A method for reasoning and completing knowledge graph of dishonest information and related equipment

Abstract

The present invention provides a method for reasoning and completing a knowledge graph of untrustworthy information and related equipment, which constructs an early warning index system of untrustworthy information according to the characteristics of untrustworthy data; obtains the untrustworthy data of untrustworthy enterprises according to the early warning index system of untrustworthy information and converts it into knowledge triples to construct a knowledge graph of untrustworthy information for training and a knowledge graph of untrustworthy information to be completed; combines association analysis and self-attention mechanism to obtain the embedding vector representation matrix of the knowledge graph of untrustworthy information for training to train the reasoning and completion model, and obtains the trained reasoning and completion model; inputs the knowledge graph of untrustworthy information to be completed into the trained reasoning and completion model, obtains the completed knowledge triples, and completes them into the knowledge graph of untrustworthy information to be completed. This method realizes the reasoning and completion of the knowledge graph of untrustworthy information, provides data support for the early warning of the knowledge graph of untrustworthy information, and further improves the supervision efficiency of untrustworthy information.

Description

Technical Field

[0001] The present invention relates to the technical field of knowledge graph completion, and in particular to a method for reasoning and completing a knowledge graph of dishonest information and related equipment. Background Art

[0002] With the continuous optimization of the business environment and the substantial reduction of market entry barriers, new domestic enterprises and new foreign investments have emerged, market opportunities have increased significantly, and the supervision of market players is also facing new problems and challenges.

[0003] How to strengthen and improve the control of commercial banks' real estate development loans and strengthen the regulation and supervision of the real estate market is a severe challenge facing government regulatory authorities. To this end, a scientific and effective means of handling information on the default of market entities is urgently needed. Summary of the invention

[0004] The present invention provides a method for reasoning and completing a knowledge graph of untrustworthy information and related equipment, the purpose of which is to improve the quality of data completion and enhance the efficiency of untrustworthy information supervision.

[0005] In order to achieve the above-mentioned object, the present invention provides a method for reasoning and completing a knowledge graph of dishonest information, comprising:

[0006] Step 1: Construct a dishonest information early warning indicator system based on the characteristics of dishonest data. The dishonest information early warning indicator system includes multiple dishonest enterprises and their upstream and downstream chain enterprises, multiple dishonest attributes of enterprises, and multiple dishonest factors;

[0007] Step 2: Obtain the dishonest data of dishonest enterprises according to the dishonest information early warning indicator system, and convert the dishonest data of dishonest enterprises into knowledge triples to construct a knowledge graph of dishonest information for training and a knowledge graph of dishonest information to be completed. The knowledge graph of dishonest information for training contains multiple complete knowledge triples, and the knowledge graph of dishonest information to be completed contains knowledge triples to be completed.

[0008] Step 3: Use association analysis to obtain the mutual correlation of each dishonesty factor, obtain the relationship weight matrix of each dishonesty factor, and transform the initial embedding representation matrix of the node of the training dishonesty information knowledge graph through the self-attention mechanism to obtain the query vector, key vector and value vector of the self-attention mechanism;

[0009] Step 4: Combine the relationship weight matrix with the query vector, key vector, and value vector of the self-attention mechanism to obtain the embedding vector representation matrix of the training default information knowledge graph, and input the embedding vector representation matrix into the constructed reasoning completion model for training to obtain the trained reasoning completion model;

[0010] Step 5, input the knowledge graph of untrustworthy information to be completed into the trained inference completion model, perform inference completion on the knowledge triples to be completed in the knowledge graph of untrustworthy information to be completed, obtain the completed knowledge triples, and complete the completed knowledge triples into the knowledge graph of untrustworthy information to be completed, to obtain the knowledge graph inference completion result.

[0011] More specifically, step 1 includes:

[0012] Analyze the characteristics of dishonest data and identify multiple dishonest enterprises and their upstream and downstream chain enterprises;

[0013] Establish multiple dishonest attributes of dishonest enterprises, including financial data, facts of dishonest behavior, shareholder information, controlled enterprises, indirectly held enterprises, foreign investment enterprises, main members, and basic industrial and commercial information data;

[0014] Identify multiple dishonest factors of dishonest enterprises, including breach of contract records, administrative penalties, associated risks, own risks, shareholder holdings, and investment shareholding ratios;

[0015] A dishonest information early warning indicator system is constructed based on multiple dishonest enterprises and their upstream and downstream chain enterprises, multiple enterprise dishonest attributes, and multiple dishonest factors.

