A group risk identification method suitable for an O2O credit mode

Abstract

A kind of group risk identification method suitable for O2O credit mode, the method includes: step S1, the multi-source data in O2O credit business is collected, multi-source data is preprocessed, and data set is constructed;Step S2, relationship graph is constructed based on data set;Step S3, the correlation strength of each applicant node in relationship graph, user similarity, user bridge point value and attribute bridge point value are calculated;Step S4, the address information is parsed into latitude and longitude, and address clustering is carried out by DBSCAN algorithm, and address community set is formed;Step S5, generate propagation label set;Step S6, based on the initial label system and propagation label set, the risk data of applicant is output.The present application has the advantages of risk control system optimization.

Description

Technical Field

[0001] This invention relates to the field of risk identification technology, and in particular to a group risk identification method applicable to the O2O credit model. Background Technology

[0002] In recent years, small and medium-sized rural commercial banks have leveraged their deep understanding of local markets, industries, and customers, along with their long-established community relationships, to enhance their core competitiveness and deepen localized services through the O2O lending model. The O2O lending model, by integrating efficient online outreach with in-depth offline verification, has significantly improved the coverage and experience of financial services. However, the dual-channel nature of this model also makes it a high-risk scenario for operations, necessitating the use of professional models to address these risks. Summary of the Invention

[0003] The purpose of this invention is to provide a group risk identification method applicable to the O2O credit model, which has the advantage of optimizing the risk control system.

[0004] To achieve the above objectives, this invention provides a group risk identification method applicable to the O2O lending model. The method includes: Step S1, collecting multi-source data from O2O lending business, preprocessing the multi-source data, and constructing a dataset; wherein, the multi-source data includes application information, address information, device information, rejection information, post-loan performance, and third-party risk information; Step S2, constructing a relationship graph based on the dataset; Step S3, calculating the association strength, user similarity, user bridge point value, and attribute bridge point value of each applicant node in the relationship graph; Step S4, parsing the address information into latitude and longitude, and performing address clustering using the DBSCAN algorithm to form an address community set; calculating the community risk concentration based on the address community set to form a community risk concentration information set; and based on the application... Information, rejection information, post-loan performance, third-party risk information, relationship graph indicators, and community risk concentration information are used to generate an initial tagging system. The initial tagging system includes propagable tags and non-propagable tags. The community risk concentration is obtained by clustering address information in the relationship graph, with each cluster representing a community. Step S5: Using the propagable tags in the initial tagging system as the propagation source, tag propagation is performed based on the similarity of all users, the bridging point values ​​of all users, and the community risk concentration information set to generate a propagation tag set. Step S6: Based on the initial tagging system and the propagation tag set, the applicant's risk data is output.

[0005] Preferably, in step S3, the formula for calculating user similarity is as follows:

[0006] ;

[0007] Where A is a single user; B is a single user who is not A; Indicates the attribute type; Indicates the attribute value; Indicates the attribute type weight, 1 / the number of attributes for the user; This represents the weight of values ​​with the same attribute, 1 / the number of different values ​​for the same attribute; This indicates that attribute values ​​share a weight, which is 1 / the number of users with the same value; if If the risk propagation threshold is exceeded, then user A and user B are considered to be sufficiently similar in the direction of risk propagation.

[0008] Preferably, in step S3, the number of common attributes between two users is recorded as the association strength between the users; if user A and user B share the same contact phone number, the association strength is at least 1; if they simultaneously share the contact phone number, device fingerprint, and address, the association strength increases accordingly.

[0009] Preferably, in step S3, the user bridge point value is used to measure the likelihood that a user node will act as a bridging node or an intermediary node in the graph.

[0010] The higher the bridge value, the stronger the user's connection between different relational subgraphs, and the more risky characteristics it may have, such as acting as an intermediary, facilitator, or central contact person. It can be defined as the frequency with which a node appears in the shortest path of all node pairs.

[0011] User bridge point value formula as follows:

[0012] ;

[0013] in, For nodes arrive The total number of shortest paths, For the nodes passed through in these paths The number of nodes is equivalent to that of a single user.

[0014] Preferably, step S4 includes: step S41, clustering the address information to divide the same address community; step S42, calculating the core risk data for each address community; and step S43, organizing the core risk data to form community risk concentration information.

