Multi-source data transaction risk prediction method and system based on space-time diagram convolutional network

Abstract

The invention provides a multi-source data transaction risk prediction method and system based on a space-time diagram convolutional network, and relates to the technical field of data processing, and the method comprises the steps: collecting multi-source heterogeneous transaction data, carrying out the preprocessing, obtaining multi-source transaction data, extracting key features, forming a preliminary feature set, and obtaining multi-view data fusion features; and capturing space-time dynamic characteristics in transaction data by adopting a space-time diagram convolutional network, deploying the trained risk prediction model to a transaction system, inputting and receiving real-time transaction data, predicting the risk level of the transaction, carrying out comparative analysis on an actual risk event and a model prediction result, identifying errors and deviations, and carrying out risk prediction. And updating and optimizing the risk prediction model. By analyzing the real estate market data and the transaction data in real time, the potential real estate transaction risk is accurately predicted, so that investors and managers are helped to formulate a reasonable risk management strategy.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a multi-source data transaction risk prediction method and system based on a spatiotemporal graph convolutional network (ST-GCN). Background Art

[0002] With the continuous development of financial markets and the diversification of trading activities, the prediction and management of trading risks has become one of the core needs to ensure the healthy and stable operation of the market. In traditional trading risk prediction methods, risk analysis usually relies on a single data source, such as transaction amount or transaction frequency. However, with the increase in data types and complexity, a single data source is difficult to fully reflect the multidimensional characteristics of trading behavior, resulting in limited accuracy and timeliness of risk prediction. Therefore, how to integrate multi-source heterogeneous data and extract effective features has become a key issue in improving the effect of trading risk prediction.

[0003] However, existing traditional methods can only analyze a single data source or feature, ignoring the potential relationship and interaction between multi-dimensional data, resulting in insufficient generalization of prediction models in complex trading environments. Therefore, developing an effective method that can process multi-source heterogeneous data and integrate information from different data sources has become an urgent need in trading risk prediction. Summary of the invention

[0004] In view of this, in order to solve the above problems, the purpose of the present invention is to propose a multi-source data transaction risk prediction method and system based on spatiotemporal graph convolutional network, aiming to improve the accuracy and real-time performance of transaction risk prediction by fusing multi-source heterogeneous data and effectively utilizing spatiotemporal features. Specifically, the present invention collects and preprocesses transaction data from different sources, extracts multi-dimensional features, and uses a spatiotemporal graph convolutional network model to capture the spatiotemporal dependencies in transaction behaviors, thereby achieving accurate prediction of transaction risks.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] Based on the above objectives, in a first aspect, the present invention provides a multi-source data transaction risk prediction method based on a spatiotemporal graph convolutional network, comprising the following steps:

[0007] Collect multi-source heterogeneous transaction data and perform preprocessing to obtain multi-source transaction data;

[0008] Extract key features such as transaction time, transaction amount, transaction record and geographic location from multi-source transaction data to form a preliminary feature set;

[0009] Use multimodal learning to perform deep feature fusion on the preliminary feature set of multi-source transaction data, and dynamically adjust the weight of each key feature in the feature fusion process to obtain multi-view data fusion features;

[0010] The spatiotemporal graph convolutional network is used to capture the spatiotemporal dynamic features in transaction data, construct a spatial relationship graph between transactions, extract the spatial dependency of transaction behaviors, perform time series analysis on each transaction node, and identify the time series pattern of transaction behaviors;

[0011] The multi-view data fusion features and spatiotemporal features are input into the deep learning model for training to obtain a risk prediction model, which outputs the risk prediction value of the transaction;

[0012] The trained risk prediction model is deployed to the trading system, real-time trading data is input and received, the risk level of the transaction is predicted, and the actual risk events are compared and analyzed with the model prediction results to identify errors and deviations, and update and optimize the risk prediction model.

[0013] As a further solution of the present invention, the multi-source heterogeneous data deploys the trained risk prediction model into the trading system, inputs the multi-perspective data fusion features and spatiotemporal features into the deep learning model for training, obtains the risk prediction model, and the model outputs the risk prediction value of the transaction.

[0014] As a further solution of the present invention, key features of transaction time, transaction amount, transaction record and geographical location are extracted from multi-source transaction data to form a preliminary feature set, including the following steps:

[0015] Extract time features from timestamp data, extract amount features from transaction amount data, and extract frequency features from transaction records;

[0016] The geographical location of the transaction is encoded as a numerical feature, and the regional density feature is obtained by calculating the transaction density in the area where the transaction is located;

[0017] The extracted time features, amount features, frequency features, and regional density features are transformed, the feature values ​​are converted into standard normal distribution, and the features extracted from different data sources are spliced ​​to form a preliminary feature set.

[0018] As a further solution of the present invention, multimodal learning is used to perform deep feature fusion on the preliminary feature set of multi-source transaction data, and the weight of each key feature in the feature fusion process is dynamically adjusted, including the following steps:

[0019] The features of different sources are represented as multimodal vectors, where the time feature vector is set to T, the amount feature vector is A, and the frequency feature vector is F;

[0020] Concatenate the feature vectors of the preliminary feature set to form a preliminary feature vector X0:

[0021] X0=[T;A;F];

[0022] Initialize the weights of different data sources: the initial weight α of the time feature T , the initial weight of the amount feature α A , the initial weight α of the frequency feature F , satisfying α T +α A +α F =1;

[0023] The weights are dynamically adjusted based on the attention mechanism, where:

[0024]

[0025] In the formula, α i is the weight of the i-th feature, i∈{T,A,F}; w i is the weight vector associated with the i-th feature; w j is the weight vector associated with the j-th feature; X0 is the preliminary feature vector.

