An overdue risk assessment method, device, equipment and readable storage medium

Abstract

The application discloses a method for overdue risk assessment, which obtains coordinate data of a user in a specified period, mines features of the coordinates according to associated financial information on one hand, and proposes a composition scheme on the other hand, taking the user and the coordinates as nodes, connecting the users according to a space-time correlation between the users and the coordinates to build a bidirectional edge, the connected users often have similarities, information can be transmitted to the users through the latitude and longitude coordinates themselves, or information can be transmitted to the users through neighbors, so as to comprehensively assess the risk of the users as much as possible, the method builds a graph structure of the users and the coordinates through the latitude and longitude coordinates, and then models the users through a graph neural network model, so that the risk of the users can be effectively modeled, and the overdue risk identification capability of the users is improved. The application also discloses an overdue risk assessment device, equipment and a readable storage medium, which have corresponding technical effects.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, apparatus, device, and readable storage medium for assessing overdue risk. Background Technology

[0002] Overdue risk assessment refers to evaluating a user's future risk of defaulting on payments based on various dimensions of user information, using machine learning or deep learning methods to predict the probability of default. In existing technologies, the latitude and longitude coordinates of a user's locations are crucial for risk assessment. Specifically, current methods often encode users by identifying points of interest (POIs) near their geographical location, and then use neural networks to extract information from these POIs to characterize the user's overdue risk. While this method can identify user risk to some extent, the risk information it reflects is relatively limited in practical applications, leading to less accurate predictions.

[0003] In conclusion, how to more comprehensively assess users' delinquency risks and improve the accuracy of modeling are technical problems that urgently need to be solved by those skilled in the art. Summary of the Invention

[0004] The purpose of this invention is to provide a method, apparatus, device, and readable storage medium for assessing overdue risk, so as to more comprehensively assess users' overdue risk and improve the accuracy of modeling.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] A method for assessing overdue risk includes:

[0007] Obtain the user's coordinate data within a specified period, and perform data preprocessing to obtain coordinate values;

[0008] The user is taken as a user node, and the corresponding risk information is obtained as a node feature; the coordinate value is taken as a coordinate node, and the location-related financial information is obtained as a node feature.

[0009] Determine the spatiotemporal relationships between users and between users and coordinate values, construct bidirectional edges between nodes, generate graph data, and set user tags based on overdue status;

[0010] The graph neural network is invoked to train the user to identify overdue risks based on the graph data.

[0011] The system receives graph data from the target user as target graph data, calls the pre-trained graph neural network to identify overdue risks from the target graph data, and generates an evaluation result.

[0012] Optionally, determining the spatiotemporal relationships between users and between users and coordinate values, and constructing bidirectional edges between nodes, includes:

[0013] Based on the number of days the user stays at the coordinate value, construct a bidirectional edge between the user node and the coordinate node;

[0014] Based on the simultaneous spatial and temporal dwell time of the users, construct bidirectional edges between the user nodes.

[0015] Optionally, constructing a bidirectional edge between user nodes based on the simultaneous spatiotemporal dwell time of the users includes:

[0016] If the number of times the first user and the second user appear at the first coordinate is greater than the first threshold, then a high-frequency connection edge is created between the first user node and the second user node, and the number of times is used as the feature of the edge.

[0017] If the first user and the second user have overlapping locations within the same time period, a spatiotemporal connection edge is created between the first user node and the second user node, and the number of times the locations overlap is used as the feature of the edge.

[0018] Optionally, the step of using the coordinate values ​​as coordinate nodes and obtaining location-related financial information as node features includes:

[0019] Use the coordinate values ​​as coordinate nodes;

[0020] Extract the financial attribute interest point data corresponding to the coordinate values ​​and generate a numerical vector;

[0021] Determine the overdue user characteristics historically associated with the coordinate values, and use them as an overdue user vector;

[0022] The numerical vector is combined with the overdue user vector to form the node feature corresponding to the coordinate node.

[0023] Optionally, the step of extracting the financial attribute interest point data corresponding to the coordinate values ​​and generating a numerical vector includes:

[0024] Determine the coordinates and points of interest encoded by geographic location hash values;

[0025] Based on the location of the points of interest, extract the financial attribute points of interest that are located within the preset range of the coordinate values;

[0026] The numerical vector is generated based on the financial attribute interest points.

[0027] Optionally, obtaining the user's latitude and longitude coordinates and performing data preprocessing to obtain coordinate values ​​includes:

[0028] Obtain the user's latitude and longitude coordinates and filter out outliers;

[0029] The latitude and longitude coordinates are blurred and aligned to obtain the coordinate values.

