CN113849580A - Subject rating prediction method and device, electronic equipment and storage medium - Google Patents
Subject rating prediction method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN113849580A CN113849580A CN202111143690.2A CN202111143690A CN113849580A CN 113849580 A CN113849580 A CN 113849580A CN 202111143690 A CN202111143690 A CN 202111143690A CN 113849580 A CN113849580 A CN 113849580A
- Authority
- CN
- China
- Prior art keywords
- node
- neural network
- graph
- rating
- classification
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/28—Databases characterised by their database models, e.g. relational or object models
- G06F16/284—Relational databases
- G06F16/285—Clustering or classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/28—Databases characterised by their database models, e.g. relational or object models
- G06F16/284—Relational databases
- G06F16/288—Entity relationship models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/03—Credit; Loans; Processing thereof
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- General Engineering & Computer Science (AREA)
- Marketing (AREA)
- Development Economics (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- Data Mining & Analysis (AREA)
- Human Resources & Organizations (AREA)
- Game Theory and Decision Science (AREA)
- Technology Law (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The application provides a subject rating prediction method, a subject rating prediction device, electronic equipment and a storage medium, wherein the method comprises the following steps: acquiring a node identifier of a target subject; inquiring a target main body node corresponding to the node identification from the knowledge graph; and carrying out classification rating prediction on the target subject nodes by using a pre-trained graph convolution neural network model to obtain the classification rating of the target subject nodes. By inquiring the target main body nodes corresponding to the node identifications from the knowledge graph and using a pre-trained graph convolution neural network model to predict the classification ratings of the target main body nodes, the classification ratings of the target main body nodes are obtained, the phenomenon that the classification ratings are predicted by using a traditional regression model or multiple regression integration algorithms is avoided, the influence and the risk of an associated company of a debt issue main body can be noticed by using the graph convolution neural network model, and therefore the accuracy of rating the debt issue main body is effectively improved.
Description
Technical Field
The embodiment of the application relates to the technical field of deep learning and knowledge maps, in particular to a subject rating prediction method and device, electronic equipment and a storage medium.
Background
The subject rating refers to a classification rating related to credit performed by taking an economic subject such as an enterprise, a company, or an individual as an object, and specifically includes: a subject credit rating, debt/bond rating, loan rating or transaction rating, and the like.
Most of the existing subject rating methods are based on traditional regression models or multiple regression integration algorithms and the like, according to information data such as financial indexes or basic operating conditions of debt companies, various regression models are used and the integration algorithms are combined to calculate the information data, so that the classification rating of the debt companies is predicted, and the default risk of the debt subjects is evaluated.
Disclosure of Invention
An object of the embodiments of the present application is to provide a subject rating prediction method, device, electronic device, and storage medium, which are used to solve the problem of insufficient accuracy in rating a debt subject.
The embodiment of the application provides a subject rating prediction method, which comprises the following steps: acquiring a node identifier of a target subject; inquiring a target main body node corresponding to the node identification from the knowledge graph; and carrying out classification rating prediction on the target subject nodes by using a pre-trained graph convolution neural network model to obtain the classification rating of the target subject nodes. In the implementation process, the target subject nodes corresponding to the node identifications are inquired from the knowledge graph, and the pre-trained graph convolution neural network model is used for carrying out classification rating prediction on the target subject nodes to obtain the classification ratings of the target subject nodes, so that the problem that the traditional regression model or multiple regression integration algorithms are used for predicting the classification ratings is avoided, the graph convolution neural network model can pay more attention to the influence and risk of the associated companies of the debt issue subject, and the accuracy of rating the debt issue subject is effectively improved.
Optionally, in this embodiment of the present application, performing classification rating prediction on target subject nodes using a pre-trained graph convolution neural network model includes: acquiring feature data and relation data of a target subject node; and carrying out classification rating prediction on the target subject nodes by using the graph convolution neural network model according to the feature data and the relation data of the target subject nodes. In the implementation process, the target subject nodes are classified, graded and predicted by using the graph convolution neural network model according to the feature data and the relation data of the target subject nodes, the situation that the classification grading is predicted by using a traditional regression model or multiple regression integration algorithms is avoided, the influence and the risk of an associated company of the debt subject can be more noticed by using the graph convolution neural network model, and the grading accuracy of the debt subject is effectively improved.
