Similarity-based risk identification method, device, equipment and storage medium

Abstract

This invention discloses a similarity-based risk identification method, apparatus, device, and storage medium. The method extracts and generates a corresponding association similarity matrix based on the correspondence between target objects and objects to be extracted from preset data. Then, it iteratively calculates the risk value using a known preset risk label vector, thereby outputting a converged risk value for the target object, facilitating risk identification by the user. Using an iterative calculation method with an association similarity matrix and preset risk label vectors, instead of the traditional scorecard model, not only allows for the definition of different similarity matrices and risk labels based on different application scenarios, improving the flexibility of the risk identification method, but also improves the accuracy of risk identification by calculating the converged risk value through an iterative algorithm, solving the technical problem of low user risk identification accuracy in some existing scenarios.

Description

Technical Field

[0001] The present invention relates to the technical field of financial technology (Fintech), and in particular to a risk identification method, device, equipment and computer-readable storage medium based on similarity. Background Art

[0002] With the development of computer technology, more and more technologies are applied in the financial field. The traditional financial industry is gradually transforming to Fintech, and model training technology is no exception. However, due to the security and real-time requirements of the financial industry, higher requirements are also placed on model training technology. The existing personal loan risk identification method is mainly based on the scoring card model, that is, first obtain some user samples with risk identification, and extract the risk-related factors corresponding to the samples, and then use the supervised algorithm to train the model after the relevant technical processing of the risk-related factors, so that the risk results output by the model are consistent with the known risk results, and then the trained model can be used to score the risk of the user. The existing scoring card model requires the risk labels and risk factors of the known sample customers, and then uses the supervised learning algorithm to find customers with similar risks. Therefore, in the absence of external information directly related to the customer, the existing risk identification method is difficult to identify the user's risk, which reduces the accuracy of risk identification. Summary of the invention

[0003] The main purpose of the present invention is to propose a risk identification method, device, equipment and computer-readable storage medium based on similarity, aiming to solve the technical problem of low accuracy of user risk identification in some existing scenarios.

[0004] To achieve the above object, the present invention provides a similarity-based risk identification method, which comprises the following steps:

[0005] Upon receiving the risk identification instruction, determining the target object and the to-be-extracted relationship object in the preset data, and generating an association similarity matrix corresponding to the target object and the to-be-extracted relationship object based on the corresponding relationship between the target object and the to-be-extracted relationship object in the preset data;

[0006] The association similarity matrix is ​​normalized, and the preset risk label vector and the normalized association similarity matrix are iterated based on an iterative algorithm to obtain the risk value of the target object;

[0007] When the risk value of the target object reaches a convergence condition, the risk value corresponding to the target object is output to determine whether the target object has a risk.

[0008] Optionally, when receiving the risk identification instruction, determining the target object and the to-be-extracted relationship object in the preset data, and generating an association similarity matrix corresponding to the target object and the to-be-extracted relationship object based on the corresponding relationship between the target object and the to-be-extracted relationship object in the preset data specifically includes:

[0009] Upon receiving the risk identification instruction, determining the target object and the relationship object to be extracted in the preset data, and determining the associated data table corresponding to the target object and the relationship to be extracted in the preset data;

[0010] Based on the association data table and the matrix formula, an association similarity matrix corresponding to the target object and the to-be-extracted relationship object is generated, wherein the matrix formula is:

[0011] S=D -1 / 2 WD -1 / 2 Or S = D -1 W, where D is a diagonal matrix and W is a vector in the association data table.

[0012] Optionally, the step of normalizing the association similarity matrix, and iterating the preset risk label vector and the normalized association similarity matrix based on an iterative algorithm to obtain the risk value of the target object specifically includes:

[0013] Normalizing the association similarity matrix and determining the matrix type corresponding to the association similarity matrix;

[0014] When the matrix type corresponding to the association similarity matrix is ​​the first type, a first iteration formula corresponding to the first type is determined as the iteration algorithm, wherein the first iteration formula is:

[0015] P(t)=kSP(t-1)+(1-k)Y, where t is the iteration round, P (t) is the risk value vector of the node of the iterated t round, P0=Y, Y is the preset risk label vector, S is the associated similarity matrix, and k is the preset constant;

[0016] Based on the iterative algorithm, the preset risk label vector and the normalized associated similarity matrix are iterated to obtain the risk value of the target object.

