A method, system and medium for identifying second-hand car financing guarantee information
By constructing a relational knowledge graph and a graph convolutional neural network, the problems of information loss and sensitivity in feature selection in the identification of used car financing guarantee fraud are solved, achieving fast and accurate fraud identification, reducing implementation costs, and improving risk avoidance capabilities.
Patent Information
- Application Number
- CN202210353298.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-04-06
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2042-04-06
AI Technical Summary
Existing used car financing guarantee fraud detection technologies frequently suffer from information gaps when dealing with new users, feature selection is sensitive and implementation costs are high, rule maintenance is difficult, and it is hard to achieve fast and accurate risk identification.
A fraud detection model for loan applications is constructed using relational knowledge graphs and graph convolutional neural networks. By learning the vector representations of graph nodes, the similarity of graph nodes is used to identify fraudulent loan applications.
It reduces reliance on feature design, improves the accuracy and ease of implementation of identification, lowers the technical requirements for practitioners, achieves low-cost fraud identification, and enhances the risk avoidance capabilities of financial institutions.
Smart Images

Figure CN114757760B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of financial information and data processing, and in particular relates to a method, system, medium, equipment and terminal for identifying second-hand car financing guarantee information. Background Art
[0002] At present, in the used car financing guarantee industry, guarantee agencies provide loan assistance services to banks and guarantee services to customers by collecting basic customer information, loan product information, vehicle information, third-party credit data, customer emergency contact information, and guarantor information. In this process, the guarantee agency assumes the responsibility of overdue repayment for the customer, that is, if the customer commits fraudulent loan fraud, it will cause significant asset losses to the guarantee agency. Therefore, before issuing a loan, it is of great significance to accurately identify these fraudulent behaviors, strengthen approval measures for loans suspected of fraud, or refuse to issue loans, which can reduce the level of non-performing loan assets and improve the risk management level of guarantee agencies.
[0003] Risk prevention in used car finance mainly focuses on solving the problem of fraud and loan fraud. It is necessary to identify malicious loan fraud and eliminate customers who are likely to have no repayment ability. It is essentially a customer classification problem. At the same time, due to the particularity of risk identification, the data distribution of this classification problem is extremely unbalanced, and the benefits brought by a high-quality customer are often much lower than the losses caused by a risky customer. Therefore, there are higher requirements for the recognition ability of the model. In addition, in actual risk business scenarios, there is often a requirement for interpretability in the identification of risks. Whether the model can provide a basis for judgment when making judgments is of great significance. In this context, in order to achieve a fast and accurate loan approval process, the requirements for risk identification models are getting higher and higher.
[0004] There are two main technologies for identifying fraud in used car financing guarantees. One is an analysis method based on feature engineering. Its main technical solution is to collect comprehensive information data of applicants, conduct feature exploration and feature design on the data, learn the features using machine learning methods such as logistic regression, and then use the learned model to calculate the credit score of each user, and set a specific score to identify fraudulent users. However, since this method is overly dependent on the designed features and feature analysis methods, it has high business and technical requirements for practitioners, which makes it difficult to implement such methods and the effectiveness is difficult to guarantee. Another method is a rule-based method. Its main technical solution is to screen applicants through pre-set rules and identify fraudulent users, such as blacklist rules, and define users appearing in the blacklist as fraudulent users. However, this method is completely dependent on the set rules, requires real-time update and maintenance of a large number of rule databases, and requires a lot of manpower, resulting in the problems of difficult rule data maintenance, difficult real-time update, and high implementation costs.
[0005] Through the above analysis, the problems and defects of the prior art are as follows:
[0006] (1) Existing technologies for identifying used car financing guarantee fraud often encounter the problem of missing information when facing new users, and are unable to conduct accurate assessments. At the same time, such methods are very sensitive to the selection of features and have very high business and technical requirements.
[0007] (2) Existing technologies for identifying used car financing guarantee fraud require real-time updating and maintenance of a large number of rule databases, which are difficult to maintain, difficult to update in real time, and have high implementation costs.
[0008] The difficulty of solving the above problems and defects is: reducing the impact of feature design on model effects and designing new model methods for data information extraction. Summary of the invention
[0009] In view of the problems existing in the prior art, the present invention provides a method, system, medium, device and terminal for identifying second-hand car financing guarantee information.
