Fund exchange information management system based on big data analysis

Abstract

The invention relates to the field of financial information management, and discloses a fund exchange information management system based on big data analysis, and the system comprises the steps: obtaining multi-source fund exchange data based on a data collection interface, carrying out the data cleaning and standardization processing, and constructing a fund information basic database; extracting fund exchange feature parameters from the fund information basic database through a preset feature extraction rule, and generating a fund feature data set; and based on the fund feature data set, mining a fund exchange association relationship by adopting an association analysis algorithm, and establishing a fund exchange relationship graph. According to the method, the association relationship among the transaction subjects is fully mined by constructing the fund exchange relationship graph, and the risk assessment is performed in combination with the fund exchange characteristic parameters and the relationship graph characteristics, so that the problem of inaccurate risk assessment caused by neglecting the association relationship of the transaction subjects in a traditional mode is solved, and the accuracy of risk level judgment is improved.

Description

Technical Field

[0001] This invention relates to the field of financial information management technology, specifically to a fund transaction information management system based on big data analysis. Background Technology

[0002] In the field of fund transfer information management, with the diversification of transaction scenarios and the explosive growth of transaction data, existing management methods often have limitations in assessing fund risks. Traditional risk assessments often rely on single-dimensional transaction data (such as based solely on the amount or frequency of a single transaction), failing to fully explore the relationships between different transaction entities and lacking in-depth integrated analysis of historical transaction patterns and real-time transaction characteristics. This results in difficulties in accurately identifying potential risks hidden in complex transaction networks, such as abnormal fund transfers between related entities, and insufficient accuracy in risk level determination. Consequently, the effectiveness of subsequent management strategies is affected, and timely and reliable risk prevention and control support for fund transfers cannot be provided. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a fund transfer information management system based on big data analysis, which solves the problems mentioned in the background section.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a method for managing fund transfer information based on big data analysis, comprising:

[0005] Data on fund transactions from multiple sources is acquired through a data acquisition interface, and the data is cleaned and standardized to build a basic database of fund information.

[0006] By using preset feature extraction rules, feature parameters of fund transactions are extracted from the basic database of fund information to generate a fund feature dataset.

[0007] Based on a dataset of fund characteristics, an association analysis algorithm is used to mine the relationships between fund transactions and establish a fund transaction relationship graph.

[0008] Based on the fund transfer relationship map and the preset risk assessment model, the risk level of fund transfer information is determined and the risk assessment results are output.

[0009] Based on the risk assessment results, corresponding fund transfer management strategies are triggered to achieve dynamic management of fund transfer information.

[0010] Preferably, the process of acquiring multi-source fund transfer data based on a data acquisition interface, performing data cleaning and standardization, and constructing a basic database of fund information specifically includes:

[0011] The multi-source fund transfer data includes bank transaction data, third-party payment data, and internal corporate accounting data, which are obtained through API interfaces, direct database connections, and file import methods.

[0012] The acquired multi-source fund transfer data is format-validated, data with incorrect format is removed, missing values ​​are handled using the mean imputation method, and outliers are identified and removed using the Z-score method.

[0013] The processed data is standardized and mapped according to the fields of transaction time, transaction entity, and transaction amount, and the data format and measurement unit are unified to build a basic database of funds information.

[0014] Preferably, by using preset feature extraction rules, feature parameters of fund transactions are extracted from the basic database of fund information to generate a fund feature dataset, specifically including:

[0015] The preset feature extraction rules include time feature rules, amount feature rules, and subject feature rules;

[0016] Based on time-feature rules, the hourly segment, number of days in a week, and monthly cycle characteristics of the transactions are extracted.

[0017] Calculate the single transaction amount, daily cumulative transaction amount, and transaction amount volatility based on the rules of amount characteristics;

[0018] Based on subject feature rules, extract the industry attributes, geographical information, and transaction frequency characteristics of the transaction subjects;

[0019] By integrating the extracted feature parameters, a structured financial feature dataset is generated.

