Loan risk monitoring method based on transaction data analysis

Abstract

The invention relates to the technical field of loan risk monitoring based on transaction data analysis, and discloses a loan risk monitoring method based on transaction data analysis. According to the scheme, the transaction record set containing the amount, the time, the longitude and the latitude is constructed, the longitude and the latitude are mapped into the plane coordinates through the self-created function, the problem that geographic information is difficult to be used for real-time monitoring is solved, and accurate expression of transaction data is achieved. And then, the amount difference, the time interval and the spatial distance of the continuous transaction pairs are calculated, a risk numerical function fusing the amount change rate and the displacement rate is constructed, the transaction volatility is effectively reflected, and the problem that multi-dimensional data interaction is difficult to quantify is solved. The early warning sensitivity is flexibly regulated and controlled by setting the risk threshold, the sensitivity and the stability are both considered, and the identification efficiency of loan transaction risk linkage is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of loan risk monitoring based on transaction data analysis, and in particular to a loan risk monitoring method based on transaction data analysis. Background Art

[0002] With the rapid development of information technology and internet finance, banks and other financial institutions have accumulated vast amounts of transaction data from their loan operations. This data primarily encompasses multiple dimensions, including transaction amounts, transaction times, and transaction locations. Existing loan risk monitoring methods primarily rely on traditional statistical regression, discriminant analysis, empirical formulas, and some algorithms based on machine learning, data mining, and neural networks. These methods typically utilize extensive historical data for training and model optimization, hoping to capture and predict trends in loan defaults or risk interactions. However, existing methods exhibit limitations and shortcomings in data processing, feature extraction, and model evaluation.

[0003] Currently, traditional loan risk assessment methods often employ a standard credit scoring system, compiling statistics and comprehensive scores based on borrowers' credit histories, lending behaviors, and repayment status. While these methods are relatively simple, they lack the ability to capture the real-time, dynamic, and multidimensional interactive information reflected in transaction data. Some systems employ empirical formulas or rule-based systems, which can be subjective and ambiguous in integrating data from different dimensions, and can easily overlook the inherent connections between transaction amounts, temporal, and spatial variations. Although some institutions have introduced machine learning and data mining techniques in recent years to predict loan risk using decision trees, logistic regression, support vector machines, and deep neural network models, these methods generally suffer from two issues: First, model parameters rely on extensive prior data training, resulting in low transparency and difficulty in interpretation. Second, these algorithms require high data preprocessing requirements, making it difficult to achieve uniform standards across different data sources. Furthermore, the multidimensional heterogeneity of the data often results in model instability and insufficient robustness in practical applications. Furthermore, existing technologies often utilize existing map projections or geographic coordinate conversion methods, such as Mercator and UTM projections, for the processing of transaction location data. While these methods are widely used in geographic information systems (GIS), their application in loan risk monitoring suffers from complex data conversion and limited accuracy. Furthermore, they struggle to seamlessly integrate location data with quantitative analysis of other financial data, such as transaction amounts and times. Existing technologies typically utilize location information only as auxiliary features, failing to develop mathematical models that can directly quantify the interactions between multidimensional information in transaction data. In transaction data analysis applications, quantifying the changes between consecutive transactions in terms of amount, time, and location, and converting these changes into easily understandable and processable quantitative data using unified mathematical expressions, remains a pressing challenge. Existing technologies typically model each data dimension separately, lacking a unified framework for integrating multidimensional data. Firstly, due to the varying scales of the data in each dimension, direct comparison and summation often introduce significant errors. Secondly, traditional methods often rely on implicit statistical parameters and fail to accurately characterize the real-time dynamics of transaction data, resulting in low real-time and accurate risk warnings.

[0004] Therefore, this case aims to propose a loan risk monitoring method based on transaction data analysis. By integrating multi-source data, calculating transaction variance, normalizing multidimensional data matrices, and employing graph-theoretic topological methods, and further modeling interaction intensity within a dynamic window and constructing abnormal links, this method quantitatively expresses the inherent connections between data dimensions. This technical solution not only addresses the inability to uniformly handle different data dimensions in traditional methods, but also provides greater sensitivity to real-time dynamic risk changes. Summary of the Invention

[0005] The present invention provides a loan risk monitoring method based on transaction data analysis, which helps solve the problems mentioned in the above background technology.

[0006] The present invention provides the following technical solution: a loan risk monitoring method based on transaction data analysis, comprising:

[0007] Set the transaction record set to T = {T1, T2, ..., T n};

[0008] Among them, T i is the i-th transaction record; n is the total number of transaction records, and n≥2;

[0009] Each transaction record contains: transaction amount Q i , represents the amount of the i-th transaction, in yuan; transaction time τ i , represents the time when the i-th transaction occurs, in seconds, starting from the predetermined initial time τ0 = 0; transaction position represents the latitude of the location where the i-th transaction occurred, in degrees; λ i The longitude of the location where the i-th transaction occurred, in degrees;

[0010] Set the coordinate mapping function to convert the latitude and longitude data into plane coordinates, specifically:

[0011]

[0012] Among them, x i represents the horizontal coordinate of the plane obtained by the coordinate mapping function for the i-th transaction; i represents the plane vertical coordinate of the i-th transaction obtained by the coordinate mapping function; π is the circumference of a circle;

[0013] For any continuous trading pair (T i ,T i+1 ), calculate the following three difference quantities:

[0014] Amount difference ΔQ i :ΔQ i =|Q i+1 -Q i |; where ΔQ i It represents the difference in the amount of the i-th pair of consecutive transactions, in yuan;

[0015] Time interval Δτ i :Δτ i =τ i+1 -τ i ; where Δτ i Indicates the time difference between the i-th pair of consecutive transactions, in seconds;

[0016] Plane space distance d i : Among them, d i represents the Euclidean distance between the i-th pair of consecutive transactions in the plane;

[0017] Construct the initial risk value function and calculate the initial risk value, specifically:

[0018]

[0019] Among them, R0 is the initial risk value; is the rate of change of amount per unit time; is the spatial displacement rate per unit time;

[0020] Set the initial risk threshold to K0;

[0021] If R0>K0, it is preliminarily judged that the loan has hidden risks of risk linkage;

[0022] If R0≤K0, it can be preliminarily judged that there is no risk linkage risk in the loan.

[0023] Optionally, it also includes constructing a multidimensional data interaction matrix, specifically:

[0024] The calculated differences of each continuous trading pair are arranged into the original data matrix M according to the following formula:

[0025]

[0026] Among them, row index i corresponds to the i-th continuous trading pair; column index ΔQ i , Δτ i d i They correspond to the amount difference, time interval, and spatial distance respectively;

[0027] For the j-th column of the original data matrix M, let:

[0028]

[0029] Among them, m j is the minimum value of the jth column of the original data matrix M; M j The maximum value of the jth column of the original data matrix M; when j = 1, it represents the amount difference, when j = 2, it represents the time interval, and when j = 3, it represents the spatial distance;

[0030] Calculate the normalized value for each element of the original data matrix M, specifically:

[0031]

[0032] in, Represents the data in the i-th row and j-th column of the normalized original data matrix M.

[0033] Optionally, it also includes constructing a normalized interaction matrix vector, specifically:

[0034] Set the normalized original data matrix to:

[0035]

[0036] in, Represents the normalized value of the amount difference of the i-th transaction pair; Represents the normalized value of the time interval of the i-th trading pair; Represents the normalized value of the spatial distance of the i-th trading pair;

[0037] Let the normalized vector be: in, is the i-th normalized vector.