Optionally, in this embodiment of the present application, before querying a node from the knowledge-graph to identify a corresponding target subject node, the method further includes: and establishing a knowledge graph by taking the node identification of the target main body as a node, taking the relation data between the target main bodies as an edge and taking the characteristic data of the target main body as the characteristics of the node. In the implementation process, the classification rating, the feature data and the relationship data of the sample nodes in the knowledge graph are used for training the graph convolution neural network model, and the graph convolution neural network model is used for rating the debt main body, so that the accuracy rate of the graph convolution neural network model for rating the debt main body is improved.
Optionally, in this embodiment of the present application, after the establishing the knowledge graph, the method further includes: obtaining a graph convolution neural network; the weight values of the edges in the knowledge-graph are reconstructed using a graph convolutional neural network. In the implementation process, the graph convolution neural network is used for reconstructing the weighted values of the edges in the knowledge graph, the classification rating, the characteristic data and the relationship data of the sample nodes in the knowledge graph are used for training the graph convolution neural network model, and the graph convolution neural network model is used for rating the debt main body, so that the accuracy rate of the graph convolution neural network model for rating the debt main body is improved.
Optionally, in this embodiment of the present application, after reconstructing an edge in a knowledge graph using a graph convolution neural network, the method further includes: obtaining classification ratings of a plurality of sample nodes, and feature data and relation data of each sample node; and training the graph convolution neural network by taking the characteristic data and the relation data of the plurality of sample nodes as training data and the classification grades of the plurality of sample nodes as training labels to obtain a graph convolution neural network model. In the implementation process, the characteristic data and the relation data of the plurality of sample nodes are used as training data, the classification grades of the plurality of sample nodes are used as training labels, the convolutional neural network is trained, and the convolutional neural network model obtained through training is used for grading the debt subjects, so that the grading accuracy of the convolutional neural network model for the debt subjects is improved.
Optionally, in an embodiment of the present application, training a graph convolution neural network includes: predicting the classification rating of the sample node by using a graph convolution neural network according to the characteristic data and the relation data of the sample node to obtain the prediction rating of the sample node; calculating a loss value between the prediction rating of the sample node and the classification rating of the sample node; and training the graph convolution neural network according to the loss value.
Optionally, in this embodiment of the present application, before training the graph convolution neural network, the method further includes: the convolution kernel in the graph convolution neural network is modified. In the implementation process, the convolution kernel in the graph convolution neural network is modified, so that the graph convolution neural network model has practical significance, and the accuracy of the graph convolution neural network model in predicting the main body rating is improved.
An embodiment of the present application further provides a subject rating prediction apparatus, including: the node identification acquisition module is used for acquiring the node identification of the target main body; the target node query module is used for querying a target main body node corresponding to the node identification from the knowledge graph; and the classification rating obtaining module is used for predicting the classification rating of the target subject node by using a pre-trained graph convolution neural network model to obtain the classification rating of the target subject node.
Optionally, in an embodiment of the present application, the classification rating obtaining module includes: the characteristic relation acquisition module is used for acquiring characteristic data and relation data of the target main body node; and the classification and rating prediction module is used for performing classification and rating prediction on the target main body nodes by using the graph convolution neural network model according to the feature data and the relation data of the target main body nodes.
Optionally, in an embodiment of the present application, the subject rating prediction apparatus further includes: and the knowledge graph establishing module is used for establishing the knowledge graph by taking the node identification of the target main body as a node, taking the relation data between the target main bodies as a side and taking the characteristic data of the target main bodies as the characteristics of the node.
Optionally, in an embodiment of the present application, the subject rating prediction apparatus further includes: the graph convolution network acquisition module is used for acquiring a graph convolution neural network; and the edge weight value reconstruction module is used for reconstructing the weight values of the edges in the knowledge graph by using the graph convolution neural network.
Optionally, in an embodiment of the present application, the subject rating prediction apparatus further includes: the training data acquisition module is used for acquiring the classification grades of a plurality of sample nodes, and the characteristic data and the relation data of each sample node; and the network model obtaining module is used for training the graph convolution neural network by taking the characteristic data and the relation data of the plurality of sample nodes as training data and the classification grades of the plurality of sample nodes as training labels to obtain the graph convolution neural network model.
Optionally, in this embodiment of the present application, the network model obtaining module includes: the classification rating prediction module is used for predicting the classification rating of the sample node by using the graph convolution neural network according to the characteristic data and the relation data of the sample node to obtain the prediction rating of the sample node; the rating loss calculation module is used for calculating a loss value between the prediction rating of the sample node and the classification rating of the sample node; and training the graph convolution neural network according to the loss value.