[0010] The present invention is implemented as follows: a method for identifying used car financing guarantee information is implemented by performing an ontology design of a relational knowledge graph based on the collected basic information data, relational data and vehicle information data of the loan applicant, constructing a relational knowledge graph, and using a graph convolutional neural network to establish a vector representation learning model for graph nodes in the relational knowledge graph. Fraud identification rules for loan applications are defined based on the vector representation of graph nodes to identify whether the loan application is a fraudulent application.
[0011] Furthermore, the basic information data of the loan applicant includes: basic information provided by the applicant, credit information, historical information of the institution and other institutions, and a comprehensive data set for each applicant is constructed.
[0012] Furthermore, the construction of the relational knowledge graph uses the applicant's contact information as a graph node, and the applicant and other people or other organizations are connected by edges to indicate that the applicant has a relationship with them, while the user's basic information and other behavioral information are used as attribute data of the node.
[0013] Furthermore, the use of a graph convolutional neural network to establish a fraud identification model for loan applications to identify whether the loan application is fraudulent includes: based on the applicant's knowledge graph, using the applicant's basic information, behavioral information and other information to initialize the vector representation of the network nodes; using a graph convolution operator to perform nonlinear changes on the vector; using a defined loss function and a stochastic gradient descent method to supervise and constrain the changed vector to obtain a learned node vector representation; and calculating the similarity between the node vector and the graph vector.
[0014] Furthermore, the second-hand car financing guarantee information identification method comprises the following steps:
[0015] The first step is to collect the basic information, credit information, and historical information of the institution and other institutions provided by the applicant, and comprehensively build a comprehensive data set for each applicant;
[0016] The second step is to build a knowledge graph, using the applicant's contact information as a graph node. The applicant and other people or other institutions are connected by edges to indicate that the applicant has a relationship with them. At the same time, the user's basic information and other behavior information are used as the attribute data of the node;
[0017] The third step is to use the applicant’s basic information, behavior information and other information to initialize the network nodes as vectors based on the applicant’s knowledge graph; use the graph convolution operator to perform nonlinear changes on the vectors; use the defined loss function and stochastic gradient descent method to supervise and constrain the changed vectors to obtain the learned node vector representation;
[0018] The fourth step is to calculate the similarity between the node vector and the graph vector, and output the nodes with similarity less than 0.5 as fraudulent nodes.
[0019] Furthermore, the second step of constructing the knowledge graph specifically includes:
[0020] 1) Construct binary data of applicants and related information, and find related information based on each applicant to form binary data in the following format:
[0021] [[P, R1], [P, R2], …, [P, Rn]], where P represents the applicant, and R1, R2, …, Rn represent information related to the applicant P;
[0022] 2) Use the binary data from 1) to construct the applicant’s knowledge graph.
[0023] Further, the third step specifically includes:
[0024] 1) Initialize the graph node vector and select the valid information of the node as follows: customer monthly income, age, number of mobile phone numbers associated with the ID card, ethnicity, and information equivalent to whether the credit report is white household. Then, the valid information is initialized according to the formula
[0025] Normalize it and finally use formula a i =(δ i ,δ 2 ,…,δ m ) to get the vector initialization representation of the node; where δ j represents the normalized value of information j, x j represents the value of information j before normalization, min(·) represents the minimum value, max(·) represents the maximum value, n represents the total number of nodes in the relational knowledge graph, a i The vector initialization representation representing the node i of the relational knowledge graph, m represents the total number of data types of valid information;
[0026] 2) Use the graph convolution operator to perform nonlinear changes on the vector. The graph convolution operator definition formula is
[0027] As shown;
[0028] Where g(i) represents the vector representation of node i after transformation at layer l, a j represents the vector representation of the neighbor node, f(j) represents the conversion function, σ is the activation function of the operator, and c ij represents the degree of node i, N i Represents the neighbors of node i, including itself, represents the vector representation of node j at layer l, Represents the parameter matrix that the transformation operator needs to learn;
[0029] 3) Use the defined loss function and stochastic gradient descent method to supervise the changed vector; use the formula
[0030] and The loss function shown uses the gradient descent method to optimize the parameters of the model;
[0031] Where q represents the vector representation of the graph, The exponential transformation of the vector label representing the graph, g(i) and g(j) represent the vectorized representation of the fraudulent node i and the normal node j respectively, y i ,y j Represent the true label values of nodes i and j respectively.