[0038] Optionally, it also includes topology construction and node definition, specifically:

[0039] For each normalized vector Set the corresponding node N i , specifically:

[0040] All nodes constitute a node set V = {N1, N2, ..., N n-1};

[0041] For any N i ,N j ∈V and i≠j, calculate respectively:

[0042]

[0043] Set the total difference between nodes to:

[0044] Among them, D ij Represents node N i and N j Total differences in three-dimensional normalized data; Represents node N i and N j Differences in the dimension of amount difference; Represents node N i and N j Differences in the time interval dimension; Represents node N i and N j Differences in the spatial distance dimension.

[0045] Optionally, it also includes establishing edges and topology graphs between nodes, specifically:

[0046] Set the fixed threshold for edge establishment to δ;

[0047] For any i≠j, if: D ij <δ, then at node N i With node N j Establish undirected edges between them, and record the edge set as: E={(N i ,N j )|D ij <δ, and i≠j};

[0048] Set the topology graph to: G = (V, E); where G is an undirected graph consisting of a node set V and an edge set E.

[0049] Optionally, dynamic relationship calculation and interaction intensity modeling are also included, specifically:

[0050] Set the interaction strength function and for each edge (N i ,N j )∈E calculates the interaction intensity I ij , specifically:

[0051]

[0052] Among them, I ij For node N i With N j the intensity of interaction between them;

[0053] For node N i , set its adjacent node set to

[0054] Node N i The total interaction intensity I i Set to:

[0055] The entire transaction sequence is divided into multiple time windows at fixed time intervals, and the length of the time window is ΔT;

[0056] In any time window, let the number of nodes contained in the window be N w , calculate the arithmetic mean of the total interaction intensity of all nodes in the time window, specifically:

[0057] Among them, I mean is the average interaction strength of all nodes in the current time window.

[0058] Optionally, it also includes abnormal interaction pattern recognition and risk linkage triggering, specifically:

[0059] For node Ni In a single time window, set the abnormal interaction degree to: A i =I i -I mean ;

[0060] Among them, A i For node N i The difference between the total interaction intensity and the average interaction intensity in the time window;

[0061] Set the abnormality judgment threshold to T A ;

[0062] When for node N i Satisfy A i ≤T A When , the node is determined to be a normal node;

[0063] When for node N i Satisfy A i >T A When , the node is determined to be an abnormal node, and the abnormal node set is recorded as

[0064] In topology graph G, set the abnormal link set to E A , specifically:

[0065] Among them, E A Represents the set of all edges connecting abnormal nodes.

[0066] Optionally, it also includes risk warning report generation and visualization map construction, specifically:

[0067] For abnormal node set Each node N in i , record the following data: original transaction number i; normalized data vector Total node interaction intensity I i Abnormal interaction A i ;

[0068] Set the overall risk linkage value R total It is the cumulative sum of abnormal interactions, specifically:

[0069]

[0070] And set the warning threshold to T R ;

[0071] When R is satisfied total >T R When , the entire system is judged to have a high loan risk linkage situation;

[0072] Construct a two-dimensional graph, where the horizontal axis uses the transaction record number i; the vertical axis uses the total interaction intensity of each node I i ;

[0073] For each node N i , shown as a solid circle in the figure, the horizontal coordinate of the point is i and the vertical coordinate is I i ;

[0074] For abnormal nodes Marked with a red solid circle;

[0075] For each abnormal link (N i ,N j )∈E A , connect the corresponding nodes with straight lines in the figure, the thickness of the connecting line depends on I ij The value is determined by the line type. A larger value will draw a thicker line, and a smaller value will draw a thinner line.

[0076] The present invention has the following beneficial effects:

[0077] 1. By establishing a transaction record set, defining specific data elements for each transaction, and employing a proprietary latitude-longitude plane mapping function, this solution achieves precise quantitative representation of raw loan transaction data. Specifically, in the first step, the amount, time, and latitude-longitude information for each transaction are clearly recorded. Using a proprietary function, these latitude-longitude coordinates are converted to plane coordinates. This step addresses the issues of ambiguous geographic information representation and projection methods unsuitable for real-time transaction monitoring in traditional methods, ensuring that transaction location data is accurately reflected on a plane, thus providing a precise spatial basis for subsequent risk calculations. Subsequently, by calculating the amount difference, time interval, and plane spatial distance for any pair of consecutive transactions, the solution directly calculates the rate of change of consecutive transactions. This function then constructs a preliminary risk function, which accumulates the rate of change of amount per unit time and the rate of spatial displacement to form a comprehensive indicator reflecting loan transaction volatility. This indicator addresses the inability of traditional risk monitoring methods to directly quantify the interactive effects of multidimensional data, making the detection of overall loan transaction volatility more intuitive and real-time. Furthermore, by setting preliminary risk thresholds, the system can fine-tune the sensitivity of risk warnings based on historical data and actual business needs. If the threshold is set low, it can capture small fluctuations and quickly trigger an early warning, thus avoiding the spread of risks; if it is set high, the early warning will only be triggered when abnormal fluctuations are obvious, reducing the false alarm rate. This flexible control mechanism solves the problem of insensitive risk warnings or frequent false alarms in traditional methods, ensuring that the early warning mechanism is both highly sensitive and can avoid interference from non-real risks. Overall, through data recording and self-created mapping, continuous difference calculation, and the construction of preliminary risk numerical functions, this solution achieves quantitative detection of hidden dangers of loan transaction risk linkage. This method makes the loan risk monitoring process highly accurate, real-time, and controllable, which helps financial institutions detect potential loan risks in advance, thereby taking timely intervention measures to ensure the stable operation of loan business and reduce possible financial risk losses.

[0078] 2. By constructing a multidimensional data interaction matrix, this solution achieves the goal of uniformly processing and quantifying continuous transaction difference data across multiple dimensions. Specifically, the three differences (amount difference, time interval, and spatial distance) calculated between consecutive transaction pairs are first arranged in a predetermined order into a raw data matrix. This step overcomes the problem in traditional methods where independent processing of each dimension of data prevents a direct representation of multidimensional data interactions. This allows each row of data to simultaneously reflect the changes in multiple dimensions of a pair of consecutive transactions. Next, each column of the raw data matrix is ​​normalized. By setting minimum and maximum values ​​for each column and using a normalization formula to convert the raw data to a uniform numerical range between 0 and 1, this method effectively eliminates the effects of dimensional differences between amount, time, and space data. This step addresses the data bias and scaling issues inherent in traditional mixed comparisons of multidimensional data, making subsequent data comparisons and comprehensive calculations more accurate and reliable. Furthermore, normalization ensures that information from different dimensions can be compared and integrated under the same standard, laying a solid foundation for the construction of further data interaction models. The normalized multidimensional data matrix directly provides a unified and standardized data input for subsequent risk warning analysis and topological structure construction, thereby greatly improving the overall data consistency and computational stability of the risk monitoring system. The effect of this step is that by integrating continuous transaction difference data into a structured, normalized matrix, the system can accurately capture the intrinsic connections and comprehensive characteristics between data of each dimension, thereby improving the accuracy of overall risk detection and early warning. At the same time, the normalized unified scale prevents the data from being disturbed by the dimension or numerical range of the original data during subsequent processing, thereby improving the robustness and interpretability of the model. Overall, by constructing and normalizing the multidimensional data interaction matrix, this solution effectively solves the problems of inconsistent data dimensions and difficult comparisons, thereby realizing the refined and standardized processing of multi-source transaction data, which will help to build a more accurate and reliable loan risk monitoring system in the future, improve the early warning system's ability to capture risk linkage conditions, and ensure the scientific nature and stability of the early warning results.

