CN115860906A - Credit risk identification method, credit risk identification device and storage medium - Google Patents
Credit risk identification method, credit risk identification device and storage medium Download PDFInfo
- Publication number
- CN115860906A CN115860906A CN202211466157.4A CN202211466157A CN115860906A CN 115860906 A CN115860906 A CN 115860906A CN 202211466157 A CN202211466157 A CN 202211466157A CN 115860906 A CN115860906 A CN 115860906A
- Authority
- CN
- China
- Prior art keywords
- loan
- vector
- customer
- credit risk
- client
- 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
- 238000000034 method Methods 0.000 title claims abstract description 59
- 238000003860 storage Methods 0.000 title claims abstract description 15
- 239000013598 vector Substances 0.000 claims abstract description 258
- 239000011159 matrix material Substances 0.000 claims abstract description 105
- 238000006243 chemical reaction Methods 0.000 claims abstract description 34
- 230000009466 transformation Effects 0.000 claims abstract description 26
- 238000012545 processing Methods 0.000 claims abstract description 17
- 230000001131 transforming effect Effects 0.000 claims abstract description 4
- 230000006870 function Effects 0.000 claims description 76
- 230000007246 mechanism Effects 0.000 claims description 18
- 238000004364 calculation method Methods 0.000 claims description 14
- 230000004913 activation Effects 0.000 claims description 9
- 238000004590 computer program Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 230000000644 propagated effect Effects 0.000 claims description 4
- 230000008569 process Effects 0.000 description 8
- 238000010586 diagram Methods 0.000 description 6
- 238000013135 deep learning Methods 0.000 description 4
- BQCADISMDOOEFD-UHFFFAOYSA-N Silver Chemical compound [Ag] BQCADISMDOOEFD-UHFFFAOYSA-N 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 229910052709 silver Inorganic materials 0.000 description 3
- 239000004332 silver Substances 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 125000004432 carbon atom Chemical group C* 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000002265 prevention Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012502 risk assessment Methods 0.000 description 1
- 230000007480 spreading Effects 0.000 description 1
- 238000003892 spreading Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Abstract
The application provides a credit risk identification method, a credit risk identification device and a credit risk identification storage medium, wherein the method comprises the following steps: constructing a customer group information knowledge graph and a customer vector; calculating a customer vector to determine a feature transformation vector for each customer, and transforming the feature transformation vector by using a first model parameter matrix to obtain a transformed feature transformation vector; processing nodes in the passenger group information knowledge graph to determine influence vectors, and calculating a propagation chain sequence to determine propagation vectors; constructing a credit risk function based on the converted feature conversion vector and the propagation vector; constructing a target loss function, and calculating the value of the first model parameter matrix, the value of the depth model parameter matrix and the credit risk probability; and interpreting the credit risk probability by using the first model parameter matrix and the depth model parameter matrix. By adopting the credit risk identification method, the credit risk identification device and the credit risk identification storage medium, the problem that the identification result cannot be explained in the conventional credit risk identification method is solved.
Description
Technical Field
The application relates to the technical field of data processing, in particular to a credit risk identification method, a credit risk identification device and a credit risk identification storage medium.
Background
The financial risk prevention mainly aims at solving the credit risk cheating problem. Not only are malicious fraudulent lending activities identified, but also customers who are likely to have no repayment capability are rejected. The method is essentially a classification problem for the clients, the data distribution of the classification problem is extremely unbalanced due to the specificity of risk identification, and the income brought by a high-quality client is often far lower than the loss caused by a risk client, so that the identification capability of the model is more required. In addition, in an actual risk business scene, the identification of risks often has interpretability requirements, and the model has very important significance for giving judgment basis when judging. In this context, the demand on risk identification models is increasing in order to achieve a fast and accurate loan approval process. The existing credit risk identification method is mainly used for building a calculation model based on a deep learning method and capturing the relationship among customers by building a customer relationship network.
However, the credit risk identification method adopts a deep learning mode, and the variable parameters are hidden in the network, so that the identification result cannot be explained.
Disclosure of Invention
In view of the above, an object of the present application is to provide a credit risk identification method, apparatus and storage medium, so as to solve the problem that the identification result cannot be interpreted in the existing credit risk identification method.
In a first aspect, an embodiment of the present application provides a credit risk identification method, including:
the method comprises the steps of obtaining client information of a plurality of loan clients, constructing a client group information knowledge graph and a client vector of each loan client based on the client information of the loan clients, wherein the client information comprises a plurality of characteristic variables, and the client vector is used for representing values of the plurality of characteristic variables corresponding to the loan clients;
aiming at each loan customer, calculating a customer vector corresponding to the loan customer by using a self-attention mechanism to determine a feature transformation vector, and transforming the feature transformation vector by using a first model parameter matrix with unknown values to obtain a transformed feature transformation vector;
processing the node corresponding to the loan client in the consumer group information knowledge graph by using the graph breadth attention model to determine an influence vector corresponding to the loan client, and calculating a propagation chain sequence corresponding to the loan client by using the graph depth attention model and the influence vector to determine a propagation vector corresponding to the loan client, wherein the influence vector is used for representing the influence degree of different neighbors on the loan client, and the propagation vector is obtained after the vector of the loan client represents the propagation of the neighbors;
constructing a credit risk function corresponding to the loan client based on the converted feature conversion vector and the propagation vector;
constructing a target loss function based on credit risk functions corresponding to different loan clients, and calculating the value of a first model parameter matrix, the value of a depth model parameter matrix corresponding to an image depth attention model and credit risk probabilities of the different loan clients when the output result of the target loss function is minimum;
and explaining the credit risk probability of different loan clients by using the first model parameter matrix and the depth model parameter matrix.
Optionally, the customer information includes credit risk customer identification, and the customer group information knowledge map and the customer vector of each loan customer are constructed based on the customer information of a plurality of loan customers, including: for each loan client, coding the value of each characteristic variable in the client information of the loan client by using the evidence weight to generate a characteristic variable code of the loan client; splicing a plurality of characteristic variable codes corresponding to different characteristic variables to generate a customer vector of the loan customer; extracting customer association relations among different loan customers from customer information of the loan customers, wherein the customer association relations comprise call contact relations among the different customers, common guarantor relations, common identification number relations and common loan merchant relations; each loan client corresponds to a node, and different nodes with client association relation are connected by a non-directional connecting line; and based on the credit risk customer identification, distinguishing and displaying the nodes corresponding to the credit risk customers and the nodes corresponding to the non-credit risk customers to obtain the customer group information knowledge graph.