[0016] More specifically, step 2 includes:

[0017] Set keywords for entities and relationships, and perform keyword matching on all dishonest enterprises through the enterprise public information platform according to the dishonest information early warning indicator system, and obtain dishonest data of target dishonest enterprises;

[0018] Perform entity recognition on the dishonest data of the dishonest enterprise to obtain multiple dishonest enterprise entities and enterprise attribute-related entities;

[0019] Extracting the relationships between the plurality of enterprise entities and the plurality of entity attributes through syntactic analysis to obtain the enterprise attribute of each entity and the relationship between the entities;

[0020] For each of the multiple entities, the entity, the enterprise attributes of the entity and the relationship between the entities are structured to obtain multiple knowledge triples;

[0021] All knowledge triples are stored in a graph database, and a knowledge graph of dishonest information for training and a knowledge graph of dishonest information to be completed are constructed. The knowledge graph of dishonest information for training contains multiple complete knowledge triples, and the knowledge graph of dishonest information to be completed contains knowledge triples to be completed.

[0022] To further explain, knowledge triples include two categories:

[0023] <enterprise node, relationship, enterprise node>;

[0024] <Enterprise node, enterprise attribute, attribute value node>.

[0025] More specifically, step 3 includes:

[0026] Use correlation analysis to obtain the relevant correlation between each dishonesty factor and obtain the relationship weight matrix of each dishonesty factor ;

[0027] Extract enterprise nodes, relationships, enterprise attributes, and attribute value nodes in the training untrustworthy information knowledge graph, obtain the labels of the relationships between enterprise nodes and enterprise nodes, and the relationships between enterprise nodes and attribute value nodes in the training untrustworthy information knowledge graph, and establish indexes to generate the node initial embedding representation matrix ;

[0028] Initially embed the node into the representation matrix Transformed through the self-attention mechanism, the query vector of the self-attention mechanism is obtained , key vector Sum value vector .

[0029] Specifically, the expression of the embedding vector representation matrix of the training default information knowledge graph is:

[0030]

[0031] in, represents the embedding vector representation matrix, Indicates the dimension of the key vector.

[0032] Specifically, the inference completion model includes an input layer, a hidden layer, a softmax layer, and a prediction layer which are connected sequentially.

[0033] The present invention also provides a knowledge graph reasoning and completion device for the supervision of dishonest information, comprising:

[0034] A construction module is used to construct a dishonest information early warning indicator system based on the characteristics of dishonest data. The dishonest information early warning indicator system includes multiple dishonest enterprises and their upstream and downstream chain enterprises, multiple dishonest attributes of enterprises, and multiple dishonest factors;

[0035] A conversion module is used to obtain the dishonest data of dishonest enterprises according to the dishonest information early warning indicator system, and convert the dishonest data of dishonest enterprises into knowledge triples to construct a knowledge graph of dishonest information for training and a knowledge graph of dishonest information to be completed. The knowledge graph of dishonest information for training contains multiple complete knowledge triples, and the knowledge graph of dishonest information to be completed contains knowledge triples to be completed.

[0036] The transformation module is used to obtain the mutual correlation between each dishonesty factor, obtain the relationship weight matrix of each dishonesty factor, and transform the initial embedding representation matrix of the node of the training dishonesty information knowledge graph through the self-attention mechanism to obtain the query vector, key vector and value vector of the self-attention mechanism;

[0037] The training module is used to combine the relationship weight matrix with the query vector, key vector and value vector of the self-attention mechanism to obtain the embedding vector representation matrix of the training default information knowledge graph and input it into the constructed reasoning completion model for training to obtain the trained reasoning completion model;

[0038] The completion module is used to input the knowledge graph of untrustworthy information to be completed into the trained reasoning completion model, perform reasoning and completion on the knowledge triples to be completed in the knowledge graph of untrustworthy information to be completed, obtain the completed knowledge triples, and complete the completed knowledge triples into the knowledge graph of untrustworthy information to be completed to obtain the knowledge graph reasoning completion result.

[0039] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, a method for reasoning and completing a knowledge graph of dishonest information is implemented.

[0040] The present invention also provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements a method for reasoning and completing a knowledge graph of dishonest information when executing the computer program.