[0015] Preferably, step S41 includes: obtaining the address information corresponding to each applicant, and parsing the address information into latitude and longitude coordinates; for any two applicants, their address coordinates are as follows:

[0016] ;

[0017] Calculate the spatial distance between the two applicants based on latitude and longitude coordinates:

[0018] ,in, The distance is spherical; the address coordinates of each applicant are used as input, and the DBSCAN clustering algorithm is used for clustering.

[0019] For any applicant i, its neighborhood set is represented as:

[0020] Where Eps represents the radius parameter and MinPts represents the minimum number of samples. When the neighborhood set of applicant i satisfies: Then applicant i is taken as the core point; applicants who meet the conditions of density reachability or density connection with the core point in terms of spatial distance are assigned to the same address community.

[0021] This results in a set of address communities:

[0022] , of which each This represents an address community. Applicants who are not assigned to any community are not included in the community risk concentration calculation.

[0023] Preferably, step S42 includes: for any address community Calculate the number of applicants within the community: Calculate the maximum diameter of the community at this address:

[0024] ;

[0025] in, This represents the maximum distance between the address coordinates of any two applicants within the community; the risk concentration is calculated based on the risk identifiers of applicants within the community, as follows:

[0026] This indicates that applicant i has met the overdue flag; otherwise... ;

[0027] This indicates that applicant i has a relationship conflict flag; otherwise, RC i =0;

[0028] This indicates that applicant i has hit the cross-validation flag; otherwise... ;

[0029] Then, address community The concentration of overdue labels is:

[0030] ;

[0031] The concentration of relationship conflict labels is:

[0032] ;

[0033] The cross-validation label concentration is:

[0034] ;

[0035] The portfolio risk concentration is calculated as follows:

[0036] The concentration of instances within the community that simultaneously match both the overdue flag and the cross-validation flag is:

[0037] ;

[0038] The concentration of simultaneously hitting overdue, cross-validation, and relationship conflict indicators within the community is:

[0039] ;

[0040] Where OD represents overdue, RC represents relational conflict, and XC represents cross-validation.

[0041] Preferably, step S43 includes: summarizing the number of people in each address community, the maximum diameter of the community, the concentration of overdue tags, the concentration of relationship conflict tags, the concentration of cross-validation tags, the concentration of combined risk, the number of relationship graph groups within the community, and the percentage of people in the largest group, to form the community risk concentration information for that address community.

[0042] ;

[0043] in, This represents the community risk concentration information for the k-th address community;

[0044] After performing the above calculations on all address communities, a set of community risk concentration information is obtained:

[0045] The community risk concentration information set is used to characterize the degree of risk aggregation in communities with different addresses, and serves as input data for the subsequent generation of the initial label system and the execution of label propagation.

[0046] Preferably, in step S5, the initial labeling system is pre-set with propagable labels and non-propagable labels, and each initial label corresponds to a label name, label category, risk level, and validity period; wherein, the risk level is used to characterize the degree of risk corresponding to the initial label, and the validity period corresponds to the risk level.

[0047] Preferably, in step S6, for any applicant, the initial tags and propagation tags matched by the applicant are summarized; if the applicant matches at least one initial tag or propagation tag, the applicant is output as an applicant with a risk indicator; when the same applicant matches multiple tags, the applicant's risk level is determined according to the highest risk level among the matched tags.

[0048] RiskLevel(B) = max{L(Tag_1), L(Tag_2)...L(Tag_n)}

[0049] Where RiskLevel(B) represents the risk level of applicant B, and L(Tag_n) represents the risk level of the nth tag hit by applicant B; when the tag hit by the applicant includes a propagation tag, the risk level of the propagation tag is determined according to the risk level of its source initial tag; the applicant's risk validity period is determined according to the validity period corresponding to the highest risk level among its hit tags; if multiple tags have the same highest risk level, the validity period corresponding to the highest risk level can be used as the applicant's risk validity period; the final output risk data includes: applicant identifier; hit initial tag; hit propagation tag; propagation source applicant; user similarity; propagation path; risk level; tag validity period.

[0050] In summary, compared with existing technologies, the group risk identification method for O2O credit models provided by this invention has the following beneficial effects:

[0051] First, the group risk identification method proposed in this invention, applicable to the O2O credit model, has the advantages of high multi-source data fusion, strong identification ability, and accurate identification of intermediary nodes.