[0030] Optionally, after determining the spatiotemporal relationships between users and between users and coordinate values, and constructing bidirectional edges between nodes, the method further includes:

[0031] Determine if an isolated node exists;

[0032] If they exist, construct self-loop connections for the isolated nodes to generate graph data.

[0033] An overdue risk assessment device, comprising:

[0034] The preprocessing unit is used to acquire the user's coordinate data within a specified period and perform data preprocessing to obtain coordinate values;

[0035] The node unit is used to take the user as a user node and obtain the corresponding risk information as a node feature; and to take the coordinate value as a coordinate node and obtain the location-related financial information as a node feature.

[0036] The connection unit is used to determine the spatiotemporal relationships between users and between users and coordinate values, construct bidirectional edges between nodes, generate graph data, and set user tags according to the overdue status.

[0037] The identification training unit is used to call the graph neural network to train the user to identify overdue risks based on the graph data.

[0038] The evaluation unit is used to receive graph data from the target user as target graph data, call the pre-trained graph neural network to identify overdue risks from the target graph data, and generate evaluation results.

[0039] A computer device, comprising:

[0040] Memory, used to store computer programs;

[0041] A processor is used to implement the steps of the above-described overdue risk assessment method when executing the computer program.

[0042] A readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described overdue risk assessment method.

[0043] The method provided in this invention acquires the coordinate data of a user within a specified period. Starting from the location, it mines the features of the coordinates based on associated financial information. Furthermore, it proposes a graph construction scheme, using users and coordinates as nodes. Based on the spatiotemporal correlation between users and between users and coordinate values, bidirectional edges are constructed to connect users. These connected users often have similarities, and information can be transmitted to users both through latitude and longitude coordinates and through neighbors, in order to comprehensively assess user risk. This method constructs a graph structure of users and coordinates using latitude and longitude coordinates, connecting users in a graph form through latitude and longitude information. Then, a graph neural network model is used to model users, effectively modeling user risk and thus improving the ability to identify user delinquency risk.

[0044] Accordingly, embodiments of the present invention also provide an overdue risk assessment device, equipment, and readable storage medium corresponding to the above-described overdue risk assessment method, which have the above-described technical effects, and will not be described in detail here. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart illustrating the implementation of an overdue risk assessment method in an embodiment of the present invention.

[0047] Figure 2 This is a schematic diagram of coordinate data in an embodiment of the present invention;

[0048] Figure 3 This is a schematic diagram of a task in an embodiment of the present invention;

[0049] Figure 4 This is a schematic diagram of a high-frequency connection in an embodiment of the present invention;

[0050] Figure 5 This is a schematic diagram of a mapping scheme based on high-frequency connectivity and spatiotemporal connectivity in an embodiment of the present invention;

[0051] Figure 6 This is a schematic diagram of the structure of an overdue risk assessment device according to an embodiment of the present invention;

[0052] Figure 7 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0053] The core of this invention is to provide an overdue risk assessment method that can more comprehensively assess a user's overdue risk and has high modeling accuracy.

[0054] To enable those skilled in the art to better understand the present invention, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] Please refer to Figure 1 , Figure 1 This is a flowchart of an overdue risk assessment method according to an embodiment of the present invention. This method can improve the ability of the risk control score in the original scheme to distinguish between positive and negative samples by using location information data. The method mainly includes the following steps:

[0056] S101. Obtain the user's coordinate data (latitude and longitude) within a specified period, and perform data preprocessing to obtain coordinate values;

[0057] Obtain user coordinate data within a specified period, where the specified period refers to the statistical timeframe. For example, if user tags (whether an item is overdue) are generated monthly, then the smallest unit of data is one month. Since this application involves modeling historical information of coordinate data, it is recommended that the training data period be at least one year. The validation set can be one month prior to the training set, and the test set spans all remaining months of the entire dataset. For instance, if the entire dataset spans from January 2020 to December 2021, then the data from January 2020 to December 2020 can be used as the training set, the data from January 2021 as the validation set, and the remaining data as the test set. It should be noted that the data referred to here is tag data. For example, the validation set may only contain tags from January 2021, but the latitude and longitude data used for mapping can be all data up to January 2021 to ensure information richness.