[0026] As a further solution of the present invention, obtaining multi-view data fusion features includes the following steps:

[0027] Feature fusion is performed based on the features adjusted by dynamic weights to obtain multi-view data fusion features; the feature fusion formula is:

[0028] X f =α T T+α A A+α F F;

[0029] Where, X f is the fused feature vector; α T is the weight of the time feature; T is the time feature vector; α A is the weight of the amount feature; A is the amount feature vector; α F is the weight of the frequency feature; F is the frequency feature vector; α T T+α A A+α F F represents the weighted sum fusion of time features, amount features and frequency features according to weights to form the final fusion feature vector.

[0030] As a further solution of the present invention, a spatiotemporal graph convolutional network is used to capture the spatiotemporal dynamic features in the transaction data, and the spatiotemporal graph of the transaction behavior is constructed, and the spatiotemporal features are extracted using the spatiotemporal graph convolutional network; wherein the spatiotemporal graph is constructed including the following steps:

[0031] According to the time feature (T), amount feature (A) and frequency feature (F) extracted from the multimodal vector, the nodes of the space-time graph are determined, wherein the transaction records are mapped to nodes (V) in the space-time graph;

[0032] The relationship between the time order and amount similarity between transaction records is mapped as edges (E) in the space-time graph;

[0033] The spatiotemporal graph (G) consists of a set of nodes (V) and a set of edges (E).

[0034] As a further solution of the present invention, the edge set (E) includes time edges, amount edges and frequency edges, wherein:

[0035] Time edge: If transaction i occurs before transaction j, add a time edge e ij ∈E time ;

[0036] Amount edge: If the amounts of transactions i and j are similar, add an amount edge e ij ∈E emount ;

[0037] Frequency edge: If the frequencies of transactions i and j are similar, add a frequency edge e ij ∈E frequency .

[0038] As a further solution of the present invention, spatiotemporal features in multi-source transaction data are extracted based on a spatiotemporal graph convolutional network, comprising the following steps:

[0039] Construct the adjacency matrix, graph convolution layer and temporal convolution layer of the spatiotemporal graph convolutional network;

[0040] Through multi-layer graph convolution and time series convolution, the spatiotemporal features of transaction behavior are extracted, where:

[0041]

[0042] Where, X st is the final extracted spatiotemporal feature matrix, is the temporal feature matrix of the Lth layer; L is the number of layers of the graph convolutional network.

[0043] As a further solution of the present invention, training to obtain a risk prediction model comprises the following steps:

[0044] Extracting spatiotemporal features X based on Spatio-Temporal Graph Convolutional Network (ST-GCN) st , combined with multi-view data to obtain comprehensive features X cd, where the multi-perspective data includes user behavior data, geographic location data, and transaction device data; where:

[0045] X cd =[X st ;X user ;X location ;X device ];

[0046] Where, X st is the spatiotemporal feature matrix; X user is the user behavior feature matrix; X location is the geographic location feature matrix; X device is the transaction equipment feature matrix;

[0047] A risk prediction model is built based on the attention mechanism network to predict the risk classification results. The loss function is selected and the model is trained using gradient descent. The model is verified using a validation set. The trained model is saved to the file system and deployed.

[0048] As a further solution of the present invention, when comparing and analyzing the actual risk events with the model prediction results, a binary classification model is used for calculation, and the prediction result is the risk probability, where:

[0049]

[0050] Where, X real_time It is the feature matrix extracted in real time; is the predicted risk probability;

[0051] Record each actual risk event and the corresponding prediction results, calculate the error and deviation between the model prediction results and the actual risk events, monitor using the recall performance evaluation indicator, regularly evaluate and update the data set, and retrain the model using the updated data set.

[0052] In a second aspect, the present invention provides a multi-source data transaction risk prediction system based on a spatiotemporal graph convolutional network, comprising:

[0053] Data collection module: used to collect multi-source heterogeneous transaction data and perform pre-processing to obtain multi-source transaction data;

[0054] Feature extraction module: used to extract key features from multi-source transaction data to form a preliminary feature set;

[0055] Feature fusion module: used to perform deep feature fusion on the preliminary feature set of multi-source transaction data using multimodal learning, and dynamically adjust the weight of each key feature in the feature fusion process to obtain multi-view data fusion features;

[0056] Spatiotemporal feature extraction module: It is used to capture the spatiotemporal dynamic features in transaction data using a spatiotemporal graph convolutional network, construct a spatial relationship graph between transactions, extract the spatial dependency of transaction behaviors, perform time series analysis on each transaction node, and identify the time series pattern of transaction behaviors;

[0057] Risk prediction module: used to input multi-view data fusion features and spatiotemporal features into the deep learning model for training to obtain a risk prediction model, which outputs the risk prediction value of the transaction;

[0058] Real-time monitoring module: used to deploy the trained risk prediction model into the trading system, input and receive real-time trading data, predict the risk level of the transaction, and compare and analyze the actual risk events with the model prediction results, identify errors and deviations, and update and optimize the risk prediction model.

[0059] In another aspect of the present invention, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, any one of the above-mentioned methods for predicting multi-source data transaction risks based on a spatiotemporal graph convolutional network according to the present invention is executed.