[0079] 3. By constructing a normalized interaction matrix vector, this solution converts continuous transaction difference data from matrix form to vector form, providing standardized, unified basic data for subsequent topology map construction. Specifically, the three data points in each row of the normalized raw data matrix represent the normalized values ​​of the amount difference, time interval, and spatial distance between consecutive transaction pairs. This step addresses the problem of data bias caused by different dimensions and numerical ranges in the raw transaction data during direct comparison and calculation, ensuring that data from different dimensions are mapped to the same numerical range. Next, by setting a normalization vector for each transaction pair, the normalized data in each row is organically combined into a unified three-dimensional vector. Through this vectorization step, the previously independent and dispersed multidimensional data is unified into a structured data unit with complete information and high comparability. Because the amount, time, and spatial data in their original state vary significantly in magnitude and dimension, the normalization process standardizes each dimension to a unified numerical range. This eliminates errors caused by different dimensions and ensures the consistency and accuracy of subsequent calculations. The construction of a normalized interaction matrix vector organically combines multidimensional data, previously dispersed across rows in the matrix, into a unified, standardized vector. This structured data unit facilitates subsequent topological graph construction and interactive relationship modeling, providing a clear and unified mathematical foundation for node generation and edge calculation. By standardizing vector data, subsequent risk linkage and anomaly detection algorithms can directly base their calculations on the distances or differences between vectors, reducing the complexity and error accumulation caused by inconsistent data dimensions, thereby improving the accuracy and robustness of the overall system in capturing risk fluctuations. All normalized vectors are derived through explicit mathematical expressions and standardized processing, ensuring the clarity and meaning of each data unit. This provides a solid foundation for subsequent topological construction and facilitates the interpretation and validation of model results. Overall, by constructing a normalized interaction matrix vector, this step effectively addresses the dimensional discrepancies and loose data structure inherent in multidimensional transaction data, achieving standardized and vectorized data representation. This provides a unified, accurate, and efficient data foundation for subsequent risk analysis, topological structure construction, and anomaly detection, thereby improving the accuracy and practicality of loan risk monitoring methods.

[0080] 4. Through topological structure construction and node definition, this solution enables intuitive representation of the interrelationships between normalized transaction data, providing a structured and quantitative foundation for subsequent analysis within the risk monitoring system. This step primarily involves the following operations: First, each normalized vector is assigned to a node. This step solves the problem of assigning continuous transaction data units to structured network nodes, enabling intuitive conversion of transaction pair data into nodes within a network graph. Furthermore, all nodes are grouped into a node set, establishing a unified graphical model for the entire transaction data. After node construction, the solution calculates the difference between any two nodes in each dimension. Subsequently, the solution superimposes these differences across these dimensions to construct a total internode difference value, which reflects the total difference between any two nodes in the three-dimensional normalized data. By explicitly converting each normalized vector into a node and forming a node set, this approach addresses the problem of fragmented information in the raw transaction data, making it difficult to intuitively represent its multidimensional interactions. Through node definition, all transaction data is incorporated into a unified network structure, enabling a holistic network representation of risk relationships. Secondly, the step of calculating the three-dimensional differences in amount, time, and space between any two nodes and superimposing them as a total difference addresses the inability to directly compare and integrate multidimensional data. Traditional methods often struggle to simultaneously account for the impact of data from different dimensions. However, this solution, through a simple summation operation after normalization, enables a comprehensive evaluation of all data dimensions on the same scale, improving the fairness and accuracy of data comparison. Furthermore, the calculation of the total difference between nodes provides a clear mathematical basis for the subsequent topological graph-based interaction strength calculation and anomalous node identification. This quantitative calculation not only intuitively demonstrates the similarities or differences between nodes but also provides a unified and standardized quantitative input for subsequent steps such as network edge establishment, interaction indicators, and anomaly determination, thereby enhancing the transparency and interpretability of the entire risk monitoring system. Overall, through the node construction and inter-node difference superposition steps, this solution not only addresses the information dispersion and dimensionality inconsistency issues encountered in traditional multidimensional transaction data processing, but also lays a solid foundation for building an efficient and transparent risk warning model, thereby improving the overall performance and accuracy of the loan risk monitoring system.

[0081] 5. By establishing edges between nodes and constructing a complete topological graph, this solution achieves the key goal of constructing a network structure from normalized multidimensional transaction data, thereby providing an intuitive and quantitative representation of interactive relationships for risk detection. The specific steps include: first, setting a fixed threshold for edge establishment to determine whether the normalized difference between any two nodes is sufficiently small, thereby determining whether to establish an undirected edge between the corresponding nodes. This setting enables the system to effectively identify and connect data points with high similarity—that is, consecutive transaction pairs with small differences. This directly addresses the difficulty of intuitively representing multidimensional data relationships in traditional methods. When the value of δ is low, only edges between nodes with extremely small differences are established. This results in a sparse topological graph, with each node connected only to other highly similar nodes. This helps to accurately identify transaction pairs with truly anomalous interaction patterns. However, it may miss some data with moderate similarity but still potentially risky associations, resulting in insufficient capture of some risk chains. On the contrary, when the value of δ is high, more edges are established between nodes, which will greatly increase the number of edges in the graph, making the topological graph structure denser and improving the overall interaction intensity; however, this situation may also introduce too much noise data, causing some non-abnormal nodes to be incorrectly connected, affecting the accurate judgment of abnormal patterns. Therefore, by reasonably setting and adjusting the value of the threshold δ, the solution solves the problem of how to strike a balance between ensuring the authenticity of graph recognition and avoiding noise interference in multidimensional data. Specifically, this step establishes a fixed threshold as the edge establishment judgment standard, while removing the impact of excessive differences on the network, ensuring that transaction nodes with similar characteristics can be connected in time to form a complete multidimensional interactive network. Subsequently, when judging if any two nodes meet D ij <δ, at node N i With node N j Undirected edges are established between them, and all edges that meet the conditions are combined into an edge set E, ultimately constructing an undirected topological graph G = (V, E). The effects of this step are: first, using a fixed threshold screening mechanism, multidimensional interaction data is converted into a graph structure, intuitively presenting the similarities and interaction characteristics between the data; second, through fine-tuning the threshold, the graph can effectively capture highly similar abnormal interactions while avoiding misjudgments caused by noisy data; finally, the constructed topological graph provides a rigorous mathematical foundation for subsequent graph-theory-based interaction strength calculations and abnormal link identification, thereby improving the accuracy, robustness, and real-time performance of the loan risk warning system as a whole.

[0082] 6. Through the dynamic relationship calculation and interaction strength modeling steps, this solution realizes the refined modeling and dynamic monitoring of the interaction relationship between each node in the transaction data, thereby improving the risk warning system's ability to respond to loan risk fluctuations. First, by setting the interaction strength function, the interaction strength of each edge between nodes is calculated. This step solves the problem that the traditional method cannot quantitatively describe and directly compare the relationship between nodes, and expresses the similarity between multidimensional data with clear numerical values, thereby enhancing the transparency and interpretability of the model. Secondly, for each node N i Define its adjacent node set The total interaction strength of each node is then calculated. This accumulation effectively consolidates the interaction between a node and all surrounding nodes into a single value, resolving the problem of fragmented interaction information being difficult to centrally represent. This step allows the model to clearly reflect the interaction level of a node within the overall trading network, providing a basis for subsequent anomaly detection. Subsequently, this approach divides the entire trading sequence into multiple time windows at fixed intervals of ΔT, enabling segmented statistics of local dynamic relationships. This time window configuration captures subtle fluctuations in trading data over time. Furthermore, by summarizing the interaction strength of all nodes within a window, the average interaction strength within that window is calculated. This step addresses the issues of data volatility and unstable anomaly detection caused by too few or too many statistical samples at a single moment. A smaller ΔT value can capture subtle fluctuations within a shorter time period, but may result in larger fluctuations in the average value due to insufficient data samples. A larger ΔT value can better reflect long-term trends but may also smooth out short-term anomalies. In practical applications, the optimal window length should be determined through pre-experimental data fitting based on trading frequency and risk monitoring requirements to ensure sufficient data and sensitive detection of sudden anomalies. Overall, by setting an interaction strength function and calculating the interaction strength of each edge, the problem of unclear quantification and inability to directly compare interactions between nodes was resolved. By defining the total interaction strength of nodes, local interaction information was effectively integrated. Furthermore, through time window division and mean statistics, the instability caused by insufficient or excessive local dynamic data samples was resolved. Ultimately, this dynamic relationship calculation and interaction strength modeling step enabled the system to accurately capture interaction fluctuations in loan transaction data over different time periods, providing a scientific and transparent quantitative basis for risk warnings, thereby improving the speed and accuracy of the warning system's response to loan risk fluctuations and abnormal behavior.

