Method, device and equipment for analyzing user credit based on graph neural network

Abstract

The invention relates to the technical field of credit analysis, in particular to a method, a device and equipment for analyzing user credit based on a graph neural network, and the method comprises the steps: fusing the multi-source data of a user in a period of time, such as object transaction data, equipment use data, credit data, associated user data and other data of different dimensions; according to the method and the device, the credit condition of the user is determined by analyzing the nodes of different dimensions and the credit characteristics of the nodes, so that the analysis accuracy of the credit condition of the user can be improved, and the processing accuracy and efficiency of user credit decision are improved; for example, the lease admission analysis accuracy is improved, the guarantee deposit adjustment efficiency and accuracy are improved, and the overdue risk prediction accuracy and reliability are improved.

Description

Technical Field

[0001] The present invention relates to the field of credit analysis technology, and in particular to a method, device, and apparatus for analyzing user credit based on a graph neural network. Background Art

[0002] Based on the advantages of item leasing, such as light assets, low cost, and adaptability to diverse needs, more and more users like to meet their item usage needs through leasing.

[0003] In practice, rental access mechanisms based on user credit assessments have become a common means of preventing and controlling credit risk. However, because different items vary in price and quality, the corresponding user rental credit requirements will also vary.

[0004] Currently, the credit of users is usually analyzed based on static data of a single dimension, such as traditional credit scoring, which cannot accurately determine the credit status of users, resulting in an inability to make accurate credit decisions based on the user's credit. Therefore, it is urgent to propose a technical solution that can accurately determine the credit status of users and make accurate credit decisions based on the user's credit status. Summary of the Invention The present invention provides a method, device and equipment for analyzing user credit based on graph neural network, which can accurately determine the user's credit status and thus make accurate credit decisions based on the user's credit status.

[0005] In order to solve the above technical problems, the first aspect of the embodiments of the present invention discloses a method for analyzing user credit based on a graph neural network, the method comprising: Collecting multi-source data of a target user within a historical time period, wherein the multi-source data of the target user includes object transaction data of the target user, device usage data of the target user, and / or credit data of the target user, and / or associated user data of the target user; Analyzing, based on the multi-source data of the target user, all first nodes corresponding to the target user and the credit characteristics of each first node, wherein all first nodes include the user node and all second nodes of the target user; wherein all second nodes include the device nodes of all target devices used by the target user within the historical time period and / or the unit nodes of all units associated with the target user within the historical time period and / or the associated user nodes of users associated with the target user within the historical time period; A credit analysis result of the target user is determined based on the credit characteristics of all the second nodes and the credit characteristics of the user node.

[0006] As an optional implementation manner, in the first aspect of the present invention, determining the credit analysis result of the target user based on the credit characteristics of all the second nodes and the credit characteristics of the user node includes: Performing a neighborhood aggregation operation on the credit characteristics of all second nodes and the credit characteristics of the user node to obtain a neighborhood-associated credit characteristic between the user node and each second node; Based on the neighborhood-associated credit feature of the user node, performing an update operation on the predetermined user credit feature of the user node to obtain a credit update feature of the user node; According to a predetermined first credit weight matrix, an analysis operation is performed on the credit update feature of the user node to obtain the credit weight feature of the user node, and a binary classification operation is performed on the credit weight feature of the user node to obtain a credit analysis result of the target user.

[0007] A second aspect of an embodiment of the present invention discloses a device for analyzing user credit based on a graph neural network, the device comprising: a collection module, configured to collect multi-source data of a target user within a historical time period, wherein the multi-source data of the target user includes object transaction data of the target user, device usage data of the target user, and / or credit data of the target user, and / or associated user data of the target user; an analysis module, configured to analyze, based on the multi-source data of the target user, all first nodes corresponding to the target user and the credit characteristics of each first node, wherein all first nodes include the user node and all second nodes of the target user; wherein all second nodes include the device nodes of all target devices used by the target user within the historical time period and / or the unit nodes of all units associated with the target user within the historical time period and / or the associated user nodes of users associated with the target user within the historical time period; A determination module is used to determine a credit analysis result of the target user based on the credit characteristics of all the second nodes and the credit characteristics of the user node.

[0008] As an optional implementation, in the second aspect of the present invention, the specific manner in which the determination module determines the credit analysis result of the target user based on the credit characteristics of all the second nodes and the credit characteristics of the user node includes: Performing a neighborhood aggregation operation on the credit characteristics of all second nodes and the credit characteristics of the user node to obtain a neighborhood-associated credit characteristic between the user node and each second node; Based on the neighborhood-associated credit feature of the user node, performing an update operation on the predetermined user credit feature of the user node to obtain a credit update feature of the user node; According to a predetermined first credit weight matrix, an analysis operation is performed on the credit update feature of the user node to obtain the credit weight feature of the user node, and a binary classification operation is performed on the credit weight feature of the user node to obtain a credit analysis result of the target user.

[0009] A third aspect of the present invention discloses a credit analysis device, comprising: a memory storing executable program code; a processor coupled to a memory; The processor calls the executable program code stored in the memory to execute some or all of the steps in any one of the methods for analyzing user credit based on a graph neural network disclosed in the first aspect of the present invention.

[0010] The fourth aspect of the present invention discloses a computer storage medium, which stores computer instructions. When the computer instructions are called, they are used to execute some or all of the steps in any one of the methods for analyzing user credit based on graph neural networks disclosed in the first aspect of the present invention.

[0011] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: In an embodiment of the present invention, by fusing data of different dimensions such as multi-source data of a user over a period of time, such as object transaction data, device usage data, credit data, and associated user data, nodes of different dimensions and credit feature analysis of nodes are performed to determine the user's credit status, the accuracy of the analysis of the user's credit status can be improved, thereby improving the accuracy and efficiency of the processing of user credit decisions, such as improving the analysis accuracy of leasing access, improving the efficiency and accuracy of deposit adjustment, and improving the accuracy and reliability of overdue risk prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0013] Figure 1 This is a flow chart of a method for analyzing user credit based on a graph neural network disclosed in an embodiment of the present invention; Figure 2 This is a flow chart of another method for analyzing user credit based on graph neural network disclosed in an embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of a device for analyzing user credit based on a graph neural network disclosed in an embodiment of the present invention; Figure 4 2 is a schematic diagram of the structure of another device for analyzing user credit based on a graph neural network disclosed in an embodiment of the present invention; Figure 5 The figure is a schematic structural diagram of a credit analysis device disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0014] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0015] The terms "first," "second," and so on, in the description and claims of the present invention and the accompanying drawings are used to distinguish between different objects, not to describe a specific order. Furthermore, the terms "including," "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, apparatus, product, or device comprising a series of steps or elements is not limited to the listed steps or elements but may optionally include steps or elements not listed, or may optionally include other steps or elements inherent to the process, method, product, or device.

[0016] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present invention. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute a separate or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0017] This invention discloses a method, apparatus, and device for analyzing user credit based on a graph neural network. By fusing multi-source data from a user over a period of time, such as object transaction data, device usage data, credit data, and associated user data, the method analyzes nodes and their credit characteristics in different dimensions to determine the user's credit status. This method improves the accuracy of credit analysis and, consequently, the accuracy and efficiency of credit decision-making. For example, this method improves the accuracy of lease approval analysis, the efficiency and accuracy of deposit adjustments, and the accuracy and reliability of overdue risk prediction. These are described in detail below.

[0018] Example 1 See also Figure 1 , Figure 1 This is a flow chart of a method for analyzing user credit based on a graph neural network disclosed in an embodiment of the present invention. The method can be applied to any scenario where user credit analysis is required, such as an item rental scenario, and the scenario is provided with a corresponding device. The device includes a credit analysis device, a credit determination system (a local system or a cloud system), and a credit determination server (a local server or a cloud server). The credit analysis device includes but is not limited to mobile phones, computers, smart bracelets, and other devices that can perform credit data analysis. Figure 1 As shown, the method may include the following operations: 101. Collect multi-source data of a target user within a historical time period, where the multi-source data of the target user includes object transaction data of the target user, device usage data of the target user, and / or credit data of the target user, and / or associated user data of the target user.

