A double-atlas-based examination data flow conversion method applied to industrial and commercial management

Abstract

The present application relates to the technical field of computer data processing, and particularly relates to a double-graph-based examination and approval data flow transfer method applied to industrial and commercial management, comprising the following steps: S1, obtaining examination and approval application data, extracting an amount feature, an operator identifier and a payee identifier in the examination and approval application data; S2, obtaining a historical amount feature, performing time-series clustering on the amount feature and the historical amount feature, and constructing a time-series graph containing clustering nodes and connecting edges. In the present application, a first evaluation value extracted from the time-series graph and a second evaluation value extracted from an interaction graph are constructed into a two-dimensional feature vector, and a deviation of a vector included angle and a resonance reference angle is calculated to perform attenuation operation, thereby quantifying concurrent intensity of two types of situations of fund splitting tendency and physical terminal cooperation, filtering interference of single-dimensional isolated data, and improving accuracy of identification of composite abnormal data.

Description

Technical Field

[0001] This invention relates to the field of computer data processing technology, and in particular to a dual-map-based approval data flow method for industrial and commercial administration. Background Technology

[0002] In existing business administration or corporate financial approval processes, to ensure the security of funds and data, information systems are typically used to investigate anomalies and manage the flow of approval application data. Existing approval systems generally record the distribution of application funds over time or independently monitor the network environment of the terminal submitting the application, serving as a reference for subsequent manual or machine review.

[0003] However, existing technologies typically isolate or simply linearly superimpose features from different dimensions, making it difficult to effectively quantify the correlation strength when fund splitting features and physical terminal collaboration features occur concurrently. When faced with isolated abnormal data of a single dimension, they are easily affected by noise, resulting in low accuracy in identifying composite abnormal data. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides a dual-graph-based approval data flow method for business administration, aiming to improve the problem that existing technologies usually isolate and judge features of different dimensions or simply superimpose them linearly.

[0005] This invention provides the following technical solution: a method for data flow in business administration based on dual-graphology approval, comprising the following steps: S1. Obtain approval application data, and extract the amount characteristics, operator identifier, and payee identifier from the approval application data; S2. Obtain historical amount features, perform time-series clustering on the amount features and the historical amount features, construct a time-series graph containing cluster nodes and connecting edges, and calculate the first evaluation value based on the distribution density of the cluster nodes in the time-series graph. S3. Extract the co-occurrence features of the operator identifier and the payee identifier based on the system log, map the operator identifier and the payee identifier to the identifier nodes in the interaction graph and construct connected edges, calculate the connectivity between the identifier node corresponding to the operator identifier and the identifier node corresponding to the payee identifier, and generate a second evaluation value. S4. Construct a two-dimensional feature vector from the first evaluation value and the second evaluation value, calculate the vector magnitude and the included angle of the two-dimensional feature vector, and generate an aggregation coefficient based on the vector magnitude and the included angle; S5. Based on the aggregation coefficient, perform a product operation on the first evaluation value and the second evaluation value to determine the routing level; S6. Generate an approval route path based on the routing level and the preset approval routing rules, and send the approval application data according to the approval route path.

[0006] Preferably, in S2, constructing the time-series graph containing clustered nodes and connecting edges specifically includes the following steps: Analyze the numerical and temporal data contained in the monetary features and the historical monetary features; In the feature space composed of the numerical dimension data and the time dimension data, the feature distance between the monetary feature and the historical monetary feature is calculated. Based on the feature distance, clustering is performed to generate multiple clusters, and each cluster is mapped to a corresponding cluster node; Extract the center timestamp of each cluster, and construct the connection edges between the cluster nodes according to the order of the center timestamps to generate a time series graph.

[0007] Preferably, in S2, the calculation of the first evaluation value based on the distribution density of the cluster nodes in the time series graph specifically includes the following steps: Extract the time interval corresponding to each connection edge in the time series graph; Based on each of the time intervals, the number of nodes clustered within a preset unit time window for each of the clustering nodes is calculated, and the number of nodes clustered is used as the distribution density. Extract the monetary features associated with each of the cluster nodes, and calculate the discrete variance of the monetary features; Calculate the ratio of the distribution density to the discrete variance, and perform normalization on the ratio to generate a first evaluation value.

[0008] Preferably, in step S3, calculating the connectivity between the identifier node corresponding to the operator identifier and the identifier node corresponding to the payee identifier specifically includes the following steps: Analyze the co-occurrence features and extract device attribute data; Map the operator identifier, the payee identifier, and the device attribute data to the identifier nodes respectively; Calculate the feature overlap between the device attribute data, establish the connected edges between the identifier nodes that meet the preset overlap threshold, and generate an interaction graph; Search the interactive graph for the target connectivity path between the identifier node corresponding to the operator identifier and the identifier node corresponding to the payee identifier; The number of path edges contained in the target connected path is counted, and the reciprocal of the number of path edges is calculated as the connectivity.