[0032] Furthermore, the fourth step calculates the similarity between the node vector and the graph vector, and outputs nodes with a similarity less than 0.5 as fraudulent nodes. The similarity calculation formula is where s i represents the similarity between the node vector and the graph vector, q and g(i) represent the graph vector and the vector of node i respectively, ‖q‖ and ‖g(i)‖ represent the norm of the graph vector and the vector of node i respectively.
[0033] Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the used car financing guarantee information identification method.
[0034] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to execute the steps of the second-hand car financing guarantee information identification method.
[0035] Another object of the present invention is to provide an information data processing terminal, which is used to implement the second-hand car financing guarantee information identification method.
[0036] Another object of the present invention is to provide a second-hand car financing guarantee information identification system for implementing the second-hand car financing guarantee information identification method, the second-hand car financing guarantee information identification system comprising:
[0037] The data collection module is used to collect basic information provided by applicants, credit information, and historical information of the institution and other institutions, and comprehensively construct a comprehensive data set for each applicant;
[0038] The knowledge graph construction module is used to use the applicant's contact information as a graph node. The applicant and other people or other institutions are connected by edges to indicate that the applicant has a relationship with them. At the same time, the user's basic information and other behavior information are used as the attribute data of the node;
[0039] The node vectorization representation module is used to initialize the vector representation of network nodes based on the applicant's knowledge graph and the applicant's basic information, behavior information and other information; use the graph convolution operator to perform nonlinear changes on the vector; use the defined loss function and stochastic gradient descent method to supervise and constrain the changed vector to obtain the learned node vectorization representation;
[0040] The similarity calculation module is used to calculate the similarity between the node vector and the graph vector, and output the nodes with similarity less than 0.5 as fraudulent nodes.
[0041] Combining all the above technical solutions, the advantages and positive effects of the present invention are as follows:
[0042] The basic data of the applicant is used for normalization processing, the relationship network nodes are initialized, and on this basis, the relationship information of the nodes is learned using a graph convolutional neural network to obtain a vector representation of the basic node information and the relationship information. The vector representation of the relationship knowledge graph is calculated based on the vector representation of all graph nodes, and the fraudulent user is identified by calculating the similarity between the graph node vector and the graph vector. The method of identifying fraud in second-hand car financing guarantees based on graph computing proposed by the present invention uses the basic information of the user and the relationship information of the user for fraud identification at the same time, which reduces the dependence on basic information, thereby reducing the sensitivity to feature selection and improving the ease of implementation. At the same time, this method has higher accuracy, thereby helping financial institutions to identify fraudulent loan applications faster and more accurately. The present invention enables financial institutions to identify fraud faster and more accurately for applicants based on the model provided by the present invention and in combination with existing data, improve the ability of financial institutions to avoid fraud risks, promote financial institutions to establish a sound risk assessment system, and be conducive to the stability of the financial market.
[0043] The present invention solves the problems and defects of the current prior art and has the following technical significance:
[0044] First, it provides a new method for fraud identification using user relationship data in the field of used car financing guarantees.
[0045] Second, it reduces the impact of feature design on model effects and relatively lowers the technical requirements for practitioners.
[0046] Third, by using this method to identify fraud, it helps used car financing guarantee institutions to identify fraud in a low-cost manner and avoid economic losses caused by fraud.
[0047] In addition, the method of the present invention further has the following advantages and technical effects:
[0048] 1) The similarity obtained by the present invention is used as a threshold to distinguish risky users from ordinary users, which has the following advantages: the information type is simple and easy to distinguish. The data information is binary information, the information type is simple, and it is easy to identify and distinguish.
[0049] 2) Strong objectivity. The effective information is normalized, and all information is represented between (0,1). There is no problem of artificially increasing or decreasing the weight of certain information due to the numerical range of the information itself, which improves the objectivity of the model.
[0050] 3) Strong semantic representation capability. Graph convolution updates the original vector. Based on the contact itself, the contact representation is updated with the information of the people and organizations related to the contact, expanding the scope of information covered by the contact node itself and improving the semantic representation capability of the contact itself.