Optionally, the calculating the customer vector corresponding to the loan customer by using a self-attention mechanism to determine a target feature transformation vector includes: determining the product of the characteristic variable code and the self-attention model parameter matrix as a weight coefficient aiming at each characteristic variable code of the loan client; determining a variable characteristic vector corresponding to the characteristic variable code based on the weight coefficient; and combining a plurality of variable feature vectors corresponding to different feature variable codes of the loan client to generate a feature conversion vector of the loan client.
Optionally, after processing the node corresponding to the loan client in the knowledge graph of the information about the loan client by using the graph breadth attention model to determine the influence vector corresponding to the loan client, the method includes: and in the customer group information knowledge graph, determining a propagation chain sequence corresponding to the loan customer according to the connection relation between the node corresponding to the loan customer and other nodes.
Optionally, the step of determining a propagation vector corresponding to the loan client by calculating the propagation chain sequence corresponding to the loan client by using the map depth attention model and the influence vector includes: taking the influence vector as an input parameter of the image depth attention model; calculating a propagation chain sequence by using the image depth attention model after the parameters are input, and respectively obtaining a right propagation vector and a left propagation vector; the right-hand propagation vector and the left-hand propagation vector are combined to form the propagation vector of the loan client.
Optionally, constructing a credit risk function corresponding to the loan client based on the converted feature conversion vector and the propagation vector, including: converting the propagation vector by using a second model parameter matrix with unknown values to obtain a converted propagation vector; combining the converted feature conversion vector and the converted propagation vector to generate a node representation vector; and constructing a credit risk function by taking the node representation vector as a parameter of the activation function.
Optionally, constructing the target loss function based on the credit risk functions corresponding to different loan clients includes: and (4) taking credit risk functions corresponding to different loan clients as parameters of a two-classification cross entropy loss function to construct a target loss function.
Optionally, the interpreting credit risk probabilities of different loan clients by using the first model parameter matrix and the depth model parameter matrix includes: carrying out weight normalization processing on the first model parameter matrix and the depth model parameter matrix to respectively obtain a variable importance coefficient and a path importance coefficient; and (4) explaining the credit risk probability of different loan clients by using the variable importance coefficient and the path importance coefficient.
In a second aspect, an embodiment of the present application further provides a credit risk identification apparatus, where the apparatus includes:
the information calculation module is used for acquiring the customer information of a plurality of loan customers, constructing a customer group information knowledge graph and a customer vector of each loan customer based on the customer information of the plurality of loan customers, wherein the customer information comprises a plurality of characteristic variables, and the customer vector is used for representing the values of the plurality of characteristic variables corresponding to the loan customers;
the first vector acquisition module is used for calculating a customer vector corresponding to each loan customer by using a self-attention mechanism to determine a feature conversion vector, and converting the feature conversion vector by using a first model parameter matrix with unknown values to obtain a converted feature conversion vector;
the second vector acquisition module is used for processing the nodes corresponding to the loan clients in the information knowledge graph of the loan clients by using the graph attention model to determine the influence vectors corresponding to the loan clients, calculating the propagation chain sequence corresponding to the loan clients by using the graph depth attention model and the influence vectors to determine the propagation vectors corresponding to the loan clients, wherein the influence vectors are used for representing the influence degrees of different neighbors on the loan clients, and the propagation vectors are vectors obtained after the vector representation of the loan clients is propagated by the neighbors;
the risk function building module is used for building a credit risk function corresponding to the loan client based on the converted feature conversion vector and the propagation vector;
the coefficient calculation module is used for constructing a target loss function based on credit risk functions corresponding to different loan clients, and calculating the value of a first model parameter matrix, the value of a depth model parameter matrix corresponding to the map depth attention model and the credit risk probabilities of the different loan clients when the output result of the target loss function is the minimum value;
and the result interpretation module is used for interpreting the credit risk probability of different loan clients by utilizing the first model parameter matrix and the depth model parameter matrix.
In a third aspect, the present application further provides a computer-readable storage medium, on which a computer program is stored, where the computer program is executed by a processor to perform the steps of the credit risk identification method as described above.
The embodiment of the application brings the following beneficial effects:
according to the credit risk identification method, the credit risk identification device and the credit risk identification storage medium, a customer vector and a customer base information knowledge spectrogram can be obtained according to customer information, the characteristic vector of each customer and the relation between different customers are described, the similarity between different characteristic variables of the same loan customer is determined by using a self-attention mechanism to form a characteristic conversion vector, meanwhile, the influence degree of different neighbors in the customer base information knowledge spectrogram on a current node and the importance degree of different risk propagation chains corresponding to the current node are mined by using a graph attention model to form a propagation vector, the characteristic conversion vector and the propagation vector are processed to form a credit risk function, a first model parameter matrix and a depth model parameter matrix are determined by using the credit risk function, the first model parameter matrix can explain an identification result from the angle of the characteristic variables, the depth model parameter matrix can explain the identification result from the angle of the propagation path, and compared with the credit risk identification method in the prior art, the problem that the identification result cannot be explained in the prior art is solved.
In order to make the aforementioned objects, features and advantages of the present application more comprehensible, preferred embodiments accompanied with figures are described in detail below.
Drawings
To more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiments 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 those skilled in the art can also obtain other related drawings based on the drawings without inventive efforts.
FIG. 1 illustrates a flow chart of a credit risk identification method provided by an embodiment of the application;
FIG. 2 illustrates a schematic diagram of a customer group knowledge-graph provided by an embodiment of the present application;
FIG. 3 illustrates a block diagram of a credit risk identification model provided by an embodiment of the present application;
fig. 4 shows a schematic structural diagram of a credit risk identification device provided in an embodiment of the application.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, 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 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, presented in the accompanying drawings, is not intended to limit the scope of the claimed application, but is merely representative of selected embodiments of the application. Every other embodiment that can be obtained by a person skilled in the art without making creative efforts based on the embodiments of the present application falls within the protection scope of the present application.
It is noted that before the present application, financial risk prevention was mainly aimed at solving the problem of credit risk fraud. Not only are malicious fraudulent lending activities identified, but also customers who are likely to have no repayment capability are rejected. The method is essentially a classification problem for the clients, the data distribution of the classification problem is extremely unbalanced due to the specificity of risk identification, and the income brought by a high-quality client is often far lower than the loss caused by a risk client, so that the identification capability of the model is more required. In addition, in an actual risk business scene, the identification of risks often has interpretability requirements, and the model has very important significance for giving judgment basis when judging. In this context, the demand on risk identification models is increasing in order to achieve a fast and accurate loan approval process. The existing credit risk identification method is mainly used for building a calculation model based on a deep learning method and capturing the relationship among customers by building a customer relationship network. However, the credit risk identification method adopts a deep learning mode, and the variable parameters are hidden in the network, so that the identification result cannot be explained.