[0009] Preferably, in step S3, generating the second evaluation value specifically includes the following steps: Obtain the preset connectivity weight coefficients; Calculate the product of the connectivity degree and the connectivity weight coefficient to obtain the initial value; The initial value is input into a preset normalization function to perform interval mapping processing, generating a second evaluation value.

[0010] Preferably, in step S4, generating the aggregation coefficient based on the vector magnitude and the included angle specifically includes the following steps: Obtain a preset first reference coordinate axis, and map the first evaluation value and the second evaluation value to a first reference component along the direction of the first reference coordinate axis and a second orthogonal component in the vertical direction, respectively, to generate a two-dimensional feature vector; Calculate the vector magnitude of the two-dimensional feature vector and the angle between the two-dimensional feature vector and the first reference coordinate axis; Calculate the angular deviation between the included angle and the preset resonance reference angle; The angle deviation value is input into a preset angle attenuation function to extract the corresponding bias weight, and the bias weight is multiplied by the vector magnitude to obtain the aggregation coefficient.

[0011] Preferably, in step S5, determining the routing level specifically includes the following steps: Calculate the product of the first evaluation value, the second evaluation value, and the aggregation coefficient to obtain the product value; Obtain a preset threshold interval set, and compare the product value with the preset threshold interval set; Determine the target threshold range into which the product value falls; Extract the level labels that have been pre-mapped to the target threshold range, and use the level labels as the routing level.

[0012] Preferably, in step S6, generating the approval route path based on the routing level and the preset approval routing rules specifically includes the following steps: Parse the preset approval routing rules and extract a mapping table containing the mapping relationship between level identifiers and routing node sets; Using the routing level as the query keyword, the corresponding set of target routing nodes is retrieved by matching and extracting the mapping table. Obtain a preset topology order, and connect each target routing node in the target routing node set in an orderly manner according to the preset topology order to generate an approved routing path.

[0013] The present invention has the following beneficial effects: 1. This invention constructs a two-dimensional feature vector by extracting the first evaluation value from the time series graph and the second evaluation value from the interaction graph, and performs attenuation calculation by calculating the deviation between the vector angle and the resonance reference angle. This quantifies the concurrent intensity of two types of situations: fund splitting tendency and physical terminal collaboration, filters out the interference of isolated data in a single dimension, and improves the accuracy of identifying composite abnormal data.

[0014] 2. This invention converts multi-dimensional evaluation values ​​into discrete routing levels through product operations and threshold comparisons, and performs serialization assembly on target routing nodes based on topological order, transforming continuous evaluation features into a deterministic node flow queue, reducing the uncertainty of cross-departmental flow paths and improving the orderliness of multi-level approval data distribution. Attached Figure Description

[0015] Figure 1 This is a flowchart of a dual-graph-based approval data flow method for industrial and commercial administration proposed in this invention. Detailed Implementation

[0016] The technical solutions in 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.

[0017] In a first embodiment of the present invention, the present invention provides a method for data flow of approval based on dual-graphology applied to industrial and commercial administration, such as... Figure 1 As shown, it includes the following steps: S1. Obtain approval application data and extract the amount characteristics, operator identifier, and payee identifier from the approval application data; Specifically, the system receives approval application data streams from front-end business terminals via an interface. The received data stream is typically a structured message. The system parses the structured message using a lexical analyzer, identifying and mapping the amount field node, operator information node, and payee information node within the message. The string sequences from the operator and payee information nodes are extracted and converted into fixed-length unique numerical identifiers using a hash algorithm, serving as the operator identifier and payee identifier, respectively. In this scheme, the fixed-length numerical value generated by the hash algorithm is preferably a 64-bit unsigned integer to ensure the global uniqueness of personnel and organizational entity identifiers under a massive user base.

[0018] For the parsed amount field nodes, the system reads the original business amount values. Because the original business amount values ​​in business administration approval scenarios vary greatly, ranging from miscellaneous travel reimbursements to large-scale project purchases, directly performing subsequent time-series clustering would cause gradient shifts in distance calculations in the feature space. Therefore, the system performs logarithmic smoothing and normalization mapping algorithms on the original business amount values ​​to extract standardized amount features. The specific formula for calculating the amount feature extraction is as follows: .