[0017] Optionally, after the step of normalizing the association similarity matrix and determining the matrix type corresponding to the association similarity matrix, the method further includes:

[0018] When the matrix type corresponding to the association similarity matrix is ​​the second type, a second iterative formula corresponding to the second type is determined as the iterative algorithm, wherein the second iterative formula is:

[0019] P e (t) = S T P e (t-1), where P e (t) is the risk value vector when the target object is a user and the relation object to be extracted is an entity;

[0020] Based on the iterative algorithm, the preset risk label vector and the normalized associated similarity matrix are iterated to obtain the risk value of the target object.

[0021] P e (t) = SP p (t-1), where P e (t) is the risk value vector when the target object is an entity and the relation object to be extracted is a user;

[0022] Based on the iterative algorithm, the preset risk label vector and the normalized associated similarity matrix are iterated to obtain the risk value of the target object.

[0023] Optionally, when the risk value of the target object reaches a convergence condition, the step of outputting the risk value corresponding to the target object so as to determine whether the target object has a risk specifically includes:

[0024] Determine whether the change value corresponding to the risk value of the target object is less than a preset change threshold;

[0025] When the change value corresponding to the risk value of the target object is less than the change threshold, the risk value corresponding to the target object is output to determine whether the target object has a risk.

[0026] Optionally, after the step of determining whether the change value corresponding to the risk value of the target object is less than a preset change threshold, the step further includes:

[0027] When the change value corresponding to the risk value of the target object is not less than the change threshold, determining whether the number of iterations of the target object reaches a preset number threshold;

[0028] When the iteration number of the target object reaches a preset number threshold, the risk value corresponding to the target object is output to determine whether the target object has a risk.

[0029] In addition, to achieve the above-mentioned purpose, the present invention further provides a risk identification device based on similarity, the risk identification device based on similarity comprising:

[0030] A similarity matrix generation module is used to determine the target object and the to-be-extracted relationship object in the preset data when receiving the risk identification instruction, and generate an association similarity matrix corresponding to the target object and the to-be-extracted relationship object based on the corresponding relationship between the target object and the to-be-extracted relationship object in the preset data;

[0031] A risk value iteration module, used for normalizing the association similarity matrix, and iterating the preset risk label vector and the normalized association similarity matrix based on an iterative algorithm to obtain the risk value of the target object;

[0032] The risk value output module is used to output the risk value corresponding to the target object when the risk value of the target object reaches a convergence condition, so as to determine whether the target object has a risk.

[0033] Optionally, the similarity matrix generation module specifically includes:

[0034] A data table generating unit, configured to determine the target object and the relationship object to be extracted in the preset data when receiving a risk identification instruction, and determine an associated data table corresponding to the target object and the relationship to be extracted in the preset data;

[0035] A matrix generating unit is used to generate an association similarity matrix corresponding to the target object and the to-be-extracted relationship object based on the association data table and the matrix formula, wherein the matrix formula is: S = D -1 / 2 WD -1 / 2 Or S = D - 1 W, where D is a diagonal matrix and W is a vector in the association data table.

[0036] In addition, to achieve the above-mentioned purpose, the present invention also provides a similarity-based risk identification device, which includes: a memory, a processor, and a similarity-based risk identification program stored in the memory and executable on the processor, and the similarity-based risk identification program, when executed by the processor, implements the steps of the similarity-based risk identification method as described above.

[0037] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, on which a similarity-based risk identification program is stored, and when the similarity-based risk identification program is executed by a processor, the steps of the similarity-based risk identification method as described above are implemented.

[0038] The present invention discloses a risk identification method based on similarity, which determines a target object and a relationship object to be extracted in preset data when a risk identification instruction is received, and generates an association similarity matrix corresponding to the target object and the relationship object to be extracted based on the corresponding relationship between the target object and the relationship object to be extracted in the preset data; the association similarity matrix is ​​normalized, and the preset risk label vector and the normalized association similarity matrix are iterated based on an iterative algorithm to obtain the risk value of the target object; when the risk value of the target object reaches a convergence condition, the risk value corresponding to the target object is output to determine whether the target object has a risk. In the above manner, the present invention extracts and generates a corresponding association similarity matrix based on the corresponding relationship between the target object and the relationship object to be extracted in the preset data, and then performs iterative calculation of the risk value in combination with the known preset risk label vector, thereby outputting the converged risk value corresponding to the target object, which is convenient for users to judge the target object for risk identification. Using the iterative calculation method of the associated similarity matrix and the preset risk label vector instead of the traditional scoring card model can not only define different similarity matrices and risk identifiers based on different application scenarios to improve the flexibility of the risk identification method, but also obtain the convergent risk value through iterative algorithm calculation, thereby improving the accuracy of risk identification and solving the technical problem of low user risk identification accuracy in some existing scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present invention;

[0040] Figure 2 It is a flowchart of the first embodiment of the risk identification method based on similarity of the present invention.