[0019] In the embodiment of the present invention, the target user is optionally any user for whom credit analysis is required. The historical time period is any time period for which credit analysis is required, such as the past week or the past half month.

[0020] In an embodiment of the present invention, the target user's object transaction data optionally includes object transaction contract data. Object transaction contract data includes, but is not limited to, historical contract performance data and deposit ratio data. Historical contract performance data includes historical fee performance data and / or historical quantity performance data. Historical fee performance data indicates the on-time payment of all fees required for all transaction objects within a historical time period; historical quantity performance data indicates the on-time payment of fees under contracts signed for all transaction objects within a historical time period. Furthermore, the target user's object transaction data also includes object type, / or object quantity, and / or object price. An object can be understood as a tangible item and / or a non-tangible service. Tangible items can include devices or non-devices. In the case of devices, device usage data can be collected by sensors installed on the corresponding target device and fed back by the target device. Terms such as "lease," "rent," and "purchase" that imply interaction between two parties can be understood as transactions.

[0021] In the embodiment of the present invention, optionally, the device usage data of the target device includes one or more of operation specification data, usage frequency data, usage duration data, and the like.

[0022] In embodiments of the present invention, the target user's credit data optionally includes the target user's personal credit data and / or the credit data of an organization associated with the target user. Optionally, the target user's associated user data includes one or more of the following: the number of associated users, the credit data of each associated user, the device usage data of each associated user, and the object transaction data of each associated user. For descriptions of other associated user data, except the number of associated users, please refer to the description of the target user's related data and are not further elaborated here.

[0023] In the embodiment of the present invention, the multi-source data may be normalized. The normalization method may be Min-Max or other methods, as long as normalization is achieved to eliminate dimension, thereby better performing feature analysis and improving the efficiency and accuracy of user credit analysis.

[0024] 102. Analyze the credit characteristics of all first nodes and each first node corresponding to the target user based on the multi-source data of the target user, wherein all first nodes include the user node of the target user and all second nodes.

[0025] In this embodiment of the present invention, optionally, all second nodes include device nodes of all target devices used by the target user within a historical time period and / or unit nodes of all units associated with the target user within the historical time period and / or associated user nodes of users associated with the target user within the historical time period. All second nodes are nodes that have an associated relationship with a user node.

[0026] In an embodiment of the present invention, optionally, different types of nodes have corresponding credit characteristics that are different from each other. Specifically, when it is a user-type node, its credit characteristics may include but are not limited to one or more of the following: historical fulfillment rate, equipment operation specification score, credit score, etc. The historical fulfillment rate includes the quantity historical fulfillment rate and / or the fee historical fulfillment rate. The fee historical fulfillment rate is used to indicate the proportion of all fees that need to be paid for the transaction object during the historical time period that are paid on time; the quantity historical fulfillment rate is used to indicate the proportion of all contracts signed for the transaction object during the historical time period that are paid on time. When it is an equipment-type node, its credit characteristics may include but are not limited to one or more of the following: equipment type, lease duration, equipment value, etc. When it is a unit-type node, its credit characteristics may include but are not limited to one or more of the following: registered capital, credit score, industry risk level, and industry type.

[0027] 103. Determine a credit analysis result of the target user based on the credit characteristics of all second nodes and the credit characteristics of the user node.

[0028] It can be seen that implementation Figure 1 The described method determines the user's credit status by fusing multi-source data of different dimensions, such as object transaction data, device usage data, credit data, and associated user data, over a period of time, to analyze nodes and their credit characteristics in different dimensions. This can improve the accuracy of the analysis of the user's credit status, thereby improving the accuracy and efficiency of the processing of user credit decisions, such as improving the analysis accuracy of leasing access, improving the efficiency and accuracy of deposit adjustments, and improving the accuracy and reliability of overdue risk predictions.

[0029] In the embodiment of the present invention, optionally, determining a credit analysis result of a target user based on the credit characteristics of all second nodes and the credit characteristics of the user node includes: Performing a neighborhood aggregation operation on the credit features of all second nodes and the credit features of the user node to obtain a neighborhood-associated credit feature between the user node and each second node; Based on the neighborhood-associated credit features of the user node, an update operation is performed on the user credit features of the predetermined user node to obtain a credit update feature of the user node; According to a predetermined first credit weight matrix, an analysis operation is performed on the credit update feature of the user node to obtain the credit weight feature of the user node, and a binary classification operation is performed on the credit weight feature of the user node to obtain a credit analysis result of the target user.

[0030] In the embodiment of the present invention, further optionally, a neighborhood aggregation operation is performed on the credit characteristics of all second nodes and the credit characteristics of the user node to obtain the neighborhood-associated credit characteristics between the user node and each second node, including: For any first node, performing a linear transformation operation on the credit feature of the first node based on a predetermined second credit weight matrix to obtain a credit transformation feature of the first node; For any second node, determining an influence weight between the target user's user node and the second node based on the credit transformation feature of the target user's user node and the credit transformation feature of the second node; Generate a neighborhood credit feature corresponding to each second node according to the credit transformation feature of each second node and the influence weight corresponding to the second node; Neighborhood aggregation is performed on the neighborhood credit features corresponding to all second nodes to obtain the neighborhood-associated credit features between the user node and each second node.

[0031] In an embodiment of the present invention, optionally, a first node and its credit characteristics are analyzed using a graph neural network. Graph neural networks include, but are not limited to, one or more networks capable of implementing user credit analysis according to the present invention, such as recursive graph neural networks, graph convolutional neural networks, graph attention neural networks, graph autoencoders, and spatiotemporal graph neural networks. Specifically, for any first node, a credit characteristic matrix corresponding to the node is constructed, and a linear transformation operation is performed on the credit characteristic matrix of the first node based on the corresponding second credit weight matrix to obtain the credit transformation characteristics of the first node. Optionally, the second credit weight matrices corresponding to different types of nodes may be the same or different. For example, a user-type node, whose credit characteristics include historical contract fulfillment rate, device operation specification score, and credit rating, and whose constructed credit characteristic matrix is ​​[0.89 0.95 0.85] and the second credit weight matrix is ​​[0.35 0.4 0.25], then the credit transformation characteristics of the first node are 0.35 × 0.89 + 0.4 × 0.95 + 0.25 × 0.85 = 0.904.

[0032] In an embodiment of the present invention, optionally, for any second node, the credit transformation feature of the second node is concatenated with the credit transformation feature of the user node to obtain a credit concatenated feature corresponding to the second node; a dot product operation is performed on the credit concatenated feature corresponding to the second node based on a predetermined attention vector feature to obtain a credit weight feature corresponding to the second node; the credit weight feature is then analyzed using a first preset activation function to obtain a credit optimization feature corresponding to the second node; and finally, the credit optimization feature is normalized (e.g., softmax) to obtain the influence weight between the user node of the target user and the second node. The first preset activation function includes, but is not limited to, any activation function capable of achieving feature activation, such as a Leaky ReLU activation function, a ReLU activation function, a Parametric ReLU activation function, and an Exponential Linear Unit activation function, and is not limited here. For example, the credit transformation feature of user A is 0.955, the credit transformation feature of device node A is 0.938, and the corresponding attention vector feature is 0.35. The concatenated feature is 0.938 + 0.955 = 1.893, and the credit optimization feature after the Parametric ReLU activation function is LeakyReLU (0.59 * 1.893) = 0.663. After normalization, the influence weight is 0.663 / (0.663 + 0.32) = 0.674.

[0033] In an embodiment of the present invention, optionally, for any second node, the credit transformation characteristic of the second node and the influence weight corresponding to the second node are multiplied to obtain the corresponding neighborhood credit characteristic, and the neighborhood credit characteristics corresponding to all second nodes are added to achieve neighborhood aggregation, thereby obtaining the neighborhood-associated credit characteristic between the user node and each second node. For example, if the credit transformation characteristics of device node A, device node B, unit A, and unit B are 0.938, 0.896, 0.942, and 0.899, respectively, and the corresponding influence weights are 0.25, 0.35, 0.2, and 0.2, respectively, then the neighborhood-associated credit characteristic of user A is 0.938*0.25+0.896*0.35+0.942*0.2+0.899*0.2=0.9163.