[0019] In this extraction formula, This represents the standardized monetary characteristic of the final output. This represents the parsed original business transaction amount, which corresponds one-to-one with the actual amount incurred in the business registration approval scenario. Using logarithmic operations with a base of the natural constant effectively smooths out the magnitude difference between large sums and small amounts of money. This represents the preset system historical maximum amount threshold parameter. In this solution, The parameter is preferably 10,000,000. The number 1 is a smoothing constant, used to prevent the logarithmic function from losing its mathematical meaning when the original business amount is 0.

[0020] After calculating the standardized amount characteristics, the system loads the amount characteristics, operator identifier, and payee identifier into the feature contiguous address segment in the system memory, thus completing the feature extraction operation of the application data.

[0021] By performing hash transformation and logarithmic smoothing normalization on the original data, the format differences and numerical bias of heterogeneous business messages are mitigated, thereby improving the numerical stability of subsequent time-series clustering and graph search processes.

[0022] S2. Obtain historical amount features, perform time-series clustering of amount features and historical amount features, construct a time-series graph containing cluster nodes and connecting edges, and calculate the first evaluation value based on the distribution density of cluster nodes in the time-series graph. Furthermore, in S2, constructing a temporal graph containing cluster nodes and connecting edges specifically includes the following steps: Analyze the numerical and temporal dimensions of the monetary and historical monetary features. In the feature space composed of numerical and time dimension data, calculate the feature distance between monetary features and historical monetary features; Clustering is performed based on feature distance to generate multiple clusters, and each cluster is mapped to a corresponding cluster node; Extract the center timestamp of each cluster, construct connection edges between the cluster nodes according to the order of the center timestamps, and generate a time series graph.

[0023] Furthermore, in S2, calculating the first evaluation value based on the distribution density of cluster nodes in the time series graph specifically includes the following steps: Extract the time intervals corresponding to each connection edge in the time series graph; Based on each time interval, the number of nodes clustered within a preset unit time window for each cluster node is calculated, and the number of nodes clustered is used as the distribution density. Extract the monetary features associated with each cluster node and calculate the discrete variance of the monetary features; Calculate the ratio of distribution density to discrete variance, and normalize the ratio to generate the first evaluation value.

[0024] Specifically, the system reads the standardized monetary features stored in system memory and, based on the operator identifier extracted in the previous step, retrieves the historical monetary features submitted by that operator from the archive database. The system parses the current and historical monetary features, separating the numerical dimension data reflecting the size of the transaction amount, and simultaneously extracts the system timestamps corresponding to the current approval application data and historical archive data as time dimension data. To avoid the absolute value of the timestamp being too large and causing inaccuracy in distance calculation, the system uses a preset baseline time origin as a reference, converting the time dimension data into a relative number of days relative to the baseline time origin. In the two-dimensional feature space composed of numerical and time dimension data, the system calculates the feature distance between the current and historical monetary features. The specific formula for calculating the feature distance is as follows: .

[0025] In this feature distance calculation formula, Indicates the feature distance. and These represent the numerical dimension data for the current amount feature and the historical amount feature, respectively. and These represent the current and historical time dimension data after conversion to relative days, respectively. and These represent numerical weights and time weights, respectively, used to balance differences in the amount of capital and the time span. In this scheme, The preferred parameter is 0.6. The optimal parameter is 0.4.

[0026] The system executes a density clustering algorithm based on calculated feature distances, dividing feature points in the feature space whose distances are less than a preset clustering radius into multiple clusters. In this scheme, the preset clustering radius is preferably 1.5. Each cluster represents a batch of business transactions with similar amounts occurring within a specific time period in business administration approvals. The system maps each cluster as an independent cluster node. Subsequently, the system calculates the average value of all time-dimensional data within each cluster as the center timestamp, and constructs directed connections between adjacent cluster nodes according to the chronological order of the center timestamps, thereby generating a time-series graph in memory that reflects the evolution logic of historical fund flows.

[0027] For the generated time series graph, the system extracts the time intervals corresponding to the connecting edges between adjacent cluster nodes. Based on each time interval, the system calculates the number of node clusters within a preset unit time window for each cluster node using a sliding window approach, and uses this number of clusters as the distribution density. In this scheme, the unit time window is preferably 7 days to align with the common weekly settlement pattern of industrial and commercial transactions. Subsequently, the system extracts all standardized monetary features associated with each cluster node and calculates the discrete variance of these monetary features. The specific formula for discrete variance is as follows: . in the formula Represents discrete variance. This represents the total number of monetary features within a cluster node. Represents a sequence of positive integer indices. Indicates the first A specific monetary value. This represents the arithmetic mean of all monetary features within the cluster node. The smaller the discrete variance, the closer the approved amount within that node is to a certain fixed value.