Optionally, in an embodiment of the present application, the subject rating prediction apparatus further includes: and the convolution kernel modification module is used for modifying the convolution kernel in the graph convolution neural network.
An embodiment of the present application further provides an electronic device, including: a processor and a memory, the memory storing processor-executable machine-readable instructions, the machine-readable instructions when executed by the processor performing the method as described above.
Embodiments of the present application also provide a computer-readable storage medium having a computer program stored thereon, where the computer program is executed by a processor to perform the method as described above.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are required to be used in the embodiments of the present application will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.
FIG. 1 is a flow chart diagram illustrating a subject rating prediction method provided by an embodiment of the present application;
FIG. 2 is a schematic diagram of a constructed knowledge-graph as provided by an embodiment of the present application;
FIG. 3 is a schematic diagram illustrating weight values of constructed reconstructed edges provided by an embodiment of the present application;
FIG. 4 is a schematic flow chart diagram of a convolutional neural network for training graph provided by an embodiment of the present application;
fig. 5 is a schematic structural diagram of a subject rating prediction apparatus provided in an embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. The components of the embodiments of the present application, generally described and illustrated in the figures herein, can be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present application, as presented in the figures, is not intended to limit the scope of the embodiments of the application, as claimed, but is merely representative of selected embodiments of the application. All other embodiments obtained by a person skilled in the art based on the embodiments of the present application without any inventive step are within the scope of the embodiments of the present application.
Before introducing the subject rating prediction method provided by the embodiment of the present application, some concepts related to the embodiment of the present application are introduced:
knowledge map (Knowledge Graph) refers to a method or a tool for linking Knowledge together according to a certain rule and showing the Knowledge in the form of map, is called a Knowledge domain visual Knowledge domain mapping map in the book intelligence field, is a series of different graphs for displaying the relationship between the Knowledge development process and the structure, describes Knowledge resources and carriers thereof by using a visual technology, and excavates, analyzes, constructs, draws and displays the Knowledge and the mutual link between the Knowledge and the carriers.
A Graph Convolutional neural Network (GCN) is a Convolutional neural Network that processes topological relation data of a Non-Euclidean Structure (Non Euclidean Structure). The non-euclidean structured topological relation data here is, for example: the data of the topological relation formed by the organization and the organization or the person and the person in the social network, the data of the topological relation formed by the router and the router or the switch and the switch in the information network, and the like.
It should be noted that the subject rating prediction method provided in the embodiment of the present application may be executed by an electronic device, where the electronic device refers to a device terminal having a function of executing a computer program or the server described above, and the device terminal includes, for example: smart phones, Personal Computers (PCs), tablet computers, Personal Digital Assistants (PDAs), or Mobile Internet Devices (MIDs), etc. The server is, for example: x86 server and non-x 86 server, non-x 86 server includes: mainframe, minicomputer, and UNIX server.
Application scenarios applicable to the subject rating prediction method are described below, where the application scenarios include, but are not limited to, the field of financial risk prevention and control, specifically for example: subject credit rating, debt rating, bond rating, loan rating, or transaction rating, etc., for a company or individual. It can be noted that the default of credit in the actual application scenario is not only related to the financial or basic situation of the debt subject itself, but also closely related to the stakeholders, the holding companies, the guaranty companies, the companies with common high administration, and even the related companies of the related companies. The traditional default model only considers the factors such as the financial condition and the operating condition of the debt subject, neglects the negative influence of the associated company on the debt subject, and therefore the accuracy rate of grading the debt subject is low.
Please refer to a flow chart diagram of a subject rating prediction method provided by the embodiment of the present application shown in fig. 1; the main idea of the subject rating prediction method is that a target subject node corresponding to a node identification is inquired from a knowledge graph, a pre-trained graph convolution neural network model is used for carrying out classification rating prediction on the target subject node, classification rating of the target subject node is obtained, the problem that a traditional regression model or multiple regression integration algorithms are used for predicting the classification rating is avoided, influence and risk of an associated company of a debt subject can be noticed by using the graph convolution neural network model, and therefore the accuracy of rating the debt subject is effectively improved. The subject rating prediction method may include:
step S110: and acquiring the node identification of the target subject.