[0052] Second, the group risk identification method proposed in this invention, applicable to the O2O credit model, has the advantages of sufficient characterization of address space risk, controllable label propagation, interpretable results, and easy continuous optimization. Attached Figure Description

[0053] Figure 1 This is a flowchart illustrating a group risk identification method applicable to the O2O credit model proposed in this invention. Detailed Implementation

[0054] The following will be combined with the appendix in the embodiments of the present invention. Figure 1 The technical solutions, structural features, objectives and effects achieved in the embodiments of the present invention will be described in detail.

[0055] It should be noted that the accompanying drawings are in a very simplified form and use non-precise proportions. They are only used to facilitate and clarify the purpose of illustrating the embodiments of the present invention, and are not intended to limit the implementation conditions of the present invention. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportional relationship, or adjustments to the size should still fall within the scope of the technical content disclosed in the present invention, provided that they do not affect the effects and objectives that the present invention can produce.

[0056] It should be noted that, in this invention, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only the expressly listed elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0057] like Figure 1 As shown, this invention provides a method for identifying group risks applicable to the O2O credit model, the method comprising:

[0058] Step S1: Collect multi-source data from O2O lending business, preprocess the multi-source data, and construct a dataset;

[0059] The multi-source data includes application information, address information, device information, rejection information, post-loan performance, and third-party risk information; application information refers to the information of each applicant, which can be understood as an individual user.

[0060] Step S2: Construct a relationship graph based on the dataset;

[0061] Step S3: Calculate the association strength, user similarity, user bridge point value, and attribute bridge point value of each applicant node in the relationship graph;

[0062] Step S4: The address information is parsed into latitude and longitude, and the addresses are clustered using the DBSCAN algorithm to form an address community set; the community risk concentration is calculated based on the address community set to form a community risk concentration information set; and an initial tagging system is generated based on application information, rejection information, post-loan performance, third-party risk information, relationship graph indicators, and the community risk concentration information set; the initial tagging system includes propagable tags and non-propagable tags (details below).

[0063] Community risk concentration is obtained by clustering address information in the relationship graph. Each cluster is a community. The density-based clustering algorithm DBSCAN is used for clustering.

[0064] Step S5: Using the propagable tags in the initial tag system as the propagation source, perform tag propagation based on the set of information on all user similarity, all user bridge point values ​​and community risk concentration, and generate a set of propagation tags;

[0065] Step S6: Based on the initial tagging system and the propagation tagging set, output the applicant's risk data.

[0066] In a specific embodiment, step S1 includes the following application information: pre-loan application data;

[0067] Pre-loan application data includes: application ID, loan application date, product ID, product name, ID number, customer number, mobile phone number, residential address, registered address, contact person's phone number, contact person's relationship, spouse's contact person's phone number, contract number, device fingerprint, online application channel, offline customer acquisition channel, account manager or merchant channel identifier, etc.

[0068] The rejection information also includes: each product application rejection record, decision serial number, product number, rejection rule name, and details of the rejection reason.

[0069] Post-loan performance also includes: post-loan repayment data;

[0070] Post-loan repayment data includes: IOU number, contract number, customer number, IOU amount, IOU balance, overdue status, overdue start date, overdue days, overdue period, cumulative overdue period, maximum overdue days, overdue performance of the first period, the first two periods, and the first three periods.

[0071] Third-party risk information includes: personal credit records, special lists, judicial data, anti-fraud data, communication risk data, and device risk data.

[0072] Address information includes text address fields such as residential address, registered address, business address, offline due diligence address, and merchant address.

[0073] In step S1, the data preprocessing is specifically as follows:

[0074] Fields such as ID card number, mobile phone number, contact person's mobile phone number, device number, customer number, and contract number are cleaned, empty-removed, anonymized, or encoded. The applicant's ID card number is used as a user node, and mobile phone number, contact person's mobile phone number, device fingerprint, and address code are used as attribute nodes or attribute values. The number of different ID card numbers corresponding to the same attribute value is counted, invalid one-to-one relationships are eliminated, and one-to-many attributes that can form a relationship network are retained. ID card numbers and attribute values ​​are numbered to form an ID card attribute detail table, an ID card attribute number table, and an attribute corresponding to an ID card quantity table. Rejection reasons are standardized and mapped, categorized into risk categories such as multiple loans and behavioral issues. Post-loan overdue data is tagged according to the first, second, and third periods and the number of overdue days. Address text is geocoded to obtain longitude, latitude, and address precision levels, and the longitude and latitude are converted into spatial coordinates for calculating spherical distance.