[0034] In an embodiment of the present invention, optionally, the user credit feature of the target user includes the credit feature of the target user, or the credit transformation feature of the target user, or the average credit feature of the former two. Specifically, the user credit feature of the user node is spliced ​​with the neighborhood-associated credit feature of the user node to obtain the credit splicing feature of the user node, and the credit splicing feature of the user node is multiplied by the third credit weight matrix to obtain the credit weight feature of the user node, and then the credit weight feature of the user node is activated by the second preset activation function to obtain the credit update feature of the user node. The second preset activation function may be the same as or different from the first preset activation function. For the description of the second preset activation function, please refer to the above description of the first preset activation function and will not be repeated here.

[0035] In the embodiment of the present invention, optionally, the credit update feature of the user node is multiplied by the first credit weight matrix to obtain the credit weight feature of the user node.

[0036] It should be noted that the first, second, and third credit weight matrices are distinct and are all derived from training the graph neural network. The first credit weight matrix is ​​the weight matrix for the output layer of the graph neural network, while the second and third credit weight matrices are the weight matrices for the intermediate layers of the graph neural network.

[0037] As can be seen, embodiments of the present invention can also perform neighborhood aggregation on the credit characteristics of a user node and its associated nodes, then update the relevant credit characteristics of the user node based on the neighborhood aggregation results, and finally analyze the updated credit characteristics with the weights of the model output layer to obtain the user credit analysis results, thereby improving the accuracy of the credit analysis results and normalizing them for more efficient corresponding credit decision analysis, thereby improving the analysis efficiency and accuracy of user credit decisions, thereby facilitating improved dynamic response efficiency of credit decisions and timely risk control. Furthermore, by performing linear transformation on the credit characteristics of all nodes, then analyzing the credit impact of the credit characteristics of associated nodes on the user node based on the linearly transformed credit characteristics, and finally performing a comprehensive analysis of the credit impact of all associated nodes, the accuracy and reliability of the analysis of the user's credit characteristics are improved, thereby further facilitating improved accuracy and reliability of user credit analysis.

[0038] In an optional embodiment, after analyzing the credit characteristics of all first nodes and each first node corresponding to the target user based on the multi-source data of the target user, the method may further include the following steps: Analyzing, based on the multi-source data of the target user, all first nodes whose association relationships have changed among all first nodes and the association relationships between all the first nodes; Constructing nodes and node associations in a credit graph corresponding to the target user based on all first nodes whose associations have changed and the associations between all first nodes; Based on the credit analysis results of the target user and the credit characteristics of all first nodes, the node association relationship in the credit graph corresponding to the target user is updated.

[0039] In this optional embodiment, if the target user is not a new user, their credit map can be pre-built or constructed currently, preferably pre-built. If it is pre-built, the constructed credit map is updated. If the target user is a new user, their credit map is constructed currently. The credit map is constructed using the target user, the units and / or users associated with the target user, and / or the devices used by the target user as nodes, and the relationships between the target user and them as edges.

[0040] In this optional embodiment, the association relationship optionally includes at least one of a device sharing relationship, a device usage relationship, and a credit relationship. Specifically, a device sharing relationship indicates that the target user and the associated user use the same device; a device usage relationship indicates that the target user uses a certain device and / or that the devices used by the target user exhibit similar usage behavior, such as if the user uses a printer and a projector with substantially the same frequency and / or duration; and a credit relationship indicates that the target user is associated with an organization, such as if the user is a shareholder of the organization. Furthermore, optionally, when scenario one and / or scenario two occur, both indicate a change in the corresponding association relationship. Scenario one includes but is not limited to the target user newly using the target device, and / or the target user's target device adding a new user, and / or the target user's association with a new organization. Scenario two includes but is not limited to the similar usage behavior between currently used target devices. In this case, for scenario two, a corresponding first node is added to the credit graph, and the credit graph is updated; for scenario two, a connecting line is drawn between the corresponding first nodes on the credit graph, and the credit graph is updated. If the credit characteristics of the corresponding second node change, the weight of the association relationship between the first node and the corresponding second node decreases, which means that the impact on the target user is reduced.

[0041] It can be seen that this optional embodiment constructs a credit map of the user and dynamically updates the credit map based on the user's multi-source data, so as to accurately and clearly capture the explicit and implicit relationships between the user and other users, devices, and units, thereby helping to improve the accuracy and reliability of user credit analysis.

[0042] In another optional embodiment, the method may further include the following steps: Determine whether there is target data in the multi-source data that does not meet the preset data conditions; When it is determined that no such node exists, the above-mentioned operation of analyzing the credit characteristics of all first nodes and each first node corresponding to the target user based on the multi-source data of the target user is performed; When it is determined that the data exists, the target data is filtered out from the multi-source data; Performing a data optimization operation on the target data according to the data type of the target data, and performing the above-mentioned operation of analyzing the credit characteristics of all first nodes and each first node corresponding to the target user based on the multi-source data of the target user; The process of determining whether there is target data in the multi-source data that does not meet the preset data conditions includes: When the multi-source data is device usage data of a target user, the device usage data of the target user includes operation frequency data performed by the target user on each target device within a historical time period. For any target device, a sorting operation is performed on the operation frequency data of the target device based on a preset sorting method to obtain sorted operation frequency data of the target device; in the sorted operation frequency data of the target device, a first operation frequency at a first position and a second operation frequency at a second position are respectively determined according to the preset sorting method; a third operation frequency is generated based on the first operation frequency and the second operation frequency, an abnormal lower limit frequency is determined based on the third operation frequency and the first operation frequency, and an abnormal upper limit frequency is determined based on the third operation frequency and the second operation frequency; it is determined whether there is a target operation frequency outside the data range formed by the abnormal lower limit frequency and the abnormal upper limit frequency in the operation frequency data of the target device; when it is determined that the target operation frequency exists, it is determined that there is target data that does not meet the preset data condition in the multi-source data, and the target data is the target operation frequency; When the multi-source data is the credit data of the target user, determine whether there is target credit data in the credit data of the target user that does not meet the preset credit conditions. When it is determined that the target credit data exists, determine that there is target data in the multi-source data that does not meet the preset data conditions, and the target data is the target credit data.

[0043] In this optional embodiment, for any target device, its operation frequency data consists of the frequency of operations performed by the target user on the target device on each day during a historical time period. The operation frequency data is binned, i.e., the operation frequency data is divided into multiple subsets based on a preset sorting method, based on a time window (e.g., times / day). The preset sorting method includes sorting from largest to smallest or from smallest to largest.

[0044] In this optional embodiment, specifically, the first position and the second position may be frequency positions directly specified in the operating frequency data, or may be determined based on the number of frequencies in the operating frequency data. For example, for sorting from small to large, the first position may be the 25th percentile frequency position of the sorted operating frequency data of the target device, and the second position may be the 75th percentile frequency position of the sorted operating frequency data of the target device. For another example, for sorting from large to small, the first position may be the 75th percentile frequency position of the sorted operating frequency data of the target device, and the second position may be the 25th percentile frequency position of the sorted operating frequency data of the target device.

[0045] In this optional embodiment, the third operating frequency may be the difference between the second operating frequency and the first operating frequency, the abnormal lower limit frequency may be the difference between the first operating frequency and the third operating frequency at the first preset multiple, and the abnormal upper limit frequency may be the difference between the second operating frequency and the third operating frequency at the second preset multiple. For example, if user A's daily device operation frequencies over the past five days are 5, 6, 7, 8, and 20, respectively, and both the first and second preset multiples are 1.5, then the first operating frequency is calculated to be 6, the second operating frequency is 8, and the third operating frequency is 2. Therefore, the abnormal lower limit frequency is 6-1.5*2=3, and the abnormal upper limit frequency is 8+1.5*2=11. At this point, only the frequency 20 exceeds 11, so 20 is the abnormal target operating frequency.