[0060] In another aspect of the present invention, a computer-readable storage medium is provided, storing computer program instructions, which, when executed, implement any of the above-mentioned multi-source data transaction risk prediction methods based on spatiotemporal graph convolutional networks according to the present invention.

[0061] Compared with the prior art, the multi-source data transaction risk prediction method and system based on spatiotemporal graph convolutional network proposed in the present invention has the following beneficial effects:

[0062] 1. Improved risk prediction accuracy: Through the integration of multi-source data and multimodal learning, the system can capture more comprehensive and detailed risk factors, improve the accuracy of risk identification, and significantly improve the accuracy of the risk prediction model through comprehensive data collection, multimodal feature fusion and spatiotemporal feature extraction. The spatiotemporal feature extraction module enables the system to identify the temporal and spatial changes of transaction behavior, further improving the accuracy of risk identification.

[0063] 2. Enhanced real-time response capabilities: By deploying real-time prediction models, the system can process and analyze transaction data in real time, and can quickly identify and respond to potential transaction risks.

[0064] 3. Dynamic adjustment and optimization are achieved: Through continuous monitoring and regular optimization, the model is ensured to adapt to the changing trading environment and maintain stable and reliable forecasting performance.

[0065] 4. Realized comprehensive information utilization: The combination of multi-source data and multimodal learning enables the system to make full use of various information and improve the overall risk prediction effect.

[0066] In summary, the present invention realizes high-precision risk identification and real-time early warning, and comprehensively improves the risk management capabilities of enterprises and financial institutions. The system's multi-source data integration, intelligent feature extraction, continuous optimization and real-time response functions not only improve the accuracy and effectiveness of risk prediction, but also significantly reduce operating costs, improve user trust and the scientific nature of business decisions. In summary, the system has significant beneficial effects in the field of risk management and provides strong support and guarantee for enterprises and financial institutions.

[0067] These and other aspects of the present application will be more concise and understandable in the following description of the embodiments. It should be understood that the above general description and the following detailed description are only exemplary and explanatory and cannot limit the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the following briefly introduces the drawings required for use in the exemplary embodiments or related technical descriptions. The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0069] Figure 1 This is a flowchart of a multi-source data transaction risk prediction method based on a spatiotemporal graph convolutional network according to an embodiment of the present invention.

[0070] Figure 2 This is a flowchart of dynamically adjusting weights in a multi-source data transaction risk prediction method based on a spatiotemporal graph convolutional network in an embodiment of the present invention. DETAILED DESCRIPTION

[0071] Below, the present application is further described in conjunction with the accompanying drawings and specific implementation methods. It should be noted that, under the premise of no conflict, the various embodiments or technical features described below can be arbitrarily combined to form a new embodiment.

[0072] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the embodiments of the present invention are further described in detail below in combination with specific embodiments and with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0073] It should be noted that all expressions using "first" and "second" in the embodiments of the present invention are intended to distinguish two non-identical entities or non-identical parameters with the same name. It can be seen that "first" and "second" are only for the convenience of expression and should not be understood as limitations on the embodiments of the present invention. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, other steps or units inherent to a process, method, system, product or device that includes a series of steps or units.

[0074] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0075] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0076] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0077] In view of the obvious deficiencies of traditional real estate transaction risk management methods in the current complex and ever-changing market environment, such as data lag, strong subjectivity, and difficulty in accurately modeling and analyzing complex relationships, this paper proposes a multi-source data transaction risk prediction method and system based on spatiotemporal graph convolutional networks. By analyzing real estate market data and transaction data in real time and combining advanced machine learning algorithms, it accurately predicts potential real estate transaction risks, thereby helping investors and managers to formulate reasonable risk management strategies.

[0078] See also Figure 1 As shown, an embodiment of the present invention provides a multi-source data transaction risk prediction method based on a spatiotemporal graph convolutional network, the method comprising the following steps:

[0079] Step S10: Collect multi-source heterogeneous transaction data and perform preprocessing to obtain multi-source transaction data.

[0080] In this step, the collected multi-source heterogeneous data mainly deploys the trained risk prediction model into the trading system, inputs the multi-view data fusion features and spatiotemporal features into the deep learning model for training, obtains the risk prediction model, and the model outputs the risk prediction value of the transaction. Then, the collected data is preprocessed, specifically including the following aspects:

[0081] (1) Data cleaning: remove noise, duplicate values, and missing values ​​in the data to ensure data quality.

[0082] (2) Data formatting: unify the format of data from different sources.

[0083] (3) Data synchronization: Time alignment of data from different data sources to ensure that all data have the same time reference.

[0084] Through the above preprocessing, structured multi-source transaction data is obtained.

[0085] Step S20: extract key features of transaction time, transaction amount, transaction record and geographic location from multi-source transaction data to form a preliminary feature set.

[0086] In this step, key features such as transaction time, transaction amount, transaction record, and geographic location are extracted from multi-source transaction data to form a preliminary feature set, including the following steps:

[0087] Extract time features from timestamp data, extract amount features from transaction amount data, and extract frequency features from transaction records;

[0088] The geographical location of the transaction is encoded as a numerical feature, and the regional density feature is obtained by calculating the transaction density in the area where the transaction is located;

[0089] The extracted time features, amount features, frequency features, and regional density features are transformed, the feature values ​​are converted into standard normal distribution, and the features extracted from different data sources are spliced ​​to form a preliminary feature set.

