A distributed storage method and system for dynamic index query of a debtor relationship diagram

Abstract

The present application relates to the technical field of financial credit data processing and distributed storage, and discloses a distributed storage method and system for dynamic index query of a debtor relationship graph.The method comprises: processing existing multi-source associated data using a scene attribute completion and cross-source conflict resolution method to generate a debtor associated node panoramic attribute matrix; constructing a debtor multi-level associated value directed graph based on the matrix; generating a multi-dimensional evaluation table of implicit key nodes through a graph implicit node value walk evaluation algorithm; and combining distributed sharding storage and dynamic index services to generate a debtor associated node communication coping scheme through a communication coping suggestion matching algorithm and archiving the scheme.The present application solves the problem of priority misjudgment of implicit associated nodes, and realizes accurate integration of multi-source data and efficient distributed storage and query.

Description

Technical Field

[0001] This invention relates to the field of financial credit data processing and distributed storage technology, and more specifically, to a distributed storage method and system for dynamic index query of debtor relationship graphs. Background Technology

[0002] Distributed storage technology for dynamic indexing of debtor relationship graphs is one of the core technologies supporting the efficient operation of financial risk control. It primarily integrates multi-source data from credit systems, telecom operator platforms, social security institutions, third-party data providers, and risk control systems, including basic debtor information, call records, employment information, and risk control interaction records. By constructing a relationship graph containing nodes such as debtors, contacts, and employers, and the relationships between these nodes, combined with a distributed storage architecture, it achieves efficient storage and rapid retrieval of large-scale data. Simultaneously, relying on a dynamic indexing mechanism, it responds to data updates in real time, such as adding call records or risk control records, ensuring that risk control personnel can obtain the latest relationship network information of debtors through queries, providing data support for identifying risk control breakthroughs and formulating outreach strategies. The core features of this technology are: multi-source data fusion, which can integrate data from different sources, eliminating data isolation, and integrating structured and unstructured data; storage architecture scalability, adapting to massive data and high-concurrency query requirements through distributed deployment; real-time index updates, ensuring the timeliness of relationship network information; and intuitive relationship presentation, clearly displaying the strength and type of node relationships in a graph structure.

[0003] However, existing distributed storage technologies for dynamic indexing of debtor relationship graphs still suffer from critical issues in practical applications, including misjudgment and inefficient reach of implicitly related nodes. This problem stems from two core limitations: First, existing technologies primarily calculate node association strength based on explicit indicators such as call frequency and direct relationship types (e.g., spouse, colleague), failing to integrate contextual information like neighbor relationships, knowledge of debtor income or whereabouts, or non-traditional data such as non-privacy interaction information, delivery addresses, and part-time attendance explicitly authorized by users in specific business scenarios (e.g., loan applications). This results in seemingly weakly related, but actually crucially impactful, implicit nodes being labeled as low-priority. Second, existing dynamic indexing mechanisms only respond to data existence updates, such as adding nodes or edges, without triggering node priority reassessment for non-traditional data updates, creating a data update blind spot. The aforementioned issues directly lead to risk control personnel being misled by system labeling, prioritizing strong but ineffective nodes, such as contacts in high-frequency business dealings, while missing implicit nodes that could provide information on debtors' whereabouts, income, or willingness to repay. This not only wastes a lot of communication time and manpower but also reduces the ability to store high-quality data. Summary of the Invention

[0004] To address the aforementioned problems, the present invention provides the following technical solution:

[0005] A method for cross-configuring highly reusable Software-as-a-Service (SaaS) functional modules includes:

[0006] Step S10: Process the existing multi-source associated data using the scenario-based attribute completion and cross-source conflict resolution methods to generate a panoramic attribute matrix of debtor associated nodes;

[0007] Specifically, the methods for contextual attribute completion and cross-source conflict resolution are as follows:

[0008] Each associated entity in the cleaned multi-source association data is identified as a node and a node number is generated. Basic attributes are matched for each node, including name and mobile phone number. The basic attributes are extracted and anonymized from legal identity information fragments in the cleaned multi-source association data. Based on the interaction features and scene information in the cleaned multi-source association data, scene-based attributes are completed for each node to obtain a preliminary attribute set.

[0009] Identify conflicting data in the initial attribute set; calculate the data source credibility of the conflicting data; combine the data source credibility and data update time of the conflicting data to determine the optimal attribute value, prioritizing attribute values ​​with higher data source credibility; mark each attribute that has undergone conflict resolution with a conflict resolution flag, which includes whether conflict processing was performed and the basis for processing, and also mark each attribute with a credibility score; after completing scenario-based attribute completion and cross-source conflict resolution, obtain the attribute set after conflict resolution.

[0010] Specifically, the debtor-related node panoramic attribute matrix is ​​a two-dimensional matrix structure. The row dimension is the node number, which covers debtor nodes, contact person nodes, company nodes, and community role nodes. The column dimension is the attribute dimension, which includes basic attributes, traditional association attributes, scenario-based attributes, conflict resolution markers, and credibility scores. Each cell of the matrix stores the attribute value and update timestamp.

[0011] Step S20: Based on the attribute values, credibility, and timeliness of the panoramic attribute matrix of the debtor's related nodes, a multi-level directed graph of debtor's related value is formed by using a dynamic weighted association graph construction and association value labeling mechanism.

[0012] Specifically, the dynamic weighted association graph construction and association value labeling mechanism is as follows:

[0013] Based on the nodes and node attribute values ​​extracted from the panoramic attribute matrix of debtor-related nodes, a value vertex set of a multi-level related value directed graph of debtors is generated.

[0014] After completing the construction of the value vertex set, the traditional association attributes and scenario-based attributes in the panoramic attribute matrix of debtor-related nodes are used to determine the association relationships between nodes and generate directed edges.

[0015] For each generated directed edge, label it with an associated value tag and an information transmission coefficient;

[0016] Perform edge weight calculation operation, and combine the attribute credibility score, the value weight of the associated value tag and the information transmission coefficient in the panoramic attribute matrix of the debtor's associated nodes to calculate the edge weight of each directed edge;

[0017] After the edge weights are calculated, the node level is labeled for each node based on the node features of the value vertex set and the edge features of the value directed edge set, thus generating a node level set.

[0018] Based on the association duration of all edges in the value directed edge set, calculate the global average association duration and generate a graph global duration marker;

[0019] Generate a multi-level value graph of debtors;

[0020] Specifically, the debtor's multi-level associated value directed graph is a graph structure containing a set of value vertices, a set of value directed edges, a set of edge weights, a set of node levels, and a global timeliness marker. The specific structure is as follows: The value vertex set, where each vertex contains a node number, core attributes, node influence radius, and latest update time; the value directed edge set, where each edge contains an edge type, associated value label, information transmission coefficient, and associated timeliness; the edge weight set, where each edge weight contains a calculation factor and the specific values ​​of each factor; the node level set, where each node contains a level label and the level determination criteria; and the global timeliness marker, containing the global average associated timeliness and the number of edges used to calculate that timeliness.

[0021] Step S30: Based on the global structure and value attributes of the multi-level associated value directed graph, the hidden node value walk evaluation algorithm on the graph is used to obtain a multi-dimensional evaluation table of hidden key nodes.

[0022] Specifically, the algorithm for evaluating the value of hidden nodes in a graph is as follows:

[0023] Perform walk parameter initialization operation, and determine the walk probability calculation parameters and iteration termination conditions based on the extracted edge weight set and value directed edge set, and generate walk parameter set;

[0024] Taking the core node in the debtor's multi-level associated value directed graph as the starting point, and based on the walking probability coefficient in the walking parameter set, it iterates through the graph according to the set probability. During the process, the value walking score of each hidden node is accumulated in real time until the iteration number threshold or the convergence error threshold is met, and the value walking score of each hidden node is output.

[0025] Based on the global structure of the debtor's multi-level associated value directed graph, and combining the extracted value directed edge set with the graph's global timeliness marker, the core association degree, value redundancy degree, and graph global timeliness matching degree of each hidden node are calculated to generate a global dimension evaluation set.

[0026] By integrating the information from the set of hidden nodes, the value walk score, and the global dimension evaluation set, a multi-dimensional evaluation table of hidden key nodes is generated according to a preset dimension structure.

[0027] Specifically, the multi-dimensional evaluation table for hidden key nodes is a two-dimensional structured table. The row dimension represents each hidden node in the set of hidden nodes; the column dimension includes eight core evaluation dimensions. The eight core evaluation dimensions include: the first column is the basic information of the node, the second column is the value walk score, the third column is the core relevance, the fourth column is the value redundancy, the fifth column is the associated value tag and timeliness, the sixth column is the high-value association path, the seventh column is the sensitive attributes and risk warnings, and the eighth column is the priority recommendation level.

[0028] Step S40: Based on the multi-dimensional assessment table of hidden key nodes, combined with distributed sharded storage and dynamic indexing services, the table data and storage resources are collaboratively processed through the communication and response suggestion matching algorithm driven by the assessment table to generate communication and response plans for debtor-related nodes and complete distributed storage archiving.