The target subject refers to an economic subject such as an enterprise, a company or an individual (individual operation or individual household); the node identification of the target subject refers to the node unique identification of the target subject on the knowledge graph, and the node identification that can be used by the company of the enterprise includes but is not limited to: company name (business name), organization code, unified social credit code, or taxpayer identification number, etc.
There are many embodiments of the step S110, including but not limited to: the first acquisition mode is that node identification of a target main body sent by other terminal equipment is received, and the node identification of the target main body is stored in a file system, a database or mobile storage equipment; the second obtaining method obtains a node identifier of a target subject stored in advance, specifically for example: acquiring a node identifier of a target subject from a file system, or acquiring a node identifier of the target subject from a database, or acquiring a node identifier of the target subject from a mobile storage device; in the third obtaining mode, software such as a browser is used for obtaining the node identification of the target subject on the internet, or other application programs are used for accessing the internet to obtain the node identification of the target subject.
After step S110, step S120 is performed: and inquiring the nodes from the knowledge graph to identify the corresponding target subject nodes.
It is understood that before the use of the knowledge graph, the knowledge graph needs to be constructed, and the process of constructing the knowledge graph is too complicated, and thus, the process of constructing the knowledge graph is described in detail later.
After step S120, step S130 is performed: and carrying out classification rating prediction on the target subject nodes by using a pre-trained graph convolution neural network model to obtain the classification rating of the target subject nodes.
The implementation of step S130 may include: and obtaining characteristic data and relation data of the target subject node by using the graph convolution neural network model. And performing classification rating prediction on the target main body nodes by using the graph convolution neural network model according to the feature data and the relation data of the target main body nodes to obtain the classification rating of the target main body nodes. Optionally, after obtaining the classification rating of the target subject node, the breach risk of the target subject, i.e. the breach probability of the target subject, may also be evaluated according to the classification rating.
In the implementation process, the target subject nodes corresponding to the node identifications are inquired from the knowledge graph, and the pre-trained graph convolution neural network model is used for carrying out classification rating prediction on the target subject nodes to obtain the classification ratings of the target subject nodes, so that the problem that the traditional regression model or multiple regression integration algorithms are used for predicting the classification ratings is avoided, the graph convolution neural network model can pay more attention to the influence and risk of the associated companies of the debt issue subject, and the accuracy of rating the debt issue subject is effectively improved.
It is understood that before using the knowledge graph, a knowledge graph needs to be constructed, and the process of constructing the knowledge graph may include:
step S121: and acquiring node identification and feature data of the target subject and relationship data between the target subject and the target subject.
The embodiment of step S121 described above is, for example: here, with a company or a business as a target subject, the node identification of the target subject may select a company name (business name), an organization code, a unified social credit code, or a taxpayer identification number, and so on. The feature data refers to attribute data of a target subject such as a company or a business, for example: asset liability rates, inventory turnover rates, accounts receivable/total assets, credit line reductions, employee population, official document aborts, registration addresses, official network addresses, mailbox addresses, legal representatives and company enterprise profiles, among others. The relationship data between the target subject and the target subject is, for example: equity relations (stock holding ratio), guaranty relations (guaranty amount/net property), pledge relations (pledge amount/net property), common high line and common shareholder, and so on.
Optionally, after obtaining the relationship data, the node identifier, and the feature data, the relationship data, the node identifier, and the feature data may be preprocessed, for example: the method comprises the steps of taking financial conditions and operational conditions of enterprises and conditions of related parties thereof as factors influencing implicit rating judgment, filling missing values in the factors according to preset logic rules for node characteristic data in a knowledge graph according to screened debt subjects and processing related factors such as open financial indexes and operational condition indexes of related companies, and transforming the filled factors, wherein specific transformation operations comprise calculating ranking percentage of industrial factors, calculating the factor relative to the current year and term change and calculating the ranking percentage of the change values. The preset logic rules refer to that different types of data processing modes in the knowledge graph are different, industry ranking percentage can be obtained for continuous data, and original data can be adopted as node characteristics for discrete data. For the relationship data, the equity relationship uses the holding ratio as the weight of the edge, the guaranty relationship uses the guaranty amount/total capital of the insured company as the weight of the edge, and the pledge relationship uses the pledge amount/total capital of the exporting company as the weight of the edge. Reconstructing the weight values for the edges in the knowledge graph, for example: if there are multiple edges of relationships between the first company and the second company, the multiple edges of relationships may be merged into one edge, and the maximum weight of the multiple edges may be taken as the edge weight.