[0083] 7. By setting up the steps of abnormal interaction pattern recognition and risk linkage triggering, this solution realizes the quantitative monitoring of abnormal node interaction in the trading network, thus providing a clear and real-time trigger mechanism for risk warning. Specifically, first, within a single time window, by calculating the abnormal interaction degree of each node, the problem of how to compare local interaction information with the overall trend to determine whether the node is abnormal is solved. This step enables the status of each node to be represented by a clear numerical value A. i The expression directly reflects the degree of deviation of the node from the local average level. Then, the scheme sets the abnormality judgment threshold T A , by A i With T A The comparison solves the problem of how to objectively judge which nodes have abnormal interactions in a complex network. i >T A The node is judged as an abnormal node, otherwise it is a normal node; a reasonable T A If the value is too low, most nodes will be identified as abnormal, which will make the set of abnormal nodes too large and generate a large number of false alarms. On the contrary, if T A If the value is too high, only the nodes with extreme deviations will be judged as abnormal, which may miss some real risk signals. Therefore, this setting provides a solution to flexibly adjust parameters based on historical data and actual risk transmission characteristics to balance detection sensitivity and accuracy. Establish abnormal link set E A This solves the problem of how to intuitively express the linkages between abnormal nodes. This step allows the risk linkage structure to be clearly revealed, helping risk managers to grasp the overall risk diffusion path. Overall, by setting and calculating abnormal interaction levels and identifying abnormal nodes and links, the problem of how to accurately and quantitatively determine node anomalies in multi-dimensional interactive data is solved, and the intuitive disclosure and real-time monitoring of risk linkages between nodes are achieved. This step enables the system to promptly identify and respond to risk areas, improving the sensitivity and accuracy of the loan risk monitoring and early warning system, and providing risk management departments with a scientific, transparent, and adjustable basis for risk assessment.

[0084] 8. Through the key step of risk warning report generation and visualization map construction, this solution realizes the accurate recording of information of abnormal transaction nodes, quantitative analysis of risk linkage level and intuitive graphical display, thereby effectively solving the problems of unclear warning information, insufficient quantitative indicators and poor visualization effect in traditional loan risk warning systems, and provides risk management personnel with real-time, transparent and adjustable warning basis. First, this step records key information such as the original transaction serial number, normalized data vector, total node interaction intensity and abnormal interaction degree for each node in the abnormal node set. This approach solves the problems of incomplete node data recording and information confusion in traditional systems, and ensures that the detailed data of each node is fully preserved after being judged as abnormal, providing a solid data foundation for subsequent traceability analysis of risk events. Then, this step constructs the overall risk linkage value R total The calculation is to accumulate all abnormal interactions to form an overall risk index, and set the warning threshold T R To judge whether the system is in a high-risk state. This design solves the problem that the warning indicators in the traditional early warning system lack quantitative basis and the parameters are difficult to adjust. R The system balances sensitivity and false alarm rate, capturing subtle, early-stage anomalous interactions in transaction data while avoiding frequent false alarms caused by short-term fluctuations. This ensures the real-time nature of risk warnings while enhancing the reliability of warning information, providing decision makers with a quantitative reference that accurately reflects the actual level of risk transmission. Finally, by constructing a two-dimensional graph with the transaction record number on the horizontal axis and the total interaction intensity of each node on the vertical axis, and visually marking nodes and anomalous links with color and line thickness, this solves the problem of unclear graphical displays and the inability to intuitively reflect data interactions in traditional systems. In the graph, normal nodes are displayed as solid circles, while anomalous nodes are marked with red solid circles. Furthermore, links connecting anomalous nodes are marked with lines of varying thickness to represent interaction intensity, visually illustrating the diffusion path of risk linkages. This visualization not only facilitates risk management personnel to quickly locate anomalous risk areas but also provides timely feedback on risk transmission, providing decision support for risk intervention. In summary, by recording detailed node information, constructing an overall risk linkage value, and intuitively building a risk map, this step effectively transforms the risk signals implicit in multi-dimensional transaction data into quantitative and visual early warning information, significantly improving the system's monitoring accuracy, response speed, and reliability of loan risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0086] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0087] Example, see Figure 1 , a loan risk monitoring method based on transaction data analysis, comprising:

[0088] Set the transaction record set to T = {T1, T2, ..., T n};

[0089] Among them, T i is the i-th transaction record; n is the total number of transaction records, and n≥2;

[0090] Each transaction record contains: transaction amount Q i , represents the amount of the i-th transaction, in yuan; transaction time τ i , represents the time when the i-th transaction occurs, in seconds, starting from the predetermined initial time τ0 = 0; transaction position represents the latitude of the location where the i-th transaction occurred, in degrees; λ i The longitude of the location where the i-th transaction occurred, in degrees;

[0091] Set the coordinate mapping function to convert the latitude and longitude data into plane coordinates, specifically:

[0092]

[0093] Among them, x i y represents the horizontal coordinate of the plane obtained by the coordinate mapping function for the i-th transaction; i represents the plane vertical coordinate of the i-th transaction obtained by the coordinate mapping function; π is the circumference of a circle;

[0094] For any continuous trading pair (T i ,T i+1 ), calculate the following three difference quantities:

[0095] Amount difference ΔQ i :ΔQ i =|Q i+1 -Q i |; where ΔQ i It represents the difference in the amount of the i-th pair of consecutive transactions, in yuan;

[0096] Time interval Δτ i :Δτ i =τi+1 -τ i ; where Δτ i Indicates the time difference between the i-th pair of consecutive transactions, in seconds;

[0097] Plane space distance d i : Among them, d i represents the Euclidean distance between the i-th pair of consecutive transactions in the plane;

[0098] Construct the initial risk value function and calculate the initial risk value, specifically:

[0099]

[0100] Among them, R0 is the initial risk value, which is the cumulative value of the change rate of all consecutive transactions and is used to reflect the volatility of loan transactions; is the rate of change of amount per unit time; is the spatial displacement rate per unit time;

[0101] Set the initial risk threshold to K0. If K0 is set too low, the sensitivity to changes in transaction amount and spatial distance will increase, which may easily cause R0 to exceed K0, resulting in a higher risk alert rate and more false positives. If K0 is set too high, it will only be judged as a risk when the transaction data fluctuates abnormally, which may make the initial risk warning less sensitive and there is a risk of missing reports. Therefore, in actual application, a reasonable value of K0 should be determined based on the loan product, transaction frequency and historical fluctuation characteristics, so as to improve the warning accuracy while avoiding unnecessary alarm interference.

[0102] If R0>K0, it is preliminarily judged that the loan has hidden risks of risk linkage;

[0103] If R0≤K0, it can be preliminarily judged that there is no risk linkage risk in the loan.