[0041] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0042] 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.

[0043] like Figure 1 As shown, Figure 1 It is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present invention.

[0044] The similarity-based risk identification device in the embodiment of the present invention may be a PC or a server device, on which a Java virtual machine runs.

[0045] like Figure 1As shown, the similarity-based risk identification device may include: a processor 1001, such as a CPU, a network interface 1004, a user interface 1003, a memory 1005, and a communication bus 1002. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory, or a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0046] Those skilled in the art will understand that Figure 1 The device structure shown in the figure does not constitute a limitation of the device, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0047] like Figure 1 As shown, the memory 1005 as a computer storage medium may include an operating system, a network communication module, a user interface module, and a similarity-based risk identification program.

[0048] exist Figure 1 In the device shown, the network interface 1004 is mainly used to connect to the background server and communicate data with the background server; the user interface 1003 is mainly used to connect to the client (user end) and communicate data with the client; and the processor 1001 can be used to call the similarity-based risk identification program stored in the memory 1005, and perform the operations in the following similarity-based risk identification method.

[0049] Based on the above hardware structure, an embodiment of the risk identification method based on similarity of the present invention is proposed.

[0050] Reference Figure 2 , Figure 2 The flowchart of the first embodiment of the risk identification method based on similarity of the present invention is as follows. The risk identification method based on similarity includes:

[0051] Step S10, upon receiving the risk identification instruction, determining the target object and the to-be-extracted relationship object in the preset data, and generating an association similarity matrix corresponding to the target object and the to-be-extracted relationship object based on the corresponding relationship between the target object and the to-be-extracted relationship object in the preset data;

[0052] The existing personal loan risk identification method is mainly based on the scoring card model. First, some samples with risk target Y are obtained, and the risk-related factors X corresponding to the samples are extracted. Then, after X is processed by relevant technologies, the model F is trained using a supervised algorithm so that the risk assessment result of F(X) is as consistent as possible with the known risk target Y. Then, the trained model F can be applied to score the risk of customers. Among them, personal credit risk management is to rate, classify, report and manage risks through risk identification, measurement, monitoring and control procedures, maintain a balanced development of risks and benefits, and improve the economic benefits of loans. For personal credit business, risk management is to combine various factors that cause personal defaults, identify the risk of personal defaults, and manage and control them throughout the process. Customers are specific individuals who apply for and use loans, and are identified by some ID (such as ID number, mobile phone number, etc.). The core idea of ​​the existing personal loan risk identification method is to use X information directly related to customers, and then identify many sets with similar risk performance through cross-combination. The customers in the set have the same attributes or behavioral performance, and the result of this cross-combination is a specific description of the set. Therefore, the existing scoring card model needs to obtain the risk label Y and risk factor X of the sample customers and use supervised learning to find customer groups with similar risk performance. However, in some scenarios, the conditions that the scoring card model relies on may not be met, thereby reducing the accuracy of user risk identification and even making it difficult to identify risks.