Based on the above, the embodiment of the application provides a credit risk identification method to realize the interpretation of the credit risk identification result.
Referring to fig. 1, fig. 1 is a flowchart illustrating a credit risk identification method according to an embodiment of the present application. As shown in fig. 1, the credit risk identification method provided in the embodiment of the present application is applied to a credit risk identification model, and the method includes:
and step S101, obtaining the customer information of a plurality of loan customers, and constructing a customer group information knowledge map and a customer vector of each loan customer based on the customer information of the plurality of loan customers.
In this step, the customer information may refer to user information of the loan customer, and the customer information is used as a basis for evaluating whether the customer loan is risky, wherein the customer may refer to the loan customer.
Exemplary, customer information includes, but is not limited to: personal information, liability information, credit history information, social relationship information, and historical loan information.
Personal information includes, but is not limited to: name, gender, ID card number, mobile phone number, school calendar, and home address.
Asset liability information includes, but is not limited to: the sum of cash withdrawal times in the last 12 months, monthly income and the total charge in the last 5 months.
The credit history information includes, but is not limited to: whether the identity card hits a p2p list, the loan application times of all institutions in the last 3 months, the loan application times of non-silver institutions in the last 12 months, the loan application times of non-silver institutions in the last 3 months and the non-silver institution number of loan application in the last 12 months.
Social relationship information includes, but is not limited to: the total number of the contact persons, the contact times of the service personnel in the last 3 months, the silent time length of the mobile phone and the service time length of the mobile phone.
Historical loan information includes, but is not limited to: loan amount, type of the cooperative merchant, merchant level, loan proportion. The type of the cooperative merchant can point to which banking institution applies for loan, if applying for loan to the industrial and commercial bank, the type of the cooperative merchant is the industrial and commercial bank, and the merchant level can point to the rating of the banking institution.
Each client corresponds to own client information, the client information comprises a plurality of characteristic variables, and information which plays an important role in credit risk evaluation can be selected from the client information as the characteristic variables, for example: and selecting the monthly income in the asset liability information from the customer information as a characteristic variable.
A customer base information knowledgegraph may refer to a knowledgegraph describing multiple customers from a credit risk perspective.
The customer group information knowledge graph is a graph which takes a single customer as a node and connects different nodes according to the relationship between the customers.
The customer base knowledge-graph will be described with reference to fig. 2.
Fig. 2 shows a schematic diagram of a customer group information knowledge-graph provided by an embodiment of the application.
As shown in fig. 2, each circle represents a node, each node represents a loan client, different nodes are connected by a line, and if two nodes are connected, the two nodes represent that a client association relationship exists between different clients corresponding to the two nodes. The black nodes indicate that the client corresponding to the nodes is a credit risk client, and the white nodes indicate that the client corresponding to the nodes is a non-credit risk client.
The customer vector can be a vector for representing the values of the characteristic variables of the customer, and the customer vector is used for representing the values of a plurality of characteristic variables corresponding to the loan customer.
For example, taking the ith customer as an example, the customer vector of the customer can be expressed as:
n i ＝[w i1 ,w i2 ,w i3 ,…,w in ]。
wherein i represents the customer number, w in And the characteristic variable code corresponding to the nth characteristic vector representing the client.
In the embodiment of the application, the client information of the loan client can be obtained from two channels, on one hand, the client information can be obtained from a loan application form filled by the loan client, and on the other hand, the client information can be obtained from the historical loan client information of a bank. And after the customer information is obtained, numbering is carried out on each customer, a customer vector corresponding to each customer is constructed for each customer, and a customer group information knowledge graph of the customers is constructed.
In an alternative embodiment, the customer information includes credit risk customer identification, and the customer group information knowledge map and the customer vector of each loan customer are constructed based on the customer information of a plurality of loan customers, including: for each loan client, coding the value of each characteristic variable in the client information of the loan client by using the evidence weight to generate a characteristic variable code of the loan client; splicing a plurality of characteristic variable codes corresponding to different characteristic variables to generate a customer vector of the loan customer; extracting customer association relations among different loan customers from customer information of the loan customers, wherein the customer association relations comprise call contact relations among the different customers, common guarantor relations, common identification number relations and common loan merchant relations; each loan client corresponds to a node, and different nodes with client association relation are connected by a non-directional connecting line; and based on the credit risk customer identification, distinguishing and displaying the nodes corresponding to the credit risk customers and the nodes corresponding to the non-credit risk customers to obtain the customer group information knowledge graph.
Here, the Evidence Weight may refer to WOE (Weight of Evidence), also known as variable encoding.
The characteristic variables may refer to variables in the customer information that play an important role in credit risk assessment. The characteristic variables are variables that describe the credit risk of the customer.
The characteristic variable encoding may refer to WOE encoding of the value of the characteristic variable.
The customer vector can refer to a vector formed by combining multiple WOE codes corresponding to characteristic variable values of a single customer.
A call contact relationship may refer to a relationship formed as a result of a call being made between two clients.
A common guarantor relationship may refer to a relationship formed as a result of a common guarantor existing between two customers.
The common identification number relationship may refer to a relationship formed due to the presence of a common identification number in different loan applications of two customers.
A common loan merchant relationship may refer to a relationship that results from two customers applying for loans from the same merchant.
Specifically, first, feature variables are selected from the customer information, and all of the 3 variables in the property and debt information, the 5 variables in the credit history information, the 4 variables in the social relationship information, and the 4 variables in the history loan information listed in the above example are taken as the feature variables, and 16 feature variables in total are selected. Taking the customer a as an example, WOE encoding is respectively performed on the values of the 16 characteristic variables of the customer a by using the evidence weight, 16 characteristic variable codes can be obtained, and the 16 characteristic variable codes are spliced together according to a fixed sequence to generate a customer vector a of the customer a.
Further, if 4 types of client associations are extracted from the client information for each client, each client is regarded as one node, and if the client and another client have any one of the 4 types of client associations, the node corresponding to the client and the node corresponding to another client having a client association with each other are connected by a line, and if the client and another client have a plurality of client associations, the nodes corresponding to both are connected by a plurality of lines, for example: when there are 3 relations between two clients, the nodes corresponding to the two clients are connected by using 3 lines, and the lines connecting the nodes are indicated in a non-directional manner.
Furthermore, if a client has fraud in the past loan process, the client is marked as a historical fraud client, and a credit risk client identifier is added to the client information of the client to mark the historical fraud of the client. When a customer group information knowledge spectrogram is constructed, if credit risk customer identification exists in customer information of a certain customer, a node corresponding to the customer is displayed in a special color to be distinguished from an ordinary customer, for example: the orange node indicates that the customer is a historical fraudulent customer, and the blue node indicates that the customer is a normal customer.