[0046] In this optional embodiment, for the target operation frequency, different abnormal causes may have different corresponding data optimization methods. Specifically, for causes that are obviously caused by errors or untrustworthy sources, such as when the operation frequency reaches 2000 times in one day, they will be deleted. For causes that are caused by accidental fluctuations, such as when the user's operation frequency suddenly increases to 30 times on one day, but other days are normal, the average or median operation frequency of each day in the historical time period will be used as the operation frequency of the day when the abnormality occurred, such as the average operation frequency within 7 days.

[0047] In this optional embodiment, different target credit data may have different corresponding data optimization standards and data optimization processing methods. Specifically, if the registered capital of a unit is negative, it is marked as invalid; if the credit score exceeds 100 points, it is marked as abnormal and automatically corrected to 100 points. Furthermore, for abnormal situations that cannot be corrected, manual review is performed. Furthermore, the accuracy of the abnormality standard is analyzed regularly or irregularly, and the abnormality standard is optimized based on the accuracy.

[0048] It can be seen that this optional embodiment performs data anomaly analysis on multi-source data, and when anomalies are analyzed, optimizes the abnormal data according to the corresponding data type and the cause of the anomaly. For example, the user performs an anomaly analysis on the operation frequency of each device operated by the user within a period of time, and performs corresponding operations on the abnormal operation frequency according to the cause of the anomaly. This can not only remove obviously abnormal operation frequencies, but also correct occasional abnormal operation frequencies, thereby improving the data quality of multi-source data and ensuring the accuracy and reliability of the device's operation frequency data, which is conducive to further improving the accuracy and reliability of user credit analysis, and further conducive to improving the accuracy of generating and executing corresponding operations for users.

[0049] In yet another optional embodiment, before determining the credit analysis result of the target user based on the credit characteristics of all second nodes and the credit characteristics of the user node, the method may further include the following steps: Determining whether there is a corresponding missing credit feature among the credit features of all first nodes; When it is determined that there is a corresponding missing credit feature, obtaining the corresponding missing credit feature for the target user based on the missing type of the missing credit feature, and performing the above-mentioned operation of determining the credit analysis result of the target user based on the credit features of all second nodes and the credit features of the user node; Wherein, according to the missing type of the missing credit feature, the corresponding missing credit feature is obtained for the target user, including: When the missing type of missing credit feature indicates that the credit feature of the user node does not include the historical fulfillment rate of the target user, the historical fulfillment data of other users matching the target user is obtained, and the historical fulfillment rate of the target user is determined based on the historical fulfillment data of other users; When the missing type of missing credit feature is used to indicate that the credit feature of the user node does not include the device operation specification score of the target user, the target user's behavior pattern data is extracted based on the device operation data of the target device used by the target user in the historical time period; based on the target user's behavior pattern data, the behavior pattern data and device operation specification scores of multiple users matching the target user's behavior pattern data are determined from the user behavior pattern database; the behavior pattern data of each matching user is analyzed with the behavior pattern data of the target user to obtain the behavior analysis weight of each matching user; and the device operation specification score of the target user is determined based on the behavior analysis weight and the device operation specification score of each matching user.

[0050] In this optional embodiment, the target user's historical fulfillment rate can be determined based on the historical fulfillment data of other users. Specifically, the median or mean historical fulfillment rate among the historical fulfillment data of other users can be selected as the target user's historical fulfillment rate. The other users can be users who traded the same type of object as the target user, users who have previously traded for the same object, or users who are currently sharing the same object with the target user.

[0051] In this optional embodiment, the target user's behavior pattern data may optionally include the target user's device operation time distribution, such as the frequency of operations during daily peak hours, and / or the target user's operation specification index data. The target user's operation specification index data include the target user's illegal operation ratio, such as 5%, and / or the target user's device operation time interval, such as 4s. Matching users are understood to be users with similar operation time distributions or similar operation specification indicators. For any matching user, the cosine similarity between the target user and the target user in terms of device operation time distribution and / or operation specification index data is calculated as the corresponding behavior analysis weight, and the device operation specification score of each matching user and the corresponding behavior analysis weight are weighted and summed to obtain the target user's device operation specification score. For example, if user B and user C are matched with user A, and their calculated behavior analysis weights are 0.9 and 0.82 respectively, and the device operation specification scores of user B and user C are 0.8 and 0.9 respectively, then user A's device operation specification score is (0.8*0.9) + (0.9*0.82) / (0.9+0.82) = 0.85.

[0052] In this optional embodiment, further, it is determined whether there is a target time period in which device operation data is missing in the historical time period; When it is determined that the target time period exists, another time period matching the target time period is determined from the historical time period, and the device operation data of the other time period is filtered from the device operation data; The missing device operation data of the target time period is determined based on the device operation data of another time period.

[0053] In this optional embodiment, the other time period that matches the target time period can be a time period that is the same as the target time period on a date before or after the target time period. For example, if the device operation data for 11:00 AM to 1:00 PM on June 3 is missing, the average of the device operation data for 11:00 AM to 1:00 PM on June 2 and the device operation data for 11:00 AM to 1:00 PM on June 4 can be used as the device operation data for 11:00 AM to 1:00 PM on June 3.

[0054] It can be seen that this optional embodiment first analyzes the actual situation of the credit characteristics before conducting user credit analysis. If there are any missing situations, the corresponding missing credit characteristics are obtained based on the missing situations. For example, when it is analyzed that the user is missing the historical fulfillment rate and / or the device operation specification score, the missing credit characteristics of the user are analyzed and filled in by combining the historical fulfillment data of similar users and / or the behavioral pattern data of the device. This ensures the correctness of the user's relevant credit characteristics while ensuring the integrity of the credit characteristics, further improving the accuracy and reliability of the user credit analysis.

[0055] Example 2 See also Figure 2 , Figure 2 This is a flow chart of another method for analyzing user credit based on graph neural networks disclosed in an embodiment of the present invention. The method can be applied to any scenario where user credit analysis is required, such as an item rental scenario, and the scenario is provided with a corresponding device. The device includes one of a credit analysis device, a credit determination system (a local system or a cloud system), and a credit determination server (a local server or a cloud server). The credit analysis device includes but is not limited to mobile phones, computers, smart bracelets, and other devices that can perform credit data analysis. Figure 2 As shown, the method may include the following operations: 201. Collect multi-source data of a target user within a historical time period, where the multi-source data of the target user includes the target user's object transaction data, the target user's device usage data, and / or the target user's credit data, and / or the target user's associated user data.

[0056] 202. Analyze all first nodes and credit characteristics of each first node corresponding to the target user based on the multi-source data of the target user, wherein all first nodes include the user node and all second nodes of the target user; wherein all second nodes include the device nodes of all target devices used by the target user in the historical time period and / or the unit nodes of all units associated with the target user in the historical time period and / or the associated user nodes of users associated with the target user in the historical time period.

[0057] 203. Determine a credit analysis result of the target user based on the credit characteristics of all second nodes and the credit characteristics of the user node.

[0058] In the implementation of the present invention, for other descriptions of steps 201 to 203, please refer to the detailed description of steps 101 to 103 in the first embodiment, which will not be repeated here.

[0059] 204. Analyze the credit analysis result of the target user and the pre-determined credit analysis conditions to obtain a first credit analysis result of the target user.

[0060] In the embodiment of the present invention, optionally, the credit analysis condition may include a preset credit threshold. Specifically, the credit analysis result of the target user is compared with the preset credit threshold to obtain a first credit analysis result.

[0061] 205. Analyze the credit characteristics of at least one first node and the predetermined credit analysis conditions of the first node to obtain a second credit analysis result of the target user.

[0062] In the embodiment of the present invention, the credit analysis condition of each first node may include a corresponding preset credit threshold, wherein different types of nodes have different credit analysis conditions.

[0063] 206. Generate a credit decision matching the target user based on the first credit analysis result and the second credit analysis result of the target user; wherein the credit decision corresponding to the target user is used to perform control operations on the target user's transaction behavior and / or transaction data.