[0028] The system calculates the ratio of distribution density to discrete variance, and performs a normalization process based on a logistic function on the ratio to generate the final first evaluation value. The specific formula for calculating the evaluation value is as follows: In this normalization formula, This represents the first evaluation value of the output, whose value is strictly mapped to the range of 0 and 1. This represents the calculated distribution density. This represents the discrete variance. Using the ratio of distribution density to discrete variance can effectively amplify the abnormal characteristics of high-frequency and fixed-amount fund flows. When multiple approval applications with very similar amounts appear in a short period of time, the distribution density increases and the discrete variance decreases, resulting in a significant increase in this ratio. This allows for precise mathematical quantification of the tendency to artificially split funds into smaller units. This represents a preset minimum zero-prevention constant used to prevent division overflow when the variance is 0. In this solution... The preferred value is 0.001. This represents the sensitivity adjustment coefficient, used to control the numerical sensitivity of the model to specific frequency behaviors. In this scheme... The preferred version is 2.0.

[0029] By performing clustering and density-variance ratio normalization calculations on the monetary value and time dimension, the high-frequency and fixed-amount fund flow patterns in the approval process were quantified, thereby effectively extracting the first evaluation value that can characterize the abnormal features of the business.

[0030] S3. Extract the co-occurrence features of operator identifier and payee identifier based on system logs, map operator identifier and payee identifier to identifier nodes in the interaction graph and construct connected edges, calculate the connectivity between the identifier node corresponding to the operator identifier and the identifier node corresponding to the payee identifier, and generate a second evaluation value. Furthermore, in S3, calculating the connectivity between the identifier node corresponding to the operator identifier and the identifier node corresponding to the payee identifier specifically includes the following steps: Analyze co-occurrence features and extract device attribute data; Map the operator identifier, payee identifier, and device attribute data to identifier nodes respectively; Calculate the feature overlap between device attribute data, establish connected edges between identified nodes that meet the preset overlap threshold, and generate an interaction graph; Search the interactive graph for the target connectivity path between the identifier node corresponding to the operator identifier and the identifier node corresponding to the payee identifier; Count the number of path edges contained in the target connected path, and calculate the reciprocal of the number of path edges as the connectivity.

[0031] Furthermore, in S3, generating the second evaluation value based on connectivity specifically includes the following steps: Obtain the preset connectivity weight coefficients; Calculate the product of connectivity and connectivity weight coefficient to obtain the initial value; The initial values ​​are input into a preset normalization function to perform interval mapping processing, generating a second evaluation value.

[0032] Specifically, the system retrieves pre-stored system logs and reads the operator identifier and payee identifier extracted from the approval application data. The system searches the system logs for the historical behavior of these two identifiers, extracting co-occurrence features of the operator and payee identifiers. The system parses these co-occurrence features to extract device attribute data reflecting the network environment and hardware information of the operating terminal. In actual business operations, this device attribute data specifically includes the login Internet Protocol address, Media Access Control address, and operating system version fields. Subsequently, the system instantiates and maps the operator identifier and payee identifier to operator identifier nodes and payee identifier nodes respectively in memory space, and instantiates and maps all extracted device attribute data to independent device identifier nodes. The system establishes initial association edges between the operator identifier node and the device identifier nodes it has used, and between the payee identifier node and the device identifier nodes it has used.

[0033] To construct a network topology reflecting hidden relationships between entities, the system calculates the feature overlap between the attribute data of various devices. Since logins at different times may experience network fluctuations or changes in hardware fingerprints, the system extracts multiple underlying environmental feature fields from any two device identifier nodes and performs an intersection-union (IoU) operation. The specific formula for calculating feature overlap is as follows: In this formula for calculating the degree of feature overlap, Indicates the degree of feature overlap. This indicates the number of intersections of environmental feature fields with identical values ​​in two device attribute data sets. This represents the total number of the union of all environmental feature fields contained in the two device attribute data mentioned above. The system compares the calculated feature overlap with a preset overlap threshold. In this scheme, the preset overlap threshold is preferably 0.5. When the feature overlap is greater than or equal to the preset overlap threshold, the system establishes an undirected connected edge between the corresponding two device identifier nodes, thereby generating an interaction graph in memory that contains the underlying shared relationships between personnel entities and physical devices.

[0034] After generating the interaction graph, the system uses a breadth-first search algorithm to find the shortest target connected path between the node corresponding to the operator's identifier and the node corresponding to the payee's identifier. The system counts the number of edges in this shortest target connected path and calculates the reciprocal of the number of edges as the connectivity. The specific connectivity calculation formula is as follows: In this formula, Indicates connectivity. This represents the actual number of edges in the shortest target connected path. The number 1 is an inverse proportional constant used to convert the increasing number of edges into a decreasing connectivity strength. When two nodes are connected through highly overlapping device nodes, the fewer the number of edges, the higher the connectivity. If no connected path can be found in the entire interaction graph, the system sets the connectivity to 0.