[0075] In a specific embodiment, the relationship graph in step S2 includes three elements: entity, relationship, and attribute.

[0076] An entity, also known as a node, is used to represent objects in reality; a relationship, also known as an edge, is a semantic association that connects entities; and an attribute is additional information about an entity or relationship.

[0077] The relationship graph abstracts customers as entities, namely the individuals who apply. Therefore, the relationship graph only includes individuals and micro-enterprises, with micro-enterprises represented by the business owner's ID number. Similar attributes are abstracted as relationships; if two customers share the same attribute, a relationship exists between them, connected by edges in the graph. Entity attributes include mobile phone number, contact person's mobile phone number, address, etc.

[0078] The relational graph constructed by this invention includes a bipartite graph connecting user nodes and attribute nodes, as well as a user projection graph that retains only user nodes.

[0079] In the bipartite graph representation, user nodes represent applicants; attribute nodes represent attribute values ​​such as mobile phone number, contact mobile phone number, device fingerprint, address, and contact relationship; the edge between a user node and an attribute node indicates that the user has that attribute.

[0080] In the user projection graph representation, nodes represent applicants; if two applicants share at least one attribute, an edge is established between the two applicants; the weight of the edge can be determined by factors such as the number of shared attributes, attribute type weight, attribute scarcity, and attribute temporal consistency.

[0081] In a specific embodiment, in step S3, user similarity is a directional variable, that is, the similarity between user A and user B can be different from the similarity between user B and user A.

[0082] User similarity considers the number of user attribute types, the number of user attribute values, and the number of attributes shared with another user. It is calculated comprehensively using attribute weights, the weights of values ​​within the same attribute, and the weight of the same value on a single user. As mentioned earlier, attributes generally refer to additional information about entities or relationships. In a relationship graph, attributes are used to characterize the descriptive features of entities or relationships, providing supplementary information about them, and typically consist of an attribute name and an attribute value.

[0083] The formula for calculating user similarity is as follows:

[0084] ;

[0085] Where A is a single user; B is a single user who is not A; Indicates the attribute type; Indicates the attribute value; Indicates the attribute type weight, 1 / the number of attributes for the user; This represents the weight of values ​​with the same attribute, 1 / the number of different values ​​for the same attribute; This indicates that attribute values ​​share a weight, which is 1 / the number of users with the same value; if If the risk propagation threshold is exceeded, then user A and user B are considered to be sufficiently similar in the direction of risk propagation.

[0086] In a specific embodiment, in step S3, the number of common attributes between two users is recorded as the association strength between the users. If user A and user B share the same contact's mobile phone number, the association strength is at least 1; if they simultaneously share the contact's mobile phone number, device fingerprint, and address, the association strength increases accordingly.

[0087] The strength of association can be used to determine the closeness of a group's relationship. The higher the strength of association, the greater the likelihood that there is a genuine connection or collaborative fraud between the two applicants.

[0088] In a specific embodiment, in step S3, the user bridge point value is used to measure the probability that a certain user node is a bridge node or an intermediary node in the graph.

[0089] The higher the bridge value, the stronger the user's connection between different relational subgraphs, and the more risky characteristics such as intermediary application, organization matching, and centralized contact person may be. It can be defined as the frequency of a node appearing in the shortest path of all node pairs.

[0090] User bridge point value formula as follows:

[0091] ;

[0092] in, For nodes arrive The total number of shortest paths, For the nodes passed through in these paths The number of nodes is equivalent to that of a single user.

[0093] In a specific embodiment, in step S3, the bridge point value of the attribute bridge point value corresponds to the bridge point value of the edge, that is, the ratio of the shortest path between any two points that passes through the edge, and the edge bridge point value is assigned to the attribute that constitutes the edge.

[0094] Attribute bridging value is used to measure the bridging effect of a certain attribute between different users. For example, if a contact's mobile phone number is connected to multiple applicants and is on multiple shortest paths, then that contact's mobile phone number has a high attribute bridging value, which may represent an intermediary mobile phone number, a mobile phone number used for proxy applications, or an abnormally shared contact.