[0058] The coordinate data in the acquired historical data undergoes preprocessing to remove outliers and ensure consistent format accuracy. This embodiment does not limit the specific preprocessing methods used; for example, it may remove coordinate data with excessively low accuracy or invalid coordinates. The methods can be set according to the accuracy requirements of the actual data analysis, as detailed in the following embodiments, which will not be elaborated upon here. In this embodiment, the preprocessed coordinate data is referred to as coordinate values ​​to distinguish it from the original coordinate data.

[0059] S102. Treat users as user nodes and obtain corresponding risk information as node features; treat coordinate values ​​as coordinate nodes and obtain location-related financial information as node features.

[0060] The main objective of the method provided in this embodiment is to improve the model's ability to distinguish user delinquency risks. Therefore, users are used as nodes in the graph, and user risk information is obtained as a feature of that user node. Risk information includes, for example, a user delinquency risk score predicted by traditional methods, or recent user delinquency information. If the obtained risk information is a user delinquency risk score generated by traditional methods (referred to as the original score in this embodiment), this method can further improve the model by mining information from the coordinate data, such as... Figure 2 The diagram illustrates a coordinate data set including User_id (user ID), date (date the user stayed at that latitude and longitude), gps (latitude and longitude coordinates), and label (label indicating whether the user is overdue). This improves the score (original score) of similar samples to enhance its ability to distinguish between positive and negative samples, resulting in an improved score. Figure 3 The improved discrimination of b should be higher than that of a.

[0061] Meanwhile, the user's coordinates after data preprocessing are also used as nodes in the map construction to obtain the financial information associated with the location corresponding to the coordinates, such as the number and type of financial venues within a preset range of the location. The user's future financial risks are predicted through the associated financial attribute information of the location.

[0062] S103. Determine the spatiotemporal relationships between users and between users and coordinate values, construct bidirectional edges between nodes, generate graph data, and set user tags based on overdue status.

[0063] The nodes in this method mainly include two types: user nodes and coordinate nodes. Correspondingly, the edge types can also be divided into two categories: connections between users and latitude / longitude coordinate nodes, and connections between users. An edge from a user node to a coordinate node indicates a spatial intersection between the user and that coordinate, such as the user having stayed at that coordinate node, the user having stayed at that coordinate node within a specified time interval, or the user frequently appearing at that coordinate node. An edge between user nodes indicates a temporal and / or spatial intersection between two users, such as two users having visited a certain coordinate simultaneously, or two users having visited a coordinate of the same type simultaneously (e.g., user A went to CITIC Bank, and user B went to Bank of China during the same time period, and Bank of China and CITIC Bank are both banking institutions), or two users frequently visiting a certain coordinate. The specific spatiotemporal relationship used to establish the edges between nodes can be set according to actual evaluation needs, ensuring that the characteristics of the edges between user nodes are values ​​under the specified relationship to meet the analysis requirements under different relationships. This embodiment only introduces the above-mentioned connection relationships as examples; other connection methods can be referred to in this embodiment and will not be elaborated further.

[0064] Bidirectional edges can be constructed between users and between users and coordinate values ​​to ensure that information can be transmitted bidirectionally on the graph.

[0065] Additionally, it's important to note that after establishing the spatiotemporal relationships between users and between users and coordinate values, and constructing bidirectional edges between nodes, the original data may not guarantee that every user can be connected to another user via latitude and longitude coordinates. Therefore, the constructed graph will contain a certain number of isolated nodes. To ensure the training effect of the subsequent model, we can further perform the following steps: determine whether isolated nodes exist; if so, construct self-loop connections for the isolated nodes to generate graph data. In the final stage of graph construction, a self-loop is added to the isolated nodes to ensure normal graph training. Of course, other methods can also be used for isolated nodes, such as adding another coordinate node and connecting all isolated nodes to the new coordinate node, or optimizing the data corresponding to the isolated nodes before graph construction, etc., which are not limited here.

[0066] The stage of determining the graph's topology relies on coordinate data, while the graph's structure is determined by the way nodes are connected and the characteristics of these edges. After determining the connection methods and edge characteristics through the steps described above, a unique graph construction scheme can be determined, generating graph data. It's important to note that to retain as much information as possible, the longer the construction period, the better, in order to maximize the use of historical latitude and longitude coordinate information. For both the training and test sets, the label data can be from the current month, but the data used for graph construction can be all data up to that month. For example, assuming the data spans from January to December 2020, when training the labels for December 2020, all coordinate data from January to December 2020 needs to be used for graph construction, and the overdue data from December 2020 should be used as labels. In other words, the training data is for the entire year of 2020, but the labels only need to be from the last month.