[0020] Preferably, the generation of the structured financial feature dataset specifically includes:

[0021] Construct a feature parameter mapping matrix to map the time-feature rules extracted into a 24-dimensional vector, the number of days in a week into a 7-dimensional vector, and the monthly cycle features into a 12-dimensional vector;

[0022] The single transaction amount and daily cumulative transaction amount calculated by the amount feature rules are logarithmically transformed, and the transaction amount volatility is standardized to generate a 3-dimensional amount feature vector.

[0023] The industry attributes extracted from the main feature rules are one-hot encoded to generate industry feature vectors, the regional information is encoded by administrative divisions to generate regional feature vectors, and the transaction frequency features are binned to generate frequency feature vectors.

[0024] Establish a feature association index table, using the transaction entity ID and transaction timestamp as the joint primary key, and associate the feature vectors of each dimension in the mapping matrix;

[0025] Based on the feature association index table, the feature vectors of each dimension are weighted and concatenated according to preset weights to generate a structured financial feature dataset with unified dimensions.

[0026] Preferably, based on a financial feature dataset, an association analysis algorithm is used to mine the relationships between financial transactions and establish a financial transaction relationship graph, specifically including:

[0027] The Apriori algorithm was used to mine association rules in the financial feature dataset. Minimum support and minimum confidence thresholds were set to filter out frequent financial transaction itemsets that met the threshold conditions.

[0028] Based on frequent itemsets, direct relationships between transaction entities are extracted, and association strength indicators are calculated, including support, confidence, and lift.

[0029] Using the transaction entities as nodes and the strength of association as the edge weights, a directed fund transfer relationship graph is constructed, storing node attribute and edge attribute information.

[0030] Preferably, the minimum support threshold is dynamically determined based on the distribution characteristics of historical transaction data, and the calculation formula is as follows:

[0031] W = μ - k × σ

[0032] Where W is the minimum support threshold, μ is the mean support of the historical transaction itemset, σ is the standard deviation, and k is an adjustable parameter with a value range of [1.5, 3.0].

[0033] The minimum confidence threshold is set according to the business risk preference level, with the threshold range for low-risk businesses being [0.6, 0.7] and the threshold range for high-risk businesses being [0.8, 0.95].

[0034] Preferably, based on the fund transfer relationship graph and a pre-set risk assessment model, the risk level of fund transfer information is determined, and the risk assessment results are output, specifically including:

[0035] The pre-defined risk assessment model includes a feature layer, a model layer, and an output layer. The feature layer takes into account the characteristic parameters of fund transactions and the characteristics of the relationship graph.

[0036] The model layer uses the gradient boosting tree algorithm to build a risk assessment model, and trains the model using historical risk case data to determine the risk weight of each feature;

[0037] Based on the calculation results of the model layer, the output layer classifies the risk of fund transactions into three levels: low, medium, and high, and outputs the corresponding risk assessment results.

[0038] Preferably, the construction of the risk assessment model specifically includes:

[0039] Collect historical fund transfer data and corresponding risk event records, combine fund transfer feature parameters and relationship graph features into sample feature vectors, label the corresponding risk tags, and construct a training dataset;

[0040] Initialize the base model using the mean risk label of all samples in the training dataset as the initial prediction value;

[0041] Follow these steps to iterate until the preset number of iterations is reached:

[0042] Calculate the prediction error of the current model for each sample, and use the error value as the new training target;

[0043] Based on the sample feature vectors and the new training objective, a decision tree is trained, and the node splitting rules and leaf node output values ​​of the tree are determined.

[0044] The newly trained decision tree is weighted and fused with the previous model to update the model's prediction results;

[0045] The model performance was evaluated using 5-fold cross-validation.

[0046] The training dataset is divided into 5 equal parts, and 4 of them are selected as training subsets, with the remaining 1 part used as validation subset.

[0047] The model is trained on a training subset, and the risk prediction accuracy of the model is verified on a validation subset, with evaluation metrics recorded.

[0048] Repeat the above process until all subsets are used as validation subsets for evaluation, and take the average of the multiple evaluation results as the final performance index of the model.