[0053] In order to solve the above problems, the present invention extracts and generates a corresponding association similarity matrix based on the corresponding relationship between the target object and the relationship object to be extracted in the preset data, and then performs iterative calculation of the risk value in combination with the known preset risk label vector, thereby outputting the converged risk value corresponding to the target object, which is convenient for users to judge the target object for risk identification. Using the association similarity matrix and the preset risk label vector iterative calculation method to replace the traditional scoring card model, not only can different similarity matrices and risk identifiers be defined based on different application scenarios to improve the flexibility of the risk identification method, but also the converged risk value is calculated by the iterative algorithm, which improves the accuracy of risk identification. Specifically, the user can trigger the corresponding risk identification instruction based on the specific problem, so as to determine the target object and the relationship object to be extracted, such as when predicting the risk information of an unknown risk customer based on the known risk customer information, the corresponding relationship between customers can be extracted, or when predicting the risk information of an unknown risk customer based on the loan information of a known customer in a related bank, the corresponding relationship between customers and entities can be extracted. Risk-expressed customers are customers who have already overdue, also known as bad customers. Risk-unexpressed customers are customers who have not yet overdue, also known as good customers. An entity is a customer who is associated with an entity due to some common reasons and needs. This entity is called an aggregate entity. The aggregate entity can be a physical object or a virtual concept. For example, if several customers apply for a loan from a bank, the bank is an aggregate entity; if multiple customers click on a loan promotion link at a certain point in time, the behavior itself is also an aggregate entity. After determining the target object and the relationship object to be extracted in the preset data, the corresponding relationship between the target object and the relationship object to be extracted is extracted in the preset data, and then the association similarity matrix corresponding to the target object and the relationship object to be extracted is generated based on the corresponding relationship. Thus, the technical problem that only the association relationship between customers and customers and customers and entities is known in the prior art, but the external X information directly related to the customer is lacking, and it is difficult to characterize the customer group with similar risk performance through X is solved. In more embodiments, after the graph structure is formed, the identification vector of the customer who has already shown some risks can be combined with the identification vector of the preset part of the risk, and the risk level of all customers can be predicted by using a semi-supervised iterative algorithm under the constraint of the graph regularization criterion, and represented by the p value. Among them, p is a continuous value, which is convenient for risk management. At the same time, the p-value is the optimal non-trivial result that satisfies the graph regularization criterion. The graph regularization criterion is that for any node v in the graph G, a function f(v) is used to represent the risk of v. This f needs to meet two restrictions at the same time: 1) the f of any two nodes must be as similar as possible; 2) for nodes with known risk y, f(v) and y must be as close as possible.The graph regularization criterion is obtained by taking the weighted sum of the above two conditions for all nodes. Based on this criterion, we can evaluate different f and use iteration to find a non-trivial optimal result that meets the criterion. The non-trivial result is the result that the f(v) value of each node v on the graph structure G is not exactly the same. The semi-supervised iterative algorithm generates the graph structure G, y of some nodes, and the graph regularization criterion. For each node v in the graph, it continuously iterates and optimizes using the surrounding values ​​and the initial value, and finally obtains a non-trivial optimal result. The P value is the final f(v) of each node v in the graph structure G, which represents the risk value of this node.

[0054] Furthermore, the step S10 specifically includes:

[0055] Upon receiving the risk identification instruction, determining the target object and the relationship object to be extracted in the preset data, and determining the associated data table corresponding to the target object and the relationship to be extracted in the preset data;

[0056] Based on the association data table and the matrix formula, an association similarity matrix corresponding to the target object and the to-be-extracted relationship object is generated, wherein the matrix formula is:

[0057] S=D -1 / 2 WD -1 / 2 Or S = D -1 W, where D is a diagonal matrix and W is a vector in the association data table.

[0058] In this embodiment, when a risk identification instruction is received, the target object and the object to be extracted are determined, that is, the customer or entity is defined, and the corresponding relationship is extracted from the preset data summary. According to the extracted corresponding relationship, a graph structure between different customers or a graph structure between customers and entities is generated. Among them, the graph structure is that different customers and different aggregated entities will be associated, and the customer and the entity are regarded as nodes, the association relationship is regarded as an edge, and the association strength between the customer and the entity is set as the edge weight. Then, different graph structures G can be generated for different scenarios, and different risks can be identified from them. Then, based on the graph structure of the corresponding relationship, an association data table is generated, as shown in the following table:

[0059] Customer ID Bank ID Borrowing time C1 B1 20190101 C1 B2 20190212 C1 B3 20190313 C2 B1 20190125 C2 B6 20190607 C2 B6 20191213 ... … … Cn B2 20191102

[0060] The corresponding relationship is then converted into a customer-customer or customer-entity similarity matrix W, where the similarity can be defined according to the specific problem, thereby generating a variety of similarity matrices, such as:

[0061] 1. The customer-entity (bank) matrix is ​​generated based on the number of times a customer borrows money from a bank:

[0062]

[0063]

[0064] 2. The customer-entity (bank (month)) matrix is ​​generated based on the number of times a customer borrows money from a certain bank in a certain month:

[0065] C1 C2 … Cn C1 C2 C1 1 1 … 1 C1 1 1 C2 1 1 .... 0 C2 1 1 … … … … … … … … Cn 0 0 0 1 Cn 0 0

[0066] 3. Customer-customer matrix, based on how many banks any two customers borrow from at the same time, generates a matrix that is symmetric about the diagonal:

[0067]

[0068] Where D is a diagonal matrix, and the formula of the diagonal matrix is:

[0069] i=1,2,3...n,j=1,2,3...n,W ij is the corresponding vector in the associated data table.