[0035] After calculating the connectivity, the system obtains a preset connectivity weight coefficient. In this scheme, the preferred connectivity weight coefficient is 0.8. The system calculates the product of the connectivity and the connectivity weight coefficient to obtain an initial value. To unify the evaluation dimension and adapt to the subsequent two-dimensional mapping matrix processing requirements, the system inputs this initial value into a preset normalization function to perform interval mapping processing, generating a second evaluation value. The specific normalization formula is as follows: In this formula, This represents the second evaluation value of the output, whose value is smoothly mapped within a closed interval of 0 to 1. This represents the initial value obtained by multiplying the connectivity by the weight coefficient. The number 1 is the smoothing adjustment base, used to convert the linearly increasing initial value into a non-linearly convergent interval feature.

[0036] By performing overlap determination and graph theory search on the device attribute features in the system logs, the strength of the hidden association between business operators and fund recipients in the underlying physical operating environment is quantified, and a second evaluation value that can characterize the collaborative features of physical terminals is extracted.

[0037] S4. Construct a two-dimensional feature vector from the first evaluation value and the second evaluation value, calculate the vector magnitude and the included angle of the two-dimensional feature vector, and generate the aggregation coefficient based on the vector magnitude and the included angle. Furthermore, in S4, generating aggregation coefficients based on vector magnitude and included angle specifically includes the following steps: Obtain a preset first reference coordinate axis, and map the first evaluation value and the second evaluation value to a first reference component along the direction of the first reference coordinate axis and a second orthogonal component in the vertical direction, respectively, to generate a two-dimensional feature vector; Calculate the vector magnitude of the two-dimensional feature vector and the angle between the two-dimensional feature vector and the first reference coordinate axis; Calculate the angular deviation between the included angle and the preset resonance reference angle; The angle deviation value is input into the preset angle attenuation function to extract the corresponding bias weight, and the bias weight is multiplied by the vector magnitude to obtain the aggregation coefficient.

[0038] Specifically, the system reads the first and second evaluation values ​​generated in the aforementioned calculation steps. To identify the collaborative concurrency characteristics of the two types of underlying data streams, the system obtains a preset first reference coordinate axis and constructs a two-dimensional feature space in memory. The system maps the first evaluation value to a first reference component along the direction of the first reference coordinate axis, and maps the second evaluation value to a second orthogonal component perpendicular to the direction of the first reference coordinate axis, thereby generating a two-dimensional feature vector. Subsequently, the system calculates the vector magnitude of this two-dimensional feature vector and the geometric angle between this two-dimensional feature vector and the first reference coordinate axis. The specific formulas for calculating the magnitude and angle are as follows: and .

[0039] In the above calculation formula, The vector magnitude represents the absolute scalar strength of the superposition of two features: the tendency of funds to be split in business approval and the co-occurrence of physical terminals. This represents the first evaluation value input. This represents the second evaluation value input. It represents the angle between the two-dimensional feature vector and the first reference coordinate axis, and its value reflects the distribution tilt of the two different dimensions of data in the current business flow. This represents a preset minimum zero-prevention constant, used to prevent overflow in the division calculation of the arctangent function when the first evaluation value is 0. In this scheme... The preferred value is 0.0001. The system acquires a preset resonance reference angle and calculates the absolute angular deviation between the aforementioned included angle and the preset resonance reference angle. The specific formula for calculating the angular deviation is as follows: In this formula, This indicates the angular deviation value. This represents the preset resonance reference angle. Since both the first and second evaluation values ​​are mapped to the range of 0 to 1, when the tendency for fund splitting and abnormal coordination of physical terminals occur simultaneously and are of similar intensity, the angle between their feature vectors in two-dimensional space and the coordinate axes naturally approaches 45 degrees. This state is defined as a resonance fraud situation; therefore, in this scheme... The optimal parameter is 0.785 radians, or a 45-degree geometric angle, to accurately capture the typical resonance fraud situation when two types of indicators occur concurrently in equal proportions. The vertical lines on both sides of the symbol represent the extraction of absolute values ​​in mathematics to ensure that the angle deviation value is non-negative.