[0090] Among them, the time features extracted from the timestamp data may include the date, time, and frequency of the transaction; the amount features extracted from the transaction amount data may include the size of the transaction amount, the trend of the transaction amount, etc.; the frequency features extracted from the transaction records may include the number of transactions within a period of time, the transaction concentration, etc. Then, the features extracted from different data sources are spliced ​​to form a unified preliminary feature set.

[0091] For example, assume that there is the following real estate transaction data:

[0092] Timestamp data: 2024-10-01 12:00:00, 2024-10-05 14:30:00, 2024-10-10 09:45:00;

[0093] Transaction amount data: 500000, 750000, 600000;

[0094] Transaction records: Transaction 1, Transaction 2, Transaction 3.

[0095] After preprocessing the above real estate transaction data, feature extraction is performed to extract the following features:

[0096] 1. Time characteristics:

[0097] (1) Withdrawal transaction date: 2024-10-01, 2024-10-05, 2024-10-10;

[0098] (2) Withdrawal transaction time: 12:00:00, 14:30:00, 09:45:00;

[0099] (3) Transaction frequency: 3 transactions.

[0100] 2. Amount characteristics:

[0101] (1) Transaction amount: 500,000, 750,000, 600,000;

[0102] (2) Changes in transaction amount: relative change rate ((750,000-500,000) / 500,000, (600,000-750,000) / 750,000).

[0103] 3. Frequency characteristics:

[0104] (1) Number of transactions within a period of time: 3 times;

[0105] (2) Trading concentration: weekly trading frequency.

[0106] Then the extracted eigenvalues ​​are transformed into a standard normal distribution, and the time features, amount features and frequency features are concatenated to form a preliminary feature set.

[0107] The multi-source data transaction risk prediction method based on spatiotemporal graph convolutional network of this embodiment collects multi-source heterogeneous data and performs preprocessing, extracts key features, performs feature transformation and fusion, and forms a preliminary feature set. This method can not only accurately identify risks in transactions, but also provide real-time warning and response, and comprehensively improve risk management capabilities. Through the integration of multi-source data and multimodal learning, the system can capture more comprehensive risk factors, improve the accuracy of risk identification, and provide scientific decision-making support for enterprises and financial institutions.

[0108] Step S30: Use multimodal learning to perform deep feature fusion on the preliminary feature set of multi-source transaction data, and dynamically adjust the weight of each key feature in the feature fusion process to obtain multi-view data fusion features.

[0109] In this step, see Figure 2 As shown in FIG. 1 , multimodal learning is used to perform deep feature fusion on the preliminary feature set of multi-source transaction data, and the weight of each key feature in the feature fusion process is dynamically adjusted, including the following steps:

[0110] Step S301: Initialize a multi-modal vector.

[0111] Represent features from different sources as multimodal vectors:

[0112] The time feature vector is set to T;

[0113] The amount feature vector is set to A;

[0114] The frequency eigenvector is set to F.

[0115] Concatenate the feature vectors of the preliminary feature set to form a preliminary feature vector X0:

[0116] X0=[T;A;F].

[0117] Step S302: Initialize the weights of different data sources.

[0118] Initialize the weights of different data sources:

[0119] The initial weight of the time feature α T ;

[0120] The initial weight of the amount feature α A ;

[0121] Initial weight α of frequency features F , satisfying α T +α A +α F =1.

[0122] Step S303: Dynamically adjust weights based on the attention mechanism.

[0123] The weights are dynamically adjusted based on the attention mechanism, where the weight adjustment formula is as follows:

[0124]

[0125] In the formula, α i is the weight of the i-th feature, i∈{T,A,F}; w i is the weight vector associated with the i-th feature; wj is the weight vector associated with the j-th feature; X0 is the preliminary feature vector.

[0126] Step S304: perform feature fusion based on the features adjusted by dynamic weights.

[0127] Feature fusion is performed based on the features adjusted by dynamic weights. The feature fusion formula is:

[0128] X f =α T T+α A A+α F F;

[0129] Where, X f is the fused feature vector; α T is the weight of the time feature; T is the time feature vector; α A is the weight of the amount feature; A is the amount feature vector; α F is the weight of the frequency feature; F is the frequency feature vector; α T T+α A A+α F F represents the weighted sum fusion of time features, amount features and frequency features according to weights to form the final fusion feature vector.

[0130] In an embodiment of the present invention, comprehensive transaction information is obtained by collecting and preprocessing multi-source heterogeneous data, and representative key features are extracted from different data sources to ensure that the feature set can fully reflect the essence of the transaction. Different features are standardized through feature transformation and spliced ​​together to form a unified feature set to ensure data comparability and consistency. The attention mechanism is used to dynamically adjust the weight of each feature according to the preliminary feature vector to perform feature fusion and obtain multi-view data fusion features. The multi-view data fusion features are used to train a machine learning model for risk prediction.

[0131] For example, assume that there is the following real estate transaction data:

[0132] Timestamp data: 2024-10-01 12:00:00, 2024-10-05 14:30:00, 2024-10-10 09:45:00;

[0133] Transaction amount data: 500000, 750000, 600000;

[0134] Transaction records: Transaction 1, Transaction 2, Transaction 3.

[0135] After preprocessing the above real estate transaction data, feature extraction is performed to extract the following features:

[0136] 1. Time characteristics:

[0137] (1) Withdrawal transaction date: 2024-10-01, 2024-10-05, 2024-10-10;

[0138] (2) Withdrawal transaction time: 12:00:00, 14:30:00, 09:45:00;

[0139] (3) Transaction frequency: 3 transactions.