And step S102, calculating a customer vector corresponding to each loan customer by using a self-attention mechanism to determine a feature transformation vector, and transforming the feature transformation vector by using a first model parameter matrix with unknown values to obtain a transformed feature transformation vector.
In this step, the self-attention mechanism may be referred to as self-attention mechanism, and the self-attention mechanism is used to determine the relationship between different vectors, i.e. calculate the degree of association or similarity between two vectors.
The feature transformation vector may refer to a collection of a plurality of variable feature vectors corresponding to a single loan customer, and the feature transformation vector is used for representing the similarity between a plurality of feature variables corresponding to the loan customer.
The first model parameter matrix can refer to a matrix with unknown values of parameters, and is used for carrying out space transformation on the feature transformation vector so as to increase the learning capacity of the whole credit risk identification model, so that different nodes can be distinguished in a more complex feature space.
It should be noted that the first model parameter matrix is a matrix of N × M dimensions, where the size of N is determined by the number of feature variables, and the size of M is determined by the dimension of the output vector, i.e., the transformed spatial dimension.
In the embodiment of the application, for each loan client, the characteristic conversion vector of the loan client is obtained by calculating the client vector of the loan client by using the self-attention mechanism, and the characteristic conversion vector b of the ith loan client is used i For example, b i ＝[b i1 ,b i2 ,...,b ic ]Wherein c represents the number of characteristic variables.
The determination process of the feature transformation vector is described below with reference to fig. 3.
Fig. 3 shows a schematic structural diagram of a credit risk identification model provided by an embodiment of the application.
As shown in fig. 3, the credit risk identification model 200 includes: WOE encoding 211, a third linearity module 212, a self-attention model 213, a first linearity model 214, a fourth linearity model 215, a map breadth attention model 216, a map depth attention model 217, a second linearity model 218, a multi-feature fusion module 219, a fifth linearity model 220, and a loss function 221.
In FIG. 3, x 1 Multiple values, x, of different characteristic variables corresponding to the customer with number 1 2 A plurality of values of different characteristic variables corresponding to the number 2 customer are shown, and by analogy, the values of the plurality of characteristic variables of all the customers are input into a WOE coding module to be coded to obtain a customer vector corresponding to each customer, and the customer vector corresponding to the number 1 customer is n 1 And the customer vector corresponding to the number 2 customer is n 2 And analogizing in sequence, then inputting a plurality of customer vectors corresponding to different customers into a third linear model for processing to respectively obtain converted customer vectors n 'corresponding to the customers with the number of 1' 1 Number 2 customer corresponding converted customer vector n' 2 And analogizing to obtain the converted customer vector corresponding to each customer.
Inputting the converted customer vector corresponding to each customer into a self-attention model, calculating the customer vectors corresponding to different loan customers by using a self-attention mechanism, determining a feature conversion vector corresponding to each customer, inputting the feature conversion vector corresponding to each customer into a first linear model, converting the feature conversion vector by using a first model parameter matrix in the first linear model to obtain a converted feature conversion vector, and recording the converted feature conversion vector as B i ，B i ＝[b 1 ,b 2 ,...,b n ]Where i represents the number of the loan clients and n represents the number of feature vectors for a single loan client.
The first linear model and the third linear model may both be linear models, and both the first linear model and the third linear model are used for performing spatial transformation on the input parameters to obtain the output parameters.
The value of each parameter in the first model parameter matrix and the third model parameter matrix is unknown, and the dimension of the third model parameter matrix corresponds to the first model parameter matrix and is also determined by the number of the characteristic variables and the dimension of the output vector.
In an alternative embodiment, the calculating the corresponding customer vector of the loan customer using the self-attention mechanism to determine the feature transformation vector comprises: determining the product of the characteristic variable code and the self-attention model parameter matrix as a weight coefficient aiming at each characteristic variable code of the loan client; determining a variable characteristic vector corresponding to the characteristic variable code based on the weight coefficient; and combining a plurality of variable feature vectors corresponding to different feature variable codes of the loan clients to generate a feature conversion vector of the loan clients.
Here, the feature transformation vector may refer to a set of a plurality of variable feature vectors corresponding to a single customer, and the feature transformation vector is used to characterize similarity between a plurality of feature variables corresponding to a single customer.
The self-attention model parameter matrix may refer to a model parameter matrix in a self-attention model, which is a parameter matrix with unknown parameter values.
The self-attention model parameter matrix comprises a self-attention model parameter matrix Q and a self-attention model parameter matrix K, and the self-attention model parameter matrix Q is recorded as: w q The self-attention model parameter matrix K is written as: w k 。
The weight coefficient corresponding to the parameter matrix Q of the self-attention model is recorded as Q d The weight coefficient corresponding to the self-attention model parameter matrix K is denoted as K d 。
Specifically, the weight coefficient q d And a weight coefficient k d Can be calculated by the following formula:
q d ＝W q ×w id ；
k d ＝W k ×w id ；
after the weight coefficient is determined, the variable feature vector b is calculated by the following calculation formula of the self-attention mechanism id ：
∝ id ＝q d ·k d 。
In the above formula, b id The d characteristic variable code of the ith client corresponds to the variable characteristic vector, w id D characteristic variable code, W, representing the i-th client q And W k For a self-attention model parameter matrix, W, whose values are unknown q And W k Both of which will be w id Mapping to different linear spaces to reflect the difference between the corresponding feature vectors of different customers, q d For matching other characteristic variable codes, k d Used for being matched by other characteristic variable codes; c represents the total number of characteristic variables.
Thus, will beVariable characteristic vector b corresponding to first characteristic variable of i clients i1 And a variable feature vector b corresponding to the second feature variable of the ith client i2 And combining the variable feature vectors corresponding to other feature variables of the ith customer to obtain a feature conversion vector b of the ith customer i ＝[b i1 ,b i2 ,...,b ic ]. Since the self-attention mechanism belongs to the prior art, the description is omitted here.
And step S103, processing the nodes corresponding to the loan clients in the customer group information knowledge graph by using the graph breadth attention model to determine the influence vectors corresponding to the loan clients, and calculating the propagation chain sequence corresponding to the loan clients by using the graph depth attention model and the influence vectors to determine the propagation vectors corresponding to the loan clients.
In this step, the graph breadth attention model may refer to an attention network model, and is used to capture the influence degree of different neighbors of the current node on the graph breadth attention model.
Illustratively, the map breadth Attention model may be a GAT (Graph Attention Network) model.
The graph depth attention model may refer to a deep learning model that is used to capture the importance of different nodes on a risk propagation chain.
Illustratively, the map depth attention model may be a BILSTM model.