[0070] Based on the association data table and the matrix formula, S=D -1 / 2 WD -1 / 2 Or S = D -1 W, W is a vector in the association data table, and an association similarity matrix corresponding to the target object and the relationship object to be extracted is generated.

[0071] Step S20, normalizing the association similarity matrix, and iterating the preset risk label vector and the normalized association similarity matrix based on an iterative algorithm to obtain the risk value of the target object;

[0072] In this embodiment, in order to reduce the amount of calculation, the association similarity matrix is ​​normalized and then combined with a preset risk label vector for iteration to calculate the risk value of the target object. The preset label vector includes a label vector of a known risk customer and a label vector of an unknown risk customer.

[0073] Furthermore, the step S20 specifically includes:

[0074] Normalizing the association similarity matrix and determining the matrix type corresponding to the association similarity matrix;

[0075] When the matrix type corresponding to the association similarity matrix is ​​the first type, a first iteration formula corresponding to the first type is determined as the iteration algorithm, wherein the first iteration formula is:

[0076] P(r)=kSP(t-1)+(1-k)Y, where t is the iteration round, P (t) is the risk value vector of the node of the iterated t round, P0=Y, Y is the preset risk label vector, S is the associated similarity matrix, and k is the preset constant;

[0077] When the matrix type corresponding to the association similarity matrix is ​​the second type, a second iterative formula corresponding to the second type is determined as the iterative algorithm, wherein the second iterative formula is:

[0078] P e (t) = S T P e (t-1), where P e (t) is the risk value vector when the target object is a user and the relation object to be extracted is an entity;

[0079] P e (t) = SP a (t-1), where P e (t) is the risk value vector when the target object is an entity and the relation object to be extracted is a user;

[0080] Based on the iterative algorithm, the preset risk label vector and the normalized associated similarity matrix are iterated to obtain the risk value of the target object.

[0081] In this embodiment, according to the pre-associated known risk label vector Y, the risk label vector is shown in the following table, where 1 represents a customer label vector with risk, and 0 represents a customer label vector with unknown risk:

[0082] Customer ID Risk Label C1 1 C1 1 C1 1 C2 0 C2 0 C2 0 ... … Cn 0

[0083] For the customer-customer matrix, that is, when the target object is the customer and the relationship object to be extracted is the customer, the current iteration type is determined to be the first iteration type, where the first iteration formula is:

[0084] P(t)=kSP(t-1)+(1-k)Y;

[0085] Among them, t is the iteration round, P (t) is the risk value vector of the node of the iteration t round, P0=Y, Y is the preset risk label vector, S is the association similarity matrix, and k is the preset constant;

[0086] For a customer-entity matrix, such as a customer-entity (bank) matrix or a customer-entity (bank (month)) matrix, that is, when the target object is a customer and the relation object to be extracted is an entity, the iteration formula is: e (t) = S T Pe (t-1), where P e (t) is the risk value vector when the target object is a user and the relation object to be extracted is an entity; e (t) = SP e (t-1), where P e (t) is the risk value vector when the target object is an entity and the relationship object to be extracted is a user.

[0087] The iterative process is performed in two steps alternately until the risk value of the target object is calculated.

[0088] Iteratively use known customer risks to predict unknown customer risks. That is, in the first round of iteration, there are more unknown risk customers. Then, use the associated known risk customers of the unknown risk customers to perform a second round of iteration to reduce the number of unknown risk customers until the number of unknown risk customers is lower than the preset value, and the risk value of the target object can be obtained.

[0089] Step S30: when the risk value of the target object reaches a convergence condition, output the risk value corresponding to the target object to determine whether the target object has a risk.

[0090] In this embodiment, when it is detected that the risk value of the target object reaches the convergence condition, that is, when the iteration stop condition is met, the iteration can be stopped, and the risk value after the current iteration is obtained as the risk value corresponding to the target object, so as to identify whether the target object has risk based on the risk value corresponding to the target object. The target object can be a customer or an entity, such as identifying whether a customer has risk in taking out a loan from a bank.