Step S122: and establishing a knowledge graph by taking the node identification of the target main body as a node, taking the relation data between the target main bodies as an edge and taking the characteristic data of the target main body as the characteristics of the node.
Please refer to fig. 2, which is a schematic diagram of a constructed knowledge graph provided by the embodiment of the present application; the embodiment of step S122 is, for example: with the node identification of the target subject as a node (denoted as
) The relationship data between the target subjects is defined as an edge (represented as ε), and the feature data of the target subjects is defined as a feature of a node (represented as ε)
) The knowledge graph is built using a relational database, where the knowledge graph may be represented as
The knowledge graph is a modern theory which achieves the aim of multi-discipline fusion by combining theories and methods of applying subjects such as mathematics, graphics, information visualization technology, information science and the like with methods such as metrology introduction analysis, co-occurrence analysis and the like and utilizing a visualized graph to vividly display the core structure, development history, frontier field and overall knowledge framework of the subjects.
Optionally, after the knowledge graph is established, the weight values of the edges in the knowledge graph may be further reconstructed, and the process of reconstructing the weight values may include:
step S123: and acquiring a graph convolution neural network.
There are many embodiments of the step S123, including but not limited to: in the first mode, a graph convolution neural network is constructed from the beginning, specifically for example: acquiring graph data, the graph data comprising: the relations of holding stocks, guarantying, pledging, upstream and downstream between companies and the relations of holding stocks, legal representatives, high management, real control persons and the like between the companies and natural persons; then, features are extracted from the graph data, and after performing operations such as node classification (node classification), graph classification (graph classification), and edge prediction (link prediction), a graph convolution neural network is obtained using the features. The second way is to obtain a pre-stored atlas neural network, specifically for example: acquiring a graph volume neural network from a file system, or acquiring the graph volume neural network from a database, or acquiring the graph volume neural network from mobile storage equipment; in the third mode, a software such as a browser is used for obtaining the atlas neural network on the internet, or other application programs are used for accessing the internet to obtain the atlas neural network.
Step S124: the weight values of the edges in the knowledge-graph are reconstructed using a graph convolutional neural network.
The embodiment of step S124 described above is, for example: because the weight of partial edges in the knowledge graph is uncertain, the weight values of the partial edges in the knowledge graph can be reconstructed in a comparison learning mode of a graph neural network; in particular, a knowledge graph is given
The resulting graph neural network is denoted as fθThe weight values of the edges in the knowledge-graph can then be reconstructed by maximizing the probability p (G; theta) of the neural network model of the graph, which can be represented as theta*＝maxθp (G; theta). The specific technical process of the probability p (G; theta) of the neural network model of the graph is as follows: after learning the model fθAnd a permutation combination of all nodes in the knowledge-graph, the distribution of the graph neural network is equal to the expected probability for all permutation combinations, that is, can be expressed as p (G; θ) ═ E using the formulaπ[pθ(Xπ，Eπ)](ii) a Wherein, XπNode classes representing different permutation combinations, EπRepresenting the edges of different permutation combinations. Then, p is addedθCarrying out logarithmic resolution on (X, E) to obtain
Thus, the method can obtain the product,
wherein E isi，oRepresenting the known edges in the knowledge-graph,
representing unknown edges in the knowledge-graph. The problem can be split into two parts through the method, wherein the two parts are that the edge is given to generate the node classification, and the other part is that the edge and the generated node classification are given to generate the rest edges.
Please refer to fig. 3 for a schematic diagram of weight values of reconstructed edges constructed according to an embodiment of the present application; there are A, B, C and D four nodes in the first sub-graph, where the node direction includes: a to B, A to C, A to D, B to D. First, referring to the second subgraph, the edges B through D can be wiped off, with the remaining edges (A through B, A through C, A through D) being known edges. Secondly, a classification (i.e., a body rating) of the node D is generated using known edges (a to B, a to C, a to D), and a relationship between the classification of the current node D and its neighboring nodes (node a and node B) is obtained according to the classification of the node D. Then, a graph convolution neural network is used to predict the weight of the erased edge (i.e., the edge from B to D) according to the known edges (the edge from a to B, the edge from a to C, and the edge from a to D) and the generated node classification, thereby obtaining the relationship weight between the node D classification and the erased edge (i.e., the edge from B to D).