[0104] By establishing a transaction record set, defining specific data elements for each transaction, and employing a proprietary latitude-longitude plane mapping function, this solution achieves precise quantitative representation of raw loan transaction data. Specifically, in the first step, the amount, time, and latitude-longitude information for each transaction are clearly recorded. Using a proprietary function, these latitude-longitude coordinates are converted to plane coordinates. This step addresses the issues of ambiguous geographic information representation and projection methods unsuitable for real-time transaction monitoring in traditional methods, ensuring that transaction location data is accurately reflected on a plane, thus providing a precise spatial basis for subsequent risk calculations. Subsequently, by calculating the amount difference, time interval, and plane spatial distance for any pair of consecutive transactions, the solution directly calculates the rate of change of consecutive transactions. This function then constructs a preliminary risk function, which accumulates the rate of change of amount per unit time and the rate of spatial displacement to form a comprehensive indicator reflecting loan transaction volatility. This indicator addresses the inability of traditional risk monitoring methods to directly quantify the interactive effects of multidimensional data, making the detection of overall loan transaction volatility more intuitive and real-time. Furthermore, by setting preliminary risk thresholds, the system can fine-tune the sensitivity of risk warnings based on historical data and actual business needs. If the threshold is set low, it can capture small fluctuations and quickly trigger an early warning, thus avoiding the spread of risks; if it is set high, the early warning will only be triggered when abnormal fluctuations are obvious, reducing the false alarm rate. This flexible control mechanism solves the problem of insensitive risk warnings or frequent false alarms in traditional methods, ensuring that the early warning mechanism is both highly sensitive and can avoid interference from non-real risks. Overall, through data recording and self-created mapping, continuous difference calculation, and the construction of preliminary risk numerical functions, this solution achieves quantitative detection of hidden dangers of loan transaction risk linkage. This method makes the loan risk monitoring process highly accurate, real-time, and controllable, which helps financial institutions detect potential loan risks in advance, thereby taking timely intervention measures to ensure the stable operation of loan business and reduce possible financial risk losses.

[0105] It also includes constructing a multidimensional data interaction matrix, specifically:

[0106] The calculated differences of each continuous trading pair are arranged into the original data matrix M according to the following formula:

[0107]

[0108] Among them, row index i corresponds to the i-th continuous trading pair; column index ΔQ i , Δτ i d i They correspond to the amount difference, time interval, and spatial distance respectively;

[0109] For the j-th column of the original data matrix M, let:

[0110]

[0111] Among them, m j is the minimum value of the jth column of the original data matrix M; M j The maximum value of the jth column of the original data matrix M; when j = 1, it represents the amount difference, when j = 2, it represents the time interval, and when j = 3, it represents the spatial distance;

[0112] Calculate the normalized value for each element of the original data matrix M, specifically:

[0113]

[0114] in, Represents the data in the i-th row and j-th column of the normalized original data matrix M.

[0115] By constructing a multidimensional data interaction matrix, this solution achieves the goal of uniformly processing and quantifying continuous transaction difference data across multiple dimensions. Specifically, the three differences (amount difference, time interval, and spatial distance) calculated between consecutive transaction pairs are first arranged in a predetermined order into a raw data matrix. This step overcomes the problem in traditional methods where each dimension of data is processed independently, resulting in an inability to intuitively reflect the interactions between multidimensional data. This allows each row of data to simultaneously reflect the changes in a pair of consecutive transactions across multiple dimensions. Next, each column of the raw data matrix is ​​normalized. By setting minimum and maximum values ​​for each column and using a normalization formula to convert the raw data to a uniform range between 0 and 1, this method effectively eliminates the effects of different dimensionalities between amount, time, and space data. This step addresses the data bias and scaling issues inherent in traditional mixed comparisons of multidimensional data, making subsequent data comparisons and comprehensive calculations more accurate and reliable. Furthermore, normalization ensures that information from different dimensions can be compared and integrated under the same standard, laying a solid foundation for the construction of further data interaction models. The normalized multidimensional data matrix directly provides a unified and standardized data input for subsequent risk warning analysis and topological structure construction, thereby greatly improving the overall data consistency and computational stability of the risk monitoring system. The effect of this step is that by integrating continuous transaction difference data into a structured, normalized matrix, the system can accurately capture the intrinsic connections and comprehensive characteristics between data of each dimension, thereby improving the accuracy of overall risk detection and early warning. At the same time, the normalized unified scale prevents the data from being disturbed by the dimension or numerical range of the original data during subsequent processing, thereby improving the robustness and interpretability of the model. Overall, by constructing and normalizing the multidimensional data interaction matrix, this solution effectively solves the problems of inconsistent data dimensions and difficult comparisons, thereby realizing the refined and standardized processing of multi-source transaction data, which will help to build a more accurate and reliable loan risk monitoring system in the future, improve the early warning system's ability to capture risk linkage conditions, and ensure the scientific nature and stability of the early warning results.

[0116] It also includes constructing a normalized interaction matrix vector, specifically:

[0117] Set the normalized original data matrix to:

[0118]

[0119] in, Represents the normalized value of the amount difference of the i-th transaction pair; Represents the normalized value of the time interval of the i-th trading pair; Represents the normalized value of the spatial distance of the i-th trading pair;

[0120] Let the normalized vector be: in, is the i-th normalized vector, which serves as the basis for subsequent topological graph construction.

[0121] By constructing a normalized interaction matrix vector, this solution converts continuous transaction difference data from matrix form to vector form, providing standardized, unified basic data for subsequent topology map construction. Specifically, the three data points in each row of the normalized raw data matrix represent the normalized values ​​of the amount difference, time interval, and spatial distance between consecutive transaction pairs. This step addresses the data bias caused by different dimensions and numerical ranges in the raw transaction data during direct comparison and calculation, ensuring that data from different dimensions are mapped to the same numerical range. Next, by setting a normalized vector for each transaction pair, the normalized data in each row is organically combined into a unified three-dimensional vector. This vectorization step represents the previously independent and dispersed multidimensional data as a structured data unit with complete information and high comparability. Because the amount, time, and spatial data in their original state vary significantly in magnitude and dimension, the normalization process standardizes each dimension to a unified numerical range. This eliminates errors caused by different dimensions and ensures the consistency and accuracy of subsequent calculations. The construction of a normalized interaction matrix vector organically combines multidimensional data, previously dispersed across rows in the matrix, into a unified, standardized vector. This structured data unit facilitates subsequent topological graph construction and interactive relationship modeling, providing a clear and unified mathematical foundation for node generation and edge calculation. By standardizing vector data, subsequent risk linkage and anomaly detection algorithms can directly base their calculations on the distances or differences between vectors, reducing the complexity and error accumulation caused by inconsistent data dimensions, thereby improving the accuracy and robustness of the overall system in capturing risk fluctuations. All normalized vectors are derived through explicit mathematical expressions and standardized processing, ensuring the clarity and meaning of each data unit. This provides a solid foundation for subsequent topological construction and facilitates the interpretation and validation of model results. Overall, by constructing a normalized interaction matrix vector, this step effectively addresses the dimensional discrepancies and loose data structure inherent in multidimensional transaction data, achieving standardized and vectorized data representation. This provides a unified, accurate, and efficient data foundation for subsequent risk analysis, topological structure construction, and anomaly detection, thereby improving the accuracy and practicality of loan risk monitoring methods.

[0122] It also includes topology construction and node definition, specifically:

[0123] For each normalized vector Set the corresponding node Ni , specifically:

[0124] All nodes constitute a node set V = {N1, N2, ..., N n-1};

[0125] For any N i ,N j ∈V and i≠j, calculate respectively:

[0126]

[0127] Set the total difference between nodes to:

[0128] Among them, D ij Represents node N i and N j Total differences in three-dimensional normalized data; Represents node N i and N j Differences in the dimension of amount difference; Represents node N i and N j Differences in the time interval dimension; Represents node N i and N j Differences in the spatial distance dimension.