[0091] The present embodiment discloses a risk identification method based on similarity, by determining the target object and the relationship object to be extracted in the preset data when receiving the risk identification instruction, and generating the association similarity matrix corresponding to the target object and the relationship object to be extracted based on the corresponding relationship between the target object and the relationship object to be extracted in the preset data; normalizing the association similarity matrix, and iterating the preset risk label vector and the normalized association similarity matrix based on the iterative algorithm to obtain the risk value of the target object; when the risk value of the target object reaches the convergence condition, outputting the risk value corresponding to the target object to determine whether the target object has a risk. In the above manner, the present invention extracts and generates the corresponding association similarity matrix based on the corresponding relationship between the target object and the relationship object to be extracted in the preset data, and then performs iterative calculation of the risk value in combination with the known preset risk label vector, thereby outputting the converged risk value corresponding to the target object, which is convenient for the user to judge the target object for risk identification. Using the iterative calculation method of the associated similarity matrix and the preset risk label vector instead of the traditional scoring card model can not only define different similarity matrices and risk identifiers based on different application scenarios to improve the flexibility of the risk identification method, but also obtain the convergent risk value through iterative algorithm calculation, thereby improving the accuracy of risk identification and solving the technical problem of low user risk identification accuracy in some existing scenarios.

[0092] Furthermore, based on the first embodiment of the risk identification method based on similarity of the present invention, a second embodiment of the risk identification method based on similarity of the present invention is proposed.

[0093] In this embodiment, step S30 specifically includes:

[0094] Determine whether the change value corresponding to the risk value of the target object is less than a preset change threshold;

[0095] When the change value corresponding to the risk value of the target object is less than the change threshold, the risk value corresponding to the target object is output to determine whether the target object has a risk.

[0096] When the change value corresponding to the risk value of the target object is not less than the change threshold, determining whether the number of iterations of the target object reaches a preset number threshold;

[0097] When the iteration number of the target object reaches a preset number threshold, the risk value corresponding to the target object is output to determine whether the target object has a risk.

[0098] In this embodiment, there may be multiple convergence conditions, i.e., iteration stop conditions, which can be defined by the user according to actual specific computing conditions and business, such as:

[0099] 1) The change in the p-value between the previous and next rounds is less than the preset change threshold;

[0100] 2) The number of iterations reaches a preset threshold;

[0101] 3) When the p-value no longer has an increasing effect on distinguishing risks.

[0102] Specifically, first determine whether the change value corresponding to the risk value of the target object is less than the preset change threshold. If it is less than the change threshold, determine that the risk value of the target object has reached the convergence condition, stop iteration, and output the risk value corresponding to the target object. If it is not less than the change threshold, determine whether the number of iterations of the target object has reached the preset number threshold. If it has, stop iteration. When the first two conditions are not met, further determine whether the risk differentiation effect of the P value after iteration continues to rise, that is, calculate the evaluation index parameter Ks corresponding to the risk value after iteration, and then measure whether the differentiation effect of p continues to rise by Ks. If Ks reaches the preset threshold, the differentiation effect of P no longer continues to rise, and iteration can be stopped. First, the preset quantities include C1, C2, C3...C100, among which there are 55 0s and 45 1s. Calculate the cumulative sum of column C1, that is, the cumulative sum of column C1 = the number of 0s in each segment of column C1 / the number of all 0s, that is, 1 / 55. Then calculate the cumulative sum of column C2 = the number of 1s in each segment / the number of all 1s, such as 10 / 45. Then calculate abs(cumulative sum of column C1 - cumulative sum of column C2), that is, abs(1 / 55-10 / 45), and so on. Calculate the corresponding values ​​of 100 columns, and then take the maximum value, which is the Ks corresponding to P.

[0103] The risk identification method based on similarity provided in this embodiment can model risks for all risk identification scenarios with graph structures and missing identifiers, replacing the traditional scorecard model. In certain cases, it can identify risks that cannot be identified by traditional models, which is an effective supplement for risk identification. Similarity matrices and risk identifiers can be defined according to specific problems to achieve different business goals. Moreover, the use of iterative algorithms makes it feasible to calculate on ultra-large-scale data, and this method can achieve optimal non-trivial results.