[0029] Specifically, distributed sharded storage uses a hash sharding algorithm based on the extracted node type and priority recommendation level to divide the evaluation table into different shards of the distributed storage cluster according to the combination rule of "node type + priority recommendation level", generating a sharded storage mapping table. Distributed sharded storage refers to a compliant storage method that splits large-scale table data into multiple encrypted storage shards according to business attributes and hash algorithms. Each storage shard corresponds to an independent storage node in the distributed storage cluster, and each shard is configured with role-based access control. The hash sharding algorithm is executed as follows: using the combination string of "node type + priority recommendation level" as the hash key, the hash value is calculated using a hash function, and then modulo the total number of shards in the distributed storage cluster to obtain the shard number corresponding to the combined data, ensuring that data with the same combination rule is stored in the same shard. The sharded storage mapping table records the mapping relationship between each shard and the corresponding storage node location, the range of node numbers contained in the shard, and access permissions. Each entry includes: shard identifier, encrypted IP address of the storage node, starting range of node numbers, shard data update time, list of authorized access modules, and hash calculation rule.

[0030] Specifically, the dynamic indexing service refers to a compliant indexing mechanism used to record the mapping relationship between implicit node numbers and corresponding storage shard locations in real time. It has the capabilities of real-time index updates, access control, and fast querying. Its core structure is a partitioned encrypted hash index table. The index table is partitioned according to node type. The key in each partition is the implicit node number, and the value is the <shard identifier + encrypted IP address of the storage node + encrypted offset of the data in the shard + index update timestamp> of the shard to which the node belongs. When the data in the multi-dimensional evaluation table of the implicit key node is updated, the dynamic indexing service will automatically trigger the index update operation.

[0031] A distributed storage system for dynamic indexing and querying of debtor relationship graphs, the system comprising:

[0032] Multi-source associated data processing module: Uses scenario-based attribute completion and cross-source conflict resolution methods to process existing multi-source associated data and generate a panoramic attribute matrix of debtor associated nodes;

[0033] The related value directed graph construction module is used to construct a multi-level related value directed graph of debtors based on the attribute values, credibility, and timeliness of the panoramic attribute matrix of debtor related nodes, using a dynamic weighted association graph construction and association value annotation mechanism.

[0034] Latent node value assessment module: Based on the global structure and value attributes of a multi-level associated value directed graph, the algorithm for evaluating the value of latent nodes on the graph is used to obtain a multi-dimensional evaluation table of latent key nodes.

[0035] Precise Targeting Execution Package Generation Module: Based on the multi-dimensional evaluation table of implicit key nodes, combined with distributed sharded storage and dynamic indexing services, the module uses an evaluation table-driven communication and response suggestion matching algorithm to collaboratively process table data and storage resources, generate communication and response plans for debtor-related nodes, and complete distributed storage archiving.

[0036] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention, through multi-stage technological innovation, accurately solves the core problem of "misjudgment and inefficient reach of implicit related nodes in the existing debtor relationship graph storage and query technology". In the data processing stage, traditional and non-traditional related data are integrated through scenario-based attribute completion. Combined with cross-source conflict resolution and compliance anonymization, this not only solves the problem of missing hidden nodes caused by single dimensions and conflict distortion of multi-source data, but also ensures that data storage complies with regulations and avoids the risk of privacy leakage. In the relationship graph construction stage, the first / secondary edge classification, association value labeling, and time expiration marking of the multi-level related value directed graph of debtors are used to break through the limitation of traditional relationship graphs that only record associations without reflecting value. This clearly locates the range of hidden nodes and avoids interference from expired data. In the value assessment stage, the hidden node value walking assessment algorithm on the graph traverses the entire graph structure and calculates node scores from multiple dimensions to generate an assessment table. This solves the priority misjudgment caused by single dimensions and helps risk control personnel quickly locate high-value hidden nodes. In the storage and query stage, the combination of distributed sharded storage and dynamic indexing services not only solves the efficiency bottleneck of centralized storage of large-scale data, but also achieves rapid data retrieval and full-process traceability through real-time index updates and compliant archiving. At the same time, it separates actual execution behavior and reduces compliance risks. Overall, it achieves accurate integration of multi-source data, efficient identification of hidden nodes, and collaborative distributed storage and querying, thereby improving storage quality. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art 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.

[0038] Figure 1 This is a flowchart of a distributed storage method for dynamic index query of debtor relationship graph in this invention;

[0039] Figure 2 This is a flowchart of the algorithm for evaluating the value of hidden nodes on the graph in an embodiment of the present invention;

[0040] Figure 3 This is a layered architecture diagram of the distributed sharded storage architecture in an embodiment of the present invention;

[0041] Figure 4 This is a functional module diagram of a distributed storage method for dynamic index query of debtor relationship graphs in this invention. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only 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.

[0043] Example 1

[0044] Please see Figure 1 As shown, this embodiment provides a distributed storage method for dynamic index query of debtor relationship graphs, including:

[0045] Step S10: Process the existing multi-source associated data using the scenario-based attribute completion and cross-source conflict resolution methods to generate a panoramic attribute matrix of debtor associated nodes;

[0046] This step focuses on the compliant and precise processing of debtor relationship network data. All existing multi-source related data were obtained through legal and compliant channels, including with the authorization and consent of relevant parties, extraction from publicly available legal information platforms, or necessary and reasonable information records based on business scenarios, strictly adhering to relevant laws and regulations on data security and personal information protection. The multi-source related data includes two categories: traditional related data and non-traditional related data. Traditional related data refers to routine business data directly related to the debtor, including call records, employment records, and risk control records. Call records are summaries of legitimate communication records between the debtor and related parties, containing only non-privacy information such as call frequency and time periods, obtained through compliant communication data interfaces. Employment records refer to publicly available professional information of the debtor and related parties, including employer names and job types, obtained through legitimate enterprise information disclosure platforms or authorized professional information registration systems. Risk control records refer to records generated during past compliant risk control processes, including risk control communication times and methods, such as telephone and SMS, retrieved through internal compliant business systems. The related parties are those who are directly or indirectly related to the debtor, including colleagues, relatives, friends, and other relevant personnel who can help promote the debtor's repayment progress.

[0047] The aforementioned non-traditional related data refers to supplementary information that can indirectly reflect the debtor's relationships, including WeChat chat history summaries, express delivery address information, community group interaction records, and part-time work attendance records. Specifically, WeChat chat history summaries refer to non-privacy interaction information obtained with the authorization and consent of the related parties, containing only interaction frequency and interaction scenario tags such as "family topics" and "work topics," without involving specific chat content, and are obtained through compliant social media platform data interfaces. Express delivery address information refers to publicly available or authorized delivery address summaries of the related parties, containing only non-precise location information such as city-level administrative districts and street-level areas, and are obtained through legitimate logistics information service platforms. Community group interaction records refer to records of the related parties' interactive behavior in public community groups, including the number of interactions and the interacting parties, and are obtained through compliant group management tools authorized by the community management. Part-time work attendance records refer to the part-time work attendance information of the debtor and related parties, including attendance frequency and attendance time periods, and are obtained through an authorized part-time employment management system.

[0048] The existing multi-source associated data is initially cleaned to remove invalid data, such as data with incorrect format or duplicate records, to obtain cleaned multi-source associated data. Based on the cleaned multi-source associated data, scenario-based attribute completion and cross-source conflict resolution methods are used for processing. First, scenario-based attribute completion is performed to obtain a preliminary attribute set. Specifically, the steps of the scenario-based attribute completion operation are as follows: First, each associated entity in the cleaned multi-source association data is identified as a node, generating a node number. The node number is a unique code that distinguishes different associated entities, generated by a combination of random characters and numbers to ensure a balance between association and anonymity. Second, basic attributes are matched for each node, including name and mobile phone number. The name is anonymized, retaining only the surname and the first character of the given name. The mobile phone number is anonymized, hiding the middle four digits. The basic attributes are obtained by extracting and anonymizing legal identity information fragments from the cleaned multi-source association data. Third, based on the interaction features and scenario information in the cleaned multi-source association data, scenario-based attributes are completed for each node. Scenario-based attributes include social relationship scenario attributes and information possession scenario attributes. Among them, social relationship scenario attributes refer to attributes that describe the type of social relationship between nodes, such as "neighborhood". The attributes for "residents," "colleagues," and "relatives" are determined by analyzing the frequency of interactions in call records (e.g., monthly call frequency ≥ a set threshold 'a') to identify close relationships; interaction scene tags in community group records (e.g., frequent participation in "building notification" type interactions) to identify "neighbors"; and consistency of work unit in employment records (e.g., identical work unit names) to identify "colleagues." The set threshold 'a' is determined by statistically analyzing the average interaction frequency in historical valid social relationship data. The "information possession scenario attribute" refers to the attribute describing whether a node is aware of specific information about the debtor, such as "knowing income status" or "knowing whereabouts." This is determined by analyzing scene tags in WeChat chat record summaries (e.g., tags related to "income" or "salary") to identify "knowing income status" and related attendance information in part-time attendance records (e.g., highly overlapping with the debtor's part-time attendance periods) to identify "knowing whereabouts." After completing the scenario-based attributes, a preliminary attribute set is obtained.