[0067] In addition, user nodes need to be labeled based on their actual overdue status to optimize model prediction results. These labels can be binary classification labels indicating whether a payment is overdue, with 1 representing a positive sample (a sample with overdue behavior) and 0 representing a negative sample (a sample without overdue behavior). Furthermore, a corresponding binary label can be assigned for each month of overdue payment (mob). This binary labeling method is simple and ensures the model's discriminative power, thus guaranteeing its recognition performance. Of course, other types of labels can be set according to actual prediction and evaluation needs; this is not limited here.

[0068] S104. Call the graph neural network to train the user's overdue risk identification based on graph data;

[0069] The final constructed graph is then trained using a graph neural network to encode and transmit information within the graph, enabling the model to identify overdue risks. The trained graph neural network outputs optimized evaluation results that include information about the user's neighbors, thus providing richer information and higher discriminative performance compared to traditional evaluation results.

[0070] In this embodiment, there are no restrictions on the type and structure of the graph neural network model used. Theoretically, any graph neural network model can encode the latitude and longitude coordinate graph structure. Considering that the score of the optimized scorecard can be mainly based on the score and secondarily based on the score of the neighbor, the graph neural network SAGE (sample and aggregate), which encodes both its own and neighbor information, can be selected. Regarding the selection of the number of model layers, since the similarity of 1-hop neighbors is the highest and the similarity of 2-hop neighbors starts to decrease, the number of layers can be selected to be only 1, and should not exceed 2 layers. That is, the SAGE model with no more than two layers can be called to train the user for overdue risk identification based on graph data.

[0071] To deepen the understanding of the model's recognition and training process, this embodiment only uses the SAGE model as an example to introduce the recognition and training mechanism. Since graph data contains edges between users and between users and coordinates, the encoded user features originate from both coordinate features and features of neighboring users. The final user features need to combine these two features through a linear layer, specifically as follows: First, two SAGE models are used to encode latitude and longitude coordinate features and neighboring user features respectively, resulting in user_feature1 and user_feature2. Then, a concat operation is used to combine the two, and finally, a linear layer outputs the final score.

[0072] User_feature1=SAGE1(user_feature,gps_feature)

[0073] User_feature2=SAGE2(user_feature,neighbor_feature)

[0074] User_feature=concat(user_feature1,user_feature2)

[0075] User_score=linear(user_feature)

[0076] This graph neural network model can be used to identify overdue payment risks for all users. During model training, either each month's overdue data can be used as a label to generate a graph for training, or only the overdue data from the last month can be used as a label to generate a graph for training; there is no specific limitation. When using each month's overdue data as labels to generate graph data, the number of trained graphs is large, and the training data from later months may overlap with the training data from previous months. Therefore, to improve efficiency, it is advisable to train only the label data from the last month. During the testing phase, overdue labels can be generated from each month's overdue data, and the data used for graph construction can be all latitude and longitude data up to the month's label. For example, if the test data is label data from January 2021 to December 2021, then the latitude and longitude data up to each month of 2021 can be used to construct graphs, meaning 12 different graphs need to be built to train for user overdue payment risk identification.

[0077] S105. Receive the target user's graph data as target graph data, call the pre-trained graph neural network to identify overdue risks in the target graph data, and generate evaluation results.

[0078] Once the model is trained, the pre-trained graph neural network can be used to perform prediction and evaluation. If graph data of a target user is obtained, the process of constructing the graph data can be referred to the steps described above, and will not be repeated here. The received graph data of the target user is used as the target graph data. The pre-trained graph neural network is called to identify the overdue risk of the target graph data. The identification process can also be referred to the steps described above. The evaluation result generated by the model evaluation, namely the overdue risk score, can be obtained, and the model can be applied.

[0079] Based on the above introduction, the technical solution provided by this invention obtains the coordinate data of users within a specified period. Starting from the location, it mines the features of the coordinates based on the associated financial information. On the other hand, it proposes a graph construction scheme, using users and coordinates as nodes. It constructs bidirectional edges to connect users based on the spatiotemporal correlation between users and between users and coordinate values. These connected users often have similarities. Information can be transmitted to users through latitude and longitude coordinates themselves, or through neighbors, in order to comprehensively assess user risk as much as possible. This method constructs a graph structure of users and coordinates through latitude and longitude coordinates, connects users in the form of a graph through latitude and longitude information, and then models users through a graph neural network model, which can effectively model user risk and thus improve the ability to identify user overdue risk.