[0041] The above scheme of the present invention has the following beneficial effects:

[0042] Compared with the prior art, the present invention constructs a dishonest information early warning indicator system according to the characteristics of dishonest data; obtains dishonest data of dishonest enterprises according to the dishonest information early warning indicator system, and converts the dishonest data of dishonest enterprises into knowledge triples to construct a dishonest information knowledge graph for training and a dishonest information knowledge graph to be completed; obtains the mutual correlation between each dishonest factor, obtains the relationship weight matrix of each dishonest factor, and transforms the node initial embedding representation matrix of the training dishonest information knowledge graph through a self-attention mechanism to obtain a query vector, a key vector and a value vector of the self-attention mechanism; and transforms the relationship weight matrix with the self-attention mechanism. The query vector, key vector and value vector are combined to obtain the embedding vector representation matrix of the training untrustworthy information knowledge graph and input it into the constructed reasoning and completion model for training to obtain the trained reasoning and completion model; the untrustworthy information knowledge graph to be completed is input into the trained reasoning and completion model, and the knowledge triples to be completed in the untrustworthy information knowledge graph to be completed are reasoned and completed to obtain the completed knowledge triples, and the completed knowledge triples are added to the untrustworthy information knowledge graph to be completed to obtain the knowledge graph reasoning and completion result; the reasoning and completion of the knowledge graph is realized, which provides data support for the early warning of the untrustworthy information knowledge graph and improves the accuracy of the supervision of untrustworthy information.

[0043] Other beneficial effects of the present invention will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A schematic diagram of a flow chart of an embodiment of the present invention;

[0045] Figure 2 Schematic diagram of the structure of the reasoning completion model in an embodiment of the present invention. DETAILED DESCRIPTION

[0046] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0048] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installation", "connection" and "connection" should be understood in a broad sense, for example, it can be a locking connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0049] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0050] In view of the existing problems, the present invention provides a method for reasoning and completing a knowledge graph of dishonest information and related equipment.

[0051] like Figure 1 As shown, an embodiment of the present invention provides a method for completing a knowledge graph of dishonest information, including:

[0052] Step 1: Construct a dishonest information early warning indicator system based on the characteristics of dishonest data. The dishonest information early warning indicator system includes multiple dishonest enterprises and their upstream and downstream chain enterprises, multiple dishonest attributes of enterprises, and multiple dishonest factors;

[0053] Step 2: Obtain the dishonest data of dishonest enterprises according to the dishonest information early warning indicator system, and convert the dishonest data of dishonest enterprises into knowledge triples to construct a knowledge graph of dishonest information for training and a knowledge graph of dishonest information to be completed. The knowledge graph of dishonest information for training contains multiple complete knowledge triples, and the knowledge graph of dishonest information to be completed contains knowledge triples to be completed.

[0054] Step 3: Obtain the mutual correlation of each dishonesty factor, obtain the relationship weight matrix of each dishonesty factor, and transform the initial embedding representation matrix of the node of the dishonesty information knowledge graph for training through the self-attention mechanism to obtain the query vector, key vector and value vector of the self-attention mechanism;

[0055] Step 4: Combine the relationship weight matrix with the query vector, key vector, and value vector of the self-attention mechanism to obtain the embedding vector representation matrix of the training default information knowledge graph and input it into the constructed reasoning completion model for training to obtain the trained reasoning completion model;

[0056] Step 5, input the knowledge graph of untrustworthy information to be completed into the trained inference completion model, perform inference completion on the knowledge triples to be completed in the knowledge graph of untrustworthy information to be completed, obtain the completed knowledge triples, and complete the completed knowledge triples into the knowledge graph of untrustworthy information to be completed, obtain the knowledge graph reasoning completion result, that is, obtain a complete knowledge graph of untrustworthy information.

[0057] The embodiment of the present invention takes a real estate group company as an example to explain the above steps in detail, as follows:

[0058] Specifically, step 1 includes:

[0059] Analyze the characteristics of dishonest data and identify multiple dishonest enterprises and their upstream and downstream chain enterprises;

[0060] Establish multiple dishonest attributes of dishonest enterprises, including financial data, facts of dishonest behavior, shareholder information, controlled enterprises, indirectly held enterprises, foreign investment enterprises, main members, and basic industrial and commercial information data;

[0061] Identify multiple dishonest factors of dishonest enterprises, including breach of contract records, administrative penalties, associated risks, own risks, shareholder holdings, and investment shareholding ratios;

[0062] A dishonest information early warning indicator system is constructed based on multiple dishonest enterprises and their upstream and downstream chain enterprises, multiple enterprise dishonest attributes, and multiple dishonest factors.