[0129] Through topological structure construction and node definition, this solution enables intuitive representation of the interrelationships between normalized transaction data, providing a structured and quantitative foundation for subsequent analysis within the risk monitoring system. This step primarily involves the following operations: First, each normalized vector is assigned to a node. This resolves the issue of assigning continuous transaction data units to structured network nodes, enabling intuitive conversion of transaction pair data into nodes within a network graph. Furthermore, all nodes are grouped into a node set, establishing a unified graphical model for the entire transaction data. After node construction, the solution calculates the difference between any two nodes in each dimension. Subsequently, the solution superimposes these differences across these dimensions to construct a total internode difference value, which reflects the total difference between any two nodes in the three-dimensional normalized data. By explicitly converting each normalized vector into a node and forming a node set, this approach addresses the issue of fragmented information in the raw transaction data, making it difficult to intuitively represent its multidimensional interactions. Through node definition, all transaction data is incorporated into a unified network structure, enabling a holistic network representation of risk relationships. Secondly, the step of calculating the three-dimensional differences in amount, time, and space between any two nodes and superimposing them as a total difference addresses the inability to directly compare and integrate multidimensional data. Traditional methods often struggle to simultaneously account for the impact of data from different dimensions. However, this solution, through a simple summation operation after normalization, enables a comprehensive evaluation of all data dimensions on the same scale, improving the fairness and accuracy of data comparison. Furthermore, the calculation of the total difference between nodes provides a clear mathematical basis for the subsequent topological graph-based interaction strength calculation and anomalous node identification. This quantitative calculation not only intuitively demonstrates the similarities or differences between nodes but also provides a unified and standardized quantitative input for subsequent steps such as network edge establishment, interaction indicators, and anomaly determination, thereby enhancing the transparency and interpretability of the entire risk monitoring system. Overall, through the node construction and inter-node difference superposition steps, this solution not only addresses the information dispersion and dimensionality inconsistency issues encountered in traditional multidimensional transaction data processing, but also lays a solid foundation for building an efficient and transparent risk warning model, thereby improving the overall performance and accuracy of the loan risk monitoring system.

[0130] It also includes establishing edges and topology graphs between nodes, specifically:

[0131] A fixed threshold for edge establishment is set to δ. When δ is low, only trading pairs with minimal differences between nodes will establish edges. This results in fewer edges in the graph and a sparse topology, which is conducive to high-precision identification of highly similar anomalous interactions, but may miss some nodes with potential connections. When δ is high, more edges are established between nodes with moderate differences, increasing the number of edges in the graph and making the topology denser, thereby improving the overall interaction strength statistics. However, this may also lead to the incorporation of noise data, affecting the ability to distinguish anomalous patterns. Therefore, the appropriate choice of δ requires a balance between ensuring the authenticity of graph recognition and avoiding excessive noise. Its optimal value should be fine-tuned based on the distribution characteristics of normalized data and actual experience.

[0132] For any i≠j, if: D ij <δ, then at node N i With node N j Establish undirected edges between them, and record the edge set as: E={(N i ,N j )|D ij <δ, and i≠j};

[0133] The topology graph is set to: G = (V, E); where G is an undirected graph consisting of a node set V and an edge set E, representing a multi-dimensional interactive relationship.

[0134] By establishing edges between nodes and constructing a complete topological graph, this solution achieves the key goal of constructing a network structure from normalized multidimensional transaction data, thereby providing an intuitive and quantitative representation of interactive relationships for risk detection. The specific steps include: first, setting a fixed threshold for edge establishment to determine whether the normalized difference between any two nodes is sufficiently small, thereby determining whether to establish an undirected edge between the corresponding nodes. This setting enables the system to effectively identify and connect data points with high similarity—that is, consecutive transaction pairs with small differences. This directly addresses the difficulty of intuitively representing multidimensional data relationships in traditional methods. When the value of δ is low, only edges between nodes with extremely small differences are established. This results in a sparse topological graph, where each node is connected only to other highly similar nodes. This helps to accurately identify transaction pairs with truly anomalous interaction patterns, but may miss data with moderate similarity that still harbors potential risk associations, resulting in insufficient capture of some risk chains. On the contrary, when the value of δ is high, more edges are established between nodes, which will greatly increase the number of edges in the graph, making the topological graph structure denser and improving the overall interaction intensity; however, this situation may also introduce too much noise data, causing some non-abnormal nodes to be incorrectly connected, affecting the accurate judgment of abnormal patterns. Therefore, by reasonably setting and adjusting the value of the threshold δ, the solution solves the problem of how to strike a balance between ensuring the authenticity of graph recognition and avoiding noise interference in multidimensional data. Specifically, this step establishes a fixed threshold as the edge establishment judgment standard, while removing the impact of excessive differences on the network, ensuring that transaction nodes with similar characteristics can be connected in time to form a complete multidimensional interactive network. Subsequently, when judging if any two nodes meet D ij <δ, at node N i With node N j Undirected edges are established between them, and all edges that meet the conditions are combined into an edge set E, ultimately constructing an undirected topological graph G = (V, E). The effects of this step are: first, using a fixed threshold screening mechanism, multidimensional interaction data is converted into a graph structure, intuitively presenting the similarities and interaction characteristics between the data; second, through fine-tuning the threshold, the graph can effectively capture highly similar abnormal interactions while avoiding misjudgments caused by noisy data; finally, the constructed topological graph provides a rigorous mathematical foundation for subsequent graph-theory-based interaction strength calculations and abnormal link identification, thereby improving the accuracy, robustness, and real-time performance of the loan risk warning system as a whole.

[0135] It also includes dynamic relationship calculation and interaction intensity modeling, specifically:

[0136] Set the interaction strength function and for each edge (N i ,N j )∈E calculates the interaction intensity I ij, specifically:

[0137]

[0138] Among them, I ij For node N i With N j The interaction strength between them; the addition of 1 in the denominator is used to ensure that D ij = 0, no division by zero error will occur, and the interaction strength will increase with D ij decreases monotonically with the increase of

[0139] For node N i , set its adjacent node set to

[0140] Node N i The total interaction intensity I i Set to:

[0141] The entire transaction sequence is divided into multiple time windows at fixed time intervals. The length of the time window is set to ΔT, which is used to calculate the local dynamic relationship in segments. If the value of ΔT is small, the time window contains less transaction data, which can capture subtle interactive fluctuations in a shorter period of time. However, the small number of statistical samples may make I mean The fluctuation is large, which increases the instability of abnormal judgment; if the ΔT value is large, the amount of data in each window increases, and I mean A smoother mean value can reflect long-term trends, but short-term anomalies may be averaged out by the majority of normal data within the window, thereby reducing sensitivity to sudden events. Therefore, the selection of ΔT needs to be balanced based on trading frequency and risk warning requirements, balancing sufficient data support statistics with high sensitivity to local anomalies. In actual settings, pre-experimental data fitting can be used to determine the optimal window length.

[0142] In any time window, let the number of nodes contained in the window be N w , calculate the arithmetic mean of the total interaction intensity of all nodes in the time window, specifically:

[0143] Among them, I mean is the average interaction strength of all nodes in the current time window.