[0080] It should be noted that, based on the above embodiments, the present invention also provides corresponding improvements. In the preferred / improved embodiments, steps that are the same as or corresponding to those in the above embodiments can be referred to each other, and the corresponding beneficial effects can also be referred to each other; however, these will not be elaborated upon in the preferred / improved embodiments herein.

[0081] In this method, the nodes used to construct the graph data are user nodes and coordinate nodes. Correspondingly, edges include edges between user nodes and latitude / longitude coordinate nodes, as well as edges between user nodes. This embodiment describes one method for establishing connections between nodes; other methods can be implemented with reference to this embodiment.

[0082] In one implementation, determining the spatiotemporal relationships between users and between users and coordinate values ​​in step S103, and constructing bidirectional edges between nodes, can be performed according to the following steps:

[0083] Step S31: Based on the number of days the user stays at the coordinate value, construct a bidirectional edge between the user node and the coordinate node;

[0084] Step S32: Construct bidirectional edges between user nodes based on the simultaneous spatial and temporal stay of users.

[0085] In this context, each edge from a user node to a coordinate node represents a user's stay at that coordinate node. The edge's characteristic is the number of days the user stayed at that latitude and longitude coordinate; the more days, the greater the influence of that coordinate's attributes on the user. To further exclude users who are merely passing through the address rather than being permanent residents, edges where the user has stayed for 1 day at a particular latitude and longitude coordinate can be filtered out.

[0086] The edges between user nodes represent a relationship between users through latitude and longitude coordinates, such as two users having visited the same coordinate simultaneously, or two users frequently visiting the same coordinate. The specific relationships can be defined according to requirements. In one implementation, step S32 constructs bidirectional edges between user nodes based on the simultaneous spatial and temporal stay of users, which can be performed as follows:

[0087] Step S321: If the number of times the first user and the second user are at the first coordinate is greater than the first threshold, then create a high-frequency connection edge between the first user node and the second user node, and use the number of times as the feature of the edge.

[0088] Step S322: If the first user and the second user have overlapping positions within the same time period, then create a spatiotemporal connection edge between the first user node and the second user node, and use the number of times the positions overlap as the feature of the edge.

[0089] For user-to-user edges, it represents the connection between users through some kind of relationship. In this embodiment, two schemes are set for the connection between users: one is high-frequency connection, and the generated edge is called high-frequency connection edge, i.e., step S321; the other is spatiotemporal connection, and the generated edge is called spatiotemporal connection edge, i.e., step S322.

[0090] In this high-frequency connection, the time point factor is not considered. A connection is established as long as users frequently appear at the same latitude and longitude coordinates (i.e., they are frequently present). For example, if a user appears at a certain coordinate more than k days in a month, an edge can be established for that coordinate. k can be specified based on the data itself, for example, 5. If two people appear at a certain location-based service (LBS) for more than 5 days, then an edge is established between them. The characteristic of the edge is the number of days the two people appear at that LBS. One edge is established for each LBS, and the characteristic of one edge corresponds to the connection relationship of that LBS. The edge characteristic mainly expresses the strength of the connection between users. For high-frequency connections, this embodiment uses the number of days as the edge characteristic because the strength of the connection is not the same even if it exceeds k days. Figure 4 The diagram shows a high-frequency connection. User 2 and User 3 appeared at the same coordinate node for 5 days in the same month, while User 1 and User 3 appeared at the same coordinate node for 15 days in the same month. It is clear that User 1 and User 3 have higher behavioral similarity, and therefore User 1 and User 3 have more similar repayment abilities.

[0091] Spatiotemporal connection refers to the connection established when two users are at the same location coordinates at the same time. It restricts not the number of days a user appears, but the specific time of their appearance. If users appear at the same latitude and longitude coordinates on the same day, a spatiotemporal connection is established. Spatiotemporal connections are used to model the intersection of the movement trajectories of two individuals. The feature of an edge is the number of times the two individuals have formed a spatiotemporal connection. For example, if users a and b both appear at time1 in LBS1 and time2 in LBS2, the edge feature is 2, because appearing on the same day counts as one occurrence; this 2 represents both two occurrences and two days. For spatiotemporal connections, this paper uses the frequency as the edge feature. Similar to high-frequency connections, the more connections users establish at the same node on the same day, the more similar their behaviors are, and therefore, the more similar their repayment abilities are.