[0104] The present invention also provides a risk identification device based on similarity, and the risk identification device based on similarity includes:

[0105] A similarity matrix generation module is used to determine the target object and the to-be-extracted relationship object in the preset data when receiving the risk identification instruction, and generate an association similarity matrix corresponding to the target object and the to-be-extracted relationship object based on the corresponding relationship between the target object and the to-be-extracted relationship object in the preset data;

[0106] A risk value iteration module, used for normalizing the association similarity matrix, and iterating the preset risk label vector and the normalized association similarity matrix based on an iterative algorithm to obtain the risk value of the target object;

[0107] The risk value output module is used to output the risk value corresponding to the target object when the risk value of the target object reaches a convergence condition, so as to determine whether the target object has a risk.

[0108] Furthermore, the similarity matrix generation module specifically includes:

[0109] A data table generating unit, configured to determine the target object and the relationship object to be extracted in the preset data when receiving a risk identification instruction, and determine an associated data table corresponding to the target object and the relationship to be extracted in the preset data;

[0110] A matrix generating unit is used to generate an association similarity matrix corresponding to the target object and the to-be-extracted relationship object based on the association data table and the matrix formula, wherein the matrix formula is:

[0111] S=D -1 / 2 WD -1 / 2 Or S = D -1 W, where D is a diagonal matrix and W is a vector in the association data table.

[0112] Furthermore, the risk value iteration module specifically includes:

[0113] A matrix type determination unit, used for normalizing the association similarity matrix and determining a matrix type corresponding to the association similarity matrix;

[0114] The first algorithm determination unit is configured to determine, when the matrix type corresponding to the association similarity matrix is ​​a first type, a first iterative formula corresponding to the first type as the iterative algorithm, wherein the first iterative formula is:

[0115] P(r)=kSP(t-1)+(1-k)Y, where t is the iteration round, P (t) is the risk value vector of the node of the iterated t round, P0=Y, Y is the preset risk label vector, S is the associated similarity matrix, and k is the preset constant;

[0116] The second algorithm determination unit is configured to determine, when the matrix type corresponding to the association similarity matrix is ​​a second type, a second iterative formula corresponding to the second type as the iterative algorithm, wherein the second iterative formula is:

[0117] P e (t) = S T P e (t-1), where P e (t) is the risk value vector when the target object is a user and the relation object to be extracted is an entity;

[0118] P e (r) = SP e (t-1), where P e (t) is the risk value vector when the target object is an entity and the relation object to be extracted is a user;

[0119] The risk value iteration unit is used to iterate the preset risk label vector and the normalized association similarity matrix based on the iteration algorithm to obtain the risk value of the target object.

[0120] Furthermore, the risk value output module specifically includes:

[0121] A first judgment unit, used to judge whether a change value corresponding to the risk value of the target object is less than a preset change threshold;

[0122] The first output unit is used to output the risk value corresponding to the target object when the change value corresponding to the risk value of the target object is less than the change threshold, so as to determine whether the target object has risks.

[0123] Furthermore, the risk value output module specifically includes:

[0124] A second judgment unit is used to judge whether the number of iterations of the target object reaches a preset number threshold when the change value corresponding to the risk value of the target object is not less than the change threshold;

[0125] The second output unit is used to output the risk value corresponding to the target object when the iteration number of the target object reaches a preset number threshold, so as to determine whether the target object has risks.

[0126] The methods executed by the above-mentioned program modules can refer to the various embodiments of the risk identification method based on similarity of the present invention, which will not be described in detail here.

[0127] The present invention also provides a computer-readable storage medium.

[0128] A similarity-based risk identification program is stored on the computer-readable storage medium of the present invention. When the similarity-based risk identification program is executed by a processor, the steps of the similarity-based risk identification method described above are implemented.

[0129] Among them, the method implemented when the similarity-based risk identification program running on the processor is executed can refer to the various embodiments of the similarity-based risk identification method of the present invention, which will not be repeated here.