Please refer to fig. 4, which is a schematic flow chart of a convolutional neural network of training diagram provided in the embodiment of the present application; it can be understood that before using the convolutional neural network model for classification prediction, the convolutional neural network model needs to be trained, and the process of training the convolutional neural network model may include:
step S210: and obtaining the classification ratings of a plurality of sample nodes, and the characteristic data and the relation data of each sample node.
The embodiment of step S210 described above is, for example: the classification rating of the sample node and the feature data and the relationship data of the sample node may be obtained separately, for example: manually collecting the classification ratings of the sample nodes, and manually identifying the characteristic data and the relation data of the sample nodes of the classification ratings of the sample nodes; of course, the classification rating of the sample node and the feature data and the relationship data of the sample node may also be obtained by packing into a training data set, and the training data set is taken as an example for explanation. The training data set is obtained in a manner that includes: the first acquisition mode is that a training data set sent by other terminal equipment is received, and the training data set is stored in a file system, a database or mobile storage equipment; a second obtaining manner, obtaining a pre-stored training data set, specifically for example: acquiring a training data set from a file system, or acquiring the training data set from a database, or acquiring the training data set from a mobile storage device; in the third obtaining mode, a software such as a browser is used for obtaining the training data set on the internet, or other application programs are used for accessing the internet to obtain the training data set.
After step S210, step S220 is performed: and training the graph convolution neural network by taking the characteristic data and the relation data of the plurality of sample nodes as training data and the classification grades of the plurality of sample nodes as training labels to obtain a graph convolution neural network model.
The implementation of step S220 may include: and predicting the classification rating of the sample node by using a graph convolution neural network according to the characteristic data and the relation data of the sample node to obtain the prediction rating of the sample node. A Loss value between the prediction rating of the sample node and the classification rating of the sample node is calculated using a Cross Entropy Loss (Cross Entropy Loss, CELoss or CELoss) function. And training the graph convolution neural network according to the loss value to obtain a graph convolution neural network model.
Optionally, before the training of the atlas neural network, the convolution kernel in the atlas neural network may also be modified, and the specific way of modification is, for example: optimizing convolution kernel of graph convolution neural network, wherein the convolution calculation formula is Hl+1＝σ(L′HlWl) Wherein L ═ I (I)n-A)Cn，CnRepresenting a diagonal matrix, W, consisting of registered capital in feature data in a knowledge-graphlIs a weight parameter matrix of the l-th layer, σ (-) is a nonlinear activation function, HlThe A relation is a temporary matrix for the input value of the l hidden layer, and I is an identity matrix. It can be noted that it is commonThe convolution calculation formula in the graph neural network is Hl+1＝σ(LsymHlWl) Wherein
WlIs a weight parameter matrix of the l-th layer, σ (-) is a nonlinear activation function, HlFor the l-th layer input, D is the degree matrix and A is the adjacency matrix. The convolution method is to introduce a self-degree matrix, solve the problem that self-transmission of self-node information is not considered, and simultaneously carry out normalization operation on an adjacent matrix, wherein the normalization operation is obtained by multiplying two sides of the adjacent matrix by the degree evolution of nodes and then taking the inverse, so that different nodes have the same influence. However, in the actual research process, the weights of the companies are not the same, and companies with higher registered capital and more relationships tend to have higher influence, so that the normalization mode is not in accordance with the actual economic significance. Therefore, the convolution kernel in the graph convolution neural network is modified, so that the graph convolution neural network model has practical significance, and the accuracy of the graph convolution neural network model in predicting the main body rating is improved.
Please refer to fig. 5, which illustrates a schematic structural diagram of a subject rating prediction apparatus according to an embodiment of the present application. The embodiment of the present application provides a subject rating prediction apparatus 300, including:
a node identifier obtaining module 310, configured to obtain a node identifier of the target subject.
And a target node query module 320, configured to query nodes from the knowledge-graph to identify corresponding target subject nodes.
And the classification rating obtaining module 330 is configured to perform classification rating prediction on the target subject node by using a pre-trained graph convolution neural network model, so as to obtain a classification rating of the target subject node.
Optionally, in an embodiment of the present application, the classification rating obtaining module includes:
and the characteristic relation acquisition module is used for acquiring the characteristic data and the relation data of the target main body node.
And the classification and rating prediction module is used for performing classification and rating prediction on the target main body nodes by using the graph convolution neural network model according to the feature data and the relation data of the target main body nodes.