[0092] Specifically, a schematic diagram of a mapping scheme based on high-frequency connectivity and spatiotemporal connectivity is shown below. Figure 5As shown, circular nodes represent users, and triangular nodes represent coordinates. Solid edges indicate that users frequently appear at coordinate 1, and the characteristic of the edges is the number of days of high-frequency connections. Sparse dashed edges represent strong connections, meaning these users have high-frequency connections to a certain latitude and longitude coordinate node. Dense dashed edges represent weak connections, meaning these nodes have connected on the same day at the same latitude and longitude coordinate, and the characteristic of the edges is the number of times such connections have occurred. The diagram also contains isolated user node 5, which has not interacted with any other nodes and considers itself a neighbor.

[0093] Spatiotemporal connectivity indicates that users have similar behaviors and activities, while high-frequency connectivity indicates that users have similar residences or workplaces. It can be inferred that high-frequency connectivity is more demanding than spatiotemporal connectivity because users at the same latitude and longitude in high-frequency connectivity have more similar lifestyles. Therefore, high-frequency connectivity edges can also be called strong connectivity edges, while spatiotemporal connectivity edges can be called weak connectivity edges.

[0094] This embodiment effectively constructs a map using latitude and longitude data, and innovatively proposes a combination of high-frequency connectivity and spatiotemporal connectivity methods to effectively extract connectivity information from latitude and longitude data.

[0095] The above embodiments introduced a coordinate mapping method. This embodiment introduces the implementation method of further feature mining based on coordinates.

[0096] In one implementation, step S102 uses coordinate values ​​as coordinate nodes and obtains location-related financial information as node features. Specifically, this can be performed according to the following steps:

[0097] Step S21: Use the coordinate values ​​as coordinate nodes;

[0098] Step S22: Extract the financial attribute interest point data corresponding to the coordinate values ​​and generate a numerical vector;

[0099] Point of Interest (POI) data contains the latitude and longitude information of each POI. By combining POI data with coordinate nodes, we can obtain supporting information for that location. Through filtering, we can identify POIs with financial attributes. These POIs can be encoded using a bag-of-words model, such as determining the number of financial attribute POIs, resulting in a numerical vector. This vector not only represents the types of surrounding amenities but also their quantity. This allows us to characterize the prosperity level of the location. The more luxurious the area associated with a user, the higher their economic strength and repayment ability, thus it can be used to assess user risk. The financial attributes can be selected using features such as the quantity of five POIs: subway stations, banks, companies, shopping malls, and schools; there are no restrictions on which features to use.

[0100] Step S23: Determine the overdue user characteristics associated with the historical coordinate values, and use them as the overdue user vector;

[0101] Coordinate nodes can also be characterized by the proportion of historically connected overdue users. For each latitude and longitude coordinate that can be connected to a user, the proportion of overdue users who have historically connected to that coordinate can be statistically analyzed. If the proportion of historically connected overdue users to a given latitude and longitude coordinate is higher than the total number of connected users, then the risk coefficient of the population connected to that latitude and longitude coordinate is higher. Therefore, using the proportion of historical overdue payments as a feature of the coordinate can also be used to characterize the overdue risk of users who will connect to that coordinate in the future.

[0102] Step S24: Combine the numerical vector with the overdue user vector to form the node features corresponding to the coordinate nodes.

[0103] The numerical vector and the overdue user vector are combined as the features corresponding to the coordinate nodes. The combination method is not limited. For example, the POI feature and the historical overdue ratio feature can be concatenated into a single vector.

[0104] In one implementation, step S22 above extracts the financial attribute interest point data corresponding to the coordinate values ​​and generates a numerical vector, which can be performed according to the following steps:

[0105] Step S221: Determine the coordinate values ​​and points of interest locations encoded by geolocation hash values ​​(geoHash);

[0106] Step S222: Extract financial attribute interest points located within a preset range of coordinate values ​​based on the location of the interest points;

[0107] Step S223: Generate numerical vectors based on financial attribute interest points.

[0108] While the latitude and longitude coordinates of user-generated location information and POI data can both be highly accurate, they are measured using different systems, leading to discrepancies and inconsistencies in the latitude and longitude information for the same location, resulting in inaccurate positioning. To address this issue, this embodiment proposes converting both latitude and longitude coordinates into geographic location hash values ​​for encoding. A geographic location hash value is a method for encoding geographic areas; each area can uniquely correspond to a single geographic location hash value. By controlling the length of the geographic location hash value, connections can be established between latitude and longitude coordinates of different accuracies; typically, the length of the geographic location hash value can be controlled to 6-8 bits.