The propagation chain sequence may refer to a sequence formed by all propagation chains in the customer base information knowledge graph, wherein nodes corresponding to a single customer have a connection relationship, and the propagation chain sequence is used for representing a plurality of propagation chains having an association relationship with the customer.
The influence vector is used to characterize the degree of influence of different neighbors on the loan customer.
The propagation vector is the vector obtained after the vector representation of the loan client is propagated by the neighbors, and the vector representation is the client vector.
In the embodiment of the present application, as shown in fig. 3, a client vector corresponding to each client is input to the fourth linear model 215, the fourth linear model 215 includes a fourth model parameter matrix, the fourth model parameter matrix is also a matrix with unknown parameter values, a plurality of client vectors corresponding to different clients are multiplied by the fourth model parameter matrix, and the multiplication result is input to the graph breadth attention model 216.
The dimension of the fourth model parameter matrix is also determined by the number of feature variables and the dimension of the output vector.
At the same time, according to the customer group information knowledge graph, all neighbors of the customer are determined for each customer, namely all connecting lines with the node corresponding to the customer as an end point are found in the customer group information knowledge graph, the other end point of each connecting line in the connecting lines is the neighbor of the customer, the node number corresponding to the customer and the neighbor number form side information, and the side information e corresponding to each customer is used i Input to the map breadth attention model 216. Taking the client with the number 1 as an example, the client is connected to the client with the number 3, the client with the number 6 and the client with the number 9 respectively, and the side information corresponding to the client with the number 1 is e 1 ＝[131619]。
Through the steps, the information of the characteristic variables corresponding to each customer and the incidence relation information of the customer and other customers are input into the graph breadth attention model, so that the influence vector corresponding to each customer is determined by the graph breadth attention model.
The influence vector corresponding to a certain node is calculated by using a graph breadth attention model (GAT model) through the following formula:
in the above formula, h i ′ Representing the influence vector corresponding to the node i, and σ represents a nonlinear activation function, wherein here, the nonlinear activation function may be any one of the following functions: sigmod, tanh, reLU, ELU, and PReLU, W represents a graph breadth model parameter matrix, which participates in parameter matrixThe value of the number is unknown, h i Representing customer vector n corresponding to node i i ，h j Customer vector n corresponding to neighbor node j representing node i j ，h i ·h j Represents h i And h j Dot product calculation of the two matrices.
In an optional embodiment, after processing the node corresponding to the loan client in the knowledge graph of the information about the loan client by using the attention model to determine the influence vector corresponding to the loan client, the method further comprises: and in the customer group information knowledge graph, determining a propagation chain sequence corresponding to the loan customer according to the connection relation between the node corresponding to the loan customer and other nodes.
Specifically, for example, a customer with the number 1 is assumed to be connected to customers with the numbers 3, 4 and 9 in the customer group knowledge base, that is, the neighbors of the customer with the number 1 are the customer with the number 3, the customer with the number 4 and the customer with the number 9. The client number 3 is directly connected to the client number 13 and the client number 130, respectively, the client number 4 is directly connected to the client number 14, and the client number 9 is directly connected to the client number 19 and the client number 190, respectively, and the sequence of the numbers of the propagation chains corresponding to the client number 3 is [313, 3130 ]]And the number sequence of the propagation chain corresponding to the number 4 client is [414 ]]The propagation chain corresponding to client number 9 has the number sequence of [919, 9190]. From the three propagation chains and the propagation chain [13, 14, 19 ] of client number 1]Generating a propagation chain sequence S corresponding to the client with the number 1 1 ＝[n 13 ,n 14 ,n 19 ,n 313 ,n 3130 ,n 414 ,n 919 ,n 9190 ]. Here, the number of propagation layers takes 3 layers, i.e., a neighbor node is also included in the sequence of propagation chains when the client of number 9 also has the neighbor node.
In an alternative embodiment, the calculating the propagation chain sequence corresponding to the loan customer by using the map depth attention model and the influence vector to determine the propagation vector corresponding to the loan customer comprises: taking the influence vector as an input parameter of the image depth attention model; calculating a propagation chain sequence by using the image depth attention model after the parameters are input, and respectively obtaining a right propagation vector and a left propagation vector; the right-hand propagation vector and the left-hand propagation vector are combined to form the propagation vector of the loan client.
Taking the above example as an example, the spreading chain sequence of client number i is S i Sequence S is aligned using the BILSTM model i Carrying out information transmission calculation to obtain the transmitted vector, namely the transmission vector H i . The propagation vector can be calculated using the following formula:
H i ＝(h i1 ,h i2 ,…,h ik )；
in the above formula, H represents a vector of a client vector of a certain node after propagation through a neighbor node, k represents the number of neighbors of the node, and w t In order to take the depth model parameter matrix whose values are unknown,
h t-1 ，/>
h t+1 ，h t is the hidden state of BILSTM. The depth model parameter matrix may refer to a parameter matrix with unknown parameter values used in the map depth attention model.
The propagation vectors corresponding to the propagation chain sequences obtained by using the BILSTM model belong to the prior art, and are not described herein again.
And step S104, constructing a credit risk function corresponding to the loan client based on the converted feature conversion vector and the propagation vector.
The credit risk function may refer to a function for determining credit risk for all customers.
In an optional embodiment, constructing the credit risk function corresponding to the loan client based on the converted feature conversion vector and the propagation vector comprises: converting the propagation vector by using a second model parameter matrix with unknown values to obtain a converted propagation vector; combining the converted feature conversion vector and the converted propagation vector to generate a node representation vector; and constructing a credit risk function by taking the node representation vector as a parameter of the activation function.
Here, the second model parameter matrix may refer to a model parameter matrix in the second linear model, and the second model parameter matrix is a multidimensional matrix, and the dimension is determined by the dimension of the input and output vectors of the second linear model.
An activation function may refer to a Singmoid function.
In FIG. 3, propagation vector H is divided i Inputting the data into a second linear model, wherein the second linear model comprises a second model parameter matrix, multiplying the second model parameter matrix and the propagation vector to obtain a converted propagation vector, and then converting the converted propagation vector into H i And B i Joint Generation node representation vector E i Then E is i The following equation:
E i ＝[B i ,H i ]＝[b i1 ,b i2 ,...,b in ,h i1 ,h i2 ,…,h ik ]。
representing a node as a vector E i Substituting the activation function results in a credit risk function, which is shown below:
P i ＝Sigmoid(W f ×E i +b f )；
in the above formula, P i Representing the probability of credit risk, W f And b f All are the parameters of the activation function model with unknown parameter values. Wherein, the credit risk probability can refer to the high and low probability of the credit risk of different loan clients, the higher probability indicates the higher risk of providing the loan to the loan client, and the lower probability indicates the lower risk of providing the loan to the loan client.