[0040] The system inputs the calculated angle deviation value into a preset angle attenuation function and extracts the corresponding bias weight. The specific formula for calculating the bias weight is as follows: In this formula, This represents the extracted bias weights. (Letter) It represents the base constant of the natural logarithm. This represents the preset attenuation adjustment coefficient, used to control the steepness of the penalty applied to the weights when the actual feature angle deviates from the resonance reference angle. To achieve accurate calibration of this coefficient, the system retrieves a sample set containing historical real processing results to perform objective function optimization. The specific parameter calibration algorithm formula is as follows:

[0041] ; In this parameter calibration formula, This represents the final determined attenuation adjustment coefficient. This indicates that the set of candidate coefficients is pre-defined. Searching for candidate coefficients that maximize the subsequent summation function. Variables. In this scheme, the set of candidate coefficients. The preferred value is a discrete numerical sequence from 0.5 to 3.0 with a step size of 0.1. This represents the total number of samples in the historical sample set. This indicates the sample traversal index. Indicates the alternative coefficients The output after substituting into the system Machine-predicted routing results for each sample. Indicates the first The historical, real, manually processed results of each sample. The Kronek indicator function outputs 1 when the predicted route matches the actual processing result, and 0 otherwise. The system uses the above optimization algorithm to obtain the coefficient that maximizes the accuracy of historical route interception. In this scheme... The optimal parameter is 1.5. After obtaining the bias weight, the system multiplies this bias weight by the vector magnitude calculated above to obtain the final aggregation coefficient. The specific formula for calculating the aggregation coefficient is as follows: In this formula, This represents the calculated aggregation coefficient. When the angle deviation value is closer to 0, the bias weight is closer to the maximum value of 1, which greatly preserves the vector magnitude in the product calculation; conversely, if the evaluation value of a single dimension is isolated and too high, causing the angle to deviate significantly from the resonance reference angle, the system will exponentially reduce its magnitude through the bias weight.

[0042] By mapping independent evaluation values ​​to two-dimensional feature vectors and performing bias attenuation operations on geometric angles, the nonlinear resonance intensity between fund splitting anomalies and physical terminal coordination anomalies is quantified, thereby effectively extracting aggregation coefficients that can accurately characterize multimodal composite fraud behavior.

[0043] S5. Based on the aggregation coefficient, perform a product operation on the first evaluation value and the second evaluation value to determine the routing level; Furthermore, in S5, determining the routing level specifically includes the following steps: Calculate the product of the first evaluation value, the second evaluation value, and the aggregation coefficient to obtain the product value; Obtain a preset threshold interval set, and compare the product value with the preset threshold interval set; Determine the target threshold range into which the product value falls; Extract the level labels that have been pre-mapped to the target threshold range, and use the level labels as the routing level.

[0044] Specifically, the system reads the first evaluation value, the second evaluation value, and the aggregation coefficient generated based on vector resonance, which are obtained from the aforementioned calculation steps. To comprehensively quantify the degree of composite anomaly in the current business application, the system performs a product operation on the first evaluation value and the second evaluation value based on the extracted aggregation coefficient. The system loads these three values—the first evaluation value, the second evaluation value, and the aggregation coefficient—into the underlying calculation unit, calculates their product, and obtains the final product value. The specific formula for calculating the product is as follows: .

[0045] In the above formula for calculating consecutive products, This represents the output product value, which characterizes the overall comprehensive risk quantification score of the current business administration approval application after superimposing the tendency of fund splitting, the coordinated operation of physical equipment, and the resonance effect of the characteristics of both. This represents the first evaluation value input. This represents the second evaluation value input. This represents the input aggregation coefficient. The formula utilizes the nonlinear amplification and mutual constraint properties of multiplication operations to ensure that only when the evaluation values ​​in each dimension and the resonance aggregation coefficient are high will it output an extremely high product value, thus effectively filtering out noise and false alarms from isolated single-dimensional data.

[0046] After calculating the product value, the system obtains a pre-configured set of threshold intervals loaded into memory. This preset set of threshold intervals contains multiple consecutive and non-overlapping numerical ranges, used to discretize continuous product values ​​into specific machine decision levels. The system compares the calculated product value with each boundary value in the preset set of threshold intervals to determine which interval the product value falls into, thereby identifying the target threshold interval into which the product value falls. In this scheme, the preset set of threshold intervals is preferably divided into a first interval with numerical boundaries greater than or equal to 0 and less than 0.3, a second interval with numerical boundaries greater than or equal to 0.3 and less than 0.7, and a third interval with numerical boundaries greater than or equal to 0.7.

[0047] After determining the target threshold range, the system reads a pre-set mapping table in a relational database and extracts level labels that have a pre-established static mapping relationship with the target threshold range. These level labels correspond to different levels of rigor in the approval channel identifiers within the business registration management system. In this scheme, the level label pre-mapped to the first range is preferably number 1, representing an automatically approved lower-level channel; the level label pre-mapped to the second range is preferably number 2, representing a mid-level channel requiring manual review; and the level label pre-mapped to the third range is preferably number 3, representing a high-level channel triggering blocking and mandatory auditing. The system uses the extracted level labels as the final determined routing level and stores them within the context of the current approval application data.

