor node is most recently changed; Determining a time feature based on the first time point and the second time point; Acquire a relationship type of a connection relationship between the first node and the first neighboring node, and perform feature extraction on the relationship type to obtain a relationship type feature; Splicing the first splicing feature, the time feature, and the relationship type feature to obtain a second splicing feature; Based on the second splicing feature, the correlation between the first neighbor node and the first node is determined.

9. The method according to claim 1, characterized in that The step of selecting candidate entity node pairs from the enterprise knowledge graph includes: Obtain the name of each entity node in the enterprise knowledge graph; Based on the name of each of the entity nodes, similarity matching is performed on the names of any two entity nodes in the enterprise knowledge graph, and two entity nodes whose similarity is greater than or equal to a similarity threshold are selected as the candidate entity node pair.

10. An entity disambiguation device, characterized in that: The device comprises: An acquisition module, used to acquire an enterprise knowledge graph, wherein the enterprise knowledge graph includes a plurality of entity nodes, wherein the plurality of entity nodes include at least one of enterprise personnel and enterprises, and each entity node has corresponding key information; A selection module, configured to select a candidate entity node pair from the enterprise knowledge graph, wherein the candidate entity node pair includes a first node and a second node; a first determination module, configured to respectively obtain a plurality of auxiliary information of the first node and the second node when the first key information of the first node does not match the second key information of the second node, and determine a first probability that the first node and the second node are the same entity node based on the plurality of auxiliary information; A prediction module, used to obtain a graph neural network corresponding to the enterprise knowledge graph, and based on the graph neural network, predict a second probability that the first node and the second node are the same entity node; The second determination module is used to determine the target probability that the first node and the second node are the same entity node based on the first probability and the second probability, and to disambiguate the first node and the second node when the target probability reaches a probability threshold.

Images