[0140] 2. Amount characteristics:

[0141] (1) Transaction amount: 500,000, 750,000, 600,000;

[0142] (2) Changes in transaction amount:

[0143] From the first to the second transaction:

[0144] From the second to the third transaction:

[0145] 3. Frequency characteristics:

[0146] (1) Number of transactions within a period of time: 3 times;

[0147] (2) Trading concentration: weekly trading frequency.

[0148] Then the extracted eigenvalues ​​are transformed into a standard normal distribution, and the time features, amount features and frequency features are concatenated to form a preliminary feature set.

[0149] Then, when performing dynamic weight adjustment and feature fusion, assume that the following feature vectors are used to calculate the attention weights:

[0150] Transaction amount: X 金额 =[500000,750000,600000];

[0151] Transaction amount change: X 变化 = [0.5, -0.2];

[0152] Transaction frequency: X 频率 =[3].

[0153] These eigenvectors are then combined into an overall eigenvector X0:

[0154] X0=[500000,750000,600000,0.5,-0.2,3];

[0155] Then, initialize the weight αT , α A , α F , the initial weight vector is:

[0156]

[0157] The initial value is set to 1 / 3, and the weight of each feature is calculated according to the weight adjustment formula, where the amount is normalized and scaled to between [0,1]:

[0158]

[0159] Then, calculate the attention weights:

[0160] exp(0) = 1;

[0161] exp(1)≈2.718;

[0162] exp(0.4)≈1.491;

[0163] exp(0.5)≈1.649;

[0164] exp(-0.2)≈0.8187;

[0165] exp(3)≈20.0855;

[0166] sum:

[0167] ∑ j exp(w j ·X0)=1+2.718+1.491+1.649+0.8187+20.0855≈27.7622;

[0168] Calculate the weights and perform weighted summation on the original features:

[0169] X f =α T T+α A A+α F F=0.036·0+0.054·0.4+0.059·0.5+0.029·(-0.2)+0.724·3≈2.3153.

[0170] The specific value for dynamically adjusting weights and performing feature fusion based on the attention mechanism is approximately 2.3153.

[0171] Step S40: Use a spatiotemporal graph convolutional network to capture the spatiotemporal dynamic features in the transaction data, construct a spatial relationship graph between transactions, extract the spatial dependencies of transaction behaviors, perform time series analysis on each transaction node, and identify the time series pattern of transaction behaviors.

[0172] In this step, a spatiotemporal graph convolutional network is used to capture the spatiotemporal dynamic features in the transaction data, and a spatiotemporal graph of the transaction behavior is constructed, and the spatiotemporal features are extracted using the spatiotemporal graph convolutional network; wherein, the spatiotemporal graph construction includes the following steps:

[0173] According to the time feature (T), amount feature (A) and frequency feature (F) extracted from the multimodal vector, the nodes of the space-time graph are determined, wherein the transaction records are mapped to nodes (V) in the space-time graph;

[0174] The relationship between the time order and amount similarity between transaction records is mapped as edges (E) in the space-time graph;

[0175] The spatiotemporal graph (G) consists of a set of nodes (V) and a set of edges (E).

[0176] The edge set (E) includes time edge, amount edge and frequency edge, where:

[0177] Time edge: If transaction i occurs before transaction j, add a time edge e ij ∈E time ;

[0178] Amount edge: If the amounts of transactions i and j are similar, add an amount edge e ij ∈E emount ;

[0179] Frequency edge: If the frequencies of transactions i and j are similar, add a frequency edge e ij ∈E frequency .

[0180] In this step, spatiotemporal features in multi-source transaction data are extracted based on the spatiotemporal graph convolutional network, including the following steps:

[0181] Construct the adjacency matrix, graph convolution layer and temporal convolution layer of the spatiotemporal graph convolutional network;

[0182] Through multi-layer graph convolution and time series convolution, the spatiotemporal features of transaction behavior are extracted, where:

[0183]

[0184] Where, X st is the final extracted spatiotemporal feature matrix, is the temporal feature matrix of the Lth layer; L is the number of layers of the graph convolutional network.

[0185] Step S50: Input the multi-view data fusion features and spatiotemporal features into the deep learning model for training to obtain a risk prediction model, which outputs a risk prediction value for the transaction.

[0186] In this step, the risk prediction model is trained, including the following steps:

[0187] Extracting spatiotemporal features X based on Spatio-Temporal Graph Convolutional Network (ST-GCN) st , combined with multi-view data to obtain comprehensive features X cd , where the multi-perspective data includes user behavior data, geographic location data, and transaction device data; where:

[0188] X cd =[X st ;X user ;X location ;X device ];

[0189] Where, X st is the spatiotemporal feature matrix; X user is the user behavior feature matrix; X location is the geographic location feature matrix; X device is the transaction equipment feature matrix;

[0190] A risk prediction model is built based on the attention mechanism network to predict the risk classification results. The loss function is selected and the model is trained using gradient descent. The model is verified using a validation set. The trained model is saved to the file system and deployed.

[0191] Step S60: deploy the trained risk prediction model to the trading system, input and receive real-time trading data, predict the risk level of the transaction, and compare and analyze the actual risk events with the model prediction results, identify errors and deviations, and update and optimize the risk prediction model.

[0192] In this step, when comparing and analyzing the actual risk events with the model prediction results, a binary classification model is used for calculation, and the prediction result is the risk probability, where:

[0193]

[0194] Where, X real_time It is the feature matrix extracted in real time; is the predicted risk probability;

[0195] Record each actual risk event and the corresponding prediction results, calculate the error and deviation between the model prediction results and the actual risk events, monitor using the recall performance evaluation indicator, regularly evaluate and update the data set, and retrain the model using the updated data set.