Optionally, in an embodiment of the present application, the subject rating prediction apparatus further includes:
and the knowledge graph establishing module is used for establishing the knowledge graph by taking the node identification of the target main body as a node, taking the relation data between the target main bodies as a side and taking the characteristic data of the target main bodies as the characteristics of the node.
Optionally, in an embodiment of the present application, the subject rating prediction apparatus further includes:
and the graph convolution network acquisition module is used for acquiring the graph convolution neural network.
And the edge weight value reconstruction module is used for reconstructing the weight values of the edges in the knowledge graph by using the graph convolution neural network.
Optionally, in this embodiment of the present application, the subject rating prediction apparatus may further include:
and the training data acquisition module is used for acquiring the classification grades of the plurality of sample nodes and the characteristic data and the relation data of each sample node.
And the network model obtaining module is used for training the graph convolution neural network by taking the characteristic data and the relation data of the plurality of sample nodes as training data and the classification grades of the plurality of sample nodes as training labels to obtain the graph convolution neural network model.
Optionally, in this embodiment of the present application, the network model obtaining module includes:
and the classification rating prediction module is used for predicting the classification rating of the sample node by using the graph convolution neural network according to the characteristic data and the relation data of the sample node to obtain the prediction rating of the sample node.
And the rating loss calculation module is used for calculating a loss value between the prediction rating of the sample node and the classification rating of the sample node.
And training the graph convolution neural network according to the loss value.
Optionally, in this embodiment of the present application, the subject rating prediction apparatus may further include:
and the convolution kernel modification module is used for modifying the convolution kernel in the graph convolution neural network.
It should be understood that the apparatus corresponds to the above subject rating prediction method embodiment, and can perform the steps related to the above method embodiment, and the specific functions of the apparatus can be referred to the above description, and the detailed description is appropriately omitted here to avoid redundancy. The device includes at least one software function that can be stored in memory in the form of software or firmware (firmware) or solidified in the Operating System (OS) of the device.
An electronic device provided in an embodiment of the present application includes: a processor and a memory, the memory storing processor-executable machine-readable instructions, the machine-readable instructions when executed by the processor performing the method as above.
Embodiments of the present application also provide a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the method as above is performed. The computer-readable storage medium may be implemented by any type of volatile or nonvolatile Memory device or combination thereof, such as a Static Random Access Memory (SRAM), an Electrically Erasable Programmable Read-Only Memory (EEPROM), an Erasable Programmable Read-Only Memory (EPROM), a Programmable Read-Only Memory (PROM), a Read-Only Memory (ROM), a magnetic Memory, a flash Memory, a magnetic disk, or an optical disk.
In the embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. The apparatus embodiments described above are merely illustrative, and for example, the flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved.
In addition, functional modules of the embodiments in the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.
In this document, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions.
The above description is only an alternative embodiment of the embodiments of the present application, but the scope of the embodiments of the present application is not limited thereto, and any person skilled in the art can easily conceive of changes or substitutions within the technical scope of the embodiments of the present application, and all the changes or substitutions should be covered by the scope of the embodiments of the present application.
Claims (10)
1. A subject rating prediction method, comprising:
acquiring a node identifier of a target subject;
inquiring a target main body node corresponding to the node identification from a knowledge graph;
and performing classification rating prediction on the target subject nodes by using a pre-trained graph convolution neural network model to obtain the classification rating of the target subject nodes.
2. The method of claim 1, wherein the using a pre-trained convolutional neural network model for class rating prediction of the target subject node comprises:
acquiring feature data and relation data of the target subject node;
and performing classification rating prediction on the target subject nodes by using the graph convolutional neural network model according to the feature data and the relation data of the target subject nodes.
3. The method of claim 1, wherein prior to said querying the nodes from the knowledge-graph to identify corresponding target subject nodes, further comprising:
and establishing the knowledge graph by taking the node identification of the target main body as a node, taking the relation data between the target main bodies as an edge and taking the characteristic data of the target main body as the characteristics of the node.
4. The method of claim 3, after said establishing said knowledge-graph, further comprising:
obtaining a graph convolution neural network;
reconstructing weight values of edges in the knowledge-graph using the graph convolution neural network.
5. The method of claim 4, further comprising, after said reconstructing edges in said knowledge-graph using said graph convolutional neural network:
obtaining classification ratings of a plurality of sample nodes, and feature data and relationship data of each sample node;
and training the graph convolution neural network by taking the feature data and the relation data of the plurality of sample nodes as training data and the classification grades of the plurality of sample nodes as training labels to obtain the graph convolution neural network model.