[0109] The latitude and longitude feature mining method introduced in this embodiment uses the POIs with relevant financial attributes that the coordinates can be associated with as the features of the coordinates. At the same time, by statistically analyzing the proportion of historical overdue users connected to each latitude and longitude coordinate, the risk factors of each latitude and longitude coordinate are further accurately characterized.

[0110] Furthermore, the above embodiments do not limit the processing method for obtaining the user's latitude and longitude coordinates and performing data preprocessing in step S101. This embodiment proposes a preprocessing method, which specifically includes the following steps:

[0111] Step S11: Obtain the user's latitude and longitude coordinates and filter out abnormal values;

[0112] Outliers can include null values, invalid coordinates (such as "0.0:0.0"), and coordinate nodes with low accuracy due to unforeseen circumstances. For example, a latitude and longitude coordinate of "113:30" represents a point with longitude of 113 and latitude of 30. Since each degree represents approximately 100 km and 0.001 degrees is approximately 100 meters, this coordinate can be discarded because it covers a very large area and involves many users. Only latitude and longitude coordinates with an accuracy of at least three decimal places should be retained.

[0113] Step S12: Perform fuzzy alignment processing on the latitude and longitude coordinates to obtain the coordinate values.

[0114] Because latitude and longitude coordinates are highly accurate, very few users share the same coordinates. Therefore, during map construction, the accuracy can be ranged, i.e., blurred alignment processing. Optionally, blurred alignment processing can use geolocation hash value encoding, a method of encoding latitude and longitude information that converts a latitude and longitude coordinate into a unique string. The accuracy of the geolocation hash value can be selected as needed, offering high flexibility; generally, the first 6-8 bits of the hash value can be used. Therefore, blurred alignment processing of latitude and longitude coordinates includes: encoding the latitude and longitude coordinates using geolocation hash values ​​to generate a string; and selecting the first N bits of the string as the coordinate value according to the accuracy requirements, where N is a positive integer and can be set according to the actual accuracy requirements.

[0115] The preprocessing method provided in this embodiment can avoid interference from abnormal data while ensuring the richness of the data, thereby ensuring the accuracy of the evaluation.

[0116] Corresponding to the above method embodiments, this invention also provides an overdue risk assessment device. The overdue risk assessment device described below can be referred to in correspondence with the overdue risk assessment method described above.

[0117] See Figure 6 As shown, the device includes the following modules:

[0118] The preprocessing unit 110 is mainly used to acquire the user's coordinate data within a specified period and perform data preprocessing to obtain coordinate values;

[0119] Node unit 120 is mainly used to treat users as user nodes and obtain corresponding risk information as node features; and to treat coordinate values ​​as coordinate nodes and obtain location-related financial information as node features.

[0120] The connection unit 130 is mainly used to determine the spatiotemporal relationship between users and between users and coordinate values, construct bidirectional edges between nodes, generate graph data, and set user tags according to the overdue situation;

[0121] The recognition training unit 140 is mainly used to call the graph neural network to train users to identify overdue risks based on graph data;

[0122] The evaluation unit 150 is mainly used to receive graph data from the target user as target graph data, call a pre-trained graph neural network to identify overdue risks from the target graph data, and generate evaluation results.

[0123] Corresponding to the above method embodiments, this invention also provides a computer device. The computer device described below can be referred to in conjunction with the overdue risk assessment method described above.

[0124] The computer device includes:

[0125] Memory, used to store computer programs;

[0126] A processor is used to implement the overdue risk assessment method of the above method embodiments when executing a computer program.

[0127] For details, please refer to Figure 7 This is a schematic diagram illustrating the specific structure of a computer device provided in this embodiment. The computer device can vary significantly due to differences in configuration or performance. It may include one or more central processing units (CPUs) 322 (e.g., one or more processors) and a memory 332. The memory 332 stores one or more computer application programs 342 or data 344. The memory 332 can be temporary or persistent storage. The program stored in the memory 332 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the data processing device. Furthermore, the central processing unit 322 may be configured to communicate with the memory 332 and execute the series of instruction operations stored in the memory 332 on the computer device 301.

[0128] Computer device 301 may also include one or more power supplies 326, one or more wired or wireless network interfaces 350, one or more input / output interfaces 358, and / or one or more operating systems 341.

[0129] The steps in the overdue risk assessment method described above can be implemented using the structure of computer equipment.

[0130] Corresponding to the above method embodiments, this invention also provides a readable storage medium. The readable storage medium described below can be referred to in conjunction with the overdue risk assessment method described above.