[0051] 4) The semantic effect is significant and the interpretability is strong. In the entire graph, fraud nodes are a small sample, and most nodes are normal nodes or nodes with good debt repayment ability. Therefore, the vector representation of the graph calculated in this way is also biased towards normal nodes. If the representation of the fraud node is used to calculate the similarity with the vector representation of the graph, the similarity score at this time will be very low. BRIEF DESCRIPTION OF THE DRAWINGS
[0052] Figure 1 It is a flow chart of a method for identifying second-hand car financing guarantee information provided by an embodiment of the present invention.
[0053] Figure 2 It is a structural diagram of a second-hand car financing guarantee information identification system provided by an embodiment of the present invention;
[0054] Figure 2 In: 1. Data collection module; 2. Knowledge graph construction module; 3. Node vectorization representation module; 4. Similarity calculation module.
[0055] Figure 3 It is a flow chart for implementing the method for identifying second-hand car financing guarantee information provided by an embodiment of the present invention.
[0056] Figure 4 It is a schematic diagram of a relational knowledge graph provided by an embodiment of the present invention.
[0057] Figure 5 It is a schematic diagram of a graph convolutional neural network provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0058] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0059] In view of the problems existing in the prior art, the present invention provides a method, system, medium, device and terminal for identifying second-hand car financing guarantee information. The present invention is described in detail below with reference to the accompanying drawings.
[0060] like Figure 1 As shown, the second-hand car financing guarantee information identification method provided by the present invention includes the following steps:
[0061] S101: Data collection: collect basic information, credit information, historical information of the institution and other institutions, etc. provided by the applicant, and comprehensively construct a comprehensive data set for each applicant.
[0062] S102: Construct a knowledge graph, use the applicant's contact information as a graph node, and connect the applicant and other people or other organizations with edges to indicate that the applicant has a relationship with them. At the same time, use the user's basic information and other behavioral information as attribute data of the node.
[0063] S103: Based on the applicant's knowledge graph, the basic information, behavior information and other information of the applicant are used to initialize the vector representation of the network nodes; the graph convolution operator is used to perform nonlinear changes on the vectors; the defined loss function and stochastic gradient descent method are used to supervise and constrain the changed vectors to obtain the learned node vector representation.
[0064] S104: Calculate the similarity between the node vector and the graph vector, and output nodes with similarity less than 0.5 as fraudulent nodes.
[0065] Ordinary technicians in the industry can also use other steps to implement the second-hand car financing guarantee information identification method provided by the present invention. Figure 1 The present invention provides only a specific embodiment.
[0066] like Figure 2 As shown, the second-hand car financing guarantee information identification system provided by the present invention includes:
[0067] The data collection module 1 is used to collect the basic information, credit information, historical information of the institution and other institutions, etc. provided by the applicant, and comprehensively construct a comprehensive data set for each applicant.
[0068] The knowledge graph construction module 2 is used to use the applicant's contact information as a graph node. The applicant and other people or other organizations are connected by edges to indicate that the applicant has a relationship with them. At the same time, the user's basic information and other behavioral information are used as attribute data of the node.
[0069] The node vectorization representation module 3 is used to initialize the vector representation of the network nodes based on the applicant's knowledge graph and the applicant's basic information, behavior information and other information; use the graph convolution operator to perform nonlinear changes on the vector; use the defined loss function and stochastic gradient descent method to supervise and constrain the changed vector to obtain the learned node vectorization representation;
[0070] The similarity calculation module 4 is used to calculate the similarity between the node vector and the graph vector, and output the nodes with similarity less than 0.5 as fraudulent nodes.