[0064] In an embodiment of the present invention, optionally, the credit analysis results corresponding to different categories of nodes correspond to different credit decisions. Specifically, credit decisions include but are not limited to leasing credit decisions and / or deposit adjustment credit decisions. Specifically, leasing credit decisions include but are not limited to if the user's credit score is greater than the credit score threshold but the credit feature score of its associated unit is greater than the credit feature score threshold (belonging to a high-risk industry), the system refuses to lease equipment to the user; deposit adjustment credit decisions include but are not limited to if the user's credit score is greater than the credit score threshold and its equipment operation standardization score is greater than the equipment operation score threshold, the system automatically increases its deposit-free ratio; or, if the user's credit score is greater than the credit score threshold but its associated user is overdue, the system increases the user's deposit ratio and triggers a risk warning.

[0065] It can be seen that implementation Figure 2The described method fuses multi-source data from users over a period of time, including object transaction data, device usage data, credit data, and associated user data, to analyze nodes and their credit characteristics in different dimensions, thereby determining the user's credit status. This method improves the accuracy of credit analysis, thereby enhancing the accuracy and efficiency of credit decision-making for users. For example, it improves the accuracy of analysis for rental access, the efficiency and accuracy of deposit adjustments, and the accuracy and reliability of overdue risk predictions. Furthermore, after analyzing the user's credit score, it is compared with pre-set credit conditions. The results of the comparison between the credit characteristics of the user's own node and the credit characteristics of associated nodes and the corresponding pre-set credit conditions are combined to determine the corresponding credit decision for the user. This improves the accuracy of credit decision analysis, thereby facilitating the accuracy of credit decision execution and, in turn, further improving the accuracy of leasing items, including equipment.

[0066] In an optional embodiment, before generating a credit decision matching the target user based on the first credit analysis result of the target user and the second credit analysis result of the target user, the method may further include the following steps: Collect information about items that the target user currently needs to rent, including item type, rental quantity, rental duration, and item value; Obtaining item information of all historical items rented by the target user in the past period of time, wherein the item information of each historical item includes multiple types of item type, item rental quantity, item rental duration, and item value; Compare the item information of all historical items with the information of the item currently required for rental to obtain all target historical items whose item similarity is greater than or equal to the preset item similarity; Generate target similarity analysis results based on the item information of all target historical items and the information of the currently required rental items, and obtain the target historical parameters of the target user for all target historical items; Based on the target similarity analysis results and target historical parameters, the credit decision that matches the target user is adjusted to obtain an adjusted credit decision.

[0067] In this optional embodiment, the items herein optionally include equipment and / or non-equipment. The past period of time may be completely the same as the aforementioned historical period of time, or may partially overlap with the aforementioned historical period of time, or may be a partial period of the aforementioned historical period of time.

[0068] In this optional embodiment, optionally, when the item is not a device, the target history parameter includes a target historical fulfillment rate; when the item is a device, the target history parameter includes a target historical fulfillment rate and / or a device operation specification score. The target historical fulfillment rate includes a cost history fulfillment rate and / or a quantity history fulfillment rate. The cost history fulfillment rate is used to indicate the proportion of all fees required to be paid for all target historical items that are paid on time; the quantity history fulfillment rate is used to indicate the proportion of all contracts corresponding to all target historical items that have their fees paid on time. The higher the historical fulfillment rate and / or the higher the device operation specification score, the greater the probability of allowing the current item to be rented and / or the lower the deposit ratio required to be paid, otherwise it is greater.

[0069] It can be seen that this optional embodiment adjusts the user's credit decision by conducting a comprehensive comparative analysis of the information of the items that the user currently wants to rent, the information of the items rented in the past period of time, and the corresponding historical fulfillment rates, thereby further improving the accuracy of the final determination of the credit decision, thereby further improving the analysis accuracy of rental access, improving the efficiency and accuracy of deposit adjustment, and improving the accuracy and reliability of overdue risk prediction.

[0070] In this optional embodiment, optionally, based on the item information of all target historical items and the information of the item currently required for rental, a similarity analysis result is generated, including: For any target historical item, compare and analyze the item information of the target historical item and the information of the item currently required for rental to obtain a first similarity analysis result; For any content included in the information of the item to be rented, extract the item features of the content, and normalize the item features to obtain a corresponding first normalization result; For any content included in the item information of any target historical item, extract the item features of the content, normalize the item features to obtain a corresponding second normalized result, and analyze the first normalized result and the second normalized result to obtain an intermediate similarity analysis result; Analyze the intermediate similarity analysis results of all target historical items to obtain a second similarity analysis result, and generate a standard similarity analysis result based on the first similarity analysis result and the second similarity analysis result.

[0071] In this optional embodiment, the normalization of item features is optionally described in the first embodiment and will not be repeated here. Whether the first similarity analysis result, the second similarity analysis result, the intermediate similarity analysis result, or the target similarity analysis result, can be a quantified value, where a higher quantized value indicates a higher similarity, a greater probability of allowing the current item to be rented, and / or a lower deposit ratio required; otherwise, a higher quantized value indicates a higher similarity.

[0072] It can be seen that this optional embodiment analyzes the similarity between each historical rental item and the currently required rental item from two perspectives: item information and item characteristics. For item characteristics, the similarity between each historical rental item and the currently required rental item is analyzed separately, and then the similarity analysis results of all historical rental items are combined to jointly determine the similarity analysis results between the historical rental items and the currently required rental items. This improves the accuracy of the determination of the similarity analysis results, thereby further improving the accuracy and reliability of the adjustment of the user's credit decision.

[0073] Example 3 See also Figure 3 , Figure 3 This is a schematic diagram of the structure of a device for analyzing user credit based on a graph neural network disclosed in an embodiment of the present invention. The device can be applied to any scenario where user credit analysis is required, such as an item rental scenario. The device includes a credit analysis device, a credit determination system (a local system or a cloud system), and a credit determination server (a local server or a cloud server). The credit analysis device includes but is not limited to mobile phones, computers, smart bracelets, and other devices that can perform credit data analysis. Figure 3 As shown, the device includes: A collection module 301 is configured to collect multi-source data of a target user within a historical period, where the multi-source data of the target user includes object transaction data of the target user, device usage data of the target user, and / or credit data of the target user, and / or associated user data of the target user; An analysis module 302 is configured to analyze, based on the multi-source data of the target user, all first nodes corresponding to the target user and the credit characteristics of each first node, wherein all first nodes include the user node and all second nodes of the target user; wherein all second nodes include the device nodes of all target devices used by the target user within a historical time period and / or the unit nodes of all units associated with the target user within the historical time period and / or the associated user nodes of users associated with the target user within the historical time period; The determination module 303 is configured to determine a credit analysis result of the target user based on the credit characteristics of all second nodes and the credit characteristics of the user node.

[0074] It can be seen that implementation Figure 3The described device determines the user's credit status by fusing multi-source data of the user over a period of time, such as object transaction data, device usage data, credit data, associated user data, and other data of different dimensions, to perform node and credit feature analysis of different dimensions. This can improve the accuracy of the analysis of the user's credit status, thereby improving the accuracy and efficiency of the processing of user credit decisions, such as improving the analysis accuracy of rental access, improving the efficiency and accuracy of deposit adjustments, and improving the accuracy and reliability of overdue risk prediction.

[0075] In the embodiment of the present invention, optionally, the specific manner in which the determination module 303 determines the credit analysis result of the target user based on the credit characteristics of all second nodes and the credit characteristics of the user node includes: Performing a neighborhood aggregation operation on the credit features of all second nodes and the credit features of the user node to obtain a neighborhood-associated credit feature between the user node and each second node; Based on the neighborhood-associated credit features of the user node, an update operation is performed on the user credit features of the predetermined user node to obtain a credit update feature of the user node; According to a predetermined first credit weight matrix, an analysis operation is performed on the credit update feature of the user node to obtain the credit weight feature of the user node, and a binary classification operation is performed on the credit weight feature of the user node to obtain a credit analysis result of the target user.