[0049] Based on the initial attribute set, a cross-source conflict resolution operation is performed to obtain the attribute set after conflict resolution. Specifically, the steps of the cross-source conflict resolution operation are as follows: First, identify conflicting data in the initial attribute set, that is, data with different values ​​under the same attribute dimension of the same node. For example, the express delivery address information of a node shows that its permanent residence is in region A, while the employment record shows that the permanent residence corresponding to its work location is in region B, which is determined to be an address attribute conflict. Second, calculate the data source credibility corresponding to the conflicting data. The data source credibility ranges from 0 to 100 and adopts a weighted calculation method. The formula is: Data source credibility = (data update frequency weight × data update frequency score) + (data verification number weight × data verification number score). Among them, the data update frequency weight and the data verification number weight are adjustment coefficients, and their sum is 1. They are determined by analyzing the accuracy of historical conflict data resolution. The data update frequency score refers to the score calculated based on the number of updates of the data in a recent period. The more updates, the higher the score. The score is determined by statistical data within a set range. The number of update records within the time window is determined; the data verification score refers to the score calculated based on the number of times the data has been verified by other legitimate channels. The more verifications, the higher the score, which is determined by statistically analyzing the verification records in the compliance verification system; the third step is to determine the optimal attribute value by combining the data source credibility and data update time of the conflicting data, prioritizing the attribute value with higher data source credibility. If the data source credibility is the same, the attribute value updated by the update time is selected; the fourth step is to mark each attribute that has undergone conflict resolution with a conflict resolution mark. The conflict resolution mark includes whether conflict processing has been carried out and the basis for processing. The conflict resolution mark takes the value of "yes" or "no", such as "selecting express delivery address information with a data source credibility of 85 points, and the data update time is within the last 15 days". At the same time, a credibility score is marked for each attribute. The credibility score is the credibility score of the data source corresponding to the attribute. After completing the conflict resolution, the set of attributes after conflict resolution is obtained.

[0050] Based on the attribute set after conflict resolution, a panoramic attribute matrix of debtor-related nodes is constructed. This matrix is ​​a two-dimensional matrix structure. The row dimension represents node numbers, covering debtor nodes, contact person nodes, company nodes, and community role nodes. Contact person nodes are individual nodes associated with the debtor; company nodes are the work unit nodes of the debtor and contact person; and community role nodes are role nodes related to the debtor's community, such as grid worker nodes. All are distinguished by node numbers generated during the node identification process. The column dimension represents attribute dimensions, including basic attributes, traditional association attributes, scenario-based attributes, conflict resolution markers, and credibility scores. Basic attributes include anonymized name and phone number; traditional association attributes... The attributes include call count and relationship tags (based on employment records such as "colleague" and "superior / subordinate"). The call count is based on the monthly call count between nodes in the call record statistics, and the relationship tags are based on employment records such as "colleague" and "superior / subordinate". The contextual attributes include social relationship context attributes and information access context attributes, with specific values ​​such as "neighbor" and "knowing income status". The conflict resolution marker and credibility score are obtained through conflict resolution operations. Each cell of the matrix stores "attribute value + update timestamp", where the update timestamp refers to the time information of the last update of the attribute value, in the format of "year-month-day hour:minute", which is obtained by recording the update operation time of the attribute value.

[0051] Specifically, the steps for constructing the panoramic attribute matrix of debtor-related nodes are as follows: First, determine the row dimension range of the matrix by arranging the node numbers of all nodes in the attribute set after conflict resolution in order, which will serve as the row index of the matrix; Second, determine the column dimension range of the matrix by arranging the basic attributes, traditional related attributes, scenario-based attributes, conflict resolution markers, and credibility scores in a preset order, which will serve as the column index of the matrix; Third, fill the attribute values ​​and corresponding update timestamps of each node in the attribute set after conflict resolution into the corresponding cells of the matrix. If a node has no corresponding information for a certain attribute dimension, mark "No valid information" and add a timestamp to that cell. After filling, the panoramic attribute matrix of debtor-related nodes is obtained.

[0052] For example, if there exists a contact node with node number "J1001", its basic attributes include name "Zhang*", mobile phone number "138****5678", and update timestamp "2025-05-10 09:30"; its traditional association attributes include call count "12 times / month" (monthly call count with the debtor node), relationship tag "colleague", and update timestamp "2025-05-08 14:20"; its contextual attributes include social relationship scenario attribute "colleague", information access scenario attribute "aware of work dynamics", and update timestamp "2025-05-09 11:15"; its conflict resolution flag is "no" (no conflict resolution has been experienced), and update timestamp "2025-05-10 09:30"; its credibility score is "92", and update timestamp "2025-05-10". If the value is 09:30, then in the panoramic attribute matrix of debtor-related nodes, the intersection cell of row "J1001" and each attribute dimension column should be filled with the above attribute value and the corresponding update timestamp.

[0053] The role and effect of this step are mainly reflected in several aspects: First, through a compliant multi-source related data processing mechanism, while ensuring data legality and privacy security, it comprehensively covers the regular and supplementary data of debtor relationships, avoiding omissions of related nodes due to a single data dimension; second, through scenario-based attribute completion operations, by combining the scenario characteristics of multi-source data, it supplements each node with attributes related to social relationships and information possession, solving the problem that traditional data can only reflect direct relationships and cannot reflect the value of related scenarios, providing a scenario-based basis for subsequent value assessment of related nodes; third, the cross-source conflict resolution operation combines data source credibility calculation with update time... This approach effectively resolves contradictions between multi-source data, while simultaneously labeling conflict resolution markers and credibility scores, thus addressing the attribute distortion problem caused by multi-source data conflicts. Furthermore, the panoramic attribute matrix of debtor-related nodes, through structured storage of complete node attribute information and timestamps, not only provides accurate and comprehensive data input for subsequent steps but also lays the foundation for subsequent timeliness judgments through timestamps. Compared to ordinary attribute matrices without conflict handling and contextualized attributes, it can more accurately carry the effective information of implicitly related nodes, reducing the probability of misjudging implicitly related nodes in the debtor relationship network from the data source. This provides core data support for solving the problem of "misjudgment and inefficient reach of implicitly related nodes in the debtor relationship network."

[0054] Step S20: Based on the attribute values, credibility, and timeliness of the panoramic attribute matrix of the debtor's related nodes, a multi-level directed graph of debtor's related value is formed by using a dynamic weighted association graph construction and association value labeling mechanism.

[0055] The input variable for this step is the panoramic attribute matrix of debtor-related nodes output in step S10. All nodes, attribute values ​​of each node, attribute credibility scores, and attribute update timestamps are extracted from this matrix. The extracted data will serve as the basic data for the construction of the dynamic weighted association graph and the association value labeling mechanism, and will be used for subsequent vertex set construction, edge set generation, weight calculation, and hierarchical labeling.

[0056] The attribute values ​​include basic attributes, traditional association attributes, and scenario-based attributes recorded in the matrix; the attribute credibility score is the credibility score corresponding to each attribute in the matrix, with a value range of 0-100; the attribute update timestamp is the latest acquisition time of the attribute data recorded in the matrix cell, in the format of "year-month-day hour:minute:second", which is used for subsequent calculation of association timeliness.

[0057] Specifically, the dynamic weighted association graph construction and association value annotation mechanism is as follows: First, a value vertex set construction operation is performed. Based on the nodes and node attribute values ​​extracted from the panoramic attribute matrix of debtor-related nodes, a value vertex set of a multi-level association value directed graph of debtors is generated. The value vertex set is a collection used to store the core information and value characteristics of all related nodes. Each vertex includes a node number, core attributes, node influence radius, and latest update time. The core attributes are the basic attributes and key scenario-based attributes of the node. For example, the core attributes of a debtor node include name, ID number, and overdue amount; the core attributes of a contact node include name, mobile phone number, and social relationship scenario attributes. The node influence radius is a quantitative indicator of the range of related nodes that a node can influence, used to characterize the node's influence range in the relationship network. It is calculated by weighting the number of related nodes and the node attribute value, with the specific formula as follows:

[0058] Node influence radius = (Number of associated nodes × Parameter a) + (Node attribute value × Parameter b)

[0059] Wherein, parameter a is the weighting coefficient for the number of associated nodes, and parameter b is the weighting coefficient for the node attribute value. The values ​​of parameters a and b are both in the range of 0-1, and their sum is 1. This is determined by analyzing the correlation between the influence range of a node and the number of associated nodes and attribute value in the historical relationship network. The number of associated nodes is the total number of directly associated nodes recorded in the debtor's associated node panoramic attribute matrix. The node attribute value is the quantified value of the node's scenario-based attributes. If the node's scenario-based attributes are high-frequency risk control value attributes such as "income awareness" and "location transmission," its attribute value is parameter c. If it is a low-risk control value attribute such as "ordinary social interaction," its attribute value is parameter d. Parameter c is greater than parameter d. The values ​​of parameters c and d are both in the range of 0-10. This is determined by statistically analyzing the contribution of different scenario-based attributes to the success of risk control. The latest update time is the latest update timestamp among all attributes of the node, used to subsequently determine the timeliness of node information.

[0060] After constructing the value vertex set, the value directed edge set generation operation is performed. Based on the traditional association attributes and scenario-based attributes in the debtor-related node panoramic attribute matrix, the association relationships between nodes are determined and a value directed edge set is generated. The value directed edge set includes first-level directed edges and second-level directed edges. The first-level directed edges are direct association edges between nodes. When the debtor-related node panoramic attribute matrix records direct interaction data between node A and node B, such as call records, direct work communication records, or direct social interaction records, a first-level directed edge is generated from node A to node B, or from node B to node A. The direction of the edge is determined by the initiator and receiver of the association relationship. For example, if node A actively calls node B, a first-level directed edge is generated from node A to node B. The second-level directed edges are indirect association edges between nodes. When node A and node C do not have direct interaction data, but node A and node B are directly associated, and node B and node C are directly associated, a second-level directed edge is generated from node A to node C. The direction of the edge is determined along the path direction of the direct association.

[0061] For each generated directed edge representing value, an associated value label and information transmission coefficient are assigned. The associated value label is a label characterizing the value of the relationship represented by the edge to risk control work, including categories such as "income awareness," "location transmission," "emergency contact," "workplace connection," and "community connection." This label is determined based on the contextual attributes of nodes in the debtor-related node panoramic attribute matrix. For example, if the relationship between node A and node B corresponds to the contextual attribute "knowing the debtor's income" in the matrix, then the directed edge is labeled with the associated value label "income awareness"; if it corresponds to the contextual attribute "knowing the debtor's whereabouts," then it is labeled with the associated value label "location transmission." A directed edge can be labeled with multiple associated value tags. The information transmission coefficient is a parameter that characterizes the success rate of information transmission in the relationship represented by the edge. The value ranges from 0 to 1. The higher the success rate of information transmission, the larger the value of the information transmission coefficient. For example, when the relationship between nodes is a close family relationship such as "spouse" or "parents", the information transmission coefficient is parameter e. When the relationship is a relationship such as "ordinary colleague" or "unfamiliar neighbor", the value is parameter f. Parameter e is greater than parameter f. The values ​​of parameter e and parameter f are determined by statistically analyzing the proportion of successful information transmission in historical relationships.

[0062] Subsequently, an edge weight calculation operation is performed. Combining the attribute credibility score, the value weight of the association value tag, and the information transmission coefficient in the panoramic attribute matrix of the debtor's associated nodes, the edge weight of each directed edge is calculated. The edge weight is a quantitative indicator representing the strength of the directed edge association and its risk control value. The specific formula is as follows:

[0063] Edge weight = Attribute credibility × Value tag weight × Information transmission coefficient

[0064] The attribute credibility is the attribute credibility score recorded in the panoramic attribute matrix of the debtor's associated nodes for the relationship corresponding to the directed edge, with a value range of 0-100, and needs to be normalized to a value of 0-1 during calculation; the value tag weight is the weight corresponding to the associated value tag labeled by the directed edge. If the edge is labeled with multiple associated value tags, the average of all tag weights is taken. The weight of high-frequency risk control value tags such as "income awareness" and "location transmission" is parameter g, and the weight of low-risk control value tags such as "ordinary social interaction" is parameter h. Parameter g is greater than parameter h, and the values ​​of both parameter g and parameter h are in the range of 0-1. It is determined by statistically analyzing the influence of different associated value tags on risk control decisions; the information transmission coefficient is the information transmission coefficient labeled by the directed edge, with a value range of 0-1.

[0065] After edge weight calculation, node hierarchy labeling is performed. Based on the node features of the value vertex set and the edge features of the value directed edge set, a hierarchy is labeled for each node, generating a node hierarchy set. The node hierarchy set includes three levels: core nodes, ordinary nodes, and implicit nodes. Node types are debtor nodes, or nodes with a first-level directed edge linking them to a debtor node with associated value tags such as "spouse" or "emergency contact," and whose information transmission coefficient is greater than or equal to parameter i. The value of parameter i ranges from 0 to 1 and is determined by the identification criteria of core nodes in historical risk control. Ordinary nodes are nodes that are directly associated with debtor nodes (first-level directed edge) but are not core nodes, such as high-frequency call contacts who are not relatives or ordinary colleagues. The criterion for determination is: the node has a first-level directed edge with the debtor node and is not a core node. The hidden nodes are nodes that are indirectly related to the debtor nodes (second-level directed edges) and have certain risk control value. The determination criteria are: there is a second-level directed edge between the node and the debtor node, and the number of related value tags marked on the second-level directed edge is greater than or equal to parameter j, and the information transmission coefficient is greater than or equal to parameter k. Parameter j is the threshold for the number of related value tags, with a value range of 1-5, and parameter k has a value range of 0-1. Both are determined by statistically analyzing the contribution conditions of historical hidden nodes to the success of risk control. Each level of node is accompanied by a level determination criterion, recording the specific edge characteristics and parameter conditions for determining the level to which the node belongs.

[0066] Finally, the graph global timeliness marker generation operation is performed. Based on the association timeliness of all edges in the value directed edge set, the global average association timeliness is calculated and a graph global timeliness marker is generated. The association timeliness is a parameter characterizing the timeliness of the directed edge association relationship. It is calculated based on the update timestamp of the associated data corresponding to the edge in the panoramic attribute matrix of the debtor's associated nodes. If the interval between the update timestamp and the current time is ≤ parameter m, the association timeliness is 1; if the interval is > parameter m and ≤ parameter n, the association timeliness is parameter p; if the interval is > parameter n, the association timeliness is parameter q. Parameter m and parameter n are time interval thresholds in days. The values ​​of parameter p and parameter q are in the range of 0-1, with parameter p being greater than parameter q. All parameters are determined by analyzing the impact of the association timeliness on the risk control value. The graph global timeliness marker is the arithmetic mean of the association timeliness of all directed edges, with a value range of 0-1, and is used for subsequent overall judgment of the overall timeliness of the association relationship in the debtor's multi-level associated value directed graph.

[0067] Through the above operations, a multi-level directed graph of debtor-related value is generated. This multi-level directed graph of debtor-related value is a graph structure containing a set of value vertices, a set of value directed edges, a set of edge weights, a set of node levels, and a global time-sensitivity marker. Specifically, the structure is as follows: Value vertex set: each vertex contains a node number, core attributes, node influence radius, and latest update time; Value directed edge set: each edge contains an edge type, associated value label, information transmission coefficient, and associated time-sensitivity; Edge weight set: each edge weight contains a calculation factor and the specific values ​​of each factor; Node level set: each node contains a level label and level determination criteria; Global time-sensitivity marker: contains the global average associated time-sensitivity and the number of edges used to calculate that time-sensitivity.

[0068] The purpose and effect of this step are as follows: By constructing a dynamic weighted association graph and using an association value labeling mechanism, the structured data in the panoramic attribute matrix of debtor association nodes is transformed into a graph structure with risk control value characteristics, solving the problem that traditional relationship graphs only record node associations while ignoring association value and timeliness; the influence radius of nodes with concentrated value vertices can quantify the scope of a node's influence in the relationship network, avoiding the omission of hidden nodes due to focusing only on the number of nodes while ignoring node value; the first- or second-level edge classification and association value labels of the value directed edge concentration clearly distinguish the value differences between direct and indirect associations, especially by capturing indirect associations through second-level edges, providing key support for identifying hidden association nodes; the calculation of edge weights integrates attribute credibility, value label weight, and information transmission coefficient, ensuring that the quantitative results of association strength are more in line with risk control needs, avoiding misjudgments of association strength caused by single-dimensional judgment; the node hierarchy set clearly divides core, ordinary, and hidden nodes, directly locating the scope of hidden association nodes, laying the foundation for accurate evaluation of hidden node value in subsequent steps; the global timeliness labeling of the graph can reflect the information timeliness of the relationship network as a whole, avoiding inefficient reach caused by using expired association data. Compared to ordinary directed graphs, the debtor multi-level relational value directed graph more comprehensively carries the value, timeliness, and transmission capability of the relational relationship. It can be generalized and adapted to the identification of implicit relational nodes in different scenarios such as family, workplace, and community. It solves the core problem of "misjudgment of implicit relational nodes in the debtor relationship network" from the graph structure level, and provides complete and accurate data support for step S30 to process the graph structure and evaluate the value of implicit nodes.

[0069] For example, if in the panoramic attribute matrix of debtor-related nodes of debtor Li Si, there is a direct call record between node Zhang San (contact node) and node Wang Wu (part-time boss node) (first-level directed edge), and node Wang Wu has an employment-related record with Li Si (first-level directed edge), then a second-level directed edge is generated from Zhang San to Li Si; Zhang San's contextual attribute is "knowing Li Si's part-time job information", then the second-level directed edge is labeled with two association value tags: "location transmission" and "workplace association"; the attribute credibility of the association between Zhang San and Wang Wu is 90 (normalized to 0.9), the weight of the "location transmission" tag is 0.8, the weight of the "workplace association" tag is 0.7, the average value tag weight is 0.75, and the information transmission coefficient is 0.6 (Zhang San and Wang Wu). Five are ordinary friends, Wang Wu and Li Si are superior and subordinate (determined by comprehensive transmission success rate), then the edge weight of this second-level directed edge = 0.9 × 0.75 × 0.6 = 0.405; Zhang San is a node associated with the second-level directed edge, the number of associated value tags is 2 (≥ parameter j = 1), and the information transmission coefficient is 0.6 (≥ parameter k = 0.4), then Zhang San is marked as an implicit node, and the hierarchy determination is based on the record "second-level directed edge + number of associated value tags 2 + information transmission coefficient 0.6"; if the association timeliness of this second-level directed edge is 0.9, and the average association timeliness of other edges is 0.85, then the global average association timeliness of the graph global timeliness mark = (0.9 + sum of timeliness of other edges) / total number of edges, which is used to judge the overall information timeliness in the subsequent process.

[0070] Step S30: Based on the global structure and value attributes of the multi-level associated value directed graph, the hidden node value walk evaluation algorithm is used to obtain a multi-dimensional evaluation table of hidden key nodes.

[0071] The input variable for this step is the debtor multi-level associated value directed graph output by step S20. The value vertex set, value directed edge set, edge weight set, node level set, and graph global time marker are extracted from this graph. The extracted data will serve as the core input of the hidden node value walk evaluation algorithm on the graph, and will be used for subsequent hidden node screening, value score calculation, global dimension evaluation, and evaluation table construction.

[0072] The set of value vertices includes the node number, core attributes, node influence radius, and latest update time of each node; the set of value directed edges includes the edge type, associated value label, information transmission coefficient, and associated timeliness of each edge; the set of edge weights includes the edge weight of each edge and its corresponding calculation factor value; the set of node levels includes the level label and level determination criteria of each node, used to filter hidden nodes; and the graph global timeliness marker includes the global average associated timeliness of the graph, with a value range of 0-1, used for subsequent calculation of the graph global timeliness matching degree.

[0073] First, based on the extracted node hierarchy set, all nodes labeled "latent node" are selected to form a latent node set. This latent node set is a subset of the latent nodes in the directed graph of debtor multi-level associated value. Each element contains the node number of the latent node and its corresponding value vertex information, including core attributes, node influence radius, and latest update time. This set will be used as the evaluation object for the latent node value walk evaluation algorithm.

[0074] Specifically, the algorithm for evaluating the value of hidden nodes in a graph, such as... Figure 2 As shown, the specific process is as follows: First, the walk parameter initialization operation is performed. Based on the extracted edge weight set and value directed edge set, the walk probability calculation parameters and iteration termination conditions are determined, generating a walk parameter set. The walk parameter set includes a walk probability coefficient, an iteration number threshold, and a convergence error threshold. The walk probability coefficient is a comprehensive weight coefficient of edge weights and information transmission coefficients, used to calculate the probability of walking from the current node to an adjacent node. Specifically, it is set as: Probability of walking from node A to node B = (Edge weight from node A to node B × Information transmission coefficient from node A to node B) / ∑(Edge weight from node A to all adjacent nodes × Corresponding information transmission coefficient). The iteration number threshold is the maximum number of walk iterations, determined by statistically analyzing the average number of iterations required for convergence in historical implicit node value assessments. The convergence error threshold is the upper limit of the difference in value walk scores between two adjacent iterations. When the difference is less than this threshold, convergence is determined, with a value range of 0-5, determined by analyzing the balance between assessment accuracy requirements and computational efficiency.

[0075] The second step involves iterative walking and score calculation. Starting with the core node in the debtor's multi-level relational value directed graph, the walk proceeds iteratively through the graph according to a set probability coefficient based on the walking probability coefficients in the walking parameter set. During this process, the value walking score of each hidden node is accumulated in real time until the iteration threshold or the convergence error threshold is met, at which point the value walking score of each hidden node is output. The value walking score is a quantitative indicator representing the comprehensive risk control value of a hidden node in the relational network, ranging from 0 to 100. It is calculated by accumulating the value contribution of the edges traversed and the node's own influence radius score. The specific formula is as follows:

[0076] Value walk score = ∑ (weight of path edge × score of associated value tag × timeliness of path edge association) + (radius of node influence × parameter a)

[0077] Wherein, the weight of the path edge is the edge weight of each edge traversed during the walk; the score of the associated value label is the quantified value of the associated value label of the path edge. If the label is a high-frequency value label such as "income awareness" or "location transmission", the score is parameter b; if it is a low-frequency value label such as "ordinary association", the score is parameter c. Parameter b is greater than parameter c. The values ​​of both parameters b and c are in the range of 0-10, and are determined by statistically analyzing the contribution of different labels to the success of risk control; the association timeliness of the path edge is the association timeliness of the path edge; parameter a is the score conversion coefficient of the node influence radius, with a value range of 0-10. It is determined by mapping the node influence radius to the score range of 0-100 to ensure the reasonable contribution of the node's own influence to the score.

[0078] The third step involves performing a global dimension calculation. Based on the global structure of the debtor's multi-level associated value directed graph, and combining the extracted value directed edge set with the graph's global timeliness markers, the core correlation degree, value redundancy degree, and graph global timeliness matching degree of each latent node are calculated to generate a global dimension evaluation set. This global dimension evaluation set contains three quantified values ​​for each latent node. The core correlation degree represents the closeness of the association between a latent node and a core node, calculated using the shortest path length from the latent node to the core node and the average information transmission coefficient along the path. The specific formula is: Core Correlation Degree = (Parameter d / Shortest Path Length) × Average Information Transmission Coefficient, where parameter d is the path length weight coefficient, ranging from 1 to 10; the shortest path length is calculated using Dijkstra's algorithm; and the average information transmission coefficient is the arithmetic mean of the information transmission coefficients of all edges along the path. The core correlation degree ranges from 0 to 1. The value redundancy degree represents the degree of association between the latent node and the core node. The irreplaceability of a node in the relational network is determined by judging whether there are other nodes with the same association value label as the latent node and the core association degree difference is less than the parameter e. If they exist, the value redundancy level is "high" and if they do not exist, it is "low". The parameter e ranges from 0 to 0.2 and is determined by evaluating the tolerance of node value duplication. The graph global timeliness matching degree is the degree of fit between the timeliness of the latent node's associated edges and the graph global timeliness. The calculation method is: graph global timeliness matching degree = 1 - |average association timeliness of all associated edges of the latent node - graph global timeliness label|, with a value range of 0 to 1. The closer the value is to 1, the better the node timeliness matches the global timeliness.

[0079] After completing the above algorithm steps, the hidden key node multi-dimensional evaluation table construction operation is performed. This integrates the information from the hidden node set, value walk scores, and global dimension evaluation set, generating a hidden key node multi-dimensional evaluation table according to a preset dimensional structure. The hidden key node multi-dimensional evaluation table is a two-dimensional structured table. The row dimension represents each hidden node in the hidden node set, and the column dimension includes eight core evaluation dimensions, as follows:

[0080] The first column contains basic node information, which is a composite field containing the node number, name, mobile phone number, and node type of implicit nodes. The node type is determined based on core attributes, such as "community role" or "workplace role".

[0081] The second column is the value walk score, which is a numerical value plus detailed fields, including the total score (0-100) and the detailed scores of each calculated factor, such as "the weight score of the path edge ×× + the score of the associated value tag ×× + the score of the node influence radius ××".

[0082] The third column is the core relevance, which is a numeric and text field containing the relevance value (0-1) and the shortest path details. The shortest path details record the shortest path node sequence from the hidden node to the core node, such as "node A → node B → core node".

[0083] The fourth column is the value redundancy, which is a level + list field, containing the redundancy level and the list of alternative nodes. If an alternative node exists, the node number and core attributes of the alternative node are recorded; otherwise, it is marked "no alternative node".

[0084] The fifth column is the associated value label and timeliness, which is a composite list field containing the average associated value label and associated timeliness of all associated edges of the implicit node, such as "income awareness (0.8), whereabouts transmission (0.7), timeliness 0.8";

[0085] The sixth column is the high-value association path, which is a path detail field that records the top two high-value association paths. Each path includes the edge type, the information transmission coefficient of each edge on the path, and the total weight of the path. The total weight of the path is the sum of the weights of all edges on the path.

[0086] The seventh column contains sensitive attributes and risk warnings, which are text plus warning fields. Sensitive attributes are extracted from the core attributes of hidden nodes, such as "privacy privy" and "public officials". Risk warnings are generated based on sensitive attributes, such as "communication requires privacy protection" and "avoid frequent phone calls".

[0087] The eighth column represents the priority recommendation level, which is a level field. It is determined based on a combination of value walk score and value redundancy. If the value walk score is greater than or equal to parameter f and the value redundancy is "low", the level is "S"; if the score is greater than or equal to parameter g and the redundancy is "medium", the level is "A"; otherwise, the level is determined by the combination of score and redundancy, which is "B" or "C". Parameter f is greater than parameter g. The value range of both parameter f and parameter g is 0-100, which is determined by the score distribution of high-value nodes in historical risk control.

[0088] The role and effect of this step are as follows: The implicit node value walk assessment algorithm on the graph traverses the global structure of the debtor's multi-level relational value directed graph, avoiding the loss of relational information caused by disassembling graph data. Compared with local assessment methods, it can more comprehensively capture the value transfer path of implicit nodes in the relational network, ensuring the integrity of value assessment. The calculation of value walk score integrates edge weight, relational label, timeliness, and node influence, solving the value misjudgment problem caused by traditional single-dimensional assessment, making the risk control value quantification of implicit nodes more accurate. The multi-dimensional assessment of core correlation, value redundancy, and graph global timeliness matching further supplements the information on the node's correlation tightness, irreplaceability, and timeliness fit, providing richer decision-making basis for subsequent reach strategy matching. The structured design of the implicit key node multi-dimensional assessment table transforms complex graph data into a table format that risk control personnel can directly understand, clearly presenting the core value and usage suggestions of each implicit node. At the same time, it can quickly locate high-value nodes through priority recommendation levels, avoiding inefficient screening of massive nodes by risk control personnel. Compared to a simple list of nodes, this assessment table can more accurately meet the needs of risk control scenarios and can be generalized to assess different types of hidden nodes such as families, workplaces, and communities, thus completely solving the problem of "misjudgment of hidden related nodes" from the perspective of value assessment.

[0089] For example, for the latent node Zhang San (node ​​number: Contact_18812345678), its associated edge information is extracted from the debtor's multi-level associated value directed graph: the edge weight with the part-time boss node is 0.7, the associated value label is "location transmission" (score 8), and the associated duration is 0.9; the indirect edge weight (through the part-time boss) with Li Si (core node) is 0.6, the associated value label is "income awareness" (score 9), and the associated duration is 0.8; Zhang San's node influence radius is 5, parameter a=2, so the influence score = 5×2=10. Calculated according to the value walk score formula: ∑(0.7×8×0.9 + 0.6×9×0.8)+10=(5.04+4.32)+10=19.36 (the total score after normalization is approximately 19, this is a simplified calculation for example); In the core relevance calculation, the shortest path length is 2 (Zhang San→Part-time Boss→Li Si), the average information transmission coefficient is 0.7, the parameter d=5, so the core relevance = (5 / 2)×0.7=1.75 (0.88 after normalization), the shortest path details are "Zhang San→Part-time Boss→Li Si"; In the value redundancy determination Since no other node simultaneously possesses the labels "location transmission + income awareness" and the core correlation difference is <0.2, the redundancy level is "low," and the alternative node list is marked "no alternative node." The final priority recommendation level is determined to be "S" (assuming parameter f=15, score 19≥15 and low redundancy), and the sensitive attributes and risk warning column is marked "no sensitive attributes, telephone communication is recommended first." After complete information integration, a record of Zhang San node in the multi-dimensional evaluation table of implicit key nodes is formed.

[0090] Step S40: Based on the multi-dimensional assessment table of implicit key nodes, combined with distributed sharded storage and dynamic indexing services, the table data and storage resources are collaboratively processed through the communication and response suggestion matching algorithm driven by the assessment table to generate communication and response plans for debtor-related nodes and complete distributed storage archiving.

[0091] The input variable for this step is the multi-dimensional evaluation table of hidden key nodes output in step S30. From this table, we extract the node type, priority recommendation level, associated value tags, core correlation, value redundancy, sensitive attributes, and high-value association paths for each hidden node. This extracted data will serve as the decision-making basis for the communication response suggestion matching algorithm driven by the evaluation table. Simultaneously, a distributed storage cluster resource is introduced as the core carrier to realize the sharded storage of the multi-dimensional evaluation table of hidden key nodes, the archived storage of communication response plans, and cross-shard data retrieval. Finally, through the collaborative processing of the algorithm and storage resources, a compliant communication response plan for debtor-related nodes is generated and archived, without involving the execution of any actual communication behavior.

[0092] The distributed storage cluster resource refers to a compliant cluster system composed of multiple independent storage nodes. Each storage node has the capability for encrypted data storage and access control. Nodes are interconnected through an encrypted network, supporting distributed split storage and compliant cross-node access to data. All data storage complies with relevant regulations on personal information protection and data security. The node type is extracted from the "Basic Node Information" column of the multi-dimensional evaluation table of hidden key nodes, such as "Community Role," "Workplace Role," and "Family Related Role," which are used as the classification basis for subsequent sharded storage. The high-value association path is extracted from the "High-Value Association Path" column of the table, including the node number and node type of each node on the path. It is used to integrate communication suggestion data of related nodes across shards to form a complete communication response reference plan, such as... Figure 3 As shown.

[0093] First, the distributed sharding storage operation of the hidden key node multi-dimensional evaluation table is performed. Based on the extracted node type and priority recommendation level, a hash sharding algorithm is used to divide the evaluation table into different shards of the distributed storage cluster according to the combination rule of "node type + priority recommendation level", generating a sharding storage mapping table. The distributed sharding storage refers to a compliant storage method that splits large-scale table data into multiple encrypted storage shards according to business attributes and hash algorithms. Each storage shard corresponds to an independent storage node in the distributed storage cluster, and each shard is configured with role-based access control, allowing only the compliance review module and the communication solution generation module to access it. Specifically, the hash sharding algorithm is executed as follows: using the combination string of "node type + priority recommendation level" as the hash key, the hash value is calculated using the MD5 hash function, and then modulo the total number of shards in the distributed storage cluster to obtain the shard number corresponding to the combined data. This ensures that data with the same combination rule is stably stored in the same shard. For example, if the node type is "community role" and the priority recommendation level is "...", the shard number is calculated. The string "S" is hashed, and if the modulo result is 1, it is stored in shard 1. The string "Workplace Role" with a priority recommendation level of "A" is hashed, and if the modulo result is 2, it is stored in shard 2. This rule ensures that high-value data shards can be accurately located during subsequent scheme generation, reducing invalid data retrieval. The shard storage mapping table records the mapping relationship between each shard and the corresponding storage node location, the range of node numbers contained in the shard, and access permissions. Each entry contains "shard identifier, encrypted IP address of storage node, starting range of node number, shard data update time, list of authorized access modules, and hash calculation rule", which is used for the mapping relationship construction of subsequent dynamic indexing services, and at the same time prevents unauthorized modules from accessing shard data.

[0094] After completing the sharded storage, a dynamic index service is bound to each storage shard. The dynamic index service refers to a compliant indexing mechanism used to record the mapping relationship between implicit node numbers and corresponding storage shard locations in real time. It has real-time index updates, access control, and fast query capabilities. Its core structure is a partitioned encrypted hash index table. The index table is partitioned by "node type," and the key in each partition is the implicit node number. The implicit node number is anonymized using a "role identifier + random number" format to avoid directly associating it with personal identity information. The value is the "shard identifier + encrypted IP address of the storage node + encrypted offset of the data within the shard + index update timestamp" of the shard to which the node belongs. When the data in the multi-dimensional evaluation table of implicit key nodes is updated, such as when the node priority recommendation level changes or associated value tags are added, the dynamic index service will automatically trigger an index update operation: First, after receiving the data update request, the compliance verification module verifies whether the request source is an authorized module, such as when the evaluation table is updated. After the module passes verification, it proceeds to the second step. The second step recalculates the shard number corresponding to the node based on the node number and new attributes in the updated data. The third step compares the latest state of the shard storage mapping table, corrects the mapping value of the corresponding node number in the partition encrypted hash index table, and records the index update log, which includes the value before update, the value after update, the update operator, and the update time. The query response latency of the dynamic index service is controlled within the dynamic index query response latency threshold. The value of the dynamic index query response latency threshold is determined comprehensively based on the encrypted network bandwidth of the distributed storage cluster, the processing performance of the storage nodes, and the number of index partitions, and is typically set to 50-200 milliseconds to meet the real-time data retrieval requirements when generating the communication scheme. Simultaneously, all query operations generate query logs, which record the query module, query time, query node number, and query result for subsequent compliance auditing.

[0095] Specifically, the assessment table-driven communication response suggestion matching algorithm is as follows: The first step is to perform a basic communication suggestion matching operation. Based on the associated value tags extracted from the multi-dimensional assessment table of implicit key nodes, the algorithm queries the preset compliance communication suggestion library, matches the basic communication suggestions of each implicit node, and generates basic suggestion matching results. The compliance communication suggestion library is a predefined mapping library of "related value tags - compliance communication suggestions". All suggestions have been reviewed by legal compliance and ethics departments and include "related value tag combinations, recommended communication channels, and compliance communication script templates". The communication channels only include compliance channels, such as written notification letters and friendly information prompts. The compliance communication script templates conceal personal privacy information and only contain friendly prompts related to repayment. For example, when the related value tag combination is "income awareness + whereabouts transmission", the recommended communication channel is "written notification letter" and the compliance communication script template is "Hello, regarding the repayment of XX (debtor's surname + first name), if you are aware of their recent situation, you can assist in conveying friendly reminders about repayment. For specific details, you can guide them to consult through official compliance channels". The basic suggestion matching result is a structured list that records each hidden node number and the corresponding basic communication suggestion, including "node number, recommended communication channel, compliance communication script template, and suggestion matching basis". This result is only used for subsequent suggestion integration and does not directly trigger any communication behavior.

[0096] The second step involves providing communication timing suggestions. Based on the extracted core relevance, friendly suggestions are given for communication timing from the basic suggestion matching results, generating a timing suggestion result. These suggestions refer to compliant communication periods that do not disrupt the normal life and work of the debtor, recommended based on the closeness of the relationship between the node and the debtor. The suggestion rules are determined by the numerical range of the core relevance: if the core relevance is greater than or equal to the high threshold, it indicates a close relationship between the node and the debtor, and the suggested communication time is during non-working hours, such as 7:00 PM to 9:00 PM, avoiding legal holidays and late-night rest periods; if the core relevance is less than the high threshold but greater than or equal to the medium threshold, the suggested communication time is during lunch breaks at work; if the core relevance is less than the medium threshold, the suggested communication time is during working hours, but high-frequency working hours should be avoided; where the core relevance is... The high-relevance threshold and the medium-relevance threshold are tiered thresholds for core relevance. The high-relevance threshold is greater than the medium-relevance threshold, and both range from 0 to 1. They are determined by statistically analyzing the communication acceptance feedback of different core relevance nodes at different time periods. For example, the high-relevance threshold can be set based on the statistical result that "when the core relevance is ≥0.7, the communication acceptance rate during non-working hours reaches more than 85%". The timing suggestion results add "recommended communication time period, timing suggestion basis" fields to the basic suggestion matching results, forming structured data of "node number, recommended communication channel, compliant communication script template, recommended communication time period, suggestion matching basis, timing suggestion basis".

[0097] The third step involves performing a communication frequency suggestion operation. Based on the extracted value redundancy, reasonable suggestions are given regarding the communication frequency after the timing suggestion, generating a frequency suggestion result. The communication frequency suggestion refers to a compliant frequency recommended to avoid over-communication based on the irreplaceable value of the node. The suggestion rules are determined by the value redundancy level: If the value redundancy level is "low," meaning there are no substitute nodes, the node information has high reference value for communication response decisions, and the suggested communication frequency is the low redundancy communication frequency suggestion value, with an interval of the low redundancy communication interval suggestion value to ensure sufficient buffer time between two communications and avoid disruption; if the level is "medium," meaning there are some substitute nodes, the suggested communication frequency is the medium redundancy communication frequency suggestion value, with an interval of the medium redundancy communication interval suggestion value; if the level is "high," meaning there are multiple completely substitute nodes, the suggested communication frequency is the high redundancy communication frequency suggestion value, with an interval of the high redundancy communication interval suggestion value. The suggested communication frequency for low redundancy communication is > for medium redundancy communication is > for high redundancy communication is > for high redundancy communication is > for medium ... The recommended communication frequency is as follows: low redundancy communication interval < medium redundancy communication interval < high redundancy communication interval. The recommended communication frequency for low redundancy, medium redundancy, and high redundancy communication should range from 1 to 2 times. The recommended communication interval for low redundancy, medium redundancy, and high redundancy communication should range from 48 to 72 hours to ensure sufficient time intervals and reduce inconvenience to stakeholders. All recommended values ​​are determined by analyzing the balance between the effectiveness of information and the acceptance of stakeholders at nodes with different levels of value redundancy. For example, the recommended communication frequency for low redundancy communication can be set based on the statistical result that "the effectiveness of information in one communication at a node with low value redundancy is over 90%". The frequency recommendation results are based on the timing recommendation results and include the fields of "recommended communication frequency, recommended communication interval, and basis for frequency recommendation".

[0098] The fourth step involves implementing compliance constraints on sensitive attributes. Based on the extracted sensitive attributes, compliance constraints are added to the frequency suggestion results, generating a compliance-constrained result. These sensitive attribute compliance constraints refer to specific compliance rules for implicit nodes with sensitive attributes, such as "public officials," "guardians of minors," and "those privy to privacy information," used to mitigate compliance risks and ethical issues in communication. For example, if the sensitive attribute is "public officials," the constraint "written notification must be filed with their unit's compliance department + confirmation of explicit authorization from the other party is required before communication" is added. If the sensitive attribute is "guardians of minors," the constraint "communication content must not involve any information about the minor + communication is only allowed during periods explicitly authorized by the guardian" is added. If the sensitive attribute is "those privy to privacy information," the constraint "no details of privacy information should be mentioned in the communication script + only official, compliant, and friendly repayment reminders are conveyed." The compliance-constrained result adds "compliance constraint type, constraint execution requirements, and constraint basis" fields to the frequency suggestion result, ensuring that all communication suggestions comply with personal information protection laws and ethical norms.

[0099] After completing the above suggestion matching, a cross-shard related data linkage and integration operation is performed. Based on the extracted high-value association paths and dynamic index services, and the compliance constraints of each associated node on different shard retrieval paths across the distributed storage cluster, an association suggestion integration scheme is generated. The cross-shard related data linkage and integration refers to: First, parsing the node sequence in the high-value association path and extracting the node number and node type of each node; Second, locating the corresponding partition of the dynamic index service based on the node type, querying the storage shard location of each node, and verifying retrieval permissions through the authorization verification module before querying; Third, sending an encrypted data retrieval request to the corresponding storage node, the request containing the node number and a description of the retrieval purpose, and obtaining the compliance constraints of each node; Fourth, structurally integrating the communication suggestions of each node according to the logical order of the association path to generate an association suggestion integration scheme; The association suggestion integration scheme only records "the order of associated nodes, the core information of each node's communication suggestions, the integration logic, and the information source shard identifier", and does not contain any actual communication instructions.

[0100] Finally, the results after compliance constraints are integrated with the related recommendations to generate a communication and response plan for debtor-related nodes, and distributed storage archiving is completed. The debtor-related node communication response plan is a compliance reference document containing communication suggestions for individual implicit nodes and integrated solutions for related nodes. Each plan includes "plan identifier, target implicit node information, basic communication suggestions, recommended communication time period, recommended communication frequency and interval, compliance constraints, integrated solution for related suggestions, plan generation time, data source shard identifier, and compliance review status". After generation, the communication response plan is archived to the plan archiving shard in the distributed storage cluster according to the rule of "plan identifier + generation time". The plan archiving shard is independent of the evaluation table shard and is only accessible to the compliance review module and the communication plan management module, triggering an automatic compliance review process: First, the plan is pushed to the compliance review module, which verifies whether the plan complies with all compliance clauses; Second, if the review is passed, the "compliance review status" of the plan is updated to "passed", and the review comments and reviewer are recorded; Third, if the review fails, specific modification suggestions are marked, such as "the script template needs to further hide the debtor information", and the plan is returned to the communication plan generation module for re-optimization.

[0101] In addition, to ensure traceability, a solution archive log is generated, recording the "archiving time, storage shard location, authorized access list, review record, and modification record" for each communication and response solution. The log and solution are stored in encrypted form for subsequent compliance audits and solution optimization analysis. The dynamic index service is updated synchronously, adding a "solution identifier-storage location" mapping entry to the "solution index partition". The entry includes "solution identifier, solution archive shard identifier, encrypted IP address of storage node, encrypted offset of the solution within the shard, and index creation time", which facilitates the subsequent authorization module to quickly retrieve approved communication and response solutions.

[0102] The purpose and effect of this step are as follows: Distributed sharded storage, through a combination of "node type + priority recommendation level" rules and a hash sharding algorithm, disperses large-scale evaluation table data and communication response plans for storage. This not only solves the storage pressure and access bottleneck problems of traditional centralized storage in large-scale data scenarios, but also ensures that data storage complies with personal information protection regulations and avoids the risk of data leakage through encrypted storage and access control. Furthermore, the sharding structure based on business attributes allows subsequent solution retrieval to accurately locate high-value data shards, improving data retrieval efficiency. The dynamic indexing service, through a partitioned encrypted hash index table and a real-time update mechanism, achieves millisecond-level location tracking from implicit node numbers and communication response plan identifiers to storage locations, solving the problems of long data query paths and slow response times in distributed storage. Simultaneously, all query and update operations generate traceable logs, meeting the full-process recording requirements of compliance auditing. The assessment table-driven communication response suggestion matching algorithm generates compliant and mild communication suggestions through multi-dimensional decision-making, avoiding the problem of "emphasizing execution and neglecting compliance" in traditional communication solutions. All suggestions are presented in the form of "reference" and "hint" and do not trigger actual behavior. They only serve as the basis for subsequent communication response decisions, reducing compliance risks from the source. Cross-segment related data linkage and integration breaks down data silos between segments and integrates complete suggestion information through high-value correlation paths, providing comprehensive reference for decision-making.

[0103] Compared to traditional solutions that include actual execution steps, this approach focuses on the core objectives of "distributed storage - dynamic indexing - communication suggestion generation and archiving," completely separating actual communication and execution activities and placing the technical emphasis on efficient data storage and rapid indexing. This approach retains the technical advantages of distributed storage and dynamic indexing while ensuring that the entire process complies with laws, regulations, and public order and good morals. It provides solid data and technical support for subsequent compliant and moderate communication response decisions, while also adapting to a distributed deployment system architecture to meet the needs of high-volume, high-compliance scenarios.

[0104] For example, for the latent node Zhang San (node ​​ID anonymized: C_Comm_188, node type: community role, priority recommendation level: S), the associated value tags extracted from the latent key node multi-dimensional evaluation table are "location transmission + community interaction", with a core correlation of 0.85, value redundancy of "low", and sensitive attributes of "none". The high-value association path is "Zhang San → part-time boss (node ​​ID anonymized: C_Work_138, node type: workplace role)". First, the evaluation table data of Zhang San is hashed according to the combination rule of "community role + S". The sharding algorithm calculates the MD5 hash value of the combined string "community role + S" modulo 1, and stores it in shard 1 (encrypted IP of storage node: 192.168.1.***). The dynamic indexing service records its index entry in the "community role" partition as "key: C_Comm_188, value: shard 1 + 192.168.1.*** + encrypted offset 0x0012 + 202509121000"; the algorithm matches the basic communication suggestion as "written notification letter + community-friendly communication language", combined with a core relevance of 0.85 (≥ high core relevance). (Threshold = 0.7) The recommended communication time is 19:00-20:00. Based on the "low" value redundancy, the recommended communication frequency is 1 time, and the recommended interval for low redundancy communication is 72 hours. There are no sensitive attribute constraints. The storage location of the part-time boss is queried through the dynamic index service (the "Workplace Role" partition index shows it is stored in shard 3, encrypted IP: 192.168.1.***). After retrieving the communication suggestions, the integrated suggestion solution is generated as follows: "First, extract the location-friendly suggestion for Zhang San, then combine it with the income-friendly suggestion for the part-time boss." This process creates a complete communication and response reference. Finally, a communication and response plan for the debtor's related nodes (identifier: S_Comm_20250912_001) is generated and archived to the "Plan Archive Fragment," triggering the compliance review process. The dynamic index service adds an index entry to the "Plan Index Partition" with the key "S_Comm_20250912_001, value: Plan Archive Fragment + 192.168.1.*** + Encrypted Offset 0x0034 + 202509121010". The plan archive log records the archiving details, awaiting retrieval and reference by subsequent authorized modules.