[0049] Calculate the risk weights for each feature:

[0050] Count the number of times each feature is used in all decision tree node splits;

[0051] Calculate the information gain that a feature brings each time it participates in a split, and sum them up to obtain the total gain value of that feature;

[0052] The risk weight of a feature is the proportion of its total gain to the sum of the total gains of all features.

[0053] Pruning and optimizing the model:

[0054] Set the maximum depth of the decision tree, the minimum number of samples in each leaf node, and the minimum information gain required for node splitting;

[0055] Remove branches that do not significantly improve the model's prediction accuracy, retain key decision paths, and reduce the risk of model overfitting.

[0056] Preferably, based on the risk assessment results, corresponding fund transfer management strategies are triggered to achieve dynamic management of fund transfer information, specifically including:

[0057] When the risk assessment result is low risk, the routine monitoring strategy is implemented, and fund transaction reports are generated according to the preset cycle.

[0058] When the risk assessment result is medium risk, an enhanced monitoring strategy will be implemented to track the fund flow of related transaction entities in real time and increase the frequency of data collection.

[0059] When the risk assessment result is high risk, an early warning intervention strategy is implemented, triggering a manual review process to restrict the trading permissions of abnormal trading accounts.

[0060] A big data analytics-based fund transfer information management system includes a data processing module, a feature extraction module, a graph construction module, a risk assessment module, and a dynamic management module, wherein:

[0061] The data processing module is used to acquire multi-source fund transaction data based on the data acquisition interface, perform data cleaning and standardization processing, and build a basic database of fund information.

[0062] The feature extraction module is connected to the data processing module and is used to extract fund transaction feature parameters from the fund information database through preset feature extraction rules to generate a fund feature dataset.

[0063] The graph construction module is connected to the feature extraction module and is used to mine the relationship between funds based on the fund feature dataset and the association analysis algorithm to establish a fund relationship graph.

[0064] The risk assessment module is connected to the graph construction module and is used to assess the risk level of fund transfer information based on the fund transfer relationship graph and the preset risk assessment model, and output the risk assessment results.

[0065] The dynamic management module is connected to the risk assessment module and is used to trigger corresponding fund transfer management strategies based on the risk assessment results, thereby realizing dynamic management of fund transfer information.

[0066] This invention provides a fund transfer information management system based on big data analysis. It has the following beneficial effects:

[0067] By constructing a relationship graph of fund flows, the relationships between trading entities are fully explored, and risk assessment is conducted by combining fund flow characteristic parameters with relationship graph features. This solves the problem of inaccurate risk assessment caused by neglecting the relationships between trading entities in traditional methods, and improves the accuracy of risk level determination.

[0068] By using the gradient boosting tree algorithm to construct a risk assessment model and combining it with historical risk case data to determine the risk weight of each feature, a quantitative assessment of the risk of financial transactions is achieved. Compared with the traditional assessment method that relies on a single data dimension, this further improves the reliability of the risk assessment results.

[0069] By cleaning and standardizing data and generating structured financial feature datasets, the quality of data input into the risk assessment model was ensured, providing a data foundation for accurate risk assessment and indirectly improving the effectiveness of risk prevention and control.

[0070] The dynamic management strategy triggered by the risk assessment results makes the management measures more targeted and can be adjusted in a timely manner according to the actual risk situation, thereby improving the flexibility and responsiveness of fund transfer information management. Attached Figure Description

[0071] Figure 1 A flowchart illustrating a method for managing financial transaction information based on big data analysis;

[0072] Figure 2 This is a framework diagram of a fund transfer information management system based on big data analysis. Detailed Implementation

[0073] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0074] Example:

[0075] Please see the appendix Figure 1 -Appendix Figure 2 This invention provides a method for managing fund transfer information based on big data analysis, comprising:

[0076] Data on fund transactions from multiple sources is acquired through a data acquisition interface, and the data is cleaned and standardized to build a basic database of fund information.

[0077] By using preset feature extraction rules, feature parameters of fund transactions are extracted from the basic database of fund information to generate a fund feature dataset.