And step S105, constructing a target loss function based on credit risk functions corresponding to different loan clients, and calculating the value of the first model parameter matrix, the value of the depth model parameter matrix corresponding to the map depth attention model and the credit risk probability of the different loan clients when the output result of the target loss function is minimum.
In this step, the objective loss function may refer to a two-class cross entropy loss function, and the objective loss function is used for performing parameter optimization on the credit risk model.
And calculating the model parameters with unknown parameter values and the credit risk probability P through the target loss function.
In an alternative embodiment, constructing the target loss function based on the credit risk functions corresponding to different credit customers includes: and (4) taking credit risk functions corresponding to different loan clients as parameters of the two-classification cross entropy loss function to construct a target loss function.
In particular, different credit risk probabilities P may be assigned i Substituting the following calculation formula to obtain a target loss function:
in the above formula, the first and second carbon atoms are,
representing the result of the calculation of the target loss function, n representing the total number of nodes, λ | θ | 2 F And a parameter item of the prediction model with unknown parameter values is represented, y is a fraud label of the client, when the label corresponding to a certain client is fraud, y =1, otherwise, y =0.
Specifically, after the target loss function is constructed, the specific values of the parameters with unknown values are continuously adjusted by using a gradient descent algorithm, so that
Continuously gets smaller until->
No longer changing, i.e. taking the minimum valueStopping adjusting the values of the parameters, determining the values of the parameters as the final values of the parameters of the target loss function, wherein the final values of the parameters comprise the values of the first model parameter matrix and the depth model parameter matrix, and determining credit risk probabilities P of different customers i The value of (a).
And step S106, explaining credit risk probabilities of different loan clients by using the first model parameter matrix and the depth model parameter matrix.
In this step, the first model parameter matrix can explain the credit risk probability from the importance point of view of different characteristic variables.
The depth model parameter matrix can explain the credit risk probability from the perspective of the degree of influence of the propagation path.
In an optional embodiment, the first model parameter matrix and the depth model parameter matrix are used for explaining credit risk probabilities of different loan clients, and the method comprises the following steps: carrying out weight normalization processing on the first model parameter matrix and the depth model parameter matrix to respectively obtain a variable importance coefficient and a path importance coefficient; and (4) explaining the credit risk probability of different loan clients by using the variable importance coefficient and the path importance coefficient.
Specifically, the weight normalization processing is performed on the first model parameter matrix and the depth model parameter matrix with the determined values, so as to obtain a variable importance coefficient and a path importance coefficient respectively, and the calculation formula is as follows:
in the above formula, the first and second carbon atoms are,
variable importance indicating ith characteristic variableSex coefficient->
A path importance coefficient, representing a propagation path with the i-th client>
The i-th element, representing the first model parameter matrix, is evaluated>
Representing the ith element in the depth model parameter matrix.
When the weight normalization processing is carried out on the first model parameter matrix, firstly, the modulus operation is carried out on the parameter corresponding to each characteristic variable, then, the maximum value of all modulus operation results is selected, the quotient of the modulus operation result of the parameter corresponding to the ith characteristic variable and the maximum value is used as the univariate importance coefficient corresponding to the characteristic variable, in this way, the univariate importance coefficient corresponding to each characteristic variable can be determined, and the univariate importance coefficients corresponding to all the characteristic variables are combined together to form the variable importance coefficient.
Similarly, by adopting the weight normalization method of the variable importance coefficients, the path importance coefficients corresponding to different propagation paths of each client can be determined.
Compared with the credit risk identification method in the prior art, the method can obtain the customer vector and the customer group information knowledge spectrogram according to customer information, describe the characteristic vector of each customer and the relationship between different customers, determine the similarity between different characteristic variables of the same loan customer by using a self-attention mechanism to form a target characteristic conversion vector, simultaneously, utilize an attention model to mine the influence degree of different neighbors in the customer group information knowledge spectrogram on a current node and the importance degree of different risk propagation chains corresponding to the current node to form a propagation vector, use an activation function to process the target characteristic conversion vector and the propagation vector to form a credit risk function, and utilize the risk function to determine a first model parameter matrix and a depth model parameter matrix, wherein the first model parameter matrix can explain the identification result from the angle of the characteristic variables, and the depth model parameter matrix can explain the identification result from the angle of the propagation path, so that the identification result cannot be explained in the existing credit risk identification method is solved.
Based on the same inventive concept, a credit risk identification device corresponding to the credit risk identification method is further provided in the embodiment of the present application, and as the principle of solving the problem of the device in the embodiment of the present application is similar to the credit risk identification method in the embodiment of the present application, the implementation of the device may refer to the implementation of the method, and repeated details are not repeated.
Referring to fig. 4, fig. 4 is a schematic structural diagram of a credit risk identification apparatus according to an embodiment of the present application. As shown in fig. 4, the credit risk identification means 300 includes:
the information calculation module 301 is configured to obtain client information of a plurality of loan clients, construct a client group information knowledge graph and a client vector of each loan client based on the client information of the plurality of loan clients, where the client information includes a plurality of characteristic variables, and the client vector is used to represent values of the plurality of characteristic variables corresponding to the loan clients;
the first vector acquisition module 302 is configured to calculate, for each loan client, a client vector corresponding to the loan client by using a self-attention mechanism to determine a feature transformation vector, and convert the feature transformation vector by using a first model parameter matrix with an unknown value to obtain a converted feature transformation vector;
a second vector obtaining module 303, configured to process a node corresponding to the loan customer in the customer group information knowledge graph by using the map breadth attention model to determine an influence vector corresponding to the loan customer, and calculate a propagation chain sequence corresponding to the loan customer by using the map depth attention model and the influence vector to determine a propagation vector corresponding to the loan customer, where the influence vector is used to represent influence degrees of different neighbors on the loan customer, and the propagation vector is a vector obtained after the vector representation of the loan customer is propagated by the neighbors;
a risk function construction module 304, configured to construct a credit risk function corresponding to the loan client based on the converted feature conversion vector and the propagation vector;
the coefficient calculation module 305 is configured to construct a target loss function based on credit risk functions corresponding to different loan clients, and calculate values of a first model parameter matrix, values of a depth model parameter matrix corresponding to the map depth attention model, and credit risk probabilities of the different loan clients when an output result of the target loss function is a minimum value;
and the result interpretation module 306 is used for interpreting the credit risk probability of different loan clients by using the first model parameter matrix and the depth model parameter matrix.
The embodiment of the present application further provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the credit risk identification method in the method embodiment shown in fig. 1 may be executed.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
In the several embodiments provided in the present application, it should be understood that the disclosed system, apparatus and method may be implemented in other ways. The above-described embodiments of the apparatus are merely illustrative, and for example, the division of the units is only one logical division, and there may be other divisions when actually implemented, and for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection of devices or units through some communication interfaces, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit.
