Enterprise dishonesty knowledge graph reasoning method and system

Abstract

The embodiment of the invention provides an enterprise dishonesty knowledge graph reasoning method and system, and relates to the technical field of data processing, and the method comprises the steps: generating a target meta-task according to target dynamic information; updating the target meta-task according to the target perception information; according to the updated target meta-task, generating a target enterprise dishonesty knowledge graph reasoning result; wherein the target dynamic information comprises a mapping relation between entity change and time, and / or a mapping relation between relation change and time; the target perception information comprises target entity information, target relation information and target multi-hop neighborhood information. According to the method, the updating timeliness and integrity of the enterprise dishonesty knowledge graph can be improved, and the accuracy of the enterprise information in the enterprise dishonesty knowledge graph can be improved, so that the loss of a user caused by error information in the enterprise dishonesty knowledge graph is avoided, and the user experience is improved.

Description

Technical Field

[0001] Embodiments of the present application relate to the technical field of data processing, and in particular, to a method and system for reasoning an enterprise credit default knowledge graph. Background Art

[0002] The enterprise credit default information knowledge graph is mainly used to facilitate users to more comprehensively understand the credit details information of enterprises, so as to avoid property losses caused to individuals or enterprises due to the credit error information of enterprises.

[0003] In related technologies, there are problems such as high latency and poor integrity in the generation of enterprise credit default information knowledge graphs.

[0004] Therefore, there is an urgent need for a new technical solution to solve the above technical problems. Summary of the Invention

[0005] According to the embodiments of the present application, a method and system for reasoning an enterprise credit default knowledge graph are provided, which can improve the timeliness and integrity of the update of the enterprise credit default knowledge graph, improve the accuracy of enterprise information in the enterprise credit default knowledge graph, thereby avoiding error information in the enterprise credit default knowledge graph from causing losses to users, and further improving the user experience.

[0006] In the first aspect of the present application, a method for reasoning an enterprise credit default knowledge graph is proposed, including: Generating a target meta-task according to target dynamic information; Updating the target meta-task according to target perception information; Generating a target enterprise credit default knowledge graph reasoning result according to the updated target meta-task; wherein, the target dynamic information includes: the mapping relationship between entity changes and time, and / or, the mapping relationship between relationship changes and time; The target perception information includes: target entity information, target relationship information, and target multi-hop neighborhood information.

[0007] In some feasible embodiments, the above method further includes: Generating a target time segment task based on the K-means method according to the target time, and / or, the target event density; Capturing the target dynamic information according to the target time segment task.

[0008] In some feasible embodiments, the above target meta-task includes: a meta-test task and a meta-training task; wherein, the time segment for generating the meta-test task corresponds to the idle generation time segment of the meta-training task.

[0009] In some feasible embodiments, generating the target meta-task according to the above-mentioned target dynamic information further includes: Pruning the target long-tail relationship from the meta-training task; Introducing the target long-tail relationship into the meta-test task.

[0010] In some feasible embodiments, updating the target meta-task according to the above-mentioned target perception information includes: Generating target perception information according to the gated recurrent component and the relational graph convolutional neural network model; Among them, each target entity is provided with a corresponding target self-loop.

[0011] In some feasible embodiments, the above method further includes: Determining the target loss function according to the first target data set; Determining the target embedding feature according to the target enterprise credit-loss knowledge graph; Updating the target meta-task according to the target loss function and the target embedding feature; Among them, the first target data set includes: target interaction state data, target action data, and / or target reward data.

[0012] In some feasible embodiments, the above method further includes: Determining the target set loss according to the second target data set; Determining the target optimization function according to the target set loss; Optimizing the target meta-task according to the target optimization function; Among them, the second target data set includes: the environmental sampling query data set corresponding to each target meta-task.

[0013] In some feasible embodiments, the above method further includes: Determining the target gating value corresponding to each action according to the following formula: ; Among them, is the target gating value corresponding to each action, represents the concatenated feature of the state and the action, is the first parameter of the target gating mechanism; is the second parameter of the gating mechanism; among them, the first parameter and the second parameter are updated based on the target set loss ; Determining the selection probability corresponding to the target action according to the following formula: ; Among them, is the selection probability corresponding to the target action, is the action score function before gating; is the gating network parameter.