[0076] In the embodiment of the present invention, further optionally, the determination module 303 performs a neighborhood aggregation operation on the credit characteristics of all second nodes and the credit characteristics of the user node to obtain a specific manner of neighborhood-associated credit characteristics between the user node and each second node, including: For any first node, performing a linear transformation operation on the credit feature of the first node based on a predetermined second credit weight matrix to obtain a credit transformation feature of the first node; For any second node, determining an influence weight between the target user's user node and the second node based on the credit transformation feature of the target user's user node and the credit transformation feature of the second node; Generate a neighborhood credit feature corresponding to each second node according to the credit transformation feature of each second node and the influence weight corresponding to the second node; Performing neighborhood aggregation on the neighborhood credit features corresponding to all second nodes to obtain the neighborhood-associated credit features between the user node and each second node; The user credit characteristics of the target user include the credit characteristics of the target user or the credit transformation characteristics of the target user; The specific manner in which the determination module determines, for any second node, the influence weight between the target user's user node and the second node based on the credit conversion characteristics of the target user's user node and the credit conversion characteristics of the second node includes: For any second node, concatenate the credit transformation feature of the second node with the credit transformation feature of the user node to obtain the credit concatenation feature corresponding to the second node; Performing a dot product operation on the credit concatenation feature corresponding to the second node based on the predetermined attention vector feature to obtain a credit weight feature corresponding to the second node; The credit weight feature corresponding to the second node is analyzed through the first preset activation function to obtain the credit optimization feature corresponding to the second node, and the credit optimization feature is normalized to obtain the influence weight between the user node and the second node.

[0077] As can be seen, embodiments of the present invention can also perform neighborhood aggregation on the credit characteristics of a user node and its associated nodes, then update the relevant credit characteristics of the user node based on the neighborhood aggregation results, and finally analyze the updated credit characteristics with the weights of the model output layer to obtain the user credit analysis results, thereby improving the accuracy of the credit analysis results and normalizing them for more efficient corresponding credit decision analysis, thereby improving the analysis efficiency and accuracy of user credit decisions, thereby facilitating improved dynamic response efficiency of credit decisions and timely risk control. Furthermore, by performing linear transformation on the credit characteristics of all nodes, then analyzing the credit impact of the credit characteristics of associated nodes on the user node based on the linearly transformed credit characteristics, and finally performing a comprehensive analysis of the credit impact of all associated nodes, the accuracy and reliability of the analysis of the user's credit characteristics are improved, thereby further facilitating improved accuracy and reliability of user credit analysis.

[0078] In an optional embodiment, Figure 4 This is a schematic diagram of the structure of another device for analyzing user credit based on graph neural network disclosed in an embodiment of the present invention. Figure 4 As shown, the device may also include: Comparison module 304, used to compare the credit analysis result of the target user with the pre-determined credit analysis conditions to obtain a first credit comparison result of the target user; The comparison module 304 is further configured to compare the credit characteristics of at least one first node with a predetermined credit analysis condition of the first node to obtain a second credit comparison result of the target user; A generating module 305 is configured to generate a credit decision matching the target user based on the first credit comparison result of the target user and the second credit comparison result of the target user; The credit decision corresponding to the target user is used to perform control operations on the target user's transaction behavior and / or transaction data.

[0079] It can be seen that implementation Figure 4 After analyzing the user's credit score, the described device compares it with the preset credit conditions, and comprehensively compares the credit characteristics of its own node and the credit characteristics of associated nodes with the corresponding preset credit conditions to jointly determine the corresponding credit decision for the user, thereby improving the analysis accuracy of the credit decision, thereby facilitating the improvement of the execution accuracy of the credit decision, and further facilitating the improvement of the rental accuracy of items including equipment.

[0080] In another optional embodiment, Figure 4 As shown, the analysis module 302 is further configured to analyze all first nodes corresponding to the target user and the credit characteristics of each first node based on the multi-source data of the target user, and then analyze all first nodes with changed association relationships among all the first nodes and the association relationships between all the first nodes based on the multi-source data of the target user; like Figure 4 As shown, the device may also include: A construction module 306 is configured to construct nodes and their associations in the credit graph corresponding to the target user based on all the first nodes whose associations have changed and the associations between the first nodes; An updating module 307 is configured to update the node association relationship in the credit graph corresponding to the target user based on the credit analysis result of the target user and the credit characteristics of all the first nodes; The association relationship includes at least one of a device sharing relationship, a device usage relationship, and a credit relationship; The device sharing relationship indicates that the target user and the associated user use the same device. The device usage relationship indicates that the target user uses a certain device and / or that the devices used by the target users have similar usage behaviors. The credit relationship indicates that the target user has an associated organization. It can be seen that implementation Figure 4 The described device constructs a user's credit map and dynamically updates the credit map based on the user's multi-source data to accurately and clearly capture the explicit and implicit relationships between the user and other users, devices, and units, thereby improving the accuracy and reliability of user credit analysis.

[0081] In another optional embodiment, Figure 4 As shown, the device may also include: The first determination module 308 is configured to determine whether there is target data in the multi-source data that does not meet the preset data conditions; if it is determined that there is no target data, the analysis module 302 is triggered to perform the above-mentioned operation of analyzing the credit characteristics of all first nodes and each first node corresponding to the target user based on the multi-source data of the target user; The screening module 309 is used to screen out target data from the multi-source data when it is determined that the target data exists; The data optimization module 310 is configured to perform a data optimization operation on the target data according to the data type of the target data, and trigger the analysis module 302 to perform the above-mentioned operation of analyzing the credit characteristics of all first nodes and each first node corresponding to the target user based on the multi-source data of the target user; The specific method of the first determination module 308 determining whether there is target data that does not meet the preset data condition in the multi-source data includes: When the multi-source data is device usage data of a target user, the device usage data of the target user includes operation frequency data performed by the target user on each target device within a historical time period. For any target device, a sorting operation is performed on the operation frequency data of the target device based on a preset sorting method to obtain sorted operation frequency data of the target device, wherein the preset sorting method includes sorting from large to small or sorting from small to large. In the sorted operation frequency data of the target device, a first operation frequency at a first position and a second operation frequency at a second position are respectively determined according to the preset sorting method. A third operation frequency is generated based on the first operation frequency and the second operation frequency. An abnormal lower limit frequency is determined based on the third operation frequency and the first operation frequency, and an abnormal upper limit frequency is determined based on the third operation frequency and the second operation frequency. It is determined whether there is a target operation frequency outside the data range formed by the abnormal lower limit frequency and the abnormal upper limit frequency in the operation frequency data of the target device. When it is determined that the target operation frequency exists, it is determined that there is target data that does not meet the preset data condition in the multi-source data, and the target data is the target operation frequency. When the multi-source data is the credit data of the target user, determine whether there is target credit data in the credit data of the target user that does not meet the preset credit conditions. When it is determined that the target credit data exists, determine that there is target data in the multi-source data that does not meet the preset data conditions, and the target data is the target credit data.

[0082] It can be seen that implementation Figure 4The described device performs data anomaly analysis on multi-source data, and when anomalies are found, optimizes the abnormal data according to the corresponding data type and cause of the anomaly. For example, an anomaly analysis is performed on the operation frequency of each device operated by the user within a period of time, and corresponding operations are performed on the abnormal operation frequency according to the cause of the anomaly. This can not only remove obviously abnormal operation frequencies, but also correct occasional abnormal operation frequencies, thereby improving the data quality of multi-source data and ensuring the accuracy and reliability of the device's operation frequency data, which is conducive to further improving the accuracy and reliability of user credit analysis, and further conducive to improving the accuracy of generating and executing corresponding operations for users.