[0131] A readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the overdue risk assessment method described in the above method embodiments.

[0132] Specifically, the readable storage medium can be a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, or any other readable storage medium capable of storing program code.

[0133] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the invention.

Claims

1. A method for assessing overdue risk, characterized in that, include: Obtain the user's coordinate data within a specified period, and perform data preprocessing to obtain coordinate values; The user is taken as a user node, and the corresponding risk information is obtained as node features; The coordinate values ​​are used as coordinate nodes, and the location-related financial information is used as node features. Determine the spatiotemporal relationships between users and between users and coordinate values, construct bidirectional edges between nodes, generate graph data, and set user tags based on overdue status; The graph neural network is invoked to train the user to identify overdue risks based on the graph data. Receive graph data from the target user as target graph data, call the pre-trained graph neural network to identify overdue risk from the target graph data, and generate an evaluation result; The process of determining the spatiotemporal relationships between users and between users and coordinate values, and constructing bidirectional edges between nodes, includes: Based on the number of days the user stays at the coordinate value, construct a bidirectional edge between the user node and the coordinate node; Based on the simultaneous spatial and temporal dwell time of the users, construct bidirectional edges between the user nodes.

2. The overdue risk assessment method according to claim 1, characterized in that, The step of constructing bidirectional edges between user nodes based on the simultaneous spatial and temporal dwell time of the users includes: If the number of times the first user and the second user appear at the first coordinate is greater than the first threshold, then a high-frequency connection edge is created between the first user node and the second user node, and the number of times is used as the feature of the edge. If the first user and the second user have overlapping locations within the same time period, a spatiotemporal connection edge is created between the first user node and the second user node, and the number of times the locations overlap is used as the feature of the edge.

3. The overdue risk assessment method according to claim 1, characterized in that, The step of using the coordinate values ​​as coordinate nodes and obtaining location-related financial information as node features includes: Use the coordinate values ​​as coordinate nodes; Extract the financial attribute interest point data corresponding to the coordinate values ​​and generate a numerical vector; Determine the overdue user characteristics historically associated with the coordinate values, and use them as an overdue user vector; The numerical vector is combined with the overdue user vector to form the node feature corresponding to the coordinate node.

4. The overdue risk assessment method according to claim 3, characterized in that, The step of extracting the financial attribute interest point data corresponding to the coordinate values ​​and generating a numerical vector includes: Determine the coordinates and points of interest encoded by geographic location hash values; Based on the location of the points of interest, extract the financial attribute points of interest that are located within the preset range of the coordinate values; The numerical vector is generated based on the financial attribute interest points.

5. The overdue risk assessment method according to claim 1, characterized in that, The process of obtaining the user's latitude and longitude coordinates and performing data preprocessing to obtain coordinate values ​​includes: Obtain the user's latitude and longitude coordinates and filter out outliers; The latitude and longitude coordinates are blurred and aligned to obtain the coordinate values.

6. The overdue risk assessment method according to claim 1, characterized in that, After determining the spatiotemporal relationships between users and between users and coordinate values, and constructing bidirectional edges between nodes, the process further includes: Determine if an isolated node exists; If they exist, construct self-loop connections for the isolated nodes to generate graph data.

7. An overdue risk assessment device, characterized in that, include: The preprocessing unit is used to acquire the user's coordinate data within a specified period and perform data preprocessing to obtain coordinate values; A node unit is used to treat the user as a user node and obtain the corresponding risk information as node features. The coordinate values ​​are used as coordinate nodes, and the location-related financial information is used as node features. The connection unit is used to determine the spatiotemporal relationships between users and between users and coordinate values, construct bidirectional edges between nodes, generate graph data, and set user tags according to the overdue status. The identification training unit is used to call the graph neural network to train the user to identify overdue risks based on the graph data. The evaluation unit is used to receive graph data from the target user as target graph data, call the pre-trained graph neural network to identify overdue risk of the target graph data, and generate evaluation results; The connection unit is also used to construct a bidirectional edge between the user node and the coordinate node based on the number of days the user stays at the coordinate value; and to construct a bidirectional edge between the user nodes based on the simultaneous spatiotemporal stay of the users.

8. A computer device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the overdue risk assessment method as described in any one of claims 1 to 6.

9. A readable storage medium, characterized in that, The readable storage medium stores a computer program that, when executed by a processor, implements the steps of the overdue risk assessment method as described in any one of claims 1 to 6.

Images