[0071] like Figure 2 As shown in the figure, basic user information, credit information, etc. are collected, and information about institutions is obtained at the same time. In this step, the authenticity, accuracy and completeness of the acquired data should be guaranteed as much as possible to ensure that the collected information can accurately reflect the user's ability to repay the loan and effectively reduce the risk. The knowledge graph construction module takes users and institutions as graph nodes, and sets the relationship between them as edges. In this way, a knowledge graph is constructed. Adjacent edges represent the relationship between users or between users and institutions. By using the relevant information of graph nodes and adjacent edges, not only can the information of the contact itself be simply obtained, but also the information of the person related to the contact can be associated with the institution, which increases the risk monitoring dimension and reduces the influence of the person related to the contact's ability to repay the loan. The node vectorization representation module uses the existing contact and institution graph by vectorizing the nodes. Since all features are manually selected, some artificial noise is inevitably introduced. The influence of artificial noise is weakened to a certain extent by updating the contact vector representation through vectorization and adjacent node vectors, and the influence of the person's social environment on his ability to repay the loan is introduced, making the information source of the model more comprehensive. Finally, the similarity calculation is performed. Since most people and institutions have good loan repayment capabilities, the vector representation of the final graph is also biased towards the semantics of good debt repayment ability. If a newly input node representation is calculated to have a similarity with the vector representation of the graph, if the similarity is greater than 0.5, it means that the node representation has a high similarity with the vector representation of the graph, which means that the new node is more likely to be a normal node, otherwise it is a fraud node.
[0072] like Figure 3 As shown, the second-hand car financing guarantee information identification method provided by the present invention specifically includes the following steps:
[0073] Step 1: Data collection: collect basic information provided by applicants, credit information, historical information of the institution and other institutions, etc., and comprehensively construct a comprehensive data set for each applicant.
[0074] Step 2: Build a knowledge graph such as Figure 4As shown, the applicant's contact information (telephone number) is used as a graph node, and the applicant and other people or other organizations are connected by edges, indicating that the applicant has a relationship with them. At the same time, the user's basic information and other behavior information are used as attribute data of the node.
[0075] 1) Construct binary data of applicants and related information, and find related information based on each applicant to form binary data in the following format:
[0076] [[P, R1], [P, R2], …, [P, Rn]], where P represents the applicant, and R1, R2, …, Rn represent information related to the applicant P;
[0077] 2) If Figure 4 As shown, the binary data in 1) is used to construct the applicant's knowledge graph.
[0078] Step 3: Based on the applicant’s knowledge graph, first use the applicant’s basic information, behavior information and other information to initialize the vector representation of the network nodes, then use the graph convolution operator to perform nonlinear changes on the vectors, and finally use the defined loss function and stochastic gradient descent method to supervise and constrain the changed vectors to obtain the learned node vector representation.
[0079] 1) Graph node vector initialization. First, select the valid information of the node as follows: customer monthly income, age, number of mobile phone numbers associated with the ID card, ethnicity, and information equivalent to whether the credit report is white household. Then, normalize the valid information according to formula (1). Finally, get the vector initialization representation of the node according to formula (2):
[0080]
[0081] a i =(δ i ,δ 2 ,…,δ m ) (2)
[0082] where δ j represents the normalized value of information j, x j represents the value of information j before normalization, min(·) represents the minimum value, max(·) represents the maximum value, n represents the total number of nodes in the relational knowledge graph, a i The vector initialization representation representing the node i of the relational knowledge graph, m represents the total number of data types of valid information.
[0083] 2) Use graph convolution operators to perform nonlinear changes on vectors, such as Figure 5 As shown. The graph convolution operator is defined as shown in formulas (3) and (4):
[0084] g(i)=∑ j∈Ni a j f(j) (3)
[0085]
[0086] Where g(i) represents the vector representation of node i after transformation at layer l, a j represents the vector representation of the neighbor node, f(j) represents the conversion function, σ is the activation function of the operator, and c ij represents the degree of node i, N i Represents the neighbors of node i, including itself, represents the vector representation of node j at layer l, Represents the parameter matrix that the transformation operator needs to learn.
[0087] 3) Use the defined loss function and stochastic gradient descent method to supervise and constrain the changed vector. The present invention uses the loss function shown in formulas (5) and (6) and uses the gradient descent method to optimize the parameters of the model:
[0088]
[0089]
[0090] Where q represents the vector representation of the graph, The exponential transformation of the vector label representing the graph, g(i) and g(j) represent the vectorized representation of the fraudulent node i and the normal node j respectively, y i ,y j Represent the true label values of nodes i and j respectively.
[0091] Step 4: Calculate the similarity between the node vector and the graph vector, and output nodes with similarity less than 0.5 as fraudulent nodes.
[0092] The similarity calculation is shown in formula (7).