6. The method of claim 5, wherein training the atlas neural network comprises:
predicting the classification rating of the sample node by using the graph convolution neural network according to the feature data and the relation data of the sample node to obtain the prediction rating of the sample node;
calculating a loss value between the prediction rating of the sample node and the classification rating of the sample node;
and training the graph convolution neural network according to the loss value.
7. The method of claim 5, further comprising, prior to said training the atlas neural network:
the convolution kernel in the graph convolution neural network is modified.
8. A subject rating prediction apparatus, comprising:
the node identification acquisition module is used for acquiring the node identification of the target main body;
the target node query module is used for querying a target main body node corresponding to the node identification from the knowledge graph;
and the classification rating obtaining module is used for predicting the classification rating of the target subject node by using a pre-trained graph convolution neural network model to obtain the classification rating of the target subject node.
9. An electronic device, comprising: a processor and a memory, the memory storing machine-readable instructions executable by the processor, the machine-readable instructions, when executed by the processor, performing the method of any of claims 1 to 7.
10. A computer-readable storage medium, having stored thereon a computer program which, when executed by a processor, performs the method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111143690.2A CN113849580A (en) | 2021-09-28 | 2021-09-28 | Subject rating prediction method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111143690.2A CN113849580A (en) | 2021-09-28 | 2021-09-28 | Subject rating prediction method and device, electronic equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113849580A true CN113849580A (en) | 2021-12-28 |
Family
ID=78980305
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111143690.2A Pending CN113849580A (en) | 2021-09-28 | 2021-09-28 | Subject rating prediction method and device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113849580A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116227939A (en) * | 2023-05-04 | 2023-06-06 | 深圳市迪博企业风险管理技术有限公司 | Enterprise credit rating method and device based on graph convolution neural network and EM algorithm |
-
2021
- 2021-09-28 CN CN202111143690.2A patent/CN113849580A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116227939A (en) * | 2023-05-04 | 2023-06-06 | 深圳市迪博企业风险管理技术有限公司 | Enterprise credit rating method and device based on graph convolution neural network and EM algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107945024B (en) | Method for identifying internet financial loan enterprise operation abnormity, terminal equipment and storage medium | |
| CN110188198B (en) | Anti-fraud method and device based on knowledge graph | |
| US20190065939A1 (en) | Bayesian network based hybrid machine learning | |
| US20210192649A1 (en) | Systems & methods for automated assessment for remediation and/or redevelopment of brownfield real estate | |
| CN110704730A (en) | Product data pushing method and system based on big data and computer equipment | |
| CN111582538A (en) | Community value prediction method and system based on graph neural network | |
| US9858526B2 (en) | Method and system using association rules to form custom lists of cookies | |
| CN112163963A (en) | Service recommendation method and device, computer equipment and storage medium | |
| CN110796269A (en) | Method and device for generating model, and method and device for processing information | |
| CN115545886A (en) | Overdue risk identification method, overdue risk identification device, overdue risk identification equipment and storage medium | |
| CN113849580A (en) | Subject rating prediction method and device, electronic equipment and storage medium | |
| CN113408627A (en) | Target object determination method and device and server | |
| CN112950347A (en) | Resource data processing optimization method and device, storage medium and terminal | |
| CN116503158A (en) | Enterprise bankruptcy risk early warning method, system and device based on data driving | |
| Neumann et al. | Machine Learning-based predictions of customers’ decisions in car insurance | |
| US20220215142A1 (en) | Extensible Agents in Agent-Based Generative Models | |
| CN114897607A (en) | Data processing method and device for product resources, electronic equipment and storage medium | |
| CN115439079A (en) | Item classification method and device | |
| CN113850669A (en) | User grouping method and device, computer equipment and computer readable storage medium | |
| CN114493853A (en) | Credit rating evaluation method, credit rating evaluation device, electronic device and storage medium | |
| CN114693428A (en) | Data determination method and device, computer readable storage medium and electronic equipment | |
| CN114170000A (en) | Credit card user risk category identification method, device, computer equipment and medium | |
| US11004156B2 (en) | Method and system for predicting and indexing probability of financial stress | |
| CN111932131A (en) | Service data processing method and device | |
| CN112116253A (en) | Method, device and system for selecting central mesh point |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |