A big data-based financial institution customer value stratified management system and method

Abstract

The application discloses a financial institution customer value stratified management system and method based on big data, and particularly relates to the field of financial customer management, and is used for solving the problems of static stratification results, difficulty in reflecting structural correlation between customers and multi-period behavior changes in the existing customer value grading management process. The application maps network structure risks to the customer level to obtain customer network value gains by constructing a customer correlation network with time sequence weights between customers, identifying customer potential communities and evaluating community structure vulnerability. Meanwhile, the application decomposes the trend of multi-period transaction behavior of customers and constructs an behavior stability index. On this basis, a two-dimensional customer value stratification matrix is constructed to realize dynamic stratification of customer value, and targeted management strategies are generated according to the historical management benefits of adjacent stratified customer groups, so as to improve the accuracy of customer value identification and the scientificity of management decisions of the financial institution.

Description

Technical Field

[0001] This invention relates to the field of financial customer management technology, and more specifically, to a customer value tiering management system and method for financial institutions based on big data. Background Technology

[0002] As financial institutions continue to expand their business scale and diversify their customer types, traditional customer segmentation and management methods that rely on single financial indicators or static labels are no longer suitable for complex and ever-changing real-world business scenarios. In actual business operations, financial institution customers often form implicit risk-related networks through guarantees, joint guarantees, and related-party transactions. These relationships have time-evolutionary characteristics; once the operating conditions of a local customer change, a chain reaction may occur along the related paths, thereby affecting other customers and even the overall asset quality.

[0003] Meanwhile, customers' transaction behavior exhibits clear phases and fluctuations across different periods, making it difficult to accurately depict their long-term operational stability based solely on short-term indicators. Existing customer value assessment and management methods often focus on single-dimensional analysis, failing to unify the modeling of structural relationships between customers with the changing characteristics of their own behavior. This results in static customer segmentation results, lagging strategy adjustments, and an inability to support refined and dynamic business decisions.

[0004] Against this backdrop, there is an urgent need for a customer value segmentation and management method that can comprehensively characterize the customer network structure, risk transmission characteristics, and multi-cycle transaction behavior stability in real financial business scenarios, so as to improve the accuracy of financial institutions in identifying customer value and the pertinence of adjusting their business strategies. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a big data-based customer value tiered management system and method for financial institutions to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A big data-based method for segmenting and managing customer value in financial institutions, characterized by the following steps:

[0008] S1. Collect data on indirect credit guarantee relationships between financial institution customers and construct a heterogeneous customer relationship network with time-series weights;

[0009] S2. Perform a modularity-based overlapping community discovery analysis on the heterogeneous customer association network to identify the potential communities to which customers belong and generate a membership vector for each customer containing different connection strengths of potential communities.

[0010] S3. Extract the internal network topology of each potential community, quantify the risk transmission efficiency within the potential community based on the shortest path distribution between nodes and the variance of edge weights, and calculate the structural vulnerability index of the potential community.

[0011] S4. Aggregate the potential community membership vector of each customer with the structural vulnerability index of each potential community to obtain the customer's value gain coefficient.

[0012] S5. Perform trend decomposition on the client's multi-period financial behavior sequence and construct a behavioral stability index vector for client trading fluctuations.

[0013] S6. Construct a two-dimensional value stratification matrix consisting of a customer behavior stability index vector and a network value gain, and map customers to the corresponding grid positions in the value stratification matrix.

[0014] S7. In the value stratification matrix, based on the historical operating efficiency of the customer groups in the adjacent grids of the customer's grid, a nearest neighbor interpolation method is used to generate operating strategies for the customer.

[0015] As a further aspect of the present invention, in step S1, constructing a heterogeneous customer association network with temporal weights specifically includes:

[0016] Data on indirect credit guarantee relationships between customers are collected from credit information of financial institutions, including the guarantor customer identifier, the guaranteed customer identifier, the effective date of the guarantee, and the guarantee amount;

[0017] Treat each customer as a network node, and each guarantee relationship as a directed edge connecting the corresponding guarantor node and the guaranteed party node;

[0018] Based on the difference between the guarantee effective date and the current analysis date, the time decay coefficient of each guarantee relationship is calculated. The result of multiplying the guarantee amount by the time decay coefficient is used as the temporal weight of the directed edge in the customer association network, thus constructing a heterogeneous customer association network with temporal weights.

[0019] As a further aspect of the present invention, in step S2, identifying the potential communities to which a customer belongs and generating a membership vector for each customer containing different connection strengths among potential communities specifically includes:

[0020] On a heterogeneous customer association network, a label propagation algorithm based on modularity optimization is used to update the potential community labels of customer nodes and perform potential community segmentation on the network.

[0021] Identify nodes that are assigned more than one potential community label during the iteration process, mark them as overlapping nodes, and record all potential community sets to which the corresponding customer node belongs;

[0022] Based on the set of potential communities to which each customer node belongs, the sum of the weights of the directed edges between the customer node and its potential community is used as the connection strength component. The connection strength components corresponding to all potential communities are combined to form the customer's membership vector.

[0023] As a further aspect of the present invention, in step S3, calculating the structural vulnerability index of a potential community specifically includes:

[0024] For each identified potential community, a subnetwork corresponding to the potential community is extracted from the heterogeneous customer association network. The subnetwork contains all customer nodes of the potential community and the directed edges between them.

[0025] Calculate the shortest path length between each pair of nodes in each sub-network, where the shortest path length is the sum of the temporal weights of each edge on the path, and calculate the average shortest path length of each sub-network based on the shortest path length of all node pairs.

[0026] The variance of the temporal weights of all edges in each sub-network is calculated to obtain the edge weight variance. The scalar value obtained by linearly combining the average shortest path length and the edge weight variance is used as the structural vulnerability index of the potential community.

[0027] As a further aspect of the present invention, in step S4, the aggregation of each customer's potential community membership vector with the structural vulnerability index of each potential community to which it belongs to obtain the customer's value gain coefficient specifically includes:

[0028] For each customer, a separate analysis is performed. The connection strength components of each potential community in the current customer's potential community membership vector are aggregated with the structural vulnerability index of the corresponding potential community, and the risk allocation value of the current customer in each potential community is obtained proportionally.

[0029] The risk allocation value of the current customer across all potential communities is summed up, and the corresponding discount ratio is retrieved from the preset mapping table as the value gain coefficient for that customer.

[0030] As a further aspect of the present invention, in step S5, constructing the behavioral stability index vector of customer transaction fluctuations specifically includes:

[0031] Obtain transaction records for each customer within a preset historical time period, perform piecewise linear fitting on the transaction sequence, identify time points in the sequence where the fitting residual growth exceeds a preset mutation threshold, and mark them as potential behavioral inflection points.

[0032] Within the behavioral stages defined by continuous potential behavioral inflection points, the autocorrelation coefficients of the transaction flow in the corresponding behavioral stages are calculated. The number and distribution density of all behavioral inflection points, as well as the autocorrelation coefficients of each stage, are integrated to construct a stability index vector.

[0033] As a further aspect of the present invention, in step S6, constructing a two-dimensional value layering matrix consisting of a customer behavior stability index vector and network value gain specifically includes:

[0034] The network value gain coefficient and behavioral stability index vector of all customers are collected. The percentile quantiles of the first principal component score set of the network value gain coefficient set and the behavioral stability index vector are calculated respectively. The numerical intervals are dynamically divided based on the quantiles as the grid boundary of the value stratification matrix.

[0035] Based on the numerical range of each customer's network value gain coefficient and behavioral stability principal component score, determine their grid coordinates in the value stratification matrix, and generate a unique stratification identifier for each grid coordinate.

[0036] As a further aspect of the present invention, in step S7, generating the business strategy specifically includes:

[0037] Adjustable customers are selected based on the historical changes in each customer's operating performance. In the value stratification matrix, the grid corresponding to the adjustable customer and the coordinates of adjacent grids are retrieved based on the stratification identifier.

[0038] Obtain the median historical operating efficiency growth rate of the grid corresponding to the adjustable customer and other customer groups in adjacent grids, and select the adjacent grid with the highest median historical operating efficiency growth rate as the target grid;

[0039] Calculate the coordinate offset between the adjustable customer's corresponding grid and the target grid in the value stratification matrix. Based on the business strategy corresponding to the customer in the target grid, adjust the parameters of the business strategy corresponding to the adjustable customer according to the Manhattan distance and direction of the coordinate offset. The adjustment range is proportional to the Manhattan distance, thus generating the business strategy for the adjustable customer.

[0040] On the other hand, the present invention provides a customer value tiering management system for financial institutions based on big data, comprising:

[0041] The network construction module is used to collect data on indirect credit guarantee relationships between financial institution customers and to construct a heterogeneous customer relationship network with time-series weights by introducing guarantee amount and time decay factors.

[0042] The community discovery module is used to perform overlapping community discovery analysis based on modularity optimization in heterogeneous customer association networks, identify the potential communities to which customers belong, and generate a membership vector representing the strength of customer associations across multiple communities based on the connection weights within the community.

[0043] The vulnerability assessment module is used to extract corresponding sub-networks for each potential community and calculate the structural vulnerability index, which reflects the stability of the community structure, based on the distribution of the shortest path of nodes and the variance of edge weights.

[0044] The behavioral stability assessment module is used to aggregate the customer's community membership vector with the corresponding community's structural vulnerability index, calculate the customer's network value gain coefficient, and construct a behavioral stability index vector by combining the trend decomposition results of multi-period transaction behavior sequences.

[0045] The business strategy generation module is used to construct a customer value stratification matrix based on the network value gain coefficient and behavioral stability index, determine the customer stratification position, and generate targeted customer management strategies based on the historical operating performance of adjacent customer groups through nearest neighbor interpolation.

[0046] The technical effects and advantages of this invention, a customer value stratification management system and method for financial institutions based on big data, are as follows:

[0047] This invention introduces a customer relationship network analysis and multi-cycle behavior modeling mechanism based on big data to achieve dynamic, three-dimensional, and hierarchical management of customer value for financial institutions. Compared with traditional customer evaluation methods based on single financial indicators or static labels, it can more realistically reflect the comprehensive value and influence of customers in actual business scenarios.

[0048] This invention incorporates indirect credit guarantee relationships between customers into a unified modeling framework, combining a time decay mechanism to characterize the timeliness of risk associations and effectively reveal potential structural risk transmission paths. Through potential community identification and structural vulnerability assessment, it maps community-level risk characteristics to the customer level, enhancing the structural interpretability of customer value assessment. Simultaneously, this invention performs multi-period trend decomposition on customer transaction behavior, constructing behavioral stability indicators to effectively distinguish between short-term fluctuations and long-term operational characteristics, avoiding misjudgments caused by occasional behavioral changes. Based on this, a two-dimensional customer value stratification matrix is ​​constructed, enabling dynamic updates to customer stratification results. Furthermore, through a strategy generation mechanism based on nearest neighbor interpolation, operational strategy adjustments are made continuous, interpretable, and controllable, thereby improving the precision of financial institutions' customer operational decisions, risk identification capabilities, and overall asset allocation and management efficiency. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of a big data-based customer value segmentation management method for financial institutions according to the present invention.

[0050] Figure 2 This is a schematic diagram of the structure of a big data-based customer value tiered management system for financial institutions, as described in this invention. Detailed Implementation

[0051] 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0052] Example 1

[0053] Figure 1 This invention presents a method for segmenting and managing customer value in financial institutions based on big data, which includes the following steps:

[0054] S1. Collect data on indirect credit guarantee relationships between financial institution customers and construct a heterogeneous customer relationship network with time-series weights;

[0055] S2. Perform a modularity-based overlapping community discovery analysis on the heterogeneous customer association network to identify the potential communities to which customers belong and generate a membership vector for each customer containing different connection strengths of potential communities.

[0056] S3. Extract the internal network topology of each potential community, quantify the risk transmission efficiency within the potential community based on the shortest path distribution between nodes and the variance of edge weights, and calculate the structural vulnerability index of the potential community.

[0057] S4. Aggregate the potential community membership vector of each customer with the structural vulnerability index of each potential community to obtain the customer's value gain coefficient.

[0058] S5. Perform trend decomposition on the client's multi-period financial behavior sequence and construct a behavioral stability index vector for client trading fluctuations.

[0059] S6. Construct a two-dimensional value stratification matrix consisting of a customer behavior stability index vector and a network value gain, and map customers to the corresponding grid positions in the value stratification matrix.

[0060] S7. In the value stratification matrix, based on the historical operating efficiency of the customer groups in the adjacent grids of the customer's grid, a nearest neighbor interpolation method is used to generate operating strategies for the customer.

[0061] In S1, a heterogeneous customer association network with temporal weights is constructed.

[0062] Structured data collection and processing are conducted on existing credit data sources from financial institutions. The collected data is explicitly limited to indirect credit guarantee relationships between customers that have already occurred and have legal effect in credit transactions. Specific data collection includes, but is not limited to: unique identifiers of the guarantor and the guaranteed party, the effective date of the guarantee relationship, and the corresponding guarantee amount. Both the guarantor and guaranteed party identifiers are represented using the financial institution's internal standardized customer coding rules to ensure no duplicate names or mappings occur during subsequent processing. The guarantee effective date is based on the official effective date recorded in the credit contract or guarantee agreement, and the guarantee amount is based on the maximum guarantee amount stipulated in the contract or the actual effective guarantee amount. Each customer entity involved in a guarantee relationship is abstracted as an independent node in a customer association network, with each node corresponding to a unique customer identifier. Simultaneously, each collected guarantee relationship is mapped to a directed edge, with the starting node pointing to the guarantor customer node and the ending node pointing to the guaranteed party customer node, thus clearly expressing the directionality of guarantee liability in the network structure. Through the above methods, a complete and clearly directional customer guarantee relationship network structure is formed within the entire customer set.

[0063] After constructing the nodes and directed edges in the customer association network, a time dimension is introduced for each guarantee relationship to characterize the change in the degree of risk transmission impact of the guarantee relationship over time. For each guarantee relationship, a time decay coefficient is calculated based on the time difference between its effective date and the current analysis date. The time decay coefficient reflects the recentity of the guarantee relationship, and its setting is as follows: the closer the effective date of the guarantee relationship is to the current analysis date, the larger the time decay coefficient value; the further the effective date of the guarantee relationship is from the current analysis date, the smaller the time decay coefficient value. The decay coefficient adopts a normalized expression form. For example, in a typical implementation, the time decay coefficient corresponding to guarantee relationships with an effective date within one year can be set to 1.0, the time decay coefficient corresponding to guarantee relationships with an effective date between one and three years can be set to 0.7, and the time decay coefficient corresponding to guarantee relationships with an effective date more than three years can be set to 0.4. Subsequently, the guarantee amount corresponding to each guarantee relationship is multiplied by the above time decay coefficient, and the result is used as the time-series weight of the directed edge in the customer association network. In this way, guarantee relationships with larger guarantee amounts and more recent effective dates are represented as directed edges with higher weights in the network, while guarantee relationships with smaller guarantee amounts or more recent effective dates correspond to relatively lower weights. Ultimately, all directed edges with time-series weights, together with customer nodes, constitute a heterogeneous customer association network with time-sensitive characteristics.

[0064] In step S2, the potential communities to which the customer belongs are identified, and a membership vector containing different connection strengths of potential communities is generated for each customer.

[0065] For the established heterogeneous customer association network, a label propagation process based on modularity optimization is performed on all customer nodes in the network to achieve automatic segmentation of potential communities. Specifically, each customer node in the network is first assigned an initial potential community label. This label can be generated using the node's own ID or sequential ID to ensure the uniqueness of each node's label in the initial state. Then, an iterative label propagation phase begins. In each iteration, the potential community labels of each customer node are updated one by one according to the directed edge connections in the customer association network. The label update rule is set as follows: the current customer node prioritizes receiving the potential community label that appears most frequently among its neighboring nodes and has the largest cumulative edge weight, thus allowing high-weight connections to play a dominant role in the community formation process. To avoid excessive label fluctuations in a single iteration, a modularity optimization constraint is introduced: the label update result is only confirmed if it improves or maintains the overall modularity level of the current network. The aforementioned label propagation and modularity verification process is continuously iterated until the overall modularity change of the network is lower than a preset stable threshold in several consecutive iterations. For example, if the modularity increment is less than 0.01 for three consecutive iterations, the potential community partitioning process is considered to have converged. During the iteration process, the potential community labels obtained by the client nodes in each round are recorded synchronously. When the same client node stably obtains two or more different potential community labels in different iteration rounds, the client node is marked as an overlapping node, and the potential community labels that appear more frequently during the iteration process are recorded together as the potential community set of that client node.

[0066] After completing the potential community segmentation and obtaining the potential community set corresponding to each customer node, the association strength between the customer node and its corresponding potential community is further quantified to construct the customer's membership vector. For each customer node, its potential community set is traversed one by one. For each potential community in the set, all directed edges between the customer node and other nodes within the potential community are selected from the customer association network. The directed edges include both edges from the current customer node to other customer nodes within the community and edges from other customer nodes within the community to the current customer node, to fully reflect the bidirectional association of the customer in the potential community. Subsequently, the temporal weights of all the selected directed edges are accumulated, and the accumulated result is used as the connection strength component of the customer node in the corresponding potential community. For example, in an example scenario, if a customer node has three directed edges in a potential community with temporal weights of 50, 30, and 20 respectively, then the connection strength component of the customer in that potential community is recorded as 100. By performing the above statistical process on all potential communities to which the same customer node belongs, a set of connection strength components for that customer node in different potential communities can be obtained. Finally, according to the numbering or identification order of the potential communities, the above connection strength components are combined sequentially to form a membership vector that characterizes the degree of association of the customer node across multiple communities. This membership vector numerically and intuitively reflects the depth of the customer's participation and structural position in different potential communities.

[0067] In S3, the structural vulnerability index of potential communities is calculated.

[0068] For each identified potential community, a sub-network structure corresponding to that potential community is extracted from the heterogeneous customer association network. The sub-network is constructed as follows: using all customer nodes included in the potential community as a node set, and filtering all directed edges in the original customer association network whose start and end points are both located within this node set, forming a network structure that only reflects the internal relationships within the potential community. After sub-network extraction, the shortest path length is calculated for each node pair within the sub-network. Since the customer association network is a directed network, the shortest path calculation is only performed on node pairs with a directed reachable path to avoid interference from unconnected node pairs in the statistical results. The path length of each path is defined as the sum of the temporal weights corresponding to all directed edges on that path, where the temporal weights have already reflected the guarantee amount and time decay factors in previous steps. In the specific calculation process, for each pair of reachable nodes, a layer-by-layer expanding path search method is adopted, prioritizing paths with smaller cumulative temporal weights as candidate paths, until the shortest path length between the node pairs is determined. After calculating the shortest path for all node pairs, the shortest path lengths are summed and averaged using the number of node pairs as the base. This yields the average shortest path length for the subnetwork corresponding to the potential community. For example, in a potential community, if there are 10 valid node pairs with shortest path lengths concentrated between 80 and 120, the average shortest path length is the sum of the individual path lengths.

[0069] After calculating the average shortest path length of the potential community subnetwork, the edge weight distribution characteristics within the subnetwork are quantitatively analyzed to reflect the equilibrium of the community structure. For each potential community subnetwork, the temporal weight values ​​corresponding to all directed edges are collected to form a set of edge weights. Subsequently, statistical analysis is performed on this set of edge weights to calculate the dispersion of each edge weight relative to the overall average weight. The resulting statistical result is the edge weight variance. The larger the edge weight variance, the more likely there are a few key association edges with significantly higher or lower weights within the potential community, indicating a higher dependence of the community structure on these edges and relatively weaker overall structural stability. Conversely, when the edge weight variance is small, it indicates a more balanced distribution of relationships within the community, and a more dispersed transmission path for risks or influences within the community. After calculating the edge weight variance, it is linearly combined with the aforementioned average shortest path length to generate a scalar value that characterizes the overall structural features of the potential community. The linear combination is implemented by using the average shortest path length and the variance of edge weights as two positive contributing factors, respectively. These are then weighted and summed to obtain the final result. The proportions of each factor are set in a fixed configuration, such as equal weighting, to ensure the comparability of structural vulnerability indices among different potential communities. The resulting structural vulnerability index can simultaneously reflect the characteristics of risk transmission distance and the uneven distribution of correlations within potential communities.

[0070] In step S4, the potential community membership vector of each customer is aggregated with the structural vulnerability index of each potential community to which it belongs, to obtain the customer's value gain coefficient.

[0071] Each customer is analyzed independently. Based on the customer's obtained potential community membership vector, the degree of association between the customer and different potential communities is quantified. Specifically, the customer's potential community membership vector is first read. This vector contains the customer's connection strength components in each potential community. These connection strength components numerically reflect the cumulative weight of the guarantee relationship between the customer and other customer nodes within the corresponding potential community. Then, for each potential community involved in the customer's membership vector, the corresponding structural vulnerability index is retrieved. This index was obtained in the previous step by combining the average shortest path length and edge weight variance of the community subnetwork. Next, the structural vulnerability index of the potential community is allocated according to the proportion of the customer's connection strength component in that potential community to the sum of the connection strength components of all potential communities, resulting in the customer's risk allocation value in that potential community. For example, in a sample scenario, if a customer belongs to three potential communities, and their connection strength components in the three potential communities are 40, 30, and 30 respectively, with a total connection strength of 100, then the customer bears 40% of the risk in the first potential community and 30% in the second and third potential communities respectively. Accordingly, the structural vulnerability index of each potential community is allocated to the customer by multiplying them according to the above proportions, forming multiple risk allocation values ​​as the user's tiered operational risk value.

[0072] After calculating the risk allocation value of the current customer in each potential community, these risk allocation values ​​are aggregated to obtain a comprehensive quantitative result that reflects the overall network role characteristics of the customer. Specifically, the risk allocation values ​​of the current customer in all potential communities are summed one by one to obtain the customer's total personal risk allocation. This total amount represents the comprehensive risk level borne by the customer in the entire potential community structure as a single numerical value. Subsequently, based on this total risk allocation, the corresponding conversion ratio is retrieved from a pre-defined mapping table, and this conversion ratio is determined as the customer's value gain coefficient. The mapping table is set in a segmented interval format, with different intervals corresponding to different conversion ratios. For example, in one example implementation, the total risk allocation can be divided into three intervals: low, medium, and high. The conversion ratio for the low interval is set to 0.8, the conversion ratio for the medium interval is set to 1.0, and the conversion ratio for the high interval is set to 1.2. The interval boundaries and conversion ratios in the mapping table are configured after statistical analysis of historical customer operating data and remain fixed in actual application to ensure the comparability of value gain coefficients between different customers.

[0073] In S5, a behavioral stability index vector for customer transaction fluctuations is constructed.

[0074] For each customer, transaction records within a preset historical time period are obtained. This historical time period is set according to the financial institution's standard analysis cycle for customer business behavior, such as a continuous 24-month or 36-month analysis window to ensure coverage of the customer's complete business changes. The obtained transaction records are organized according to a uniform time granularity, such as monthly or by natural transaction cycle, ensuring continuity and comparability of the transaction sequence over time. After organizing the transaction sequence, piecewise linear fitting is performed. Specifically, the transaction data is fitted segment by segment in chronological order. Each segment uses a linear trend to describe the overall change characteristics of the transaction within that time interval, and the fitting residual for each time point is calculated simultaneously to reflect the deviation between the actual transaction volume and the linear trend. Subsequently, the magnitude of the fitting residual change is continuously monitored. When the residual at a certain time point shows a significant increase or decrease compared to the previous time point, and the magnitude exceeds a preset abrupt change threshold, that time point is determined to be a potential behavioral inflection point. The mutation threshold is determined by statistically analyzing the residual distribution of historical transaction flows. For example, a mutation threshold is set when the magnitude of residual change exceeds twice the standard fluctuation level of historical residuals, so as to ensure that the identified behavioral inflection point corresponds to the actual change in the transaction behavior structure.

[0075] After identifying potential behavioral inflection points, the customer's transaction flow sequence is further divided into stages based on these points. The time interval between two adjacent potential behavioral inflection points is considered an independent behavioral stage. For the beginning and end of the sequence, the time from the start of the analysis window to the first behavioral inflection point and the time from the last behavioral inflection point to the end of the analysis window are respectively considered independent behavioral stages. For each segmented behavioral stage, a subsequence of transaction flow within that stage is extracted, and an autocorrelation coefficient is calculated for this subsequence to characterize the continuity and stability of transaction flow between adjacent time points within that stage. The autocorrelation coefficient is calculated using a uniform time lag step, such as using adjacent time units as the lag interval, to ensure the comparability of autocorrelation results between different customers and different behavioral stages. After calculating the autocorrelation coefficients for each behavioral stage, the overall customer behavioral structure characteristics are integrated, the total number of potential behavioral inflection points is counted, the distribution density of behavioral inflection points within the analysis time window is calculated (e.g., the number of times an inflection point occurs per unit time), and the autocorrelation coefficient values ​​corresponding to each behavioral stage are summarized. Finally, the three types of information are combined according to a predefined index order to construct a stability index vector that characterizes the stability of customer trading behavior. This stability index vector reflects the comprehensive characteristics of the frequency of changes in customer trading behavior during different stages and the level of stability within each stage in a structured form.

[0076] In S6, a two-dimensional value layering matrix is ​​constructed, consisting of a customer behavior stability index vector and network value gain.

[0077] This study aggregates the calculated network value gain coefficients and corresponding behavioral stability index vectors for all customers. The network value gain coefficients are one-dimensional values ​​representing the overall influence of a customer within a potential community structure; the behavioral stability index vectors are multi-dimensional structured data representing the phased changes in customer transaction behavior. To reduce the correlation between multi-dimensional indicators and extract the most representative comprehensive features, principal component analysis (PCA) is performed on the behavioral stability index vectors for all customers. The PCA implementation involves constructing a covariance relationship based on the behavioral stability index vectors of all customers, extracting principal components from high to low feature contribution, and selecting the first principal component with the highest contribution as the comprehensive representation of customer behavioral stability, thus generating a corresponding behavioral stability principal component score for each customer. After obtaining the set of network value gain coefficients and the set of behavioral stability principal component scores, percentile quantiles are calculated for both sets of values. The quantiles are set using a proportional division principle, for example, dividing the overall numerical distribution into four intervals corresponding to the 25%, 50%, and 75% quantile positions, thus forming multiple levels of intervals such as low, low-medium, high-medium, and high. By using percentile quantiles instead of fixed thresholds, it is possible to ensure that the interval division of the value stratification matrix remains dynamically adapted under different customer sizes and different business cycles, avoiding imbalance in the stratification results due to changes in numerical distribution.

[0078] After defining the numerical interval boundaries for both dimensions of the value stratification matrix, the position of each customer within the matrix is ​​determined. For each customer, their corresponding network value gain coefficient and behavioral stability first principal component score are read and compared with the defined quantile interval boundaries to determine which specific interval number they fall into within the network value dimension and behavioral stability dimension. For example, if a customer's network value gain coefficient is in the third quantile interval and their behavioral stability principal component score is in the second quantile interval, their grid coordinates in the value stratification matrix can be determined as "third column, second row". After determining the grid coordinates for all customers, a unique stratification identifier is generated for each grid coordinate in the value stratification matrix. This identifier is generated using a regularized encoding method, such as combining and encoding in the order of "value dimension interval number - stability dimension interval number" to ensure a clear and distinguishable identification relationship between different grid positions. Subsequently, the generated stratification identifiers are mapped to all customers, so that each customer corresponds to a unique value stratification identifier. In this way, the continuous numerical characteristics of customers are uniformly mapped to a discretized and structured value hierarchy matrix, which not only ensures the comparability of hierarchical results between different customers.

[0079] In step S7, a business strategy is generated.

[0080] For all customers in the value stratification matrix, their historical operating performance changes are analyzed to screen for adjustable customers requiring adjustments to their operating strategies. These historical operating performance changes are based on customer operating indicators statistically compiled by the financial institution, such as the changes in customer net asset value, credit balance return rate, or comprehensive return indicators over multiple consecutive statistical periods. In practice, a unified historical evaluation window is selected, such as four consecutive quarters. The operating performance trends of customers within this window are compared. When a customer's operating performance shows a continuous decline or its growth rate is significantly lower than the median level of its customer group within its stratum, the customer is marked as an adjustable customer. After completing the adjustable customer screening, based on the stratification identifier generated for each customer in the previous steps, the current grid coordinates corresponding to the adjustable customer are retrieved in the value stratification matrix. Subsequently, using these grid coordinates as the center, according to the discrete structure rules of the matrix, the directly adjacent grid coordinates in the horizontal and vertical directions are determined. These adjacent grids do not include diagonal positions, only grid positions sharing a boundary, to ensure that the subsequent strategy reference range is limited to the nearest neighboring area with the smallest difference from the current customer's value stratum.

[0081] After determining the coordinates of the current grid and adjacent grids for the adjustable customer, a statistical analysis of the historical operating efficiency of customer groups within each relevant grid is conducted to select target grids as a reference for strategy migration. For the grid where the adjustable customer resides and its adjacent grids, operating efficiency growth rate data for other customers within the corresponding grid are collected within the same historical evaluation window. The operating efficiency growth rate is calculated using a standardized method, such as the percentage increase in operating efficiency indicators at the end of the evaluation period relative to the beginning of the evaluation period, to ensure comparability between different customers. Subsequently, the median value of the operating efficiency growth rate set for each customer group within a grid is calculated to represent the overall operating performance level of that customer group within that grid. Using the median instead of the average effectively reduces the interference of extreme individual customers on the statistical results, making the grid operating efficiency evaluation more robust. After obtaining the median operating efficiency growth rate for each relevant grid, these medians are compared and analyzed, and the adjacent grid with the highest median value is selected as the target grid. This target grid represents the area where, under the current value stratification matrix structure, the customer group adjacent to the adjustable customer level and with the best historical operating performance is located.

[0082] The system retrieves the current grid coordinates of the adjustable customer and the target grid coordinates, and calculates their coordinate offsets within the value stratification matrix. The coordinate offsets consist of the stratification differences in two directions, describing the necessary adjustment directions for the customer in the network value and behavioral stability dimensions. Subsequently, the Manhattan distance between the two is calculated based on the coordinate offsets. The Manhattan distance is defined as the sum of the absolute values ​​of the stratification differences in each dimension, characterizing the number of stratification steps required to migrate the current customer stratum to the target stratum. After obtaining the Manhattan distance and offset direction, the system selects the business strategy corresponding to the customer group within the target grid as the basic strategy template. This template includes, but is not limited to, parameter configurations such as product configuration structure, credit condition combinations, and customer service frequency. Next, the system determines the key dimensions for strategy adjustment based on the coordinate offset direction. For example, when the offset primarily occurs in the network value dimension, the focus is on adjusting parameters related to resource investment intensity; when the offset primarily occurs in the behavioral stability dimension, the focus is on adjusting parameters related to risk management and service rhythm. Finally, the strategy parameters are linearly adjusted based on the Manhattan distance. A larger Manhattan distance results in a greater adjustment magnitude, while a smaller Manhattan distance results in a more limited adjustment magnitude. By using the above methods, an operational strategy is generated that matches the current level of the adjustable customer and gradually transitions towards the target grid. This ensures that the strategy adjustment process is continuous, controllable, and has a clear structural basis, thereby achieving refined customer management based on the value stratification matrix.

[0083] Example 2

[0084] The difference between Embodiment 2 and Embodiment 1 is that this embodiment introduces a customer value tiered management system for financial institutions based on big data.

[0085] Figure 2 A schematic diagram of a big data-based customer value tiering management system for financial institutions is provided. This system includes:

[0086] The network construction module is used to collect data on indirect credit guarantee relationships between financial institution customers and to construct a heterogeneous customer relationship network with time-series weights by introducing guarantee amount and time decay factors.

[0087] The community discovery module is used to perform overlapping community discovery analysis based on modularity optimization in heterogeneous customer association networks, identify the potential communities to which customers belong, and generate a membership vector representing the strength of customer associations across multiple communities based on the connection weights within the community.

[0088] The vulnerability assessment module is used to extract corresponding sub-networks for each potential community and calculate the structural vulnerability index, which reflects the stability of the community structure, based on the distribution of the shortest path of nodes and the variance of edge weights.

[0089] The behavioral stability assessment module is used to aggregate the customer's community membership vector with the corresponding community's structural vulnerability index, calculate the customer's network value gain coefficient, and construct a behavioral stability index vector by combining the trend decomposition results of multi-period transaction behavior sequences.

[0090] The business strategy generation module is used to construct a customer value stratification matrix based on the network value gain coefficient and behavioral stability index, determine the customer stratification position, and generate targeted customer management strategies based on the historical operating performance of adjacent customer groups through nearest neighbor interpolation.

[0091] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0092] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0093] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0094] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0095] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0096] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0097] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0098] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0099] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for segmenting and managing customer value in financial institutions based on big data, characterized in that: Includes the following steps: S1. Collect data on indirect credit guarantee relationships between financial institution customers and construct a heterogeneous customer relationship network with time-series weights; S2. Perform a modularity-based overlapping community discovery analysis on the heterogeneous customer association network to identify the potential communities to which customers belong and generate a membership vector for each customer containing different connection strengths of potential communities. S3. Extract the internal network topology of each potential community, quantify the risk transmission efficiency within the potential community based on the shortest path distribution between nodes and the variance of edge weights, and calculate the structural vulnerability index of the potential community. S4. Aggregate the potential community membership vector of each customer with the structural vulnerability index of each potential community to obtain the customer's value gain coefficient. S5. Perform trend decomposition on the client's multi-period financial behavior sequence and construct a behavioral stability index vector for client trading fluctuations. S6. Construct a two-dimensional value stratification matrix consisting of a customer behavior stability index vector and a network value gain, and map customers to the corresponding grid positions in the value stratification matrix. S7. In the value stratification matrix, based on the historical operating efficiency of the customer groups in the adjacent grids of the customer's grid, a nearest neighbor interpolation method is used to generate operating strategies for the customer. In S3, calculating the structural vulnerability index of a potential community specifically includes: For each identified potential community, a subnetwork corresponding to the potential community is extracted from the heterogeneous customer association network. The subnetwork contains all customer nodes of the potential community and the directed edges between them. Calculate the shortest path length between each pair of nodes in each sub-network, where the shortest path length is the sum of the temporal weights of each edge on the path, and calculate the average shortest path length of each sub-network based on the shortest path length of all node pairs. The variance of the temporal weights of all edges in each sub-network is calculated to obtain the edge weight variance. The scalar value obtained by linearly combining the average shortest path length and the edge weight variance is used as the structural vulnerability index of the potential community. In step S4, the potential community membership vector of each customer is aggregated with the structural vulnerability index of each potential community to obtain the customer's value gain coefficient, specifically including: For each customer, a separate analysis is performed. The connection strength components of each potential community in the current customer's potential community membership vector are aggregated with the structural vulnerability index of the corresponding potential community, and the risk allocation value of the current customer in each potential community is obtained proportionally. Accumulate the risk allocation value of the current customer across all potential communities, and retrieve the corresponding discount ratio from the preset mapping table as the value gain coefficient for the customer; In step S7, generating the business strategy specifically includes: Adjustable customers are selected based on the historical changes in each customer's operating performance. In the value stratification matrix, the grid corresponding to the adjustable customer and the coordinates of adjacent grids are retrieved based on the stratification identifier. Obtain the median historical operating efficiency growth rate of the grid corresponding to the adjustable customer and other customer groups in adjacent grids, and select the adjacent grid with the highest median historical operating efficiency growth rate as the target grid; Calculate the coordinate offset between the adjustable customer's corresponding grid and the target grid in the value stratification matrix. Based on the business strategy corresponding to the customer in the target grid, adjust the parameters of the business strategy corresponding to the adjustable customer according to the Manhattan distance and direction of the coordinate offset. The adjustment range is proportional to the Manhattan distance, thus generating the business strategy for the adjustable customer.

2. The method for segmented customer value management of financial institutions based on big data as described in claim 1, characterized in that, In step S1, constructing a heterogeneous customer association network with temporal weights specifically includes: Data on indirect credit guarantee relationships between customers are collected from credit information of financial institutions, including the guarantor customer identifier, the guaranteed customer identifier, the effective date of the guarantee, and the guarantee amount; Treat each customer as a network node, and each guarantee relationship as a directed edge connecting the corresponding guarantor node and the guaranteed party node; Based on the difference between the guarantee effective date and the current analysis date, the time decay coefficient of each guarantee relationship is calculated. The result of multiplying the guarantee amount by the time decay coefficient is used as the temporal weight of the directed edge in the customer association network, thus constructing a heterogeneous customer association network with temporal weights.

3. The method for segmented customer value management of financial institutions based on big data as described in claim 1, characterized in that, In step S2, identifying the potential communities to which a customer belongs and generating a membership vector for each customer that contains different connection strengths among the potential communities specifically includes: On a heterogeneous customer association network, a label propagation algorithm based on modularity optimization is used to update the potential community labels of customer nodes and perform potential community segmentation on the network. Identify nodes that are assigned more than one potential community label during the iteration process, mark them as overlapping nodes, and record all potential community sets to which the corresponding customer node belongs; Based on the set of potential communities to which each customer node belongs, the sum of the weights of the directed edges between the customer node and its potential community is used as the connection strength component. The connection strength components corresponding to all potential communities are combined to form the customer's membership vector.

4. The method for segmented customer value management of financial institutions based on big data as described in claim 1, characterized in that, In S5, constructing the behavioral stability index vector for customer transaction fluctuations specifically includes: Obtain transaction records for each customer within a preset historical time period, perform piecewise linear fitting on the transaction sequence, identify time points in the sequence where the fitting residual growth exceeds a preset mutation threshold, and mark them as potential behavioral inflection points. Within the behavioral stages defined by continuous potential behavioral inflection points, the autocorrelation coefficients of the transaction flow in the corresponding behavioral stages are calculated. The number and distribution density of all behavioral inflection points, as well as the autocorrelation coefficients of each stage, are integrated to construct a stability index vector.

5. A method for segmented customer value management in financial institutions based on big data, as described in claim 1, is characterized in that... In step S6, constructing a two-dimensional value layering matrix consisting of a customer behavior stability index vector and network value gain specifically includes: The network value gain coefficient and behavioral stability index vector of all customers are collected. The percentile quantiles of the first principal component score set of the network value gain coefficient set and the behavioral stability index vector are calculated respectively. The numerical intervals are dynamically divided based on the quantiles as the grid boundary of the value stratification matrix. Based on the numerical range of each customer's network value gain coefficient and behavioral stability principal component score, determine their grid coordinates in the value stratification matrix, and generate a unique stratification identifier for each grid coordinate.

6. A big data-based customer value tiering management system for financial institutions, used to implement the big data-based customer value tiering management method for financial institutions as described in any one of claims 1-5, characterized in that, include: The network construction module is used to collect data on indirect credit guarantee relationships between financial institution customers and to construct a heterogeneous customer relationship network with time-series weights by introducing guarantee amount and time decay factors. The community discovery module is used to perform overlapping community discovery analysis based on modularity optimization in heterogeneous customer association networks, identify the potential communities to which customers belong, and generate a membership vector representing the strength of customer associations across multiple communities based on the connection weights within the community. The vulnerability assessment module is used to extract corresponding sub-networks for each potential community and calculate the structural vulnerability index, which reflects the stability of the community structure, based on the distribution of the shortest path of nodes and the variance of edge weights. The behavioral stability assessment module is used to aggregate the customer's community membership vector with the corresponding community's structural vulnerability index, calculate the customer's network value gain coefficient, and construct a behavioral stability index vector by combining the trend decomposition results of multi-period transaction behavior sequences. The business strategy generation module is used to construct a customer value stratification matrix based on the network value gain coefficient and behavioral stability index, determine the customer stratification position, and generate targeted customer management strategies based on the historical operating performance of adjacent customer groups through nearest neighbor interpolation.

Images