[0078] Based on a dataset of fund characteristics, an association analysis algorithm is used to mine the relationships between fund transactions and establish a fund transaction relationship graph.

[0079] Based on the fund transfer relationship map and the preset risk assessment model, the risk level of fund transfer information is determined and the risk assessment results are output.

[0080] Based on the risk assessment results, corresponding fund transfer management strategies are triggered to achieve dynamic management of fund transfer information.

[0081] Specifically, based on multi-source fund transfer data, a basic database is built through data cleaning and standardization. Feature parameters are extracted to generate datasets, correlation analysis is used to mine relationships and establish graphs, and risk assessment models are combined to determine risk levels. Finally, management strategies are triggered based on the results. The whole process forms a closed loop of "data collection - feature extraction - correlation mining - risk assessment - dynamic management". With the help of big data technology, systematic management of the entire fund transfer process is achieved. This realizes integrated processing of fund transfer information from data collection to management strategy execution, avoiding information gaps caused by scattered data processing, improving the integrity and consistency of fund transfer management, and enabling a more comprehensive grasp of the fund transfer situation.

[0082] Based on the data acquisition interface, multi-source fund transaction data is acquired, and data cleaning and standardization processes are performed to construct a basic database of fund information, specifically including:

[0083] The multi-source fund transfer data includes bank transaction data, third-party payment data, and internal corporate accounting data, which are obtained through API interfaces, direct database connections, and file import methods.

[0084] The acquired multi-source fund transfer data is format-validated, data with incorrect format is removed, missing values ​​are handled using the mean imputation method, and outliers are identified and removed using the Z-score method.

[0085] The processed data is standardized and mapped according to the fields of transaction time, transaction entity, and transaction amount, and the data format and measurement unit are unified to build a basic database of funds information.

[0086] Specifically, the data types and acquisition methods of multi-source fund transfer data are clearly defined. The data is cleaned through format validation, missing value handling, and outlier removal. Then, it is standardized and mapped according to unified fields to construct a standardized basic database of fund information. Acquiring data from multiple channels ensures data comprehensiveness, while rigorous data cleaning steps reduce the interference of format errors, missing values, and outliers on subsequent analysis. Standardization ensures consistency across different sources, providing a high-quality data foundation for subsequent feature extraction and risk assessment, and improving data reliability.

[0087] By extracting fund transaction feature parameters from the basic database of fund information using preset feature extraction rules, a fund feature dataset is generated, which specifically includes:

[0088] The preset feature extraction rules include time feature rules, amount feature rules, and subject feature rules;

[0089] Based on time-feature rules, the hourly segment, number of days in a week, and monthly cycle characteristics of the transactions are extracted.

[0090] Calculate the single transaction amount, daily cumulative transaction amount, and transaction amount volatility based on the rules of amount characteristics;

[0091] Based on subject feature rules, extract the industry attributes, geographical information, and transaction frequency characteristics of the transaction subjects;

[0092] By integrating the extracted feature parameters, a structured financial feature dataset is generated.

[0093] Specifically, the classification and extraction of feature parameters can capture key information about the timing, amount, and subject of financial transactions. The structured dataset facilitates subsequent algorithm processing and analysis, making the feature information easier for the model to identify and utilize, thus laying the foundation for accurately uncovering correlations and assessing risks.

[0094] The generated structured financial feature dataset specifically includes:

[0095] Construct a feature parameter mapping matrix to map the time-feature rules extracted into a 24-dimensional vector, the number of days in a week into a 7-dimensional vector, and the monthly cycle features into a 12-dimensional vector;

[0096] The single transaction amount and daily cumulative transaction amount calculated by the amount feature rules are logarithmically transformed, and the transaction amount volatility is standardized to generate a 3-dimensional amount feature vector.

[0097] The industry attributes extracted from the main feature rules are one-hot encoded to generate industry feature vectors, the regional information is encoded by administrative divisions to generate regional feature vectors, and the transaction frequency features are binned to generate frequency feature vectors.