[0048] By performing a continuous multiplication operation on the multidimensional evaluation values ​​and aggregation coefficients and comparing the threshold intervals, the system achieves accurate dimensionality reduction from continuous feature values ​​to discrete distribution states, thereby providing a definite machine decision-making basis for subsequent approval business processes.

[0049] S6. Generate an approval route path based on the routing level and the preset approval routing rules, and send the approval application data according to the approval route path.

[0050] Furthermore, in S6, generating an approval route path based on the route level and preset approval route rules specifically includes the following steps: Parse the preset approval routing rules and extract the mapping table containing the mapping relationship between level identifiers and the set of routing nodes; Use the route level as the query keyword to match and extract the corresponding set of target route nodes from the mapping table; Obtain the preset topology order, connect each target routing node in the target routing node set in an orderly manner according to the preset topology order, and generate the approved routing path.

[0051] Specifically, the system reads the routing levels stored in the context environment from the aforementioned calculation steps, and simultaneously retrieves the approval application data obtained in the initial steps. The system calls the data interface of the configuration service center to read and parse the preset approval routing rules. These preset approval routing rules are typically stored on the server in a Structured Markup Language (SML) file format. Through data deserialization, the system extracts a mapping table from the rule file containing the mapping relationship between level identifiers and routing node sets. This mapping table records the corresponding association between level identifiers of different risk levels and the approval node sets of the business administration departments.

[0052] The system uses the routing level stored in the current context as the query keyword and performs a hash matching search in the extracted mapping table. The system locates key-value pairs consistent with this routing level and extracts the corresponding set of target routing nodes. In this solution, if the input routing level is the number 3, the system preferably extracts a set of target routing nodes from the mapping table that includes legal compliance nodes, financial audit nodes, and high-level decision-making nodes, thus addressing the workflow requirements of high-risk approval processes.

[0053] After obtaining the target routing node set, the system retrieves the pre-configured directed acyclic graph data structure from the underlying layer to obtain the preset topological order. This preset topological order defines the sequential dependencies between various business nodes in the business administration approval process. Following this preset topological order, the system performs ordered connection and serialization assembly of the target routing nodes in the target routing node set to generate the approval routing path. The specific sequence assembly algorithm formula is as follows: .

[0054] In this sequence assembly formula, This indicates that the generated approval route path is at the first position. The numerical identifier of the routing node in the execution order. This represents the mathematical summation operator. This represents the total number of nodes contained in the target routing node set. This represents the traversal index for the summation operation, with its initial value being the number 1. Represents the first in the set of target routing nodes A numeric identifier for each target routing node. In this scheme, if the target routing node is a legal compliance node, its numeric identifier is preferably the number 101. This indicates that the first element is assigned according to the preset topological order. The topology execution sequence number of each target routing node. In this scheme, the topology execution sequence number corresponding to the legal compliance node is preferably the number 1. This represents the step size index of the currently generated path sequence, and its value ranges from the number 1 to... A series of positive integers. The Kronecker function is defined if and only if the node's topological execution sequence number is... Equal to the current traversal order When the condition is met, the function outputs the number 1; otherwise, it outputs the number 0. The system uses the above operations to convert the unordered set of node identifiers in memory into an ordered queue of flowing values, generating the approval routing path for this application.

[0055] After generating the approval routing path, the system encapsulates the approval application data, which includes monetary features and the original form, into a standardized business event through a message middleware. Then, according to the node sequence defined in the approval routing path, the system sends the approval application data to the corresponding business terminals in sequence to drive subsequent manual or automated review processes.

[0056] By mapping and extracting routing rules and performing ordered serialization calculations of nodes based on topology logic, the uncertainty of business application flow paths is reduced, thereby improving the orderliness of multi-level approval data flow processes.

[0057] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. 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 data flow in business administration based on dual-graphology approval processes, characterized in that: Includes the following steps: S1. Obtain approval application data, and extract the amount characteristics, operator identifier, and payee identifier from the approval application data; S2. Obtain historical amount features, perform time-series clustering on the amount features and the historical amount features, construct a time-series graph containing cluster nodes and connecting edges, and calculate the first evaluation value based on the distribution density of the cluster nodes in the time-series graph. S3. Extract the co-occurrence features of the operator identifier and the payee identifier based on the system log, map the operator identifier and the payee identifier to the identifier nodes in the interaction graph and construct connected edges, calculate the connectivity between the identifier node corresponding to the operator identifier and the identifier node corresponding to the payee identifier, and generate a second evaluation value. S4. Construct a two-dimensional feature vector from the first evaluation value and the second evaluation value, calculate the vector magnitude and the included angle of the two-dimensional feature vector, and generate an aggregation coefficient based on the vector magnitude and the included angle; S5. Based on the aggregation coefficient, perform a product operation on the first evaluation value and the second evaluation value to determine the routing level; S6. Generate an approval route path based on the routing level and the preset approval routing rules, and send the approval application data according to the approval route path.