[0093]
[0094] where s i represents the similarity between the node vector and the graph vector, q and g(i) represent the graph vector and the vector of node i respectively, ‖q‖ and ‖g(i)‖ represent the norm of the graph vector and the vector of node i respectively.
[0095] like Figure 3 The present invention uses the user's associated data to establish Figure 4The user knowledge graph shown in Figure 1 is then preprocessed based on the user knowledge graph and combined with basic user and other information, and the graph nodes are initialized with vectors. The vectors are then transformed nonlinearly using graph computing operators, as shown in Figure 1. Figure 3 As shown in the figure, by defining the loss function and optimization method, the vector representation of each node in the latent space is learned. The vectorized representation of the graph is calculated as shown in formula (6). On this basis, as shown in formula (7), the similarity between each node and the graph vector is calculated. By analyzing whether the similarity is less than a specific threshold (0.5), fraud identification of the applicant can be achieved.
[0096] like Figure 3 As shown in the figure, the most critical step in loan fraud identification is data collection. Only when the collected data has strong authenticity, integrity and validity can the accuracy of the identification result be guaranteed. The second step is to build a knowledge graph, which is built through the contact itself, the person or organization associated with the contact, and the connection between these related parties. The third step is to complete the initialization of the graph node vector, build the graph convolutional neural network, define the loss function, and complete the pre-training preparation. The fourth step is to learn the vector representation of the graph node and the complete graph, and express the contact information in a language that can be recognized by the computer. The fifth step is to calculate the similarity between the node vector and the complete graph vector; the sixth step is to compare the similarity with the threshold. If it is less than the threshold, it is considered a fraudulent loan, otherwise it is not a fraudulent loan.
[0097] like Figure 4 As shown in the figure, all nodes in the figure represent contacts. There are call, SMS and other relationships between different contacts. The relationship between contacts constitutes Figure 4 The graph network shown
[0098] like Figure 5 As shown in the figure, when K=1, node i will update the representation of node i by performing graph convolution on the nodes whose distance from node i is 1. When K=2, the nodes whose distance from node i is 2 will be selected.
[0099] The embodiment of the present invention uses real second-hand car loan data for experiments, which contains 6542 customers (470 fraudulent customers and 6072 normal customers). Each customer has 50 types of valid data, and there are 7548 mutual related information between customers. The experimental effect uses the AUC comparison indicator. The comparison results are shown in the following table. Under the same data, the method of the present invention achieves better results.
[0100] Method Type AUC Scorecard Model (Existing Methods) 80.00% Method of the present invention 83.46%
[0101] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. It can be understood by a person of ordinary skill in the art that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.
[0102] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.
Claims
1. A method for identifying second-hand car financing guarantee information, characterized in that: The second-hand car financing guarantee information identification method comprises: Based on the collected basic information data, relationship data and vehicle information data of loan applicants, the ontology of the relationship knowledge graph is designed, the relationship knowledge graph is constructed, and the graph convolutional neural network is used to establish a vector representation learning model for the graph nodes in the relationship knowledge graph. Based on the vector representation of the graph nodes, the fraud identification rules for loan applications are defined to identify whether they are fraudulent loan applications; The second-hand car financing guarantee information identification method comprises the following steps: The first step is to collect the basic information, credit information, and historical information of the institution and other institutions provided by the applicant, and comprehensively build a comprehensive data set for each applicant; The second step is to build a knowledge graph, using the applicant's contact information as a graph node. The applicant and other people or other institutions are connected by edges to indicate that the applicant has a relationship with them. At the same time, the user's basic information and other behavior information are used as the attribute data of the node; The third step is to use the applicant’s basic information, behavior information and other information to initialize the network nodes as vectors based on the applicant’s knowledge graph; use the graph convolution operator to perform nonlinear changes on the vectors; use the defined loss function and stochastic gradient descent method to supervise and constrain the changed vectors to obtain the learned node vector representation; The fourth step is to calculate the similarity between the node vector and the graph vector, and output the nodes with similarity less than 0.5 as fraudulent nodes; The third step specifically includes: 1) Initialize the graph node vector and select the valid information of the node as follows: customer monthly income, age, number of mobile phone numbers associated with the ID card, ethnicity, and information equivalent to whether the credit report is white household. Then, the valid information is initialized according to the formula Normalize it and finally use formula a i =(δ i , δ2, ..., δ m ) to get the vector initialization representation of the node; where