[0196] The multi-source data transaction risk prediction method based on the spatiotemporal graph convolutional network of the present invention aims to improve the accuracy and real-time performance of transaction risk prediction by constructing a spatiotemporal graph, extracting spatiotemporal features, constructing a risk prediction model, and real-time risk prediction and model optimization. By constructing a spatiotemporal graph, the dynamic changes of transaction behaviors in time and space are captured, and the spatiotemporal features of transaction records are extracted by multi-level convolution using ST-GCN. Combined with multi-perspective data, a risk prediction model based on the attention mechanism is constructed to achieve accurate prediction of transaction risks. After the model is deployed, its prediction accuracy and real-time performance are ensured through real-time monitoring and optimization.

[0197] The present invention captures the spatiotemporal dynamic changes of transaction behaviors and combines multi-perspective data to build an accurate and real-time risk prediction model. This method not only improves the accuracy of transaction risk prediction, but also enhances the ability to capture the complexity of transaction behaviors, and has broad application prospects and practical value.

[0198] It should be noted that the above figures are only schematic illustrations of the processes included in the method according to an exemplary embodiment of the present invention, and are not intended to be limiting. It is easy to understand that the processes shown in the above figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0199] It should be understood that, although described in a certain order, these steps are not necessarily performed in sequence in the above order. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be performed in other orders. Moreover, a part of the steps of the present embodiment may include a plurality of steps or a plurality of stages, and these steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a part of the steps or stages in other steps or other steps.

[0200] In a second aspect of the embodiments of the present invention, the present invention further provides a multi-source data transaction risk prediction system based on a spatiotemporal graph convolutional network, comprising:

[0201] Data collection module: used to collect multi-source heterogeneous transaction data and perform pre-processing to obtain multi-source transaction data;

[0202] Feature extraction module: used to extract key features from multi-source transaction data to form a preliminary feature set;

[0203] Feature fusion module: used to perform deep feature fusion on the preliminary feature set of multi-source transaction data using multimodal learning, and dynamically adjust the weight of each key feature in the feature fusion process to obtain multi-view data fusion features;

[0204] Spatiotemporal feature extraction module: It is used to capture the spatiotemporal dynamic features in transaction data using a spatiotemporal graph convolutional network, construct a spatial relationship graph between transactions, extract the spatial dependency of transaction behaviors, perform time series analysis on each transaction node, and identify the time series pattern of transaction behaviors;

[0205] Risk prediction module: used to input multi-view data fusion features and spatiotemporal features into the deep learning model for training to obtain a risk prediction model, which outputs the risk prediction value of the transaction;

[0206] Real-time monitoring module: used to deploy the trained risk prediction model into the trading system, input and receive real-time trading data, predict the risk level of the transaction, and compare and analyze the actual risk events with the model prediction results, identify errors and deviations, and update and optimize the risk prediction model.

[0207] Through the above detailed steps and examples, the multi-source data transaction risk prediction system based on the spatiotemporal graph convolutional network of the present invention is used to execute the steps of the multi-source data transaction risk prediction method based on the spatiotemporal graph convolutional network in the above embodiment, which will not be repeated here. The multi-source data transaction risk prediction system based on the spatiotemporal graph convolutional network of the present invention integrates multi-source data and multimodal learning. The system can capture more comprehensive and detailed risk factors and improve the accuracy of risk identification. It also significantly improves the accuracy of the risk prediction model through comprehensive data collection, multimodal feature fusion and spatiotemporal feature extraction. The spatiotemporal feature extraction module enables the system to identify the temporal and spatial changes of transaction behavior, further improving the accuracy of risk identification. The real-time prediction model is deployed, and the system can process and analyze transaction data in real time, and can quickly identify and respond to potential transaction risks. Through continuous monitoring and regular optimization, it ensures that the model can adapt to the changing transaction environment and maintain the stability and reliability of the prediction performance.

[0208] In summary, the present invention realizes high-precision risk identification and real-time early warning, and comprehensively improves the risk management capabilities of enterprises and financial institutions. The system's multi-source data integration, intelligent feature extraction, continuous optimization and real-time response functions not only improve the accuracy and effectiveness of risk prediction, but also significantly reduce operating costs, improve user trust and the scientific nature of business decisions. In summary, the system has significant beneficial effects in the field of risk management and provides strong support and guarantee for enterprises and financial institutions.

[0209] According to a third aspect of an embodiment of the present invention, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the method of any one of the above embodiments is implemented.

[0210] The computer device includes a processor and a memory, and may also include: an input system and an output system. The processor, memory, input system and output system may be connected via a bus or other means, and the input system may receive input digital or character information, and generate signal input related to the migration of multi-source data transaction risk prediction based on spatiotemporal graph convolutional network. The output system may include display devices such as display screens.

[0211] As a non-volatile computer-readable storage medium, the memory can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as the program instructions / modules corresponding to the multi-source data transaction risk prediction method based on the spatiotemporal graph convolution network in the embodiment of the present application. The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created by the use of the multi-source data transaction risk prediction method based on the spatiotemporal graph convolution network, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the local module via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0212] The processor may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments. The processor is generally used to control the overall operation of the computer device. In this embodiment, the processor is used to run the program code stored in the memory or process data. The processors of multiple computer devices of the computer device of this embodiment execute various functional applications and data processing of the server by running non-volatile software programs, instructions and modules stored in the memory, that is, implementing the steps of the multi-source data transaction risk prediction method based on the spatiotemporal graph convolutional network of the above method embodiment.