[0130] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0131] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0132] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0133] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A similarity-based risk identification method, characterized in that: The similarity-based risk identification method comprises the following steps: When a risk identification instruction is received, a target object and a to-be-extracted relationship object, as well as an association data table corresponding to the target object and the to-be-extracted relationship object are determined in the preset data, and an association similarity matrix corresponding to the target object and the to-be-extracted relationship object is generated based on the association data table and a matrix formula, wherein the association data table is generated based on the corresponding relationship between the target object and the to-be-extracted relationship object in the preset data, the corresponding relationship includes a relationship between customers, or a relationship between customers and entities, the association similarity matrix includes a plurality of similarity matrices defined based on different similarities, the different similarities are respectively determined based on data that can reflect risks in different aspects, and the matrix formula is: or , where D is a diagonal matrix and W is a vector in the association data table; Normalizing the association similarity matrix, determining the matrix type corresponding to the association similarity matrix, and iterating the preset risk label vector and the normalized association similarity matrix based on an iterative algorithm to obtain the risk value of the target object; When the matrix type corresponding to the association similarity matrix is ​​a matrix composed of customers, the iteration type is determined to be the first iteration type, and the first iteration formula is adopted, wherein the first iteration formula is: , where t is the iteration round, P (t) is the risk value vector of the node of the iterated t round, P0=Y, Y is the preset risk label vector, S is the associated similarity matrix, and k is the preset constant; When the matrix type corresponding to the association similarity matrix is ​​a matrix composed of customers and entities, the iteration type is determined to be the second iteration type, and the second iteration formula is used, wherein the second iteration formula is: ,in, is the risk value vector when the target object is a user and the relation object to be extracted is an entity; ,in, is the risk value vector when the target object is an entity and the relation object to be extracted is a user; When the risk value of the target object reaches a convergence condition, the risk value corresponding to the target object is output to determine whether the target object has a risk.

2. The similarity-based risk identification method according to claim 1, characterized in that: When the risk value of the target object reaches a convergence condition, the step of outputting the risk value corresponding to the target object so as to determine whether the target object has a risk specifically includes: Determine whether the change value corresponding to the risk value of the target object is less than a preset change threshold; When the change value corresponding to the risk value of the target object is less than the change threshold, the risk value corresponding to the target object is output to determine whether the target object has a risk.

3. The similarity-based risk identification method according to claim 2, characterized in that: After the step of determining whether the change value corresponding to the risk value of the target object is less than a preset change threshold, the method further includes: When the change value corresponding to the risk value of the target object is not less than the change threshold, determining whether the number of iterations of the target object reaches a preset number threshold; When the iteration number of the target object reaches a preset number threshold, the risk value corresponding to the target object is output to determine whether the target object has a risk.

4. A risk identification device based on similarity, characterized in that: The similarity-based risk identification device comprises: A similarity matrix generation module is used to determine the target object and the relationship object to be extracted in the preset data, and the association data table corresponding to the target object and the relationship object to be extracted, when receiving the risk identification instruction, and generate the association similarity matrix corresponding to the target object and the relationship object to be extracted based on the association data table and the matrix formula, wherein the association data table is generated based on the corresponding relationship between the target object and the relationship object to be extracted in the preset data, the corresponding relationship includes the relationship between customers and customers, or the relationship between customers and entities, the association similarity matrix includes multiple similarity matrices defined based on different similarities, the different similarities are determined based on data that can reflect risks in different aspects, and the matrix formula is: or , where D is a diagonal matrix and W is a vector in the association data table; A risk value iteration module is used to normalize the association similarity matrix, determine the matrix type corresponding to the association similarity matrix, and iterate the preset risk label vector and the normalized association similarity matrix based on an iterative algorithm to obtain the risk value of the target object; The risk value iteration module includes: The first algorithm determination unit is used to determine that the iteration type is the first iteration type and adopt the first iteration formula when the matrix type corresponding to the association similarity matrix is ​​a matrix composed of customers and customers, wherein the first iteration formula is: , where t is the iteration round, P (t) is the risk value vector of the node of the iterated t round, P0=Y, Y is the preset risk label vector, S is the associated similarity matrix, and k is the preset constant; The second algorithm determination unit is used to determine the iteration type as the second iteration type and adopt the second iteration formula when the matrix type corresponding to the association similarity matrix is ​​a matrix composed of customers and entities, wherein the second iteration formula is: ,in, is the risk value vector when the target object is a user and the relation object to be extracted is an entity; ,in, is the risk value vector when the target object is an entity and the relation object to be extracted is a user; The risk value output module is used to output the risk value corresponding to the target object when the risk value of the target object reaches a convergence condition, so as to determine whether the target object has a risk.

5. A risk identification device based on similarity, characterized in that: The similarity-based risk identification device includes: a memory, a processor, and a similarity-based risk identification program stored in the memory and executable on the processor. When the similarity-based risk identification program is executed by the processor, the steps of the similarity-based risk identification method as described in any one of claims 1 to 3 are implemented.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a similarity-based risk identification program, and when the similarity-based risk identification program is executed by the processor, the steps of the similarity-based risk identification method according to any one of claims 1 to 3 are implemented.

Images