[0095] In a specific embodiment, step S4 includes:

[0096] Step S41: Cluster the address information and divide the same address community;

[0097] Specifically, the address information corresponding to each applicant is obtained, and the address information is parsed into latitude and longitude coordinates; for any two applicants, their address coordinates are as follows:

[0098] ;

[0099] Calculate the spatial distance between the two applicants based on latitude and longitude coordinates:

[0100] ,in, The distance is spherical. Using the address coordinates of each applicant as input, the DBSCAN clustering algorithm is employed for clustering.

[0101] For any applicant i, its neighborhood set is represented as:

[0102] Where Eps represents the radius parameter and MinPts represents the minimum number of samples. When the neighborhood set of applicant i satisfies: Then applicant i is taken as the core point; applicants who meet the conditions of density reachability or density connection with the core point in terms of spatial distance are assigned to the same address community.

[0103] This results in a set of address communities:

[0104] , of which each This represents an address community. Applicants who are not assigned to any community are not included in the community risk concentration calculation.

[0105] Step S42: Calculate the core risk data for each address community;

[0106] Specifically, for any address community Calculate the number of applicants within the community: Calculate the maximum diameter of the community at this address:

[0107] ;

[0108] in, This represents the maximum distance between the address coordinates of any two applicants within the community;

[0109] The risk concentration is calculated based on the risk profiles of applicants within the community, as follows:

[0110] This indicates that applicant i has met the overdue flag; otherwise... ;

[0111] This indicates that applicant i has a relationship conflict flag; otherwise... ;

[0112] This indicates that applicant i has hit the cross-validation flag; otherwise... .

[0113] Then, address community The concentration of overdue labels is:

[0114] ;

[0115] The concentration of relationship conflict labels is:

[0116] ;

[0117] The cross-validation label concentration is:

[0118] ;

[0119] Furthermore, the combined risk concentration can be calculated. For example, the concentration of cases within a community that simultaneously hit both the overdue flag and the cross-validation flag is:

[0120] ;

[0121] The concentration of simultaneously hitting overdue, cross-validation, and relationship conflict indicators within the community is:

[0122] ;

[0123] Where OD represents overdue, RC represents relational conflict, and XC represents cross-validation.

[0124] Step S43: Organize core risk data to form community risk concentration information;

[0125] Specifically, the community risk concentration information for each address community is formed by summarizing the community population, maximum community diameter, overdue label concentration, relationship conflict label concentration, cross-validation label concentration, combined risk concentration, number of relationship graph groups within the community, and the percentage of the largest group.

[0126] ;

[0127] in, This represents the community risk concentration information for the k-th address community;

[0128] After performing the above calculations on all address communities, a set of community risk concentration information is obtained:

[0129] The community risk concentration information set is used to characterize the degree of risk aggregation in communities with different addresses, and serves as input data for the subsequent generation of the initial label system and the execution of label propagation.

[0130] In a specific embodiment, in step S5, the propagable tags in the initial tag system are used as the propagation source, and tag propagation is performed according to the user similarity between applicants in the relationship graph to generate a propagation tag set.

[0131] Specifically, the initial labeling system pre-sets propagable and non-propagable labels. Each initial label corresponds to a label name, label category, risk level, and validity period. The risk level is used to characterize the degree of risk corresponding to the initial label, and the validity period corresponds to the risk level.

[0132] For any applicant A, if applicant A matches the initial tag Tag_a, and the initial tag Tag_a is a propagable tag, then applicant A is identified as the tag propagation source user. In the relationship graph, find applicant B who is associated with applicant A, and calculate the user similarity between applicant A and applicant B: SIM(A,B). When applicant A and applicant B satisfy the following condition: SIM(A,B) > T, then the propagable initial tag Tag_a matched by applicant A is propagated to applicant B, generating the propagation tag corresponding to applicant B: LP_P1_Tag_a, where T represents a preset similarity threshold, and LP_P1_Tag_a represents the propagation tag formed by the initial tag Tag_a through first-level propagation.

[0133] The propagation label LP_P1_Tag_a inherits the label name, label category, risk level, and validity period of its source initial label Tag_a. For example, if the propagation label LP_P1_Tag_a originates from the initial label Tag_a, then: L(LP_P1_Tag_a) = L(Tag_a), where L(Tag_n) represents the risk level of label Tag_n.

[0134] The preset similarity threshold T is a label propagation boundary parameter, which is determined based on the hit effect of propagating labels in historical samples.