[0083] In another optional embodiment, Figure 4 As shown, the device may also include: A second judgment module 311 is configured to determine whether there is a corresponding missing credit feature among the credit features of all first nodes before the determination module 303 determines the credit analysis result of the target user based on the credit features of all second nodes and the credit features of the user node; Acquisition module 312 is configured to, when determining that there are corresponding missing credit features, acquire the corresponding missing credit features for the target user based on the missing type of the missing credit features, and trigger determination module 303 to perform the above-mentioned operation of determining the credit analysis result of the target user based on the credit features of all second nodes and the credit features of the user node; The specific method of obtaining the missing credit feature corresponding to the target user by the acquisition module 312 according to the missing type of the missing credit feature includes: When the missing type of missing credit feature indicates that the credit feature of the user node does not include the historical fulfillment rate of the target user, the historical fulfillment data of other users matching the target user is obtained; and the historical fulfillment rate of the target user is determined based on the historical fulfillment data of other users; When the missing type of missing credit feature is used to indicate that the credit feature of the user node does not include the device operation specification score of the target user, the target user's behavior pattern data is extracted based on the device operation data of the target device used by the target user in the historical time period; based on the target user's behavior pattern data, the behavior pattern data and device operation specification scores of multiple users matching the target user's behavior pattern data are determined from the user behavior pattern database; the behavior pattern data of each matching user is analyzed with the behavior pattern data of the target user to obtain the behavior analysis weight of each matching user; and the device operation specification score of the target user is determined based on the behavior analysis weight and the device operation specification score of each matching user.

[0084] It can be seen that implementation Figure 4Before performing user credit analysis, the described device first analyzes the actual situation of the credit characteristics. If there are any missing situations, the corresponding missing credit characteristics are obtained based on the missing situations. For example, when it is analyzed that the user is missing the historical fulfillment rate and / or the device operation specification score, the missing credit characteristics of the user are analyzed and filled in by combining the historical fulfillment data of similar users and / or the behavioral pattern data of the device. This ensures the correctness of the user's relevant credit characteristics while ensuring the integrity of the credit characteristics, further improving the accuracy and reliability of the user's credit analysis.

[0085] In another optional embodiment, Figure 4 As shown, the collection module 301 is further configured to collect information on items currently required to be rented by the target user, including multiple items such as item type, item rental quantity, item rental duration, and item value, before the generation module 305 generates a credit decision matching the target user based on the first credit analysis result of the target user and the second credit analysis result of the target user; and obtain item information on all historical items rented by the target user in the past period of time, wherein the item information of each historical item includes multiple items such as item type, item rental quantity, item rental duration, and item value; The analysis module 302 is further configured to compare the item information of all historical items with the information of the item currently required for rental, and obtain all target historical items whose item similarity is greater than or equal to a preset item similarity; The generating module 305 is further used to generate target similarity analysis results based on the item information of all target historical items and the information of the item currently required for rental, and obtain the target history parameters of the target user for all target historical items; like Figure 4 As shown, the device may also include: The adjustment module 313 is used to adjust the credit decision matching the target user according to the target similarity analysis result and the target historical parameters to obtain an adjusted credit decision.

[0086] In this optional embodiment, optionally, when the item is not a device, the target history parameter includes a target historical fulfillment rate; when the item is a device, the target history parameter includes a target historical fulfillment rate and / or a device operation specification score. The target historical fulfillment rate includes a cost history fulfillment rate and / or a quantity history fulfillment rate. The cost history fulfillment rate is used to indicate the proportion of all fees required to be paid for all target historical items that are paid on time; the quantity history fulfillment rate is used to indicate the proportion of all contracts corresponding to all target historical items that have their fees paid on time. The higher the historical fulfillment rate and / or the higher the device operation specification score, the greater the probability of allowing the current item to be rented and / or the lower the deposit ratio required to be paid, otherwise it is greater.

[0087] It can be seen that this optional embodiment adjusts the user's credit decision by conducting a comprehensive comparative analysis of the information of the items that the user currently wants to rent, the information of the items rented in the past period of time, and the corresponding historical fulfillment rates, thereby further improving the accuracy of the final determination of the credit decision, thereby further improving the analysis accuracy of rental access, improving the efficiency and accuracy of deposit adjustment, and improving the accuracy and reliability of overdue risk prediction.

[0088] In this optional embodiment, the specific method of generating the similarity analysis result by the generating module 305 based on the item information of all target historical items and the information of the item currently required to be rented includes: For any target historical item, compare and analyze the item information of the target historical item and the information of the item currently required for rental to obtain a first similarity analysis result; For any content included in the information of the item to be rented, extract the item features of the content, and normalize the item features to obtain a corresponding first normalization result; For any content included in the item information of any target historical item, extract the item features of the content, normalize the item features to obtain a corresponding second normalized result, and analyze the first normalized result and the second normalized result to obtain an intermediate similarity analysis result; Analyze the intermediate similarity analysis results of all target historical items to obtain a second similarity analysis result, and generate a standard similarity analysis result based on the first similarity analysis result and the second similarity analysis result.

[0089] In this optional embodiment, the normalization of item features is optionally described in the first embodiment and will not be repeated here. Whether the first similarity analysis result, the second similarity analysis result, the intermediate similarity analysis result, or the target similarity analysis result, can be a quantified value, where a higher quantized value indicates a higher similarity, a greater probability of allowing the current item to be rented, and / or a lower deposit ratio required; otherwise, a higher quantized value indicates a higher similarity.

[0090] It can be seen that this optional embodiment analyzes the similarity between each historical rental item and the currently required rental item from two perspectives: item information and item characteristics. For item characteristics, the similarity between each historical rental item and the currently required rental item is analyzed separately, and then the similarity analysis results of all historical rental items are combined to jointly determine the similarity analysis results between the historical rental items and the currently required rental items. This improves the accuracy of the determination of the similarity analysis results, thereby further improving the accuracy and reliability of the adjustment of the user's credit decision.

[0091] Example 4 See also Figure 5 , Figure 5 This is a structural diagram of a credit analysis device disclosed in an embodiment of the present invention. The credit analysis device can be applied to any scenario where user credit analysis is required, such as an item rental scenario. The credit analysis device includes but is not limited to mobile phones, computers, smart bracelets and other devices that can perform credit data analysis. Figure 5 As shown, the device may include: A memory 401 storing executable program code; a processor 402 coupled to the memory 401; Furthermore, it may also include an input interface 403 and an output interface 404 coupled to the processor 402; The processor 402 calls the executable program code stored in the memory 401 to execute the steps of the method for analyzing user credit based on a graph neural network described in Example 1 or Example 2.

[0092] Example 5 An embodiment of the present invention discloses a computer storage medium storing computer instructions. When the computer instructions are called, they are used to execute the steps of the method for analyzing user credit based on a graph neural network described in Example 1 or Example 2.

[0093] Example 6 An embodiment of the present invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute the steps of the method for analyzing user credit based on a graph neural network described in Example 1 or Example 2.

[0094] The device embodiments described above are merely illustrative. Modules described as separate components may or may not be physically separate, and components shown as modules may or may not be physical modules, i.e., they may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0095] Through the detailed description of the above embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by means of hardware. Based on this understanding, the above technical solution, in essence, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM), or other optical disk storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0096] Finally, it should be noted that the method, device and equipment for analyzing user credit based on graph neural network disclosed in the embodiments of the present invention are only preferred embodiments of the present invention, which are only used to illustrate the technical solutions of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features therein can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for analyzing user credit based on graph neural network, characterized in that: The method comprises: Collecting multi-source data of a target user within a historical time period, wherein the multi-source data of the target user includes object transaction data of the target user, device usage data of the target user, and / or credit data of the target user, and / or associated user data of the target user; Analyzing, based on the multi-source data of the target user, all first nodes corresponding to the target user and the credit characteristics of each first node, wherein all first nodes include the user node and all second nodes of the target user; wherein all second nodes include the device nodes of all target devices used by the target user within the historical time period and / or the unit nodes of all units associated with the target user within the historical time period and / or the associated user nodes of users associated with the target user within the historical time period; A credit analysis result of the target user is determined based on the credit characteristics of all the second nodes and the credit characteristics of the user node.