[0014] In some feasible embodiments, the generating the inference result of the target enterprise credit default knowledge graph according to the updated target meta-task described above includes: Determine the probability values corresponding to multiple inference paths according to the beam search algorithm; Generate the inference result of the target enterprise credit default knowledge graph according to the probability values.

[0015] In a second aspect of the present application, an inference system for an enterprise credit default knowledge graph is proposed, including: A generating unit, configured to generate a target meta-task according to the target dynamic information; An updating unit; configured to update the target meta-task according to the target perception information; An inference unit, which generates an inference result of the target enterprise credit default knowledge graph according to the updated target meta-task; Wherein, the target dynamic information includes: the mapping relationship between entity changes and time, and / or, the mapping relationship between relationship changes and time; The target perception information includes: target entity information, target relationship information, and target multi-hop neighborhood information.

[0016] The enterprise credit default knowledge graph inference method and system provided by the embodiments of the present application, wherein the method includes: generating a target meta-task according to the target dynamic information; updating the target meta-task according to the target perception information; generating an inference result of the target enterprise credit default knowledge graph according to the updated target meta-task; wherein, the target dynamic information includes: the mapping relationship between entity changes and time, and / or, the mapping relationship between relationship changes and time; the target perception information includes: target entity information, target relationship information, and target multi-hop neighborhood information. The present application can improve the timeliness and integrity of the update of the enterprise credit default knowledge graph, improve the accuracy of enterprise information in the enterprise credit default knowledge graph, thereby avoiding error information in the enterprise credit default knowledge graph and causing losses to users, and further improving the user experience.

[0017] It should be understood that the content described in the summary of the invention section is not intended to limit the key or important features of the embodiments of the present application, nor is it used to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Combined with the drawings and referring to the following detailed description, the above and other features, advantages, and aspects of the embodiments of the present application will become more obvious. In the drawings, the same or similar reference numerals represent the same or similar elements, where: Figure 1 A flowchart of a method for reasoning about an enterprise credit default knowledge graph provided by an embodiment of the present application; Figure 2 A structural diagram of a system for reasoning about an enterprise credit default knowledge graph provided by an embodiment of the present application; Figure 3 A structural diagram of an electronic device provided by an embodiment of the present application. Detailed implementation manners

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present disclosure. Apparently, the described embodiments are some but not all of the embodiments of the present disclosure. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments of the present disclosure without creative efforts shall fall within the protection scope of the present disclosure.

[0020] In addition, the term "and / or" in this article is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this article generally represents an "or" relationship between the associated objects before and after.

[0021] In the current related technologies, the methods for reasoning about an enterprise credit default knowledge graph mainly include the following two categories: one is the method based on semantic relevance, such as: TTransE, HyTE, DE-SimplE, TNTComplEx, etc.; the other is the method based on structural relevance, such as: RE-Net, TeMP, RE-GCN, CluSTeR, etc. The above methods often ignore the structural information in the knowledge graph, and there are problems that new entities or new events cannot be associated with the entire temporal knowledge graph in a timely manner, and semantic or structural connectivity cannot be achieved, resulting in problems such as high latency and poor integrity in the generation of the enterprise credit default information knowledge graph.

[0022] Based on this, in the first aspect of the embodiments of the present application, a method for reasoning about an enterprise credit default knowledge graph is proposed, which can be used for applications such as enterprise credit default knowledge graph completion and question answering based on the enterprise credit default knowledge graph. Figure 1 A flowchart of a method 100 for reasoning about an enterprise credit default knowledge graph provided by an embodiment of the present application, as Figure 1 shown, the method 100 includes: Step S1; generating a target meta-task according to target dynamic information, where the target dynamic information includes: the mapping relationship between entity changes and time, and / or, the mapping relationship between relationship changes and time.

[0023] Exemplarily, the mapping relationship between the above-mentioned entity changes and time may include: the correspondence between the enterprise name change event and the time node when the enterprise name change event occurs; the correspondence between the enterprise employee position change event and the time node when the enterprise employee position change event occurs, etc.

[0024] Exemplarily, the mapping relationship between the above-mentioned relationship changes and time may include: the correspondence between the enterprise personnel change event and the time node when the enterprise personnel change event occurs; the correspondence between the enterprise administrative penalty event and the time node when the enterprise administrative penalty event occurs, etc.

[0025] Exemplarily, the above-mentioned target meta-task may correspond to a meta-learning task.