2. The method for data flow of approval based on dual-graphology applied to industrial and commercial administration according to claim 1, characterized in that, In S2, the construction of the time-series graph containing clustering nodes and connecting edges is specifically... Includes the following steps: Analyze the numerical and temporal data contained in the monetary features and the historical monetary features; In the feature space composed of the numerical dimension data and the time dimension data, the feature distance between the monetary feature and the historical monetary feature is calculated. Based on the feature distance, clustering is performed to generate multiple clusters, and each cluster is mapped to a corresponding cluster node; Extract the center timestamp of each cluster, and construct the connection edges between the cluster nodes according to the order of the center timestamps to generate a time series graph.

3. The method for data flow of approval based on dual-graphology applied to industrial and commercial administration according to claim 1, characterized in that, In S2, the calculation of the first evaluation value based on the distribution density of the cluster nodes in the time series graph specifically includes the following steps: Extract the time interval corresponding to each connection edge in the time series graph; Based on each of the time intervals, the number of nodes clustered within a preset unit time window for each of the clustering nodes is calculated, and the number of nodes clustered is used as the distribution density. Extract the monetary features associated with each of the cluster nodes, and calculate the discrete variance of the monetary features; Calculate the ratio of the distribution density to the discrete variance, and perform normalization on the ratio to generate a first evaluation value.

4. The method for data flow of approval based on dual-graphology applied to industrial and commercial administration according to claim 1, characterized in that, In S3, calculating the connectivity between the identifier node corresponding to the operator identifier and the identifier node corresponding to the payee identifier specifically includes the following steps: Analyze the co-occurrence features and extract device attribute data; Map the operator identifier, the payee identifier, and the device attribute data to the identifier nodes respectively; Calculate the feature overlap between the device attribute data, establish the connected edges between the identifier nodes that meet the preset overlap threshold, and generate an interaction graph; Search the interactive graph for the target connectivity path between the identifier node corresponding to the operator identifier and the identifier node corresponding to the payee identifier; The number of path edges contained in the target connected path is counted, and the reciprocal of the number of path edges is calculated as the connectivity.

5. The method for data flow of approval based on dual-graphology applied to industrial and commercial administration according to claim 1, characterized in that, In S3, generating the second evaluation value specifically includes the following steps: Obtain the preset connectivity weight coefficients; Calculate the product of the connectivity degree and the connectivity weight coefficient to obtain the initial value; The initial value is input into a preset normalization function to perform interval mapping processing, generating a second evaluation value.

6. The method for data flow of approval based on dual-graphology applied to industrial and commercial administration according to claim 1, characterized in that, In S4, the generation of aggregation coefficients based on the vector magnitude and the included angle specifically includes the following steps: Obtain a preset first reference coordinate axis, and map the first evaluation value and the second evaluation value to a first reference component along the direction of the first reference coordinate axis and a second orthogonal component in the vertical direction, respectively, to generate a two-dimensional feature vector; Calculate the vector magnitude of the two-dimensional feature vector and the angle between the two-dimensional feature vector and the first reference coordinate axis; Calculate the angular deviation between the included angle and the preset resonance reference angle; The angle deviation value is input into a preset angle attenuation function to extract the corresponding bias weight, and the bias weight is multiplied by the vector magnitude to obtain the aggregation coefficient.

7. The method for data flow of approval based on dual-graphology applied to industrial and commercial administration according to claim 1, characterized in that, In S5, determining the routing level specifically includes the following steps: Calculate the product of the first evaluation value, the second evaluation value, and the aggregation coefficient to obtain the product value; Obtain a preset threshold interval set, and compare the product value with the preset threshold interval set; Determine the target threshold range into which the product value falls; Extract the level labels that have been pre-mapped to the target threshold range, and use the level labels as the routing level.

8. A method for data flow of approval based on dual-graphology applied to industrial and commercial administration according to claim 1, characterized in that, In S6, the process of generating an approval route path based on the routing level and preset approval routing rules specifically includes the following steps: Parse the preset approval routing rules and extract a mapping table containing the mapping relationship between level identifiers and routing node sets; Using the routing level as the query keyword, the corresponding set of target routing nodes is retrieved by matching and extracting the mapping table. Obtain a preset topology order, and connect each target routing node in the target routing node set in an orderly manner according to the preset topology order to generate an approved routing path.

Images