[0105] It is important to note that the data generated or collected by this invention is obtained and used in compliance with the law, and is solely for internal risk assessment and decision-making reference within financial institutions. It does not contain any external communication instructions, nor does it directly trigger any external contact. Any subsequent customer service or risk management measures based on this report must be conducted independently by authorized personnel of the relevant institution in strict compliance with national laws and regulations such as the Personal Information Protection Law. It is strictly prohibited to use this system for any form of harassment or violation of personal privacy. The primary purpose of this system is to enhance risk identification capabilities, not to guide or execute external communication.

[0106] Example 2

[0107] This embodiment, based on Embodiment 1, provides a distributed storage system for dynamic index querying of debtor relationship graphs, such as... Figure 4 As shown, it includes:

[0108] A distributed storage system for dynamic indexing and querying of debtor relationship graphs, the system comprising:

[0109] Multi-source associated data processing module: Uses scenario-based attribute completion and cross-source conflict resolution methods to process existing multi-source associated data and generate a panoramic attribute matrix of debtor associated nodes;

[0110] The related value directed graph construction module is used to construct a multi-level related value directed graph of debtors based on the attribute values, credibility and timeliness of the panoramic attribute matrix of debtor related nodes, using a dynamic weighted association graph construction and related value annotation mechanism.

[0111] Latent node value assessment module: Based on the global structure and value attributes of a multi-level associated value directed graph, the algorithm for evaluating the value of latent nodes on the graph is used to obtain a multi-dimensional evaluation table of latent key nodes.

[0112] Precise Targeting Execution Package Generation Module: Based on the multi-dimensional evaluation table of implicit key nodes, combined with distributed sharded storage and dynamic indexing services, the module uses an evaluation table-driven communication and response suggestion matching algorithm to collaboratively process table data and storage resources, generate communication and response plans for debtor-related nodes, and complete distributed storage archiving.

Claims

1. A distributed storage method for dynamic index query of debtor relationship graphs, characterized in that, The method includes: Step S10: Process the existing multi-source associated data using the scenario-based attribute completion and cross-source conflict resolution methods to generate a panoramic attribute matrix of debtor associated nodes; Step S20: Based on the attribute values, credibility, and timeliness of the panoramic attribute matrix of the debtor's related nodes, a multi-level directed graph of debtor's related value is formed by using a dynamic weighted association graph construction and association value labeling mechanism. Step S30: Based on the global structure and value attributes of the multi-level associated value directed graph, the hidden node value walk evaluation algorithm on the graph is used to obtain a multi-dimensional evaluation table of hidden key nodes. Step S40: Based on the multi-dimensional assessment table of hidden key nodes, combined with distributed sharded storage and dynamic indexing services, the table data and storage resources are processed collaboratively through the communication response suggestion matching algorithm driven by the assessment table to generate communication response plans for debtor-related nodes and complete distributed storage archiving. The mechanism for constructing the dynamic weighted association graph and labeling the association value is as follows: Based on the nodes and node attribute values ​​extracted from the panoramic attribute matrix of debtor-related nodes, a value vertex set of a multi-level related value directed graph of debtors is generated. After completing the construction of the value vertex set, the traditional association attributes and scenario-based attributes in the panoramic attribute matrix of debtor-related nodes are used to determine the association relationships between nodes and generate directed edges. For each generated directed edge, label it with an associated value tag and an information transmission coefficient; Perform edge weight calculation operation, and combine the attribute credibility score, the value weight of the associated value tag and the information transmission coefficient in the panoramic attribute matrix of the debtor's associated nodes to calculate the edge weight of each directed edge; After the edge weights are calculated, the node level is labeled for each node based on the node features of the value vertex set and the edge features of the value directed edge set, thus generating a node level set. Based on the association duration of all edges in the value directed edge set, calculate the global average association duration and generate a graph global duration marker; Generate a multi-level value graph of debtors; The algorithm for evaluating the value of hidden nodes on the graph is as follows: Perform walk parameter initialization operation, and determine the walk probability calculation parameters and iteration termination conditions based on the extracted edge weight set and value directed edge set, and generate walk parameter set; Taking the core node in the debtor's multi-level associated value directed graph as the starting point, and based on the walking probability coefficient in the walking parameter set, it iterates through the graph according to the set probability. During the process, the value walking score of each hidden node is accumulated in real time until the iteration number threshold or the convergence error threshold is met, and the value walking score of each hidden node is output. Based on the global structure of the debtor's multi-level associated value directed graph, and combining the extracted value directed edge set with the graph's global timeliness marker, the core association degree, value redundancy degree, and graph global timeliness matching degree of each hidden node are calculated to generate a global dimension evaluation set. By integrating information from the set of hidden nodes, value walk scores, and global dimension evaluation sets, a multi-dimensional evaluation table of hidden key nodes is generated according to a preset dimensional structure.

2. The distributed storage method for dynamic index query of debtor relationship diagram according to claim 1, characterized in that, The method for contextualized attribute completion and cross-source conflict resolution is as follows: Each associated entity in the multi-source associated data is identified as a node and a node number is generated; basic attributes are matched for each node, which are obtained by extracting and de-identifying legal identity information fragments in the multi-source associated data. Based on the interaction features and scene information in multi-source associated data, scene-based attributes are completed for each node to obtain a preliminary attribute set. Identify conflicting data in the initial attribute set; Calculate the data source credibility of conflicting data; combine the data source credibility of conflicting data with the data update time to determine the optimal attribute value; mark each attribute that has undergone conflict resolution with a conflict resolution flag, which includes whether conflict processing has been carried out and the basis for processing, and mark each attribute with a credibility score. After completing the scenario-based attribute completion and cross-source conflict resolution, the set of attributes after conflict resolution is obtained.

3. The distributed storage method for dynamic index query of debtor relationship diagram according to claim 2, characterized in that, The panoramic attribute matrix of the debtor-related nodes is a two-dimensional matrix structure, with the row dimension representing the node number and the column dimension representing the attribute dimension. Each cell of the matrix stores the attribute value and the update timestamp.

4. The distributed storage method for dynamic index query of debtor relationship diagram according to claim 1, characterized in that, The debtor's multi-level associated value directed graph includes a graph structure consisting of a value vertex set, a value directed edge set, an edge weight set, a node hierarchy set, and a graph global time-limited marker.

5. The distributed storage method for dynamic index query of debtor relationship diagram according to claim 1, characterized in that, The hidden key node multi-dimensional evaluation table is a two-dimensional structured table, with each row representing a hidden node in the set of hidden nodes.

6. The distributed storage method for dynamic index query of debtor relationship diagram according to claim 1, characterized in that, The distributed sharded storage, based on the extracted node type and priority recommendation level, uses a hash sharding algorithm to divide the evaluation table into different shards of the distributed storage cluster according to the combination rules of node type and priority recommendation level, generating a sharded storage mapping table.

7. The distributed storage method for dynamic index query of debtor relationship graph according to claim 1, characterized in that, The dynamic indexing service refers to a compliant indexing mechanism used to record the mapping relationship between implicit node numbers and corresponding storage shard locations in real time. The compliant indexing mechanism has a partitioned encrypted hash index table. The index table is partitioned according to node type. The key in each partition is the implicit node number, and the partition value is represented as <shard identifier + encrypted IP address of storage node + encrypted offset of data in shard + index update timestamp>. When the data in the implicit key node multi-dimensional evaluation table is updated, the dynamic indexing service will automatically trigger the index update operation.

8. A distributed storage system for dynamic index query of debtor relationship graphs, used to implement the method of any one of claims 1-7, characterized in that, The system includes: Multi-source associated data processing module: Uses scenario-based attribute completion and cross-source conflict resolution methods to process existing multi-source associated data and generate a panoramic attribute matrix of debtor associated nodes; The related value directed graph construction module is used to construct a multi-level related value directed graph of debtors based on the attribute values, credibility and timeliness of the panoramic attribute matrix of debtor related nodes, using a dynamic weighted association graph construction and related value annotation mechanism. Latent node value assessment module: Based on the global structure and value attributes of a multi-level associated value directed graph, it uses the latent node value walk assessment algorithm on the graph to obtain a multi-dimensional assessment table of latent key nodes. Precise Targeting Execution Package Generation Module: Based on the multi-dimensional evaluation table of implicit key nodes, combined with distributed sharded storage and dynamic indexing services, the module uses an evaluation table-driven communication and response suggestion matching algorithm to collaboratively process table data and storage resources, generate communication and response plans for debtor-related nodes, and complete distributed storage archiving.

Images