The functions, if implemented in software functional units and sold or used as a stand-alone product, may be stored in a non-transitory computer-readable storage medium executable by a processor. Based on such understanding, the technical solution of the present application or portions thereof that substantially contribute to the prior art may be embodied in the form of a software product stored in a storage medium and including instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.
Finally, it should be noted that: the above-mentioned embodiments are only specific embodiments of the present application, and are used for illustrating the technical solutions of the present application, but not limiting the same, and the scope of the present application is not limited thereto, and although the present application is described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: any person skilled in the art can modify or easily conceive the technical solutions described in the foregoing embodiments or equivalent substitutes for some technical features within the technical scope disclosed in the present application; such modifications, changes or substitutions do not depart from the spirit and scope of the exemplary embodiments of the present application, and are intended to be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (10)
1. A credit risk identification method, comprising:
the method comprises the steps of obtaining customer information of a plurality of loan customers, and constructing a customer group information knowledge graph and a customer vector of each loan customer based on the customer information of the plurality of loan customers, wherein the customer information comprises a plurality of characteristic variables, and the customer vector is used for representing values of the plurality of characteristic variables corresponding to the loan customers;
aiming at each loan customer, calculating a customer vector corresponding to the loan customer by using a self-attention mechanism to determine a feature transformation vector, and transforming the feature transformation vector by using a first model parameter matrix with unknown values to obtain a transformed feature transformation vector;
processing a node corresponding to the loan client in a client information knowledge graph by using a graph breadth attention model to determine an influence vector corresponding to the loan client, and calculating a propagation chain sequence corresponding to the loan client by using a graph depth attention model and the influence vector to determine a propagation vector corresponding to the loan client, wherein the influence vector is used for representing the influence degree of different neighbors on the loan client, and the propagation vector is obtained after the vector of the loan client represents the propagation of the neighbors;
constructing a credit risk function corresponding to the loan client based on the converted feature conversion vector and the propagation vector;
constructing a target loss function based on credit risk functions corresponding to different loan clients, and calculating the value of a first model parameter matrix, the value of a depth model parameter matrix corresponding to an image depth attention model and credit risk probabilities of the different loan clients when the output result of the target loss function is the minimum value;
and explaining the credit risk probability of different loan clients by using the first model parameter matrix and the depth model parameter matrix.
2. The method of claim 1, wherein the customer information includes credit risk customer identification;
the constructing of the customer group information knowledge graph and the customer vector of each loan customer based on the customer information of the plurality of loan customers comprises the following steps:
for each loan client, coding the value of each characteristic variable in the client information of the loan client by using the evidence weight to generate a characteristic variable code of the loan client;
splicing a plurality of characteristic variable codes corresponding to different characteristic variables to generate a customer vector of the loan customer;
extracting customer association relations among different loan customers from the customer information of the plurality of loan customers, wherein the customer association relations comprise call contact relations among the different customers, common guarantor relations, common identification number relations and common loan merchant relations;
each loan client corresponds to a node, and different nodes with client association relation are connected by a non-directional connecting line;
and based on the credit risk customer identification, distinguishing and displaying the nodes corresponding to the credit risk customers and the nodes corresponding to the non-credit risk customers to obtain a customer group information knowledge graph.
3. The method of claim 2, wherein the determining the feature transformation vector by computing the customer vector corresponding to the loan customer using a self-attention mechanism comprises:
determining the product of the characteristic variable code and the self-attention model parameter matrix as a weight coefficient aiming at each characteristic variable code of the loan client;
determining a variable feature vector corresponding to the feature variable code based on the weight coefficient;
and combining a plurality of variable feature vectors corresponding to different feature variable codes of the loan client to generate a feature conversion vector of the loan client.
4. The method according to claim 1, further comprising, after processing the node corresponding to the loan client in the knowledge-base of information of the loan client by using the graph breadth attention model to determine the influence vector corresponding to the loan client:
and in the customer group information knowledge graph, determining a propagation chain sequence corresponding to the loan customer according to the connection relation between the node corresponding to the loan customer and other nodes.
5. The method according to claim 1, wherein the determining the propagation vector corresponding to the loan customer by calculating the propagation chain sequence corresponding to the loan customer using the graph depth attention model and the influence vector comprises:
taking the influence vector as an input parameter of a map depth attention model;
calculating the propagation chain sequence by using the image depth attention model after the parameters are input, and respectively obtaining a right propagation vector and a left propagation vector;
and combining the right propagation vector and the left propagation vector to form the propagation vector of the loan client.
6. The method according to claim 1, wherein constructing the credit risk function corresponding to the loan client based on the converted feature conversion vector and the propagation vector comprises:
converting the propagation vector by using a second model parameter matrix with unknown values to obtain a converted propagation vector;
combining the converted feature conversion vector and the converted propagation vector to generate a node representation vector;
and constructing a credit risk function by taking the node representation vector as a parameter of the activation function.
7. The method according to claim 1, wherein the constructing the target loss function based on the credit risk functions corresponding to different lenders comprises:
and (4) taking credit risk functions corresponding to different loan clients as parameters of the two-classification cross entropy loss function to construct a target loss function.
8. The method according to claim 1, wherein the interpreting the credit risk probability of different loan clients by using the first model parameter matrix and the depth model parameter matrix comprises:
carrying out weight normalization processing on the first model parameter matrix and the depth model parameter matrix to respectively obtain a variable importance coefficient and a path importance coefficient;
and (4) explaining the credit risk probability of different loan clients by using the variable importance coefficient and the path importance coefficient.
9. A credit risk identification device, comprising:
the information calculation module is used for acquiring the customer information of a plurality of loan customers, constructing a customer group information knowledge graph and a customer vector of each loan customer based on the customer information of the plurality of loan customers, wherein the customer information comprises a plurality of characteristic variables, and the customer vector is used for representing the values of the plurality of characteristic variables corresponding to the loan customers;
the system comprises a first vector acquisition module, a second vector acquisition module and a third vector acquisition module, wherein the first vector acquisition module is used for calculating a customer vector corresponding to each loan customer by using a self-attention mechanism to determine a feature conversion vector, and converting the feature conversion vector by using a first model parameter matrix with unknown values to obtain a converted feature conversion vector;
the second vector acquisition module is used for processing the nodes corresponding to the loan clients in the information knowledge graph of the loan clients by using the graph breadth attention model to determine the influence vectors corresponding to the loan clients, calculating the propagation chain sequence corresponding to the loan clients by using the graph depth attention model and the influence vectors to determine the propagation vectors corresponding to the loan clients, wherein the influence vectors are used for representing the influence degrees of different neighbors on the loan clients, and the propagation vectors are vectors obtained after the vector representation of the loan clients is propagated by the neighbors;
the risk function building module is used for building a credit risk function corresponding to the loan client based on the converted feature conversion vector and the propagation vector;
the coefficient calculation module is used for constructing a target loss function based on credit risk functions corresponding to different loan clients, and calculating the value of a first model parameter matrix, the value of a depth model parameter matrix corresponding to the map depth attention model and the credit risk probabilities of the different loan clients when the output result of the target loss function is the minimum value;
and the result interpretation module is used for interpreting the credit risk probability of different loan clients by utilizing the first model parameter matrix and the depth model parameter matrix.