[0063] In the embodiment of the present invention, a real estate group company is taken as a dishonest enterprise, its shareholder company is the upstream chain enterprise, and its foreign investment enterprise, controlling enterprise, and indirectly held enterprise are the downstream chain enterprises. The dishonest information early warning indicator system is a tree diagram constructed with a real estate group company as the root node, and upstream and downstream chain enterprises, multiple enterprise dishonest attributes, and multiple dishonest factors as leaf nodes.

[0064] The attributes of corporate dishonesty can be subdivided into company name, number of insured persons, enterprise type, enterprise scale, paid-in capital, controlled enterprises, invested enterprises, shareholder names, key personnel, enterprise address, enterprise profile, affiliated institutions, net profit, net profit margin, total profit, weighted return on net assets, national standard industry category, enterprise official website, establishment date, district and county, city, province, total owner's equity, investment quantity, investment ratio, investment chain, indirect investment chain, shareholding ratio, latest annual report year, final profit shares, invested enterprise profit shares, net assets per share, operating cash flow per share, gross profit margin, legal representative, There are 66 labels in total, including registration number, registered capital, current ratio, registration status, taxpayer identification number, net cash flow from operating activities, business scope, unified social credit code, position, English name, total operating income, indirect investment enterprise, subscribed capital date, subscribed capital amount, subscribed capital amount or number of shares held by the investment enterprise, total liabilities, financial annual report, total assets, debt-to-asset ratio, correspondence address, quick ratio, email address, indirect shareholding ratio, first shareholding date, number of associated risks, number of own risks, number of dishonest behaviors, whether the person is subject to execution, whether the person is subject to execution in dishonesty, whether there is a bill default, and whether there is a restriction on high consumption.

[0065] Specifically, step 2 includes:

[0066] Set keywords for entities and relationships, and perform keyword matching on all dishonest enterprises through the enterprise public information platform according to the dishonest information early warning indicator system, and obtain dishonest data of target dishonest enterprises;

[0067] Perform entity recognition on the dishonest data of the dishonest enterprise to obtain multiple dishonest enterprise entities and enterprise attribute-related entities;

[0068] Extracting the relationships between the plurality of enterprise entities and the plurality of entity attributes through syntactic analysis to obtain the enterprise attribute of each entity and the relationship between the entities; for each of the plurality of entities, performing structural processing on the entity, the enterprise attribute of the entity and the relationship between the entities to obtain a plurality of knowledge triples;

[0069] All knowledge triples are stored in a graph database, and a knowledge graph of dishonest information for training and a knowledge graph of dishonest information to be completed are constructed. The knowledge graph of dishonest information for training contains multiple complete knowledge triples, and the knowledge graph of dishonest information to be completed contains knowledge triples to be completed.

[0070] In the embodiment of the present invention, it is assumed that the keyword of the entity and the relationship is set to real estate, and a plurality of dishonest real estate enterprises are crawled in the dishonest information early warning indicator system.

[0071] It should be noted that the enterprise public information platform is an open platform.

[0072] In the embodiment of the present invention, a conventional bidirectional long short-term memory network (BiLSTM) and a conditional random field (CRF) model are used to perform entity recognition on the dishonest data of dishonest enterprises; the enterprise attributes of multiple entities and the relationships between multiple entities are extracted by syntactic analysis of the text, and the enterprise attributes of each entity and the relationships between entities are obtained; finally, the knowledge triples are obtained in the form of<h,r,t> .

[0073] In the embodiment of the present invention, the graph database adopts the Neo4j graph database. Neo4j is a high-performance NOSQL graph database that stores structured data on a network structure rather than in a table. It is an embedded, disk-based Java persistence engine with complete transaction features, but it stores structured data on a network (called a graph from a mathematical point of view) rather than in a table.

[0074] Specifically, knowledge triples include two categories:

[0075] <enterprise node, relationship, enterprise node>;

[0076] <Enterprise node, enterprise attribute, attribute value node>.