δ j represents the normalized value of information j, xj represents the value of information j before normalization, min(·) represents the calculated minimum value, max(·) represents the calculated maximum value, n represents the total number of nodes in the relational knowledge graph, ai represents the vector initialization representation of the relational knowledge graph node i, and m represents the total number of data types of valid information; 2) Use the graph convolution operator to perform nonlinear changes on the vector. The graph convolution operator definition formula is and As shown; Where g(i) represents the vector representation of node i after transformation at layer l, a j represents the vector representation of the neighbor node, f(j) represents the conversion function, σ is the activation function of the operator, and c ij represents the degree of node i, N i Represents the neighbors of node i, including itself, represents the vector representation of node j at layer l, Represents the parameter matrix that the transformation operator needs to learn; 3) Use the defined loss function and stochastic gradient descent method to supervise the changed vector; use the formula and The loss function shown uses the gradient descent method to optimize the parameters of the model; Where q represents the vector representation of the graph, The exponential transformation of the vector label representing the graph, g(i) and g(j) represent the vectorized representation of the fraudulent node i and the normal node j respectively, y i ,y j Represent the true label values of nodes i and j respectively; The fourth step calculates the similarity between the node vector and the graph vector, and outputs nodes with a similarity less than 0.5 as fraudulent nodes. The similarity calculation formula is where s i represents the similarity between the node vector and the graph vector, q and g(i) represent the graph vector and the vector of node i respectively, ‖q‖ and ‖g(i)‖ represent the norm of the graph vector and the vector of node i respectively.
2. The method for identifying second-hand car financing guarantee information according to claim 1, characterized in that: The basic information data of the loan applicant includes: The basic information, credit information, and historical information of the institution and other institutions provided by the applicant are used to build a comprehensive data set for each applicant.
3. The method for identifying second-hand car financing guarantee information according to claim 1, characterized in that: The construction of the relational knowledge graph is to use the applicant's contact information as a graph node, and the applicant and other people or other organizations are connected by edges to indicate that the applicant has a relationship with them, and at the same time use the user's basic information and other behavioral information as attribute data of the node.
4. The method for identifying second-hand car financing guarantee information according to claim 1, characterized in that: The fraud identification model of loan applications established by using graph convolutional neural networks to identify whether a loan application is fraudulent includes: Based on the applicant's knowledge graph, the network nodes are initialized with vector representations using the applicant's basic information, behavioral information and other information; the vectors are nonlinearly transformed using the graph convolution operator; the transformed vectors are supervised and constrained using the defined loss function and stochastic gradient descent method to obtain the learned node vector representation; and the similarity between the node vector and the graph vector is calculated.
5. The method for identifying second-hand car financing guarantee information according to claim 1, characterized in that: The second step of constructing the knowledge graph specifically includes: 1) Construct binary data of applicants and related information, and find related information based on each applicant to form binary data in the following format: [[P, R1], [P, R2], …, [P, Rn]], where P represents the applicant, and R1, R2, …, Rn represent information related to the applicant P; 2) Use the binary data from 1) to construct the applicant’s knowledge graph.
6. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the method for identifying second-hand car financing guarantee information according to any one of claims 1 to 5.
7. A second-hand car financing guarantee information identification system implementing the second-hand car financing guarantee information identification method according to any one of claims 1 to 5, characterized in that: The second-hand car financing guarantee information identification system includes: The data collection module is used to collect basic information provided by applicants, credit information, and historical information of the institution and other institutions, and comprehensively construct a comprehensive data set for each applicant; The relationship knowledge graph construction module is used to use the applicant's contact information as a graph node. The applicant and other people or other institutions are connected by edges to indicate that the applicant has a relationship with them. At the same time, the user's basic information and other behavior information are used as the attribute data of the node; The node vectorization representation module is used to initialize the vector representation of network nodes based on the applicant's relationship knowledge graph and the applicant's basic information, behavior information and other information; use the graph convolution operator to perform nonlinear changes on the vector; use the defined loss function and stochastic gradient descent method to supervise and constrain the changed vector to obtain the learned node vectorization representation; The similarity calculation module is used to calculate the similarity between the node vector and the graph vector, and output the nodes with similarity less than 0.5 as fraudulent nodes.