[0098] Establish a feature association index table, using the transaction entity ID and transaction timestamp as the joint primary key, and associate the feature vectors of each dimension in the mapping matrix;

[0099] Based on the feature association index table, the feature vectors of each dimension are weighted and concatenated according to preset weights to generate a structured financial feature dataset with unified dimensions.

[0100] Specifically, vectorization and standardization processes make different types of feature parameters numerically comparable, the associated index table ensures accurate correspondence between features and transaction entities and time, and the weighted concatenation generates a unified dimensional dataset that avoids analytical bias caused by inconsistent feature dimensions, thereby improving the usability of feature data and the accuracy of analysis.

[0101] Based on a dataset of financial characteristics, an association analysis algorithm is used to mine the relationships between financial transactions and establish a financial transaction relationship graph, specifically including:

[0102] The Apriori algorithm was used to mine association rules in the financial feature dataset. Minimum support and minimum confidence thresholds were set to filter out frequent financial transaction itemsets that met the threshold conditions.

[0103] Based on frequent itemsets, direct relationships between transaction entities are extracted, and association strength indicators are calculated, including support, confidence, and lift.

[0104] Using the transaction entities as nodes and the strength of association as the edge weights, a directed fund transfer relationship graph is constructed, storing node attribute and edge attribute information.

[0105] Specifically, the Apriori algorithm can effectively uncover potential relationships between trading entities. The screening of frequent itemsets ensures the salience of these relationships, and the directed graph intuitively displays the relationship status and strength of fund transfers, providing a clear basis for subsequent risk assessment and helping to more accurately identify risks in related transactions.

[0106] The minimum support threshold is dynamically determined based on the distribution characteristics of historical transaction data, and the calculation formula is as follows:

[0107] W = μ - k × σ

[0108] Where W is the minimum support threshold, μ is the mean support of the historical transaction itemset, σ is the standard deviation, and k is an adjustable parameter with a value range of [1.5, 3.0].

[0109] The minimum confidence threshold is set according to the business risk preference level, with the threshold range for low-risk businesses being [0.6, 0.7] and the threshold range for high-risk businesses being [0.8, 0.95].

[0110] Specifically, the dynamically determined minimum support threshold avoids the incompatibility of fixed thresholds with different data distributions, and the tiered minimum confidence threshold meets the needs of different business risk preferences, improving the flexibility and targeting of association rule mining, and the selected frequent itemsets are more in line with actual business scenarios.

[0111] Based on the fund transfer relationship map and the pre-set risk assessment model, the risk level of fund transfer information is determined, and the risk assessment results are output, including:

[0112] The pre-defined risk assessment model includes a feature layer, a model layer, and an output layer. The feature layer takes into account the characteristic parameters of fund transactions and the characteristics of the relationship graph.

[0113] The model layer uses the gradient boosting tree algorithm to build a risk assessment model, and trains the model using historical risk case data to determine the risk weight of each feature;

[0114] Based on the calculation results of the model layer, the output layer classifies the risk of fund transactions into three levels: low, medium, and high, and outputs the corresponding risk assessment results.

[0115] Specifically, multi-feature input allows for a more comprehensive consideration of factors in risk assessment, the gradient boosting tree algorithm can effectively handle complex feature relationships, accurately determine the risk weight of each feature, and the risk level classification makes the assessment results more intuitive, facilitating the formulation and implementation of subsequent management strategies, thus improving the accuracy and practicality of risk assessment.

[0116] The construction of the risk assessment model specifically includes:

[0117] Collect historical fund transfer data and corresponding risk event records, combine fund transfer feature parameters and relationship graph features into sample feature vectors, label the corresponding risk tags, and construct a training dataset;

[0118] Initialize the base model using the mean risk label of all samples in the training dataset as the initial prediction value;

[0119] Follow these steps to iterate until the preset number of iterations is reached:

[0120] Calculate the prediction error of the current model for each sample, and use the error value as the new training target;

[0121] Based on the sample feature vectors and the new training objective, a decision tree is trained, and the node splitting rules and leaf node output values ​​of the tree are determined.