[0077] Specifically, step 3 includes:

[0078] Use correlation analysis to obtain the relevant correlation between each dishonesty factor and obtain the relationship weight matrix of each dishonesty factor ;

[0079] In the embodiment of the present invention, the relationship weight matrix Each element in Representation Node With Node The weights of the relationships between nodes, such as default records, administrative penalties, associated risks, own risks, shareholder holdings, investment shareholding ratios, etc.;

[0080] Extract enterprise nodes, relationships, enterprise attributes, and attribute value nodes in the training untrustworthy information knowledge graph, obtain the labels of the relationships between enterprise nodes and enterprise nodes, and the relationships between enterprise nodes and attribute value nodes in the training untrustworthy information knowledge graph, and establish indexes to generate the node initial embedding representation matrix ;

[0081] Initially embed the node into the representation matrix Transformed through the self-attention mechanism, the query vector of the self-attention mechanism is obtained , key vector Sum value vector , and their expressions are:

[0082]

[0083]

[0084]

[0085] in, represents the learnable parameter matrix.

[0086] Specifically, the expression of the embedding vector representation matrix of the training default information knowledge graph is:

[0087]

[0088] in, represents the embedding vector representation matrix, Represents the dimension of the key vector, the relationship weight matrix is adjusted to Match the size.

[0089] Specifically, if Figure 2 As shown, the inference completion model includes an input layer, a hidden layer, a softmax layer, and a prediction layer connected in sequence.

[0090] The reasoning completion model is based on the graph convolutional neural network, and the rule-based link prediction algorithm is integrated into the output layer of the graph convolutional neural network to form a prediction layer.

[0091] The embedding vector representation matrix is ​​input into the inference completion model for high-order feature learning and weight training. After multiple iterations, the trained inference completion model is obtained. The training process is as follows:

[0092] The input of the input layer is the embedding vector representation matrix and the adjacency matrix of the training default information knowledge graph. The calculation starts from the input layer and enters the hidden layer, which is as follows:

[0093] The eigenvector matrix at the initial stage With the adjacency matrix After the convolution operation, the next layer of feature vector matrix is ​​obtained for:

[0094]

[0095] in, represents the identity matrix, express The degree matrix of represents a nonlinear activation function, Represents the trainable parameter matrix of the current layer convolution transformation;

[0096] The eigenvector matrix at the initial stage With the adjacency matrix After multiple forward propagations, the final feature vector is obtained ;

[0097] Further, the feature vector Input to the Softmax layer for calculation to obtain the probability of node relationship in the training default information knowledge graph :

[0098]

[0099] in, , The eigenvectors are No. Wei and Di Dimensional The exponent of the base, , Represents the eigenvectors The dimension of .

[0100] The embodiment of the present invention targets all enterprise nodes in the knowledge graph of untrustworthy information to be completed, takes the knowledge triples of missing enterprise attributes as an example, screens out the enterprise attributes missing in the knowledge graph of untrustworthy information to be completed, and generates knowledge triples to be completed for all attribute value nodes corresponding to the target enterprise attributes of the enterprise node. The format of a single knowledge triple is: <enterprise node, enterprise attribute, attribute value node>; for example: if a real estate group company is missing the enterprise attribute of <region>, then knowledge triples are generated for all attribute value nodes in the enterprise attribute <region>, namely <real estate group company, region, Shenzhen>, <real estate group company, region, Hangzhou>, <real estate group company, region, Shanghai>, etc.

[0101] The knowledge triples to be completed are sequentially input into the input layer, hidden layer, and softmax layer for prediction, and the probability of each enterprise attribute value corresponding to the enterprise attribute is obtained and input into the prediction layer for inference completion. The completion process is as follows:

[0102] Enterprise attributes with probabilities greater than the preset threshold are added to the completion candidate queue. The enterprise attributes in the completion candidate queue are judged in combination with the rules defined in the rule base, and the enterprise attributes that violate the rules are eliminated. The enterprise attributes are further selected based on the probability of each enterprise attribute to obtain the completed knowledge triples, and the knowledge graph reasoning completion results of the dishonest enterprises are obtained based on the completed knowledge triples.

[0103] In the embodiment of the present invention, the basic reasoning rules in the rule base are defined as:

[0104]

[0105]

[0106] in, Indicates the judgment of the existence of the complement relationship between nodes. Indicates the probability of existence of the node enterprise attribute value completion relationship. Represents the tail node of the completed knowledge triple, Represents the relationship between the completed knowledge triples, Represents the probability of enterprise attributes.