[0026] Exemplarily, the above-mentioned target dynamic information can be captured based on the division of time segments.

[0027] In some feasible implementation manners, based on the K-means method, according to the target time, and / or, the target event density, a target time segment task can be generated; according to the target time segment task, the target dynamic information can be captured.

[0028] Exemplarily, based on the K-means method, according to the target time, and / or, the target event density, the target time segment task can be adaptively divided and generated.

[0029] Specifically, the timestamps of all events can be extracted to form the following timestamp sequence: (1); Then, based on the K-means method, all timestamps are divided into clusters, where each cluster corresponds to a time segment. The timestamps and the corresponding event densities are used as clustering features to assign the quadruple to the corresponding target time segment task according to the timestamp , where is the head entity, i.e., the subject; is the relationship type; is the tail entity, i.e., the object; is the timestamp.

[0030] Thus, based on the K-means method, the above method can automatically and accurately generate target time segment tasks according to the target time and / or the target event density for meta-reinforcement learning initialization and meta-learning task construction, which can improve the adaptability of this application to situations with uneven time and / or irregular event distributions. By capturing target dynamic information according to the target time segment tasks, any target meta-task can be designed as an inference problem for different time segments, so that the target agent can acquire general knowledge representations in multiple target time segment tasks and quickly adapt to unseen target time segment tasks.

[0031] In some feasible embodiments, the above target meta-tasks include: meta-test tasks and meta-training tasks.

[0032] It should be noted that the distributions of the above meta-test tasks and meta-training tasks are different and have a preset correlation. Among them, the above preset correlation may include: similar relationship types but specific relationships not seen before.

[0033] Among them, the time segment used to generate the meta-test task corresponds to the idle generation time segment of the meta-training task, that is, the meta-test task is generated in the time segment that has never appeared in the meta-training task, so as to achieve a distribution difference in time segments between the meta-test task and the meta-training task.

[0034] It should be noted that the latest time slice can be reserved as the meta-test task. There is a significant difference between the target event density corresponding to the test time slice and the target event density corresponding to the training time slice.

[0035] Thus, based on the above method for meta-reinforcement learning initialization and meta-learning task construction, by setting the time segment used to generate the meta-test task to correspond to the idle generation time segment of the meta-training task, it is beneficial to achieve a differential distribution between the meta-test task and the meta-training task, thereby improving the generalization ability of this application in sparse data scenarios.

[0036] In some feasible embodiments, the above step S1; generating the target meta-task according to the target dynamic information further includes: pruning the target long-tail relationship from the meta-training task; introducing the target long-tail relationship into the meta-test task.

[0037] Exemplarily, the target long-tail relationship can be introduced as part of the meta-test task, and some target long-tail relationships can be completely excluded from the meta-training task and reserved in the meta-test task to achieve a difference in relationship types between the meta-test task and the meta-training task.

[0038] Thus, the above method differentiates the relationship types between the meta-test task and the meta-training task by pruning the target long-tail relationships from the meta-training task and introducing the target long-tail relationships into the meta-test task, so as to further improve the generalization ability of the present application in sparse data scenarios.

[0039] It should be noted that in the generation of the target meta-task, the differences in time segment distribution and relationship type can be comprehensively considered. The quadruples are mainly selected based on the time slice differences, and then the differences in relationship type are considered on this basis for the selection of quadruples.

[0040] Step S2; update the target meta-task according to the target perception information, where the target perception information includes: target entity information, target relationship information, and target multi-hop neighborhood information.

[0041] Exemplarily, the above target entity information may include: all entity information traversed by the target agent. The above target relationship information may include: all relationship information traversed by the target agent. The above target multi-hop neighborhood information may include: the multi-hop neighborhood information corresponding to each entity on the historical path traversed by the target agent.

[0042] It should be noted that among them, the above target multi-hop neighborhood information can be integrated and determined based on a relational graph convolutional neural network combined with an attention mechanism.

[0043] It should be noted that a deep time series model can be used to encode the target perception information to obtain a vector corresponding to the target perception information for updating the target meta-task. Among them, updating the target meta-task may include: updating the task-level parameters corresponding to the target meta-task, and / or optimizing cross-task parameters, etc.

[0044] In some feasible embodiments, the above step S2; updating the target meta-task according to the target perception information includes: generating the target perception information according to a gated recurrent component and a relational graph convolutional neural network model; where each target entity is provided with a corresponding target self-loop.

[0045] Exemplarily, the gated recurrent component and the relational graph convolutional neural network model can be used to represent the historical path and the target multi-hop neighborhood information corresponding to the entities thereon to generate the target perception information for use as the perception of the environment by the target agent.

[0046] Specifically, the relational graph convolutional neural network can be used to integrate the neighborhood information of the nodes into the representation of the nodes to enhance the perception ability of the target agent for the environment, thereby improving the decision-making ability of the target agent. Among them, in the training stage, the target agent can be given step feedback based on a knowledge graph representation learning model with paths to further optimize the decision-making of the target agent.

[0047] It should be noted that corresponding target self-loops can be added to each entity, that is, connection triples from the entity to itself, so as to generate target neighborhood information based on the information of the central entity itself and its neighborhood information, so as to integrate the information of the entity itself and its neighborhood information into the new representation of the central entity.

[0048] Thus, the above method integrates the current entity node, historical path, relationship, and the neighborhood state around the historical path into the target perception information, that is, the environmental representation, so as to improve the timeliness and integrity of the update of the enterprise credit default knowledge graph.

[0049] In some feasible implementation manners, the above method further includes: determining a target loss function according to a first target data set; determining target embedding features according to the target enterprise credit default knowledge graph; updating the target meta-task according to the target loss function and the target embedding features; wherein, the first target data set includes: target interaction state data, target action data, and / or target reward data.

[0050] It should be noted that in reinforcement learning, each target meta-task can correspond to a specific reinforcement learning environment. Among them, for each target meta-task , a first target data set, that is, a support set, can be sampled and generated from the environment , according to the above first target data set, that is, the support set , calculate and determine the target loss function corresponding to the policy network ; extract and determine target embedding features, that is, the embedding features of entities and relationships, according to the target enterprise credit default knowledge graph; update the task-level parameters of the target meta-task according to the above target loss function and the above target embedding features, wherein the above first target data set, that is, the support set includes: state information of the interaction between the target agent and the environment, that is, target interaction state data, action information, that is, target action data, and / or reward information, that is, target reward data.

[0051] Thus, the above method can quickly adjust the model parameters according to the data in the first target data set corresponding to each target meta-task, that is, the support set to complete the update of the task-level parameters, and can improve the adaptability of this application to the specific requirements of each target meta-task; through the enhancement of the target embedding features and the update of the parameters, the dynamic mode of the current target meta-task can be accurately reflected; thus making the performance of the model reach the optimal, providing a basis for subsequent cross-task parameter update.

[0052] It should be noted that after completing the above task-level parameter update, the cross-task parameters can be globally updated to improve the comprehensive learning ability for multiple tasks, so as to further improve the generalization ability of the model for new target meta-tasks.

[0053] In some feasible embodiments, the above method further includes: determining a target set loss according to the second target data set; determining a target optimization function according to the target set loss; optimizing the target meta-task according to the target optimization function; wherein, the second target data set includes: an environment sampling query data set corresponding to each target meta-task.

[0054] It should be noted that the optimization initialization parameters can be set , after completing the task-level parameter update, for each target meta-task , the second target data set, that is, the query set, can be sampled and generated from the environment . Among them, the above-mentioned second target data set, that is, the query set is used to evaluate the performance of the model after the task-level parameter update in the target meta-task to calculate and determine the target set loss , summarize the target set losses of all the above target meta-tasks , calculate and determine the above target optimization function, that is, the global optimization target function to achieve the global update of the cross-task parameters. Among them, the current loss function is consistent with the aforementioned task-level parameter update. It should be noted that the above method can realize the dynamic adjustment of the transmission of information flow based on the gating mechanism, improve the sensitivity of the model to different features. In the process of the global update of the above cross-task parameters, the gating mechanism is applied to the path inference policy network to selectively model the nodes and edges on the inference path, which can avoid the interference of noise information on the long-tail relationship inference.

[0055] It should be noted that in some feasible embodiments, the importance of different nodes can also be dynamically adjusted by setting corresponding gating weights for the nodes of each inference path, so that the model can differentially focus on the path information related to the target relationship to improve the inference performance of the long-tail relationship.

[0056] It should be noted that in some feasible embodiments, the above method further includes: determining the target gating value corresponding to each action according to the following formula:

[0057] In some feasible embodiments, the above method further includes: determining the target gating value corresponding to each action according to the following formula: (2); Wherein, is the target gating value corresponding to each action, represents the concatenated feature of the state and the action, is the first parameter of the target gating mechanism; is the second parameter of the gating mechanism; wherein, the first parameter and the second parameter are updated based on the target set loss .

[0058] In some feasible embodiments, the above method further includes: determining the selection probability corresponding to the target action according to the following formula: (3); wherein, is the selection probability corresponding to the target action, is the action score function before gating; are the gating network parameters.

[0059] Wherein, the selection probability corresponding to the above target action can correspond to the selection probability corresponding to the final action.

[0060] Thus, the above method can accurately determine the target gating value corresponding to each action and the selection probability corresponding to the final action according to the above formulas (2) to (3), thereby improving the adaptability of the present application to complex environmental scenarios, that is, high-dimensional or sparse scenarios, further improving the generalization ability and robustness of the present method, alleviating the problem of gradient disappearance or explosion, and making the training process of generating the inference result of the target enterprise's credit default knowledge graph in the present application more stable and efficient.

[0061] It should be noted that by optimizing the policy network parameters of the target meta-task, parameter adaptation can be quickly completed on the new target meta-task.

[0062] Exemplarily, the optimized policy network parameters can be obtained through the meta-training task, and the above parameters can be quickly adapted on the already divided meta-test task set. Extract the first target data set, that is, the support set from the unseen task and the second target data set, that is, the query set , load the optimized parameters in the meta-training stage as the initial value , and implement task-level parameter adaptation based on the first target data set, that is, the support set , and implement performance inference evaluation based on the second target data set, that is, the query set .

[0063] Step S3; generating an inference result of the target enterprise's credit default knowledge graph according to the updated target meta-task.

[0064] It should be noted that the target enterprise's credit default knowledge graph may include: a temporal enterprise credit default knowledge graph.

[0065] Among them, the above-mentioned temporal enterprise credit-loss knowledge graph corresponds to a Temporal Knowledge Graph (TKG), which is a dynamic multi-relational graph data used to record evolutionary events and knowledge in the real world.

[0066] It should be noted that each fact in the temporal enterprise credit-loss knowledge graph corresponds to a timestamp, which is used to represent the occurrence time of each fact. The above-mentioned temporal enterprise credit-loss knowledge graph not only considers entities and relationships, but also considers the temporal dynamic changes of entities and relationships.

[0067] Among them, the inference generation mechanism of the above-mentioned temporal enterprise credit-loss knowledge graph may include: an interpolation mechanism, and / or, an extrapolation mechanism. Among them, the interpolation mechanism is used to infer missing facts that occur at time and the extrapolation mechanism is used to predict facts at the occurrence time , where .

[0068] In some feasible implementation manners, the above step S3; generating an inference result of the target enterprise credit-loss knowledge graph according to the updated target meta-task includes: determining probability values corresponding to multiple inference paths according to the beam search algorithm; generating an inference result of the target enterprise credit-loss knowledge graph according to the probability values.

[0069] Exemplarily, the inference model training applicable to the target agent can be completed based on the above steps S1~S2. When the target agent generates each step of decision-making, the probability distribution result of the action is generated based on the above inference model, and the beam search algorithm is used to determine the probability values corresponding to multiple inference paths. After performing a preset integration operation on the above probability values, according to the probability values, the inference result of the above target enterprise credit-loss knowledge graph is output in descending order.

[0070] Thus, the above method can implement screening out inference paths with higher accuracy from the possible inference results of multiple target enterprise credit-loss knowledge graphs based on the beam search algorithm, thereby improving the generation accuracy of the inference result of the target enterprise credit-loss knowledge graph and improving the applicability of this application to large-scale data scenarios.

[0071] Based on this, the enterprise credit default knowledge graph reasoning method proposed in this application includes: generating a target meta-task according to target dynamic information; updating the target meta-task according to target perception information; generating a target enterprise credit default knowledge graph reasoning result according to the updated target meta-task; wherein, the target dynamic information includes: the mapping relationship between entity changes and time, and / or, the mapping relationship between relationship changes and time; the target perception information includes: target entity information, target relationship information, and target multi-hop neighborhood information. This application can improve the accuracy of the representation corresponding to new entities in the graph, timely associate new entities or new events with the entire enterprise credit default knowledge graph, realize the connectivity of semantics and structure in the enterprise credit default knowledge graph, so as to realize the timely, complete, and accurate update and generation of the enterprise credit default knowledge graph reasoning result, avoid error information in the enterprise credit default knowledge graph, cause losses to users, and improve user experience.

[0072] It should be noted that for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that this application is not limited by the described action sequence, because according to this application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this application.

[0073] The above is the introduction of the method embodiments. The following further illustrates the solution of this application through device embodiments.

[0074] In the second aspect of the embodiments of this application, an enterprise credit default knowledge graph reasoning system is proposed. Figure 2 It is a structural schematic diagram of an enterprise credit default knowledge graph reasoning system 200 provided by the embodiments of this application. As Figure 2 shown, the system 200 includes: a generation unit 210, an update unit 220, and an inference unit 230.

[0075] The generation unit 210 is used to generate a target meta-task according to target dynamic information; The update unit 220; is used to update the target meta-task according to target perception information; The inference unit 230 generates a target enterprise credit default knowledge graph reasoning result according to the updated target meta-task; wherein, the target dynamic information includes: the mapping relationship between entity changes and time, and / or, the mapping relationship between relationship changes and time; The target perception information includes: target entity information, target relationship information, and target multi-hop neighborhood information.

[0076] Figure 3The following is a schematic structural diagram of an electronic device 300 provided by an embodiment of the present application. As Figure 3 shown, the electronic device 300 includes a central processing unit (CPU) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage section 308 into a random access memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the terminal device or server are also stored. The CPU 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0077] The following components are connected to the I / O interface 305: an input section 306 including a keyboard, a mouse, etc.; an output section 307 including, for example, a cathode ray tube (CRT), a liquid crystal display (LCD), etc. and a speaker; a storage section 308 including a hard disk, etc.; and a communication section 309 including a network interface card such as a LAN card, a modem, etc. The communication section 309 performs communication processing via a network such as the Internet. A drive 310 is also connected to the I / O interface 305 as needed. A removable medium 311, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 310 as needed, so that a computer program read from it can be installed into the storage section 308 as needed.

[0078] Specifically, according to the embodiments of the present application, the above method flow steps can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a machine-readable medium, and the computer program includes program codes for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network via the communication section 309, and / or installed from the removable medium 311. When the computer program is executed by the central processing unit (CPU) 301, the above functions defined in the system of the present application are executed.

[0079] It should be noted that the computer-readable medium described in this application can be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of a computer-readable storage medium can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this application, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. And in this application, a computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium can be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination of the above.

[0080] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram can represent a module, a program segment, or a part of code that contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks can occur in a different order than marked in the accompanying drawings. For example, two consecutive blocks shown can actually be executed substantially in parallel, and they can sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0081] The units or modules involved in the embodiments of the present application can be implemented in software or in hardware. The described units or modules can also be provided in a processor. Among them, the names of these units or modules do not, in some cases, constitute a limitation on the units or modules themselves.

[0082] The above description is only a preferred embodiment of the present application and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of the application involved in the present application is not limited to the technical solutions formed by the specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the foregoing application concept. For example, the technical solutions formed by mutually replacing the above features with the technical features (but not limited to) having similar functions applied in the present application.

Claims

1. An inference method for an enterprise credit default knowledge graph, characterized in that, Including: Generate a target meta-task according to the target dynamic information; Update the target meta-task according to the target perception information; Generate a target enterprise credit default knowledge graph reasoning result according to the updated target meta-task; Among them, the target dynamic information includes: the mapping relationship between entity changes and time, and / or the mapping relationship between relationship changes and time; The target perception information includes: target entity information, target relationship information, and target multi-hop neighborhood information; The updating the target meta-task according to the target perception information includes: Generate the target perception information according to the gated recurrent unit component and the relational graph convolutional neural network model; Also including: Determine the target gate value corresponding to each action according to the following formula: ； Among them, is the target gating value corresponding to each action, represents the concatenated feature of the state and the action, is the first parameter of the target gating mechanism; is the second parameter of the gating mechanism; among them, the first parameter and the second parameter are updated based on the target set loss ; Determine the selection probability corresponding to the target action according to the following formula: ； Among them, is the selection probability corresponding to the target action, is the action scoring function before gating; is the gating network parameter; Among them, the mapping relationship between entity changes and time includes: the correspondence between enterprise name change events and the time nodes when the enterprise name change events occur, and / or the correspondence between enterprise employee position change events and the time nodes when the enterprise employee position change events occur; The mapping relationship between relationship changes and time includes: the correspondence between enterprise personnel change events and the time nodes when the enterprise personnel change events occur, and / or the correspondence between enterprise administrative penalty events and the time nodes when the enterprise administrative penalty events occur; The target entity information includes: all entity information traversed by the target agent; The target relationship information includes: all relationship information traversed by the target agent; The target multi-hop neighborhood information includes: the multi-hop neighborhood information corresponding to each entity on the historical path traversed by the target agent; The updating the target meta-task includes: updating the task-level parameters corresponding to the target meta-task, and / or optimizing the cross-task parameters.

2. The method according to claim 1, characterized in that Also including: Generate target time segment tasks based on the K-means method according to the target time and / or the target event density; Capture the target dynamic information according to the target time segment tasks.

3. The method according to claim 2, wherein The target meta-task includes: meta-test tasks and meta-training tasks; Among them, the time segment used to generate the meta-test task corresponds to the idle generation time segment of the meta-training task.

4. The method according to claim 3, wherein The generating the target meta-task according to the target dynamic information further includes: Prune the target long-tail relationships from the meta-training task; Introduce the target long-tail relationships into the meta-test task.

5. The method according to any one of claims 1 to 4, characterized in that Each target entity is set with a corresponding target self-loop.

6. The method according to claim 5, wherein Also including: Determine the target loss function according to the first target data set; Determine the target embedding features according to the target enterprise credit default knowledge graph; Update the target meta-task according to the target loss function and the target embedding features; Among them, the first target data set includes: target interaction state data, target action data, and / or target reward data.

7. The method according to claim 6, characterized in that Also including: Determine the target set loss according to the second target data set; Determine the target optimization function according to the target set loss; Optimize the target meta-task according to the target optimization function; Among them, the second target data set includes: the environmental sampling query data set corresponding to each target meta-task.

8. The method according to claim 7, wherein Generating the inference result of the target enterprise credit loss knowledge graph according to the updated target meta-task includes: Determining the probability values corresponding to multiple inference paths according to the beam search algorithm; Generating the inference result of the target enterprise credit loss knowledge graph according to the probability values.

9. An enterprise credit default knowledge graph reasoning system, characterized in that, Including: A generating unit, configured to generate a target meta-task according to target dynamic information; An updating unit; For updating the target meta-task according to target perception information; An inference unit, generating an inference result of the target enterprise credit loss knowledge graph according to the updated target meta-task; Wherein, the target dynamic information includes: the mapping relationship between entity changes and time, and / or the mapping relationship between relationship changes and time; The target perception information includes: target entity information, target relationship information, and target multi-hop neighborhood information; Updating the target meta-task according to the target perception information includes: Generating the target perception information according to a gated recurrent component and a relational graph convolutional neural network model; Further including: Determining the target gate value corresponding to each action according to the following formula: ； Among them, is the target gating value corresponding to each action, represents the concatenated feature of the state and the action, is the first parameter of the target gating mechanism; is the second parameter of the gating mechanism; among them, the first parameter and the second parameter are updated based on the target set loss ; Determining the selection probability corresponding to the target action according to the following formula: ； Among them, is the selection probability corresponding to the target action, is the action score function before gating; is the gating network parameter; Wherein, the mapping relationship between entity changes and time includes: the correspondence between the enterprise name change event and the time node when the enterprise name change event occurs, and / or the correspondence between the enterprise employee position change event and the time node when the enterprise employee position change event occurs; The mapping relationship between relationship changes and time includes: the correspondence between the enterprise personnel change event and the time node when the enterprise personnel change event occurs, and / or; the correspondence between the enterprise administrative penalty event and the time node when the enterprise administrative penalty event occurs; The target entity information includes: all entity information traversed by the target agent; The target relationship information includes: all relationship information traversed by the target agent; The target multi-hop neighborhood information includes: the multi-hop neighborhood information corresponding to each entity on the historical path traversed by the target agent; Updating the target meta-task includes: updating the task-level parameters corresponding to the target meta-task, and / or optimizing the cross-task parameters.

Images