[0122] The newly trained decision tree is weighted and fused with the previous model to update the model's prediction results;

[0123] The model performance was evaluated using 5-fold cross-validation.

[0124] The training dataset is divided into 5 equal parts, and 4 of them are selected as training subsets, with the remaining 1 part used as validation subset.

[0125] The model is trained on a training subset, and the risk prediction accuracy of the model is verified on a validation subset, with evaluation metrics recorded.

[0126] Repeat the above process until all subsets are used as validation subsets for evaluation, and take the average of the multiple evaluation results as the final performance index of the model.

[0127] Calculate the risk weights for each feature:

[0128] Count the number of times each feature is used in all decision tree node splits;

[0129] Calculate the information gain that a feature brings each time it participates in a split, and sum them up to obtain the total gain value of that feature;

[0130] The risk weight of a feature is the proportion of its total gain to the sum of the total gains of all features.

[0131] Pruning and optimizing the model:

[0132] Set the maximum depth of the decision tree, the minimum number of samples in each leaf node, and the minimum information gain required for node splitting;

[0133] Remove branches that do not significantly improve the model's prediction accuracy, retain key decision paths, and reduce the risk of model overfitting.

[0134] Specifically, constructing a reasonable training dataset provides high-quality samples for model training, iterative training and cross-validation improve the model's fitting and generalization abilities, accurate calculation of feature risk weights enables the model to focus on high-risk features, and pruning optimization reduces the risk of overfitting, thus improving the overall reliability and accuracy of the risk assessment model.

[0135] Based on the risk assessment results, corresponding fund transfer management strategies are triggered to achieve dynamic management of fund transfer information, specifically including:

[0136] When the risk assessment result is low risk, the routine monitoring strategy is implemented, and fund transaction reports are generated according to the preset cycle.

[0137] When the risk assessment result is medium risk, an enhanced monitoring strategy will be implemented to track the fund flow of related transaction entities in real time and increase the frequency of data collection.

[0138] When the risk assessment result is high risk, an early warning intervention strategy is implemented, triggering a manual review process to restrict the trading permissions of abnormal trading accounts.

[0139] Specifically, differentiated management strategies avoid the waste of resources or inadequate management caused by adopting a uniform management approach for all fund transactions. Routine monitoring of low-risk transactions reduces management costs, while targeted measures for medium- and high-risk transactions can prevent and control risks in a timely manner, thereby improving the efficiency of fund transaction management and the effectiveness of risk prevention and control.

[0140] A big data analytics-based fund transfer information management system includes a data processing module, a feature extraction module, a graph construction module, a risk assessment module, and a dynamic management module, wherein:

[0141] The data processing module is used to acquire multi-source fund transaction data based on the data acquisition interface, perform data cleaning and standardization processing, and build a basic database of fund information.

[0142] The feature extraction module is connected to the data processing module and is used to extract fund transaction feature parameters from the fund information database through preset feature extraction rules to generate a fund feature dataset.

[0143] The graph construction module is connected to the feature extraction module and is used to mine the relationship between funds based on the fund feature dataset and the association analysis algorithm to establish a fund relationship graph.

[0144] The risk assessment module is connected to the graph construction module and is used to assess the risk level of fund transfer information based on the fund transfer relationship graph and the preset risk assessment model, and output the risk assessment results.

[0145] The dynamic management module is connected to the risk assessment module and is used to trigger corresponding fund transfer management strategies based on the risk assessment results, thereby realizing dynamic management of fund transfer information.

[0146] Specifically, the clear division of labor and collaborative work among the modules ensured the effective implementation of the method. The connections between modules made the data transmission and processing flow smooth, avoiding the inefficiency caused by the dispersion of functions. The system can stably and efficiently manage financial transaction information, improving the convenience and reliability of management.