[0213] It should be understood that, to the extent that they do not conflict with each other, all the embodiments, features and advantages described above for the multi-source data transaction risk prediction method based on the spatiotemporal graph convolutional network according to the present invention are also applicable to the multi-source data transaction risk prediction and storage medium based on the spatiotemporal graph convolutional network according to the present invention.

[0214] It will also be appreciated by those skilled in the art that various exemplary logic blocks, modules, circuits and algorithm steps described in conjunction with the disclosure herein can be implemented as electronic hardware, computer software or a combination of the two. In order to clearly illustrate this interchangeability of hardware and software, a general description has been given to the functions of various schematic components, blocks, modules, circuits and steps. Whether this function is implemented as software or hardware depends on specific applications and the design constraints imposed on the entire system. Those skilled in the art can implement the function in various ways for each specific application, but this implementation decision should not be interpreted as causing a departure from the disclosed scope of the embodiments of the present invention.

[0215] Finally, it should be noted that the computer-readable storage medium (e.g., memory) herein may be a volatile memory or a non-volatile memory, or may include both a volatile memory and a non-volatile memory. As an example and not by way of limitation, a non-volatile memory may include a read-only memory (ROM), a programmable ROM (PROM), an electrically programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), or a flash memory. A volatile memory may include a random access memory (RAM), which may act as an external cache memory. As an example and not by way of limitation, RAM may be obtained in a variety of forms, such as synchronous RAM (DRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct Rambus RAM (DRRAM). The storage devices of the disclosed aspects are intended to include, but are not limited to, these and other suitable types of memory.

[0216] The various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure herein may be implemented or executed using the following components designed to perform the functions herein: a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination of these components. A general purpose processor may be a microprocessor, but alternatively, the processor may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, for example, a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP, and / or any other such configuration.

[0217] The above is an exemplary embodiment disclosed by the present invention, but it should be noted that, without departing from the scope disclosed by the embodiment of the present invention, various changes and modifications may be made. The functions, steps and / or actions of the method according to the disclosed embodiment described herein do not need to be performed in any particular order. In addition, although the elements disclosed by the embodiment of the present invention may be described or required in individual form, it may also be understood as multiple unless explicitly limited to the singular.

[0218] It should be understood that, as used herein, the singular form "a" or "an" is intended to include the plural form as well, unless the context clearly supports an exception. It should also be understood that, as used herein, "and / or" refers to any and all possible combinations of one or more of the items listed in association. The serial numbers of the embodiments disclosed in the above embodiments of the present invention are for description only and do not represent the advantages and disadvantages of the embodiments.

[0219] A person skilled in the art should understand that the discussion of any of the above embodiments is only exemplary and is not intended to imply that the scope of the disclosure of the embodiments of the present invention is limited to these examples; under the idea of ​​the embodiments of the present invention, the technical features in the above embodiments or different embodiments can also be combined, and there are many other changes in different aspects of the above embodiments of the present invention, which are not provided in detail for the sake of simplicity. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present invention should be included in the protection scope of the embodiments of the present invention.

Claims

1. A multi-source data transaction risk prediction method based on spatiotemporal graph convolutional network, characterized in that: The method comprises the following steps: Collect multi-source heterogeneous transaction data and perform preprocessing to obtain multi-source transaction data; Extract key features such as transaction time, transaction amount, transaction record and geographic location from multi-source transaction data to form a preliminary feature set; Use multimodal learning to perform deep feature fusion on the preliminary feature set of multi-source transaction data, and dynamically adjust the weight of each key feature in the feature fusion process to obtain multi-view data fusion features; The spatiotemporal graph convolutional network is used to capture the spatiotemporal dynamic features in transaction data, construct a spatial relationship graph between transactions, extract the spatial dependency of transaction behaviors, perform time series analysis on each transaction node, and identify the time series pattern of transaction behaviors; The multi-view data fusion features and spatiotemporal features are input into the deep learning model for training to obtain a risk prediction model, which outputs the risk prediction value of the transaction; The trained risk prediction model is deployed to the trading system, real-time trading data is input and received, the risk level of the transaction is predicted, and the actual risk events are compared and analyzed with the model prediction results to identify errors and deviations, and update and optimize the risk prediction model.

2. The multi-source data transaction risk prediction method based on spatiotemporal graph convolutional network according to claim 1, characterized in that: The multi-source heterogeneous data deploys the trained risk prediction model into the trading system, inputs the multi-view data fusion features and spatiotemporal features into the deep learning model for training, and obtains the risk prediction model, which outputs the risk prediction value of the transaction.

3. The multi-source data transaction risk prediction method based on spatiotemporal graph convolutional network according to claim 2 is characterized in that: Extract key features such as transaction time, transaction amount, transaction record, and geographic location from multi-source transaction data to form a preliminary feature set, including the following steps: Extract time features from timestamp data, extract amount features from transaction amount data, and extract frequency features from transaction records; The geographical location of the transaction is encoded as a numerical feature, and the regional density feature is obtained by calculating the transaction density in the area where the transaction is located; The extracted time features, amount features, frequency features, and regional density features are transformed, the feature values ​​are converted into standard normal distribution, and the features extracted from different data sources are spliced ​​to form a preliminary feature set.