[0135] Iterate through all applicants who have hit the initial propagation tag and their related applicants in the relationship graph, generate propagation tags according to the above propagation rules, and form a propagation tag set.

[0136] In a specific embodiment, in step S6, the applicant's risk data is output based on the initial labeling system and the propagation label set.

[0137] Specifically, for any applicant, the initial tags and propagation tags matched by the applicant are summarized; if the applicant matches at least one initial tag or propagation tag, the applicant is output as an applicant with a risk indicator.

[0138] When the same applicant matches multiple labels, the applicant's risk level is determined according to the highest risk level among the matched labels:

[0139] RiskLevel(B) = max{L(Tag_1), L(Tag_2)...L(Tag_n)}

[0140] Where RiskLevel(B) represents the risk level of applicant B, and L(Tag_n) represents the risk level of the nth tag that applicant B hits.

[0141] When the label matched by the applicant includes a propagation label, the risk level of the propagation label is determined according to the risk level of the original label from which it originated.

[0142] The applicant's risk validity period is determined according to the validity period corresponding to the highest risk level among the labels it hits; if multiple labels have the same highest risk level, the validity period corresponding to the highest risk level can be used as the applicant's risk validity period.

[0143] The final output risk data includes: applicant identifier; initial tag hit; propagation tag hit; applicant of propagation source; user similarity; propagation path; risk level; tag validity period.

[0144] The generation of the propagation label is based on the method described in step S5.

[0145] Although the present invention has been described in detail through the preferred embodiments above, it should be understood that the above description should not be considered as a limitation of the present invention. Various modifications and substitutions to the present invention will be apparent to those skilled in the art after reading the above description. Therefore, the scope of protection of the present invention should be defined by the appended claims.

Claims

1. A method for identifying group risk applicable to O2O credit models, characterized in that, The method includes: Step S1: Collect multi-source data from O2O lending business, preprocess the multi-source data, and construct a dataset; the multi-source data includes application information, address information, device information, rejection information, post-loan performance, and third-party risk information. Step S2: Construct a relationship graph based on the dataset; Step S3: Calculate the association strength, user similarity, user bridge point value, and attribute bridge point value of each applicant node in the relationship graph; Step S4: The address information is parsed into latitude and longitude, and the addresses are clustered using the DBSCAN algorithm to form an address community set; the community risk concentration is calculated based on the address community set to form a community risk concentration information set; and an initial tagging system is generated based on application information, rejection information, post-loan performance, third-party risk information, relationship graph indicators, and the community risk concentration information set; the initial tagging system includes propagable tags and non-propagable tags; wherein, the community risk concentration is obtained by clustering the address information in the relationship graph, and each cluster is a community. Step S5: Using the propagable tags in the initial tag system as the propagation source, perform tag propagation based on the set of information on all user similarity, all user bridge point values ​​and community risk concentration, and generate a set of propagation tags; Step S6: Based on the initial tagging system and the propagation tagging set, output the applicant's risk data.

2. The method for identifying group risk applicable to the O2O credit model according to claim 1, characterized in that, In step S3, the formula for calculating user similarity is as follows: ； Where A is a single user; B is a single user who is not A; Indicates the attribute type; Indicates the attribute value; Indicates the attribute type weight, 1 / the number of attributes for the user; This represents the weight of values ​​with the same attribute, 1 / the number of different values ​​for the same attribute; This indicates that attribute values ​​share a weight, which is 1 / the number of users with the same value; if If the risk propagation threshold is exceeded, then user A and user B are considered to be sufficiently similar in the direction of risk propagation.

3. The method for identifying group risk in an O2O credit model according to claim 2, characterized in that, In step S3, the number of common attributes between two users is recorded as the association strength between the users; if user A and user B use the same contact phone number, the association strength is at least 1; if they share the contact phone number, device fingerprint and address at the same time, the association strength increases accordingly.

4. The group risk identification method applicable to the O2O credit model according to claim 2, characterized in that, In step S3, the user bridge point value is used to measure the likelihood that a user node will act as a bridge node or an intermediary node in the graph. The higher the bridge value, the stronger the user's connection between different relationship subgraphs, which may indicate risk characteristics such as intermediary application, matchmaking, and centralized contact person. It can be defined as the frequency with which a node appears in the shortest path of all node pairs. User bridge point value formula as follows: ； in, For nodes arrive The total number of shortest paths, For the nodes passed through in these paths The number of nodes is equivalent to that of a single user.