[0147] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for managing fund transfer information based on big data analysis, characterized in that, include: Data on fund transactions from multiple sources is acquired through a data acquisition interface, and the data is cleaned and standardized to build a basic database of fund information. By using preset feature extraction rules, feature parameters of fund transactions are extracted from the basic database of fund information to generate a fund feature dataset. Based on a dataset of fund characteristics, an association analysis algorithm is used to mine the relationships between fund transactions and establish a fund transaction relationship graph. Based on the fund transfer relationship map and the preset risk assessment model, the risk level of fund transfer information is determined and the risk assessment results are output. Based on the risk assessment results, corresponding fund transfer management strategies are triggered to achieve dynamic management of fund transfer information.

2. The method for managing fund transfer information based on big data analysis according to claim 1, characterized in that, Based on the data acquisition interface, multi-source fund transaction data is acquired, and data cleaning and standardization processes are performed to construct a basic database of fund information, specifically including: The multi-source fund transfer data includes bank transaction data, third-party payment data, and internal corporate accounting data, which are obtained through API interfaces, direct database connections, and file import methods. The acquired multi-source fund transfer data is format-validated, data with incorrect format is removed, missing values ​​are handled using the mean imputation method, and outliers are identified and removed using the Z-score method. The processed data is standardized and mapped according to the fields of transaction time, transaction entity, and transaction amount, and the data format and measurement unit are unified to build a basic database of funds information.

3. The method for managing fund transfer information based on big data analysis according to claim 1, characterized in that, By extracting fund transaction feature parameters from the basic database of fund information using preset feature extraction rules, a fund feature dataset is generated, which specifically includes: The preset feature extraction rules include time feature rules, amount feature rules, and subject feature rules; Based on time-feature rules, the hourly segment, number of days in a week, and monthly cycle characteristics of the transactions are extracted. Calculate the single transaction amount, daily cumulative transaction amount, and transaction amount volatility based on the rules of amount characteristics; Based on subject feature rules, extract the industry attributes, geographical information, and transaction frequency characteristics of the transaction subjects; By integrating the extracted feature parameters, a structured financial feature dataset is generated.

4. The method for managing fund transfer information based on big data analysis according to claim 3, characterized in that, The generated structured financial feature dataset specifically includes: Construct a feature parameter mapping matrix to map the time-feature rules extracted into a 24-dimensional vector, the number of days in a week into a 7-dimensional vector, and the monthly cycle features into a 12-dimensional vector; The single transaction amount and daily cumulative transaction amount calculated by the amount feature rules are logarithmically transformed, and the transaction amount volatility is standardized to generate a 3-dimensional amount feature vector. The industry attributes extracted from the main feature rules are one-hot encoded to generate industry feature vectors, the regional information is encoded by administrative divisions to generate regional feature vectors, and the transaction frequency features are binned to generate frequency feature vectors. Establish a feature association index table, using the transaction entity ID and transaction timestamp as the joint primary key, and associate the feature vectors of each dimension in the mapping matrix; Based on the feature association index table, the feature vectors of each dimension are weighted and concatenated according to preset weights to generate a structured financial feature dataset with unified dimensions.

5. The method for managing fund transfer information based on big data analysis according to claim 1, characterized in that, Based on a dataset of financial characteristics, an association analysis algorithm is used to mine the relationships between financial transactions and establish a financial transaction relationship graph, specifically including: The Apriori algorithm was used to mine association rules in the financial feature dataset. Minimum support and minimum confidence thresholds were set to filter out frequent financial transaction itemsets that met the threshold conditions. Based on frequent itemsets, direct relationships between transaction entities are extracted, and association strength indicators are calculated, including support, confidence, and lift. Using the transaction entities as nodes and the strength of association as the edge weights, a directed fund transfer relationship graph is constructed, storing node attribute and edge attribute information.

6. The method for managing fund transfer information based on big data analysis according to claim 5, characterized in that, The minimum support threshold is dynamically determined based on the distribution characteristics of historical transaction data, and the calculation formula is as follows: W = μ - k × σ Where W is the minimum support threshold, μ is the mean support of the historical transaction itemset, σ is the standard deviation, and k is an adjustable parameter with a value range of [1.5, 3.0]. The minimum confidence threshold is set according to the business risk preference level, with the threshold range for low-risk businesses being [0.6, 0.7] and the threshold range for high-risk businesses being [0.8, 0.95].

7. The method for managing fund transfer information based on big data analysis according to claim 1, characterized in that, Based on the fund transfer relationship map and the pre-set risk assessment model, the risk level of fund transfer information is determined, and the risk assessment results are output, including: The pre-defined risk assessment model includes a feature layer, a model layer, and an output layer. The feature layer takes into account the characteristic parameters of fund transactions and the characteristics of the relationship graph. The model layer uses the gradient boosting tree algorithm to build a risk assessment model, and trains the model using historical risk case data to determine the risk weight of each feature; Based on the calculation results of the model layer, the output layer classifies the risk of fund transactions into three levels: low, medium, and high, and outputs the corresponding risk assessment results.

8. The method for managing fund transfer information based on big data analysis according to claim 7, characterized in that, The construction of the risk assessment model specifically includes: Collect historical fund transfer data and corresponding risk event records, combine fund transfer feature parameters and relationship graph features into sample feature vectors, label the corresponding risk tags, and construct a training dataset; Initialize the base model using the mean risk label of all samples in the training dataset as the initial prediction value; Follow these steps to iterate until the preset number of iterations is reached: Calculate the prediction error of the current model for each sample, and use the error value as the new training target; Based on the sample feature vectors and the new training objective, a decision tree is trained, and the node splitting rules and leaf node output values ​​are determined. The newly trained decision tree is weighted and fused with the previous model to update the model's prediction results; The model performance was evaluated using 5-fold cross-validation. The training dataset is divided into 5 equal parts, and 4 of them are selected as training subsets, with the remaining 1 part used as validation subset. The model is trained on a training subset, and the risk prediction accuracy of the model is verified on a validation subset, with evaluation metrics recorded. Repeat the above process until all subsets are used as validation subsets for evaluation, and take the average of the multiple evaluation results as the final performance index of the model. Calculate the risk weights for each feature: Count the number of times each feature is used in all decision tree node splits; Calculate the information gain that a feature brings each time it participates in a split, and sum them up to obtain the total gain value of that feature; The risk weight of a feature is the proportion of its total gain to the sum of the total gains of all features. Pruning and optimizing the model: Set the maximum depth of the decision tree, the minimum number of samples in each leaf node, and the minimum information gain required for node splitting; Remove branches that do not significantly improve the model's prediction accuracy, retain key decision paths, and reduce the risk of model overfitting.

9. The method for managing fund transfer information based on big data analysis according to claim 1, characterized in that, Based on the risk assessment results, corresponding fund transfer management strategies are triggered to achieve dynamic management of fund transfer information, specifically including: When the risk assessment result is low risk, the routine monitoring strategy is implemented, and fund transaction reports are generated according to the preset cycle. When the risk assessment result is medium risk, an enhanced monitoring strategy will be implemented to track the fund flow of related transaction entities in real time and increase the frequency of data collection. When the risk assessment result is high risk, an early warning intervention strategy is implemented, triggering a manual review process to restrict the trading permissions of abnormal trading accounts.

10. A fund transfer information management system based on big data analysis, characterized in that, A method for managing fund transfer information based on big data analysis, as described in any one of claims 1-9, comprises a data processing module, a feature extraction module, a graph construction module, a risk assessment module, and a dynamic management module, wherein: The data processing module is used to acquire multi-source fund transaction data based on the data acquisition interface, perform data cleaning and standardization processing, and build a basic database of fund information. The feature extraction module is connected to the data processing module and is used to extract fund transaction feature parameters from the fund information database through preset feature extraction rules to generate a fund feature dataset. The graph construction module is connected to the feature extraction module and is used to mine the relationship between funds based on the fund feature dataset and the association analysis algorithm to establish a fund relationship graph. The risk assessment module is connected to the graph construction module and is used to assess the risk level of fund transfer information based on the fund transfer relationship graph and the preset risk assessment model, and output the risk assessment results. The dynamic management module is connected to the risk assessment module and is used to trigger corresponding fund transfer management strategies based on the risk assessment results, thereby realizing dynamic management of fund transfer information.

Images