4. The multi-source data transaction risk prediction method based on spatiotemporal graph convolutional network as claimed in claim 3, characterized in that: Use multimodal learning to perform deep feature fusion on the preliminary feature set of multi-source transaction data, and dynamically adjust the weight of each key feature in the feature fusion process, including the following steps: The features of different sources are represented as multimodal vectors, where the time feature vector is set to T, the amount feature vector is A, and the frequency feature vector is F; Concatenate the feature vectors of the preliminary feature set to form a preliminary feature vector X0: X0=[T;A;F]; Initialize the weights of different data sources: the initial weight α of the time feature T , the initial weight of the amount feature α A , the initial weight α of the frequency feature F , satisfying α T +α A +α F =1; The weights are dynamically adjusted based on the attention mechanism, where: In the formula, α i is the weight of the i-th feature, i∈{T,A,F}; w i is the weight vector associated with the i-th feature; w j is the weight vector associated with the j-th feature; X0 is the preliminary feature vector.

5. The multi-source data transaction risk prediction method based on spatiotemporal graph convolutional network according to claim 4, characterized in that: Obtaining multi-view data fusion features includes the following steps: Feature fusion is performed based on the features adjusted by dynamic weights to obtain multi-view data fusion features; the feature fusion formula is: X f =a T T+a A A+a F F; In the formula, X f is the fused feature vector; α T is the weight of the time feature; T is the time feature vector; α A is the weight of the amount feature; A is the amount feature vector; α f is the weight of the frequency feature; F is the frequency feature vector; α T T+α A A+α F F represents the weighted sum fusion of time features, amount features and frequency features according to weights to form the final fusion feature vector.

6. The multi-source data transaction risk prediction method based on spatiotemporal graph convolutional network according to claim 5, characterized in that: The spatial-temporal graph convolutional network is used to capture the spatial-temporal dynamic features in the transaction data, and the spatial-temporal graph of the transaction behavior is constructed, and the spatial-temporal graph convolutional network is used to extract the spatial-temporal features; wherein, the construction of the spatial-temporal graph includes the following steps: Determine the nodes of the space-time graph according to the time features, amount features and frequency features extracted from the multimodal vector, wherein the transaction records are mapped as nodes in the space-time graph; The relationships between transaction records in terms of time order and amount similarity are mapped as edges in the space-time graph; A spatiotemporal graph consists of a set of nodes and a set of edges.

7. The multi-source data transaction risk prediction method based on spatiotemporal graph convolutional network according to claim 6, characterized in that: The edge set includes time edge, amount edge and frequency edge, where: Time edge: If transaction i occurs before transaction j, add a time edge e ij ∈E time ; Amount edge: If the amounts of transactions i and j are similar, add an amount edge e ij ∈E emount ; Frequency edge: If the frequencies of transactions i and j are similar, add a frequency edge e ij ∈E frequency .

8. The multi-source data transaction risk prediction method based on spatiotemporal graph convolutional network according to claim 7, characterized in that: Extracting spatiotemporal features from multi-source transaction data based on spatiotemporal graph convolutional networks includes the following steps: Construct the adjacency matrix, graph convolution layer and temporal convolution layer of the spatiotemporal graph convolutional network; Through multi-layer graph convolution and time series convolution, the spatiotemporal features of transaction behavior are extracted, where: In the formula, X st is the final extracted spatiotemporal feature matrix, is the temporal feature matrix of the Lth layer; L is the number of layers of the graph convolutional network.

9. The multi-source data transaction risk prediction method based on spatiotemporal graph convolutional network according to claim 8, characterized in that: The training of the risk prediction model includes the following steps: Extracting spatiotemporal features X based on spatiotemporal graph convolutional network st , combined with multi-view data to obtain comprehensive features X cd , where the multi-perspective data includes user behavior data, geographic location data, and transaction device data; where: X cd =[X st ;X user ;X location ;X device ]; Where, X st is the spatiotemporal feature matrix; X user is the user behavior feature matrix; X location is the geographic location feature matrix; X device is the transaction equipment feature matrix; A risk prediction model is built based on the attention mechanism network to predict the risk classification results. The loss function is selected and the model is trained using gradient descent. The model is verified using a validation set. The trained model is saved to the file system and deployed.

10. A multi-source data transaction risk prediction system based on spatiotemporal graph convolutional network, characterized in that: Used to execute the multi-source data transaction risk prediction method based on spatiotemporal graph convolutional network as described in any one of claims 1 to 9, the system comprises: Data collection module: used to collect multi-source heterogeneous transaction data and perform pre-processing to obtain multi-source transaction data; Feature extraction module: used to extract key features from multi-source transaction data to form a preliminary feature set; Feature fusion module: used to perform deep feature fusion on the preliminary feature set of multi-source transaction data using multimodal learning, and dynamically adjust the weight of each key feature in the feature fusion process to obtain multi-view data fusion features; Spatiotemporal feature extraction module: It is used to capture the spatiotemporal dynamic features in transaction data using a spatiotemporal graph convolutional network, construct a spatial relationship graph between transactions, extract the spatial dependency of transaction behaviors, perform time series analysis on each transaction node, and identify the time series pattern of transaction behaviors; Risk prediction module: used to input multi-view data fusion features and spatiotemporal features into the deep learning model for training to obtain a risk prediction model, which outputs the risk prediction value of the transaction; Real-time monitoring module: used to deploy the trained risk prediction model into the trading system, input and receive real-time trading data, predict the risk level of the transaction, and compare and analyze the actual risk events with the model prediction results, identify errors and deviations, and update and optimize the risk prediction model.

Images