[0144] Through the dynamic relationship calculation and interaction strength modeling steps, this solution realizes the refined modeling and dynamic monitoring of the interaction relationship between each node in the transaction data, thereby improving the risk warning system's ability to respond to loan risk fluctuations. First, by setting the interaction strength function, the interaction strength of each edge between nodes is calculated. This step solves the problem that the traditional method cannot quantitatively describe and directly compare the relationship between nodes, and expresses the similarity between multidimensional data with clear numerical values, thereby enhancing the transparency and interpretability of the model. Secondly, for each node N i Define its adjacent node set The total interaction strength of each node is then calculated. This accumulation effectively consolidates the interaction between a node and all surrounding nodes into a single value, resolving the problem of fragmented interaction information being difficult to centrally represent. This step allows the model to clearly reflect the interaction level of a node within the overall trading network, providing a basis for subsequent anomaly detection. Subsequently, this approach divides the entire trading sequence into multiple time windows at fixed intervals of ΔT, enabling segmented statistics of local dynamic relationships. This time window configuration captures subtle fluctuations in trading data over time. Furthermore, by summarizing the interaction strength of all nodes within a window, the average interaction strength within that window is calculated. This step addresses the issues of data volatility and unstable anomaly detection caused by too few or too many statistical samples at a single moment. A smaller ΔT value can capture subtle fluctuations within a shorter time period, but may result in larger fluctuations in the average value due to insufficient data samples. A larger ΔT value can better reflect long-term trends but may also smooth out short-term anomalies. In practical applications, the optimal window length should be determined through pre-experimental data fitting based on trading frequency and risk monitoring requirements to ensure sufficient data and sensitive detection of sudden anomalies. Overall, by setting an interaction strength function and calculating the interaction strength of each edge, the problem of unclear quantification and inability to directly compare interactions between nodes was resolved. By defining the total interaction strength of nodes, local interaction information was effectively integrated. Furthermore, through time window division and mean statistics, the instability caused by insufficient or excessive local dynamic data samples was resolved. Ultimately, this dynamic relationship calculation and interaction strength modeling step enabled the system to accurately capture interaction fluctuations in loan transaction data over different time periods, providing a scientific and transparent quantitative basis for risk warnings, thereby improving the speed and accuracy of the warning system's response to loan risk fluctuations and abnormal behavior.

[0145] It also includes abnormal interaction pattern recognition and risk linkage triggering, specifically:

[0146] For node N i In a single time window, set the abnormal interaction degree to: A i =I i -I mean ;

[0147] Among them, A i For node N i The difference between the total interaction intensity and the average interaction intensity in the time window is used to reflect whether there is abnormal interaction phenomenon at the node;

[0148] Set the abnormality judgment threshold to T A If T A If the setting is low, the tolerance for deviation of the interaction strength of the nodes is small, and more nodes may be judged as abnormal, resulting in a larger set of abnormal nodes and possible false positives. A If the setting is high, only nodes whose interaction intensity significantly exceeds the local mean will be identified as abnormal, which may miss some real risk signals. A The value should be determined based on historical data and actual risk transmission characteristics to ensure that anomaly detection has high sensitivity without excessively spreading false alarm information;

[0149] When for node N i Satisfy A i ≤T A When , the node is determined to be a normal node;

[0150] When for node N i Satisfy A i >T A When , the node is determined to be an abnormal node, and the abnormal node set is recorded as

[0151] In topology graph G, set the abnormal link set to E A , specifically:

[0152] This set directly reflects the risk linkage structure formed between abnormal nodes; among them, E A It represents the set of all edges connecting abnormal nodes, directly revealing the structure of risk linkage.

[0153] By setting up the steps of abnormal interaction pattern recognition and risk linkage triggering, this solution realizes the quantitative monitoring of abnormal node interaction in the trading network, thus providing a clear and real-time triggering mechanism for risk warning. Specifically, first, within a single time window, by calculating the abnormal interaction degree of each node, the problem of how to compare local interaction information with the overall trend to determine whether the node is abnormal is solved. This step enables the status of each node to be represented by a clear numerical value A. i The expression directly reflects the degree of deviation of the node from the local average level. Then, the scheme sets the abnormality judgment threshold T A , by A i With T AThe comparison solves the problem of how to objectively judge which nodes have abnormal interactions in a complex network. i >T A The node is judged as an abnormal node, otherwise it is a normal node; a reasonable T A If the value is too low, most nodes will be identified as abnormal, which will make the set of abnormal nodes too large and generate a large number of false alarms. On the contrary, if T A If the value is too high, only the nodes with extreme deviations will be judged as abnormal, which may miss some real risk signals. Therefore, this setting provides a solution to flexibly adjust parameters based on historical data and actual risk transmission characteristics to balance detection sensitivity and accuracy. Establish abnormal link set E A This solves the problem of how to intuitively express the linkages between abnormal nodes. This step allows the risk linkage structure to be clearly revealed, helping risk managers to grasp the overall risk diffusion path. Overall, by setting and calculating abnormal interaction levels and identifying abnormal nodes and links, the problem of how to accurately and quantitatively determine node anomalies in multi-dimensional interactive data is solved, and the intuitive disclosure and real-time monitoring of risk linkages between nodes are achieved. This step enables the system to promptly identify and respond to risk areas, improving the sensitivity and accuracy of the loan risk monitoring and early warning system, and providing risk management departments with a scientific, transparent, and adjustable basis for risk assessment.

[0154] It also includes risk warning report generation and visualization map construction, specifically:

[0155] For abnormal node set Each node N in i , record the following data: original transaction number i; normalized data vector Total node interaction intensity I i Abnormal interaction A i ;

[0156] Set the overall risk linkage value R total It is the cumulative sum of abnormal interactions, specifically:

[0157]

[0158] And set the warning threshold to T R ; Balance between warning sensitivity and false alarm risk: When T R When the value is low, that is, the tolerance for risk is reduced, the smaller the total abnormal interaction cumulative value R totalThe system can trigger an early warning at the early stage of risk transmission and can capture subtle abnormal interactions. However, this will also increase the sensitivity of the early warning and make it more susceptible to interference from noise and occasional trading fluctuations, resulting in more false alarms. When T R When the value is high, only when the overall abnormal interaction cumulative value R total The warning is triggered only when a larger value is reached. This setting can avoid false alarms caused by short-term accidental fluctuations and ensure the reliability of warning information. However, it may also cause a lag in response to initial risk signals and delay intervention in sudden risks. Risk accumulation effect and response time: When T R If the value is set too low, the system may frequently detect risk linkage and trigger warnings, reflecting the cumulative effect of sporadic anomalies in transaction data, but this does not necessarily mean that a real systemic risk event has occurred; when T R If the value is set higher, the system will pay more attention to the accumulation of large-scale abnormal interactions and will not respond until the risk linkage reaches a certain scale. This will challenge the timeliness of the early warning and may fail to provide timely feedback in the early stage of risk diffusion. R The value of should be determined based on historical transaction data, risk warning experience and risk tolerance in actual business scenarios. The value of R can be established through statistical analysis of historical data. total distribution model, and then determine a reasonable quantile as T R The initial value is set and fine-tuned based on on-site feedback during actual operation;

[0159] When R is satisfied total >T R When , the entire system is judged to have a high loan risk linkage situation;

[0160] Construct a two-dimensional graph, where the horizontal axis uses the transaction record number i; the vertical axis uses the total interaction intensity of each node I i ;

[0161] For each node N i , shown as a solid circle in the figure, the horizontal coordinate of the point is i and the vertical coordinate is I i ;

[0162] For abnormal nodes Marked with a red solid circle;

[0163] For each abnormal link (N i ,N j )∈E A , connect the corresponding nodes with straight lines in the figure, the thickness of the connecting line depends on I ij The value is determined by the line type. A larger value will draw a thicker line, and a smaller value will draw a thinner line.