2. The method for analyzing user credit based on graph neural network according to claim 1, characterized in that: The step of determining a credit analysis result of the target user based on the credit characteristics of all the second nodes and the credit characteristics of the user node includes: Performing a neighborhood aggregation operation on the credit characteristics of all second nodes and the credit characteristics of the user node to obtain a neighborhood-associated credit characteristic between the user node and each second node; Based on the neighborhood-associated credit feature of the user node, performing an update operation on the predetermined user credit feature of the user node to obtain a credit update feature of the user node; According to a predetermined first credit weight matrix, an analysis operation is performed on the credit update feature of the user node to obtain the credit weight feature of the user node, and a binary classification operation is performed on the credit weight feature of the user node to obtain a credit analysis result of the target user.

3. The method for analyzing user credit based on graph neural network according to claim 2, characterized in that: The performing a neighborhood aggregation operation on the credit characteristics of all the second nodes and the credit characteristics of the user node to obtain a neighborhood-associated credit characteristic between the user node and each of the second nodes includes: For any of the first nodes, performing a linear transformation operation on the credit characteristics of the first node based on a predetermined second credit weight matrix to obtain a credit transformation characteristic of the first node; For any second node, determining an influence weight between the user node of the target user and the second node based on the credit transformation feature of the user node of the target user and the credit transformation feature of the second node; generating a neighborhood credit feature corresponding to each second node according to the credit transformation feature of each second node and the influence weight corresponding to the second node; Performing neighborhood aggregation on the neighborhood credit features corresponding to all the second nodes to obtain a neighborhood-associated credit feature between the user node and each of the second nodes; The user credit characteristics of the target user include the credit characteristics of the target user or the credit conversion characteristics of the target user; The step of determining, for any second node, the influence weight between the user node of the target user and the second node based on the credit conversion feature of the user node of the target user and the credit conversion feature of the second node includes: For any second node, concatenate the credit conversion feature of the second node with the credit conversion feature of the user node to obtain a credit concatenation feature corresponding to the second node; Performing a dot product operation on the credit concatenation feature corresponding to the second node based on the predetermined attention vector feature to obtain a credit weight feature corresponding to the second node; The credit weight feature corresponding to the second node is analyzed through a first preset activation function to obtain a credit optimization feature corresponding to the second node, and the credit optimization feature is normalized to obtain an influence weight between the user node and the second node.

4. The method for analyzing user credit based on graph neural network according to any one of claims 1 to 3, characterized in that: The method further comprises: Analyzing the credit analysis result of the target user and a predetermined credit analysis condition to obtain a first credit analysis result of the target user; analyzing the credit characteristics of at least one of the first nodes and a predetermined credit analysis condition for the first node to obtain a second credit analysis result of the target user; generating a credit decision matching the target user based on the first credit analysis result of the target user and the second credit analysis result of the target user; The credit decision corresponding to the target user is used to perform control operations on the transaction behavior and / or transaction data of the target user.

5. The method for analyzing user credit based on graph neural network according to any one of claims 1 to 3, characterized in that: After analyzing the credit characteristics of all first nodes and each first node corresponding to the target user based on the multi-source data of the target user, the method further includes: Analyzing, based on the multi-source data of the target user, all first nodes whose association relationships have changed among all the first nodes and the association relationships between all the first nodes; Constructing nodes and node associations in a credit graph corresponding to the target user based on all the first nodes whose associations have changed and the associations between the first nodes; Update the node association relationship in the credit graph corresponding to the target user based on the credit analysis result of the target user and the credit characteristics of all the first nodes; The association relationship includes at least one of a device sharing relationship, a device usage relationship, and a credit relationship; Among them, the device sharing relationship is used to indicate that the target user and the associated user use the same device, the device usage relationship is used to indicate that the target user uses a certain device and / or similar usage behaviors occur between the devices used by the target users, and the credit relationship is used to indicate that the target user has an associated unit.

6. The method for analyzing user credit based on graph neural network according to any one of claims 1 to 3, characterized in that: The method further comprises: Determining whether there is target data in the multi-source data that does not meet the preset data conditions; When it is determined that no such node exists, performing the operation of analyzing the credit characteristics of all first nodes and each first node corresponding to the target user based on the multi-source data of the target user; When it is determined that the target data exists, the target data is filtered out from the multi-source data; performing a data optimization operation on the target data according to the data type of the target data, and performing the operation of analyzing the credit characteristics of all first nodes and each first node corresponding to the target user based on the multi-source data of the target user; The step of determining whether there is target data in the multi-source data that does not meet a preset data condition includes: When the multi-source data is device usage data of the target user, the device usage data of the target user includes operation frequency data performed by the target user on each target device within a historical time period. For any target device, a sorting operation is performed on the operation frequency data of the target device based on a preset sorting method to obtain sorted operation frequency data of the target device, wherein the preset sorting method includes sorting from large to small or sorting from small to large. In the sorted operation frequency data of the target device, a first operation frequency at a first position and a second operation frequency at a second position are respectively determined according to the preset sorting method. A third operation frequency is generated based on the first operation frequency and the second operation frequency. An abnormal lower limit frequency is determined based on the third operation frequency and the first operation frequency, and an abnormal upper limit frequency is determined based on the third operation frequency and the second operation frequency. It is determined whether a target operation frequency exists in the operation frequency data of the target device and is outside a data range formed by the abnormal lower limit frequency and the abnormal upper limit frequency. If the target operation frequency exists, it is determined that target data that does not meet the preset data condition exists in the multi-source data, and the target data is the target operation frequency. When the multi-source data is the credit data of the target user, it is determined whether there is target credit data that does not meet the preset credit conditions in the credit data of the target user. When it is determined that the target credit data exists, it is determined that there is target data that does not meet the preset data conditions in the multi-source data, and the target data is the target credit data.

7. The method for analyzing user credit based on graph neural network according to any one of claims 1 to 3, characterized in that: Before determining the credit analysis result of the target user based on the credit characteristics of all the second nodes and the credit characteristics of the user node, the method further includes: Determining whether there is a corresponding missing credit feature among all the credit features of the first nodes; When it is determined that the corresponding missing credit feature exists, obtaining the corresponding missing credit feature for the target user according to the missing type of the missing credit feature, and performing the operation of determining the credit analysis result of the target user based on the credit features of all the second nodes and the credit features of the user node; The step of obtaining the corresponding missing credit feature for the target user according to the missing type of the missing credit feature includes: When the missing type of the missing credit feature indicates that the credit feature of the user node does not include the historical performance rate of the target user, obtaining historical performance data of other users matching the target user, and determining the historical performance rate of the target user based on the historical performance data of the other users; When the missing type of the missing credit feature is used to indicate that the credit feature of the user node does not include the device operation specification score of the target user, the behavior pattern data of the target user is extracted based on the device operation data of the target device used by the target user during the historical time period; based on the behavior pattern data of the target user, the behavior pattern data and device operation specification scores of multiple users matching the behavior pattern data of the target user are determined from a user behavior pattern database; the behavior pattern data of each of the matching users is analyzed with the behavior pattern data of the target user to obtain a behavior analysis weight of each of the matching users; and the device operation specification score of the target user is determined based on the behavior analysis weight and the device operation specification score of each of the matching users.

8. A device for analyzing user credit based on graph neural network, characterized in that: The device comprises: a collection module, configured to collect multi-source data of a target user within a historical time period, wherein the multi-source data of the target user includes object transaction data of the target user, device usage data of the target user, and / or credit data of the target user, and / or associated user data of the target user; an analysis module, configured to analyze, based on the multi-source data of the target user, all first nodes corresponding to the target user and the credit characteristics of each first node, wherein all first nodes include the user node and all second nodes of the target user; wherein all second nodes include the device nodes of all target devices used by the target user within the historical time period and / or the unit nodes of all units associated with the target user within the historical time period and / or the associated user nodes of users associated with the target user within the historical time period; A determination module is used to determine a credit analysis result of the target user based on the credit characteristics of all the second nodes and the credit characteristics of the user node.

9. A credit analysis device, characterized in that: The device comprises: a memory storing executable program code; a processor coupled to a memory; The processor calls the executable program code stored in the memory to execute the method for analyzing user credit based on a graph neural network as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that The computer storage medium stores computer instructions, which, when called, are used to execute the method for analyzing user credit based on a graph neural network as described in any one of claims 1 to 7.

Images