10. A computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, which computer program, when being executed by a processor, carries out the steps of the credit risk identification method according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211466157.4A CN115860906A (en) | 2022-11-22 | 2022-11-22 | Credit risk identification method, credit risk identification device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211466157.4A CN115860906A (en) | 2022-11-22 | 2022-11-22 | Credit risk identification method, credit risk identification device and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115860906A true CN115860906A (en) | 2023-03-28 |
Family
ID=85664913
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211466157.4A Pending CN115860906A (en) | 2022-11-22 | 2022-11-22 | Credit risk identification method, credit risk identification device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115860906A (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112597775A (en) * | 2020-12-16 | 2021-04-02 | 中国工商银行股份有限公司 | Credit risk prediction method and device |
| US20220076101A1 (en) * | 2020-09-04 | 2022-03-10 | Alipay (Hangzhou) Information Technology Co., Ltd. | Object feature information acquisition, classification, and information pushing methods and apparatuses |
| CN114255121A (en) * | 2021-12-27 | 2022-03-29 | 中国农业银行股份有限公司 | Credit risk prediction model training method and credit risk prediction method |
| WO2022076234A1 (en) * | 2020-10-06 | 2022-04-14 | Visa International Service Association | Detecting adversarial examples using latent neighborhood graphs |
| US20220138502A1 (en) * | 2020-08-25 | 2022-05-05 | Alipay (Hangzhou) Information Technology Co., Ltd. | Graph neural network training methods and systems |
| CN114662570A (en) * | 2022-03-04 | 2022-06-24 | 中国建设银行股份有限公司 | Business risk prediction method and device, computer equipment and storage medium |
| CN115204457A (en) * | 2022-05-30 | 2022-10-18 | 浙大城市学院 | Loan default risk prediction method based on graph attention network |
| CN115345727A (en) * | 2022-08-19 | 2022-11-15 | 中电金信软件有限公司 | Method and device for identifying fraud loan application |
-
2022
- 2022-11-22 CN CN202211466157.4A patent/CN115860906A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220138502A1 (en) * | 2020-08-25 | 2022-05-05 | Alipay (Hangzhou) Information Technology Co., Ltd. | Graph neural network training methods and systems |
| US20220076101A1 (en) * | 2020-09-04 | 2022-03-10 | Alipay (Hangzhou) Information Technology Co., Ltd. | Object feature information acquisition, classification, and information pushing methods and apparatuses |
| WO2022076234A1 (en) * | 2020-10-06 | 2022-04-14 | Visa International Service Association | Detecting adversarial examples using latent neighborhood graphs |
| CN112597775A (en) * | 2020-12-16 | 2021-04-02 | 中国工商银行股份有限公司 | Credit risk prediction method and device |
| CN114255121A (en) * | 2021-12-27 | 2022-03-29 | 中国农业银行股份有限公司 | Credit risk prediction model training method and credit risk prediction method |
| CN114662570A (en) * | 2022-03-04 | 2022-06-24 | 中国建设银行股份有限公司 | Business risk prediction method and device, computer equipment and storage medium |
| CN115204457A (en) * | 2022-05-30 | 2022-10-18 | 浙大城市学院 | Loan default risk prediction method based on graph attention network |
| CN115345727A (en) * | 2022-08-19 | 2022-11-15 | 中电金信软件有限公司 | Method and device for identifying fraud loan application |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Arundina et al. | The predictive accuracy of Sukuk ratings; Multinomial Logistic and Neural Network inferences | |
| Chi et al. | A hybrid approach to integrate genetic algorithm into dual scoring model in enhancing the performance of credit scoring model | |
| CN111260462B (en) | Transaction fraud detection method based on heterogeneous relation network attention mechanism | |
| US11263644B2 (en) | Systems and methods for detecting unauthorized or suspicious financial activity | |
| Van Thiel et al. | Artificial intelligence credit risk prediction: An empirical study of analytical artificial intelligence tools for credit risk prediction in a digital era | |
| Muslim et al. | New model combination meta-learner to improve accuracy prediction P2P lending with stacking ensemble learning | |
| Esmaeilzadeh et al. | Individuals’ cryptocurrency adoption: A proposed moderated-mediation model | |
| Mohanty et al. | Predicting fluctuations in cryptocurrencies' price using users' comments and real-time prices | |
| Klaas | Machine learning for finance: principles and practice for financial insiders | |
| CN115170319A (en) | Abnormal account detection method, and method and device for constructing graph neural network model | |
| Liu et al. | An innovative model fusion algorithm to improve the recall rate of peer-to-peer lending default customers | |
| CN115860906A (en) | Credit risk identification method, credit risk identification device and storage medium | |
| Sudrajat et al. | Factors Affecting the Use of Financial Technology Payment Systems (Case Study on ShopeePay Digital Wallet Users in Madiun City) | |
| CN115345727A (en) | Method and device for identifying fraud loan application | |
| CN115063143A (en) | Account data processing method and device, computer equipment and storage medium | |
| Gelindon et al. | Credit Risk Evaluation in Banking and Lending Sectors Using Neural Network Model | |
| Zou et al. | A STUDY OF NON-PERFORMING LOAN BEHAVIOUR IN P2P LENDING UNDER ASYMMETRIC INFORMATION. | |
| Tselekidou | A Machine Learning Approach for Micro-Credit Scoring and Limit Optimization | |
| Bundi | Risk Concentration in Networks of Banks Connected by Financial Contracts | |
| Lu et al. | Multi-layer and parallel-connected graph convolutional networks for detecting debt default in P2P networks | |
| CN113344587B (en) | Data grade determining method and device, electronic equipment and storage medium | |
| CN112967134B (en) | Network training method, risk user identification method, device, equipment and medium | |
| Çetin | The impact of ICT patents on OECD countries' banks risk indicators and discussion on the use of robotic communication and smart glasses in the banking sector | |
| CN114757723B (en) | Data analysis model construction system and method for resource element trading platform | |
| Sen et al. | The Latest Trend in India Cryptocurrency |
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 |