[0107] The embodiment of the present invention compares the effects of link prediction ranking (MRR, Mean Reciprocal Ranking) and link prediction ranking ratio Hits indicators with some classic models, such as TransE, TransH, ConvE, and R-GCN, and gives the prediction results on the WN18RR dataset and the dishonest information knowledge graph dataset, as shown in Table 1 below:

[0108] Table 1 Prediction results of the classic model on the WN18RR dataset and the dishonesty information knowledge graph dataset

[0109]

[0110] From the experimental results in Table 1 above, it can be seen that the completion method provided by the invention performs well on the WN18RR dataset and achieves the best effect on the hits@10 indicator; at the same time, on the knowledge graph dataset of dishonest information, due to the selection of associated factors of the completion method provided by the invention and the definition of the rule base, the performance indicators all achieved the best results.

[0111] Compared with the prior art, the embodiments of the present invention construct a default information early warning indicator system according to the characteristics of default data; obtain the default data of default enterprises according to the default information early warning indicator system, and convert the default data of default enterprises into knowledge triples to construct a training default information knowledge graph and a to-be-completed default information knowledge graph; obtain the mutual correlation between each default factor, obtain the relationship weight matrix of each default factor, and transform the node initial embedding representation matrix of the training default information knowledge graph through a self-attention mechanism to obtain a query vector, a key vector and a value vector of the self-attention mechanism; and transform the relationship weight matrix with the self-attention mechanism. The query vector, key vector and value vector are combined to obtain the embedding vector representation matrix of the training untrustworthy information knowledge graph and input it into the constructed reasoning and completion model for training to obtain the trained reasoning and completion model; the untrustworthy information knowledge graph to be completed is input into the trained reasoning and completion model, and the knowledge triples to be completed in the untrustworthy information knowledge graph to be completed are reasoned and completed to obtain the completed knowledge triples, and the completed knowledge triples are added into the untrustworthy information knowledge graph to be completed to obtain the knowledge graph reasoning and completion result; the reasoning and completion of the untrustworthy information knowledge graph is realized, the confidence and comprehensiveness of the untrustworthy information data are improved, and the accuracy of the untrustworthy information supervision is improved.

[0112] The present invention also provides a knowledge graph reasoning and completion device for the supervision of dishonest information, comprising:

[0113] A construction module is used to construct a dishonest information early warning indicator system based on the characteristics of dishonest data. The dishonest information early warning indicator system includes multiple dishonest enterprises and their upstream and downstream chain enterprises, multiple dishonest attributes of enterprises, and multiple dishonest factors;

[0114] A conversion module is used to obtain the dishonest data of dishonest enterprises according to the dishonest information early warning indicator system, and convert the dishonest data of dishonest enterprises into knowledge triples to construct a knowledge graph of dishonest information for training and a knowledge graph of dishonest information to be completed. The knowledge graph of dishonest information for training contains multiple complete knowledge triples, and the knowledge graph of dishonest information to be completed contains knowledge triples to be completed.

[0115] The transformation module is used to obtain the mutual correlation between each dishonesty factor, obtain the relationship weight matrix of each dishonesty factor, and transform the initial embedding representation matrix of the node of the training dishonesty information knowledge graph through the self-attention mechanism to obtain the query vector, key vector and value vector of the self-attention mechanism;

[0116] The training module is used to combine the relationship weight matrix with the query vector, key vector and value vector of the self-attention mechanism to obtain the embedding vector representation matrix of the training default information knowledge graph and input it into the constructed reasoning completion model for training to obtain the trained reasoning completion model;

[0117] The completion module is used to input the knowledge graph of untrustworthy information to be completed into the trained reasoning completion model, perform reasoning and completion on the knowledge triples to be completed in the knowledge graph of untrustworthy information to be completed, obtain the completed knowledge triples, and complete the completed knowledge triples into the knowledge graph of untrustworthy information to be completed to obtain the knowledge graph reasoning completion result.

[0118] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0119] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0120] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, a method for reasoning and completing a knowledge graph of dishonest information is implemented.

[0121] If the integrated module is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above method embodiment, which can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of each of the above method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device that can carry the computer program code to the construction device / terminal device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium. For example, a USB flash drive, a mobile hard disk, a disk or an optical disk.

[0122] The present invention also provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements a method for reasoning and completing a knowledge graph of dishonest information when executing the computer program.

[0123] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, a server, a server cluster, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0124] The processor may be a central processing unit (CPU), or other general-purpose processors, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0125] In some embodiments, the memory may be an internal storage unit of the terminal device, such as a hard disk or memory of the terminal device. In other embodiments, the memory may also be an external storage device of the terminal device, such as a plug-in hard disk, a smart memory card (SMC, Smart Media Card), a secure digital (SD, Secure Digital) card, a flash card (Flash Card), etc. equipped on the terminal device. Further, the memory may also include both an internal storage unit of the terminal device and an external storage device. The memory is used to store an operating system, an application program, a boot loader (Boot Loader), data and other programs, such as the program code of the computer program, etc. The memory may also be used to temporarily store data that has been output or is to be output.

[0126] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of the embodiment of the present invention. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0127] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the protection scope of the embodiments of the present invention. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0128] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for reasoning and completing a knowledge graph of dishonest information, characterized in that: include: Step 1: construct a dishonest information early warning indicator system based on the characteristics of dishonest data, wherein the dishonest information early warning indicator system includes multiple dishonest enterprises and their upstream and downstream chain enterprises, multiple dishonest attributes of enterprises, and multiple dishonest factors; Step 2, obtaining the dishonest data of dishonest enterprises according to the dishonest information early warning indicator system, and converting the dishonest data of the dishonest enterprises into knowledge triples to construct a dishonest information knowledge graph for training and a dishonest information knowledge graph to be completed, including: Set keywords for entities and relationships, and perform keyword matching on all dishonest enterprises through the enterprise public information platform according to the dishonest information early warning indicator system, and obtain dishonest data of target dishonest enterprises; Perform entity recognition on the dishonest data of the dishonest enterprise to obtain multiple dishonest enterprise entities and enterprise attribute-related entities; Extracting the relationships between the multiple dishonest enterprise entities and multiple enterprise attribute related entities through syntactic analysis to obtain the enterprise attributes of each of the entities and the relationships between the entities; For each of the multiple dishonest enterprise entities, structural processing is performed on the dishonest enterprise entity, the enterprise attributes of the entity, and the relationship between the entities to obtain multiple knowledge triples; All knowledge triples are stored in a graph database, and a knowledge graph of dishonest information for training and a knowledge graph of dishonest information to be completed are constructed, wherein the knowledge graph of dishonest information for training contains multiple complete knowledge triples, and the knowledge graph of dishonest information to be completed contains knowledge triples to be completed; Step 3, obtaining the mutual correlation between each dishonesty factor, obtaining the relationship weight matrix of each dishonesty factor, and transforming the node initial embedding representation matrix of the training dishonesty information knowledge graph through the self-attention mechanism to obtain the query vector, key vector and value vector of the self-attention mechanism, including: Use correlation analysis to obtain the relevant correlation between each dishonesty factor and obtain the relationship weight matrix of each dishonesty factor ; Extract the enterprise nodes, relationships, enterprise attributes and attribute value nodes in the training untrustworthy information knowledge graph, obtain the labels of the relationships between enterprise nodes and enterprise nodes, and the relationships between enterprise nodes and attribute value nodes in the training untrustworthy information knowledge graph, and create an index to generate an initial node embedding representation matrix ; The node is initially embedded in the representation matrix Transformed by the self-attention mechanism, the query vector of the self-attention mechanism is obtained , key vector Sum value vector ; Step 4: combining the relationship weight matrix with the query vector, key vector and value vector of the self-attention mechanism to obtain an embedding vector representation matrix of the training default information knowledge graph, and inputting the embedding vector representation matrix into the constructed reasoning completion model for training to obtain a trained reasoning completion model; Step 5, input the knowledge graph of untrustworthy information to be completed into the trained inference completion model, perform inference completion on the knowledge triples to be completed in the knowledge graph of untrustworthy information to be completed, obtain completed knowledge triples, and complete the completed knowledge triples into the knowledge graph of untrustworthy information to be completed, to obtain the knowledge graph inference completion result.

2. The method for reasoning and completing the knowledge graph of dishonest information according to claim 1 is characterized in that: The step 1 comprises: Analyze the characteristics of dishonest data and identify multiple dishonest enterprises and their upstream and downstream chain enterprises; Establish multiple dishonest attributes of dishonest enterprises, including financial data, facts of dishonest behavior, shareholder information, controlled enterprises, indirectly held enterprises, foreign investment enterprises, main members, and basic industrial and commercial information data; Identify multiple dishonest factors of dishonest enterprises, including breach of contract records, administrative penalties, associated risks, own risks, shareholder holdings, and investment shareholding ratios; A dishonest information early warning indicator system is constructed based on multiple dishonest enterprises and their upstream and downstream chain enterprises, multiple enterprise dishonest attributes, and multiple dishonest factors.

3. The method for reasoning and completing the knowledge graph of dishonest information according to claim 2 is characterized in that: The knowledge triples include two categories, namely: <enterprise node, relationship, enterprise node>; <Enterprise node, enterprise attribute, attribute value node>.

4. The method for reasoning and completing the knowledge graph of dishonest information according to claim 3 is characterized in that: The expression of the embedding vector representation matrix of the training dishonesty information knowledge graph is: in, represents the embedding vector representation matrix, Indicates the dimension of the key vector.

5. The method for reasoning and completing the knowledge graph of dishonest information according to claim 4 is characterized in that: The inference completion model includes an input layer, a hidden layer, a softmax layer and a prediction layer connected in sequence.

6. A knowledge graph reasoning and completion device for the supervision of dishonest information, characterized in that: include: A construction module is used to construct a dishonest information early warning indicator system according to the characteristics of dishonest data, wherein the dishonest information early warning indicator system includes multiple dishonest enterprises and their upstream and downstream chain enterprises, multiple dishonest attributes of enterprises, and multiple dishonest factors; A conversion module, used to obtain the dishonest data of dishonest enterprises according to the dishonest information early warning indicator system, and convert the dishonest data of the dishonest enterprises into knowledge triples to construct a dishonest information knowledge graph for training and a dishonest information knowledge graph to be completed, including: Set keywords for entities and relationships, and perform keyword matching on all dishonest enterprises through the enterprise public information platform according to the dishonest information early warning indicator system, and obtain dishonest data of target dishonest enterprises; Perform entity recognition on the dishonest data of the dishonest enterprise to obtain multiple dishonest enterprise entities and enterprise attribute-related entities; Extracting the relationships between the multiple dishonest enterprise entities and multiple enterprise attribute related entities through syntactic analysis to obtain the enterprise attributes of each of the entities and the relationships between the entities; For each of the multiple dishonest enterprise entities, structural processing is performed on the dishonest enterprise entity, the enterprise attributes of the entity, and the relationship between the entities to obtain multiple knowledge triples; All knowledge triples are stored in a graph database, and a knowledge graph of dishonest information for training and a knowledge graph of dishonest information to be completed are constructed, wherein the knowledge graph of dishonest information for training contains multiple complete knowledge triples, and the knowledge graph of dishonest information to be completed contains knowledge triples to be completed; The transformation module is used to obtain the mutual correlation between the various untrustworthy factors, obtain the relationship weight matrix of the various untrustworthy factors, and transform the node initial embedding representation matrix of the training untrustworthy information knowledge graph through the self-attention mechanism to obtain the query vector, key vector and value vector of the self-attention mechanism, including: Use correlation analysis to obtain the relevant correlation between each dishonesty factor and obtain the relationship weight matrix of each dishonesty factor ; Extract the enterprise nodes, relationships, enterprise attributes and attribute value nodes in the training untrustworthy information knowledge graph, obtain the labels of the relationships between enterprise nodes and enterprise nodes, and the relationships between enterprise nodes and attribute value nodes in the training untrustworthy information knowledge graph, and create an index to generate an initial node embedding representation matrix ; The node is initially embedded in the representation matrix Transformed by the self-attention mechanism, the query vector of the self-attention mechanism is obtained , key vector Sum value vector ; A training module, used to combine the relationship weight matrix with the query vector, key vector and value vector of the self-attention mechanism to obtain an embedding vector representation matrix of the training default information knowledge graph, and input the embedding vector representation matrix into the constructed reasoning completion model for training to obtain a trained reasoning completion model; The completion module is used to input the knowledge graph of untrustworthy information to be completed into the trained reasoning completion model, perform reasoning completion on the knowledge triples to be completed in the knowledge graph of untrustworthy information to be completed, obtain the completed knowledge triples, and complete the completed knowledge triples into the knowledge graph of untrustworthy information to be completed to obtain the knowledge graph reasoning completion result.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for reasoning and completing the knowledge graph of dishonest information as described in any one of claims 1 to 5 is implemented.

8. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method for reasoning and completing the knowledge graph of dishonest information as described in any one of claims 1 to 5 is implemented.

Images