[0164] Through the key step of risk warning report generation and visualization map construction, this solution achieves accurate recording of information on abnormal transaction nodes, quantitative analysis of risk linkage levels, and intuitive graphical display, thereby effectively solving the problems of unclear warning information, insufficient quantitative indicators, and poor visualization in traditional loan risk warning systems, and providing risk management personnel with real-time, transparent, and adjustable warning basis. First, this step records key information such as the original transaction serial number, normalized data vector, total node interaction intensity, and abnormal interaction degree for each node in the abnormal node set. This approach solves the problems of incomplete node data recording and information confusion in traditional systems, ensuring that the detailed data of each node is fully preserved after being judged as abnormal, providing a solid data foundation for subsequent traceability analysis of risk events. Next, this step constructs the overall risk linkage value R total The calculation is to accumulate all abnormal interactions to form an overall risk index, and set the warning threshold T R To judge whether the system is in a high-risk state. This design solves the problem that the warning indicators in the traditional early warning system lack quantitative basis and the parameters are difficult to adjust. R The system balances sensitivity and false alarm rate, capturing subtle, early-stage anomalous interactions in transaction data while avoiding frequent false alarms caused by short-term fluctuations. This ensures the real-time nature of risk warnings while enhancing the reliability of warning information, providing decision makers with a quantitative reference that accurately reflects the actual level of risk transmission. Finally, by constructing a two-dimensional graph with the transaction record number on the horizontal axis and the total interaction intensity of each node on the vertical axis, and visually marking nodes and anomalous links with color and line thickness, this solves the problem of unclear graphical displays and the inability to intuitively reflect data interactions in traditional systems. In the graph, normal nodes are displayed as solid circles, while anomalous nodes are marked with red solid circles. Furthermore, links connecting anomalous nodes are marked with lines of varying thickness to represent interaction intensity, visually illustrating the diffusion path of risk linkages. This visualization not only facilitates risk management personnel to quickly locate anomalous risk areas but also provides timely feedback on risk transmission, providing decision support for risk intervention. In summary, by recording detailed node information, constructing an overall risk linkage value, and intuitively building a risk map, this step effectively transforms the risk signals implicit in multi-dimensional transaction data into quantitative and visual early warning information, significantly improving the system's monitoring accuracy, response speed, and reliability of loan risks.

[0165] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0166] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A loan risk monitoring method based on transaction data analysis, characterized in that: include: Set the transaction record set to T = {T1, T2, ..., T n }; Among them, T i is the i-th transaction record; n is the total number of transaction records, and n≥2; Each transaction record contains: transaction amount Q i , represents the amount of the i-th transaction, in yuan; transaction time τ i , represents the time when the i-th transaction occurs, in seconds, starting from the predetermined initial time τ0 = 0; transaction position represents the latitude of the location where the i-th transaction occurred, in degrees; λ i The longitude of the location where the i-th transaction occurred, in degrees; Set the coordinate mapping function to convert the latitude and longitude data into plane coordinates, specifically: Among them, x i represents the horizontal coordinate of the plane obtained by the coordinate mapping function for the i-th transaction; i represents the plane vertical coordinate of the i-th transaction obtained by the coordinate mapping function; π is the circumference of a circle; For any continuous trading pair (T i ,T i+1 ), calculate the following three difference quantities: Amount difference ΔQ i :ΔQ i =|Q i+1 -Q i |; where ΔQ i It represents the difference in the amount of the i-th pair of consecutive transactions, in yuan; Time interval Δτ i :Δτ i =τ i+1 -τ i ; where Δτ i Indicates the time difference between the i-th pair of consecutive transactions, in seconds; Plane space distance d i : Among them, d i represents the Euclidean distance between the i-th pair of consecutive transactions in the plane; Construct the initial risk value function and calculate the initial risk value, specifically: Among them, R0 is the initial risk value; is the rate of change of amount per unit time; is the spatial displacement rate per unit time; Set the initial risk threshold to K0; If R0>K0, it is preliminarily judged that the loan has hidden risks of risk linkage; If R0≤K0, it can be preliminarily judged that there is no risk linkage risk in the loan.

2. A loan risk monitoring method based on transaction data analysis according to claim 1, characterized in that: It also includes constructing a multidimensional data interaction matrix, specifically: The calculated differences of each continuous trading pair are arranged into the original data matrix M according to the following formula: Among them, row index i corresponds to the i-th continuous trading pair; column index ΔQ i , Δτ i d i They correspond to the amount difference, time interval, and spatial distance respectively; For the j-th column of the original data matrix M, let: Among them, m j is the minimum value of the jth column of the original data matrix M; M j The maximum value of the jth column of the original data matrix M; when j = 1, it represents the amount difference, when j = 2, it represents the time interval, and when j = 3, it represents the spatial distance; Calculate the normalized value for each element of the original data matrix M, specifically: in, Represents the data in the i-th row and j-th column of the normalized original data matrix M.

3. The loan risk monitoring method based on transaction data analysis according to claim 2, characterized in that: It also includes constructing a normalized interaction matrix vector, specifically: Set the normalized original data matrix to: in, Represents the normalized value of the amount difference of the i-th transaction pair; Represents the normalized value of the time interval of the i-th trading pair; Represents the normalized value of the spatial distance of the i-th trading pair; Let the normalized vector be: in, is the i-th normalized vector.

4. A loan risk monitoring method based on transaction data analysis according to claim 3, characterized in that: It also includes topology construction and node definition, specifically: For each normalized vector Set the corresponding node N i , specifically: All nodes constitute a node set V = {N1, N2, ..., N n-1 }; For any N i ,N j ∈V and i≠j, calculate respectively: Set the total difference between nodes to: Among them, D ij Represents node N i and N j Total differences in three-dimensional normalized data; Represents node N i and N j Differences in the dimension of amount difference; Represents node N i and N j Differences in the time interval dimension; Represents node N i and N j Differences in the spatial distance dimension.

5. The loan risk monitoring method based on transaction data analysis according to claim 4, characterized in that: It also includes establishing edges and topology graphs between nodes, specifically: Set the fixed threshold for edge establishment to δ; For any i≠j, if: D ij <δ, then at node N i With node N j Establish undirected edges between them, and record the edge set as: E={(N i ,N j )|D ij <δ, and i≠j}; Set the topology graph to: G = (V, E); where G is an undirected graph consisting of a node set V and an edge set E.

6. A loan risk monitoring method based on transaction data analysis according to claim 5, characterized in that: It also includes dynamic relationship calculation and interaction intensity modeling, specifically: Set the interaction strength function and for each edge (N i ,N j )∈E calculates the interaction intensity I ij , specifically: Among them, I ij For node N i With N j the intensity of interaction between them; For node N i , set its adjacent node set to Node N i The total interaction intensity I i Set to: The entire transaction sequence is divided into multiple time windows at fixed time intervals, and the length of the time window is ΔT; In any time window, let the number of nodes contained in the window be N w , calculate the arithmetic mean of the total interaction intensity of all nodes in the time window, specifically: Among them, I mean is the average interaction strength of all nodes in the current time window.

7. A loan risk monitoring method based on transaction data analysis according to claim 6, characterized in that: It also includes abnormal interaction pattern recognition and risk linkage triggering, specifically: For node N i In a single time window, set the abnormal interaction degree to: A i =I i -I mean ; Among them, A i For node N i The difference between the total interaction intensity and the average interaction intensity in the time window; Set the abnormality judgment threshold to T A ; When for node N i Satisfy A i ≤T A When , the node is determined to be a normal node; When for node N i Satisfy A i >T A When , the node is determined to be an abnormal node, and the abnormal node set is recorded as In topology graph G, set the abnormal link set to E A , specifically: And (N i ,N j )∈E}; where E A Represents the set of all edges connecting abnormal nodes.

8. The loan risk monitoring method based on transaction data analysis according to claim 7, characterized in that: It also includes risk warning report generation and visualization map construction, specifically: For abnormal node set Each node N in i , record the following data: original transaction number i; Normalized data vector Node total interaction intensity I; abnormal interaction degree A i ; Set the overall risk linkage value R total It is the cumulative sum of abnormal interactions, specifically: And set the warning threshold to T R ; When R is satisfied total >T R When , the entire system is judged to have a high loan risk linkage situation; Construct a two-dimensional graph, where the horizontal axis uses the transaction record number i; the vertical axis uses the total interaction intensity of each node I i ; For each node N i , shown as a solid circle in the figure, the horizontal coordinate of the point is i and the vertical coordinate is I i ; For abnormal nodes Marked with a red solid circle; For each abnormal link (N i ,N j )∈E A , connect the corresponding nodes with straight lines in the figure, the thickness of the connecting line depends on I ij The value is determined by the line type. A larger value will draw a thicker line, and a smaller value will draw a thinner line.

Images