5. A method for identifying group risk in an O2O credit model according to claim 3, characterized in that, Step S4 includes: Step S41: Cluster the address information and divide the same address community; Step S42: Calculate the core risk data for each address community; Step S43: Organize core risk data to form community risk concentration information.

6. A method for identifying group risk in an O2O credit model according to claim 4, characterized in that, Step S41 includes: Obtain the address information corresponding to each applicant, and parse the address information into latitude and longitude coordinates; for any two applicants, their address coordinates are as follows: ； Calculate the spatial distance between the two applicants based on latitude and longitude coordinates: ,in, The distance is spherical; the address coordinates of each applicant are used as input, and the DBSCAN clustering algorithm is used for clustering. For any applicant i, its neighborhood set is represented as: Where Eps represents the radius parameter and MinPts represents the minimum number of samples. When the neighborhood set of applicant i satisfies: Then applicant i is taken as the core point; applicants who meet the conditions of density reachability or density connection with the core point in terms of spatial distance are assigned to the same address community. This results in a set of address communities: , of which each This represents an address community. Applicants who are not assigned to any community are not included in the community risk concentration calculation.

7. A method for identifying group risk in an O2O credit model according to claim 5, characterized in that, Step S42 includes: for any address community Calculate the number of applicants within the community: Calculate the maximum diameter of the community at this address: ； in, This represents the maximum distance between the address coordinates of any two applicants within the community; The risk concentration is calculated based on the risk profiles of applicants within the community, as follows: This indicates that applicant i has met the overdue flag; otherwise... ; This indicates that applicant i has a relationship conflict flag; otherwise, RC i =0; This indicates that applicant i has hit the cross-validation flag; otherwise... ; Then, address community The concentration of overdue labels is: ； The concentration of relationship conflict labels is: ； The cross-validation label concentration is: ； The portfolio risk concentration is calculated as follows: The concentration of instances within the community that simultaneously match both the overdue flag and the cross-validation flag is: ； The concentration of simultaneously hitting overdue, cross-validation, and relationship conflict indicators within the community is: ； Where OD represents overdue, RC represents relational conflict, and XC represents cross-validation.

8. A method for identifying group risk in an O2O credit model according to claim 6, characterized in that, Step S43 includes: The community risk concentration information for each address community is formed by summarizing the community population, maximum community diameter, overdue label concentration, relationship conflict label concentration, cross-validation label concentration, combined risk concentration, number of relationship graph groups within the community, and the percentage of the largest group's population. ； in, This represents the community risk concentration information for the k-th address community; After performing the above calculations on all address communities, a set of community risk concentration information is obtained: The community risk concentration information set is used to characterize the degree of risk aggregation in communities with different addresses, and serves as input data for the subsequent generation of the initial label system and the execution of label propagation.

9. A method for identifying group risk in an O2O credit model according to claim 7, characterized in that, In step S5, the initial labeling system is pre-set with propagable labels and non-propagable labels. Each initial label has a label name, label category, risk level, and validity period. The risk level is used to characterize the degree of risk corresponding to the initial label, and the validity period corresponds to the risk level.

10. A method for identifying group risk in an O2O credit model according to claim 8, characterized in that, In step S6 For any applicant, summarize the initial tags and propagation tags matched by that applicant; if the applicant matches at least one initial tag or propagation tag, then output the applicant as an applicant with a risk indicator. When the same applicant matches multiple labels, the applicant's risk level is determined according to the highest risk level among the matched labels: RiskLevel(B) = max{L(Tag_1), L(Tag_2)...L(Tag_n)}; Where RiskLevel(B) represents the risk level of applicant B, and L(Tag_n) represents the risk level of the nth tag that applicant B matches; When the applicant's matched tags include propagation tags, the risk level of the propagation tags is determined according to the risk level of the initial tag from which they originated; the applicant's risk validity period is determined according to the validity period corresponding to the highest risk level among the matched tags; if multiple tags have the same highest risk level, the validity period corresponding to that highest risk level can be used as the applicant's risk validity period; the final output risk data includes: applicant identifier; matched initial tag; matched propagation tag; propagation source applicant; user similarity; propagation path; risk level; tag validity period.

Images