Bank loan decision-making method and system based on big data analysis

Abstract

The invention discloses a bank loan decision-making method and system based on big data analysis, and belongs to the technical field of bank loan risk control, and the method comprises the steps: obtaining multi-source heterogeneous data at different data source nodes, carrying out the local preprocessing and local dimension reduction processing of the multi-source heterogeneous data of each data source node, and carrying out the local dimension reduction processing of the multi-source heterogeneous data; training a federated credit local model pre-constructed in each data source node by using the dimensionality-reduced multi-source heterogeneous data, obtaining a loan request of a user, sending a federated query request to a related data source node based on the loan request, selecting each related federated credit local model to process the loan request, obtaining a plurality of local feature results, and sending the local feature results to the user; and fusing the plurality of local feature results by adopting a federal credit global model to obtain a fused feature result, obtaining a real-time macroeconomic index, and obtaining a loan decision based on the real-time macroeconomic and the fused feature result. The method ensures that data cannot be processed locally, ensures data privacy security, and has a risk dynamic adjustment capability.

Description

Technical Field

[0001] The present invention belongs to the technical field of bank loan risk control, and specifically relates to a bank loan decision-making method and system based on big data analysis. Background Art

[0002] Due to the rapid development of online lending products and the gradual saturation of the market, fewer and fewer customers are applying for loans at banks, and data silos are easily generated for bank information. With the development of big data and machine learning technology, internal and external data of banks are often integrated. The internal data of banks include transaction flows and account information, and external data include behavioral data, industrial chain data, social security, and taxation. However, it is currently difficult to effectively integrate the internal and external data of banks in a safe and compliant manner. The existing integration method mainly adopts traditional centralized learning, which is prone to data privacy leakage and legal compliance risks. In addition, the use of traditional models for integration makes it difficult to adapt to market changes, economic cycle fluctuations, and changes in individual risks, and lacks the ability to dynamically adjust risks.

[0003] Therefore, based on the above-mentioned shortcomings, how to provide an effective technical solution to solve the problems existing in existing technologies, such as the difficulty in effectively integrating the internal and external data of banks in a safe and compliant manner, the risk of data privacy leakage, and the lack of the ability to dynamically adjust risks, has become a difficult problem that needs to be solved urgently in existing technologies. Summary of the Invention

[0004] The purpose of the present invention is to provide a bank loan decision-making method and system based on big data analysis to solve the above-mentioned problems existing in the prior art.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a bank loan decision-making method based on big data analysis, comprising: Acquire multi-source heterogeneous data at different data source nodes, and locally pre-process the multi-source heterogeneous data at each data source node to obtain pre-processed multi-source heterogeneous data, wherein the multi-source heterogeneous data includes enterprise data and internal bank data; Performing local dimensionality reduction processing on the pre-processed multi-source heterogeneous data at each data source node to obtain dimensionality-reduced multi-source heterogeneous data, inputting the dimensionality-reduced multi-source heterogeneous data into a pre-built federal credit local model at each data source node for training, and obtaining a trained federal credit local model; Obtaining a user's loan request, the loan request including information authorized by the applicant, sending a federated query request to relevant data source nodes based on the loan request, selecting a trained federal credit local model from each relevant data source node to process the loan request, obtaining a plurality of local feature results, and each relevant data source node uploading the plurality of local feature results to a central processor; The federal credit global model of the central processing unit is used to fuse multiple local feature results to obtain fused feature results, obtain real-time macroeconomic indicators, obtain dynamic risk preference coefficients based on the real-time macroeconomic indicators, obtain the final score based on the fused feature results and the dynamic risk preference coefficient, and output the loan decision based on the final score.

[0006] In one possible design, the multi-source heterogeneous data of each data source node is locally preprocessed to obtain preprocessed multi-source heterogeneous data, including: Perform data cleaning on the multi-source heterogeneous data of each data source node to obtain cleaned multi-source heterogeneous data; Normalize the cleaned multi-source heterogeneous data to obtain standard multi-source heterogeneous data; Feature engineering is performed on standard multi-source heterogeneous data to obtain preprocessed multi-source heterogeneous data.

[0007] In one possible design, feature engineering is performed on standard multi-source heterogeneous data to obtain pre-processed multi-source heterogeneous data, including: Perform feature extraction on standard multi-source heterogeneous data to obtain standard multi-source heterogeneous features; A feature alignment operation is performed on the standard multi-source heterogeneous features to align the features of the same entity in the standard multi-source heterogeneous data to obtain preprocessed multi-source heterogeneous data.

[0008] In one possible design, the preprocessed multi-source heterogeneous data includes structured data, unstructured data, time series data, and graph data; the dimensionality-reduced multi-source heterogeneous data includes reduced-dimensional structured data, reduced-dimensional unstructured data, 32-dimensional time features, and graph data features; performing local dimensionality reduction processing on the preprocessed multi-source heterogeneous data at each data source node to obtain the reduced-dimensional multi-source heterogeneous data includes: Based on the PCA principal component analysis method, the structured data is locally reduced in dimension to obtain the reduced-dimensional structured data; Based on the BERT model and pooling operation, the unstructured data is reduced in dimension to obtain reduced-dimensional unstructured data; The LSTM encoder is used to reduce the dimensionality of the time series data to obtain 32-dimensional time features; A semi-supervised algorithm is used to reduce the dimension of graph data and obtain graph data features.

[0009] In one possible design, a federated query request is sent to relevant data source nodes based on a loan request, including: Obtaining preset dynamic triggering rules, a node type of each data source node, and a data quality assessment result, wherein the preset dynamic triggering rules include a time triggering rule, an event triggering rule, and a data volume triggering rule, and the data quality assessment result is an assessment result of multi-source heterogeneous data in each data source node, so that the central processing unit assigns a dynamic weight to the federal credit local model of each data source node based on the node type of each data source node, the preset dynamic triggering rules, and the data quality assessment result; Select relevant data source nodes based on loan requests and data source nodes with different dynamic weights; Generate a federated query request based on the loan request and send the federated query request to the relevant data source node.

[0010] In one possible design, the trained federal credit local model of each relevant data source node is selected to process the loan request, and multiple local feature results are obtained, including: Extract features from the applicant's authorization information in the loan request to obtain authorization features; Based on the authorization features, the multi-source heterogeneous data of each relevant data source node is matched to obtain multiple local data, and features are extracted from the multiple local data to obtain multiple local features; Performing feature alignment processing on multiple local features to obtain multiple aligned local features; The authorization feature and the multiple aligned local features are respectively input into the trained federal credit local model of each relevant data source node to obtain multiple local feature results.

[0011] In a possible design, the real-time macroeconomic indicators include the PMI index and the unemployment rate; and obtaining the dynamic risk preference coefficient based on the real-time macroeconomic indicators includes: Construct a dynamic risk preference regulator and obtain historical macroeconomic indicators, wherein the historical macroeconomic indicators include historical PMI index and historical unemployment rate; Input historical macroeconomic indicators into the dynamic risk preference regulator for training, and obtain the trained dynamic risk preference regulator; Input the real-time macroeconomic indicators into the trained dynamic risk preference regulator to obtain the dynamic risk preference coefficient.

[0012] In a second aspect, the present invention provides a bank loan decision-making system based on big data analysis, which is used to implement the method described in the first aspect, including: A data acquisition module is used to acquire multi-source heterogeneous data from different data source nodes, and locally pre-process the multi-source heterogeneous data of each data source node to obtain pre-processed multi-source heterogeneous data, wherein the multi-source heterogeneous data includes enterprise data and internal bank data; A model training module is used to perform local dimensionality reduction processing on the pre-processed multi-source heterogeneous data at each data source node to obtain the reduced-dimensional multi-source heterogeneous data, and input the reduced-dimensional multi-source heterogeneous data into a pre-built federal credit local model at each data source node for training to obtain the trained federal credit local model; a local computing module configured to obtain a user's loan request, the loan request including information authorized by the applicant, send a federated query request to a relevant data source node based on the loan request, select a trained federal credit local model from each relevant data source node to process the loan request, and obtain a plurality of local feature results. Each relevant data source node then uploads the plurality of local feature results to a central processing unit; The fusion decision module is used to fuse multiple local feature results using the federal credit global model of the central processor to obtain fusion feature results, obtain real-time macroeconomic indicators, obtain dynamic risk preference coefficients based on the real-time macroeconomic indicators, obtain a final score based on the fusion feature results and the dynamic risk preference coefficient, and output a loan decision based on the final score.

[0013] In a third aspect, the present invention provides a computer device comprising a memory, a processor and a transceiver that are communicatively connected in sequence, wherein the memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the bank loan decision-making method based on big data analysis as described in any one of the above.

[0014] In a fourth aspect, the present invention provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute any one of the bank loan decision-making methods based on big data analysis as described above.

[0015] The beneficial effects of the present invention are as follows: The present invention discloses a bank loan decision-making method and system based on big data analysis, comprising: obtaining multi-source heterogeneous data at different data source nodes, preprocessing the multi-source heterogeneous data locally to obtain preprocessed multi-source heterogeneous data, performing local dimensionality reduction processing on the preprocessed multi-source heterogeneous data at each data source node, using the dimensionality reduced multi-source heterogeneous data to train a pre-built federal credit local model in each data source node to obtain a trained federal credit local model, obtaining a user's loan request, sending a federal query request to a relevant data source node based on the loan request, selecting the trained federal credit local model of each relevant data source node to process the loan request to obtain multiple local feature results, each relevant data source node uploading the multiple local feature results to a central processing unit, using the federal credit global model of the central processing unit to fuse the multiple local feature results to obtain a fused feature structure, obtaining real-time macroeconomic indicators, obtaining a dynamic risk preference coefficient based on the real-time macroeconomic indicators, obtaining a final score based on the fused feature results and the dynamic risk preference coefficient, and outputting a loan decision based on the final score. The present invention processes bank loan request data through a federal dynamic processing mechanism. The central processor's federal credit global model and the federal credit local models of each data source node cooperate with each other to ensure that the processed data does not leave the local area, ensuring data privacy and security. At the same time, it can effectively integrate multi-source heterogeneous data and finally obtain the final score based on the dynamic risk preference coefficient. It has the ability to dynamically adjust risks, reduce the risks of loan decisions, and facilitate application and promotion. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A flowchart of a bank loan decision-making method based on big data analysis provided in the first aspect of this embodiment; Figure 2 This is a module block diagram of a bank loan decision-making system based on big data analysis provided in the second aspect of this embodiment. DETAILED DESCRIPTION

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the present invention will be briefly introduced below in conjunction with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structure of the drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.

[0018] It should be understood that although the terms "first," "second," etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, a first element can be referred to as a second element, and similarly, a second element can be referred to as a first element without departing from the scope of the exemplary embodiments of the present invention.

[0019] It should be understood that the term "and / or" that may appear in this document is merely a description of the association relationship between associated objects, indicating that three relationships may exist. For example, A and / or B may indicate three situations: A exists alone, B exists alone, and A and B exist at the same time. The term " / and" that may appear in this document describes another type of association object relationship, indicating that two relationships may exist. For example, A / and B may indicate two situations: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this document generally indicates that the previous and subsequent associated objects are in an "or" relationship.

[0020] Example: like Figure 1 As shown, a first aspect of this embodiment provides a bank loan decision-making method based on big data analysis, which can be executed by, but is not limited to, a computer device or a virtual machine with certain computing resources, such as a personal computer or a smart phone, or a virtual machine. The bank loan decision-making method based on big data analysis includes, but is not limited to, the following steps: S1. Acquire multi-source heterogeneous data at different data source nodes, perform local preprocessing on the multi-source heterogeneous data at each data source node, and obtain preprocessed multi-source heterogeneous data, wherein the multi-source heterogeneous data includes enterprise data and internal bank data; Among them, internal bank data includes but is not limited to internal bank accounts, transactions and historical loan data, and corporate data includes but is not limited to corporate credit data, tax data, industrial and commercial data, social security data and open market data.

[0021] Specifically, in step S1, the multi-source heterogeneous data of each data source node is locally preprocessed to obtain preprocessed multi-source heterogeneous data, including: S11. Clean the multi-source heterogeneous data of each data source node to obtain cleaned multi-source heterogeneous data; S12. Normalize the cleaned multi-source heterogeneous data to obtain standard multi-source heterogeneous data; Among them, data cleaning includes but is not limited to processing missing values ​​and outliers, normalizing the data, eliminating the impact of data differences due to different dimensions and value ranges, and facilitating subsequent model training.

[0022] S13. Perform feature engineering on the standard multi-source heterogeneous data to obtain preprocessed multi-source heterogeneous data.

[0023] Specifically, in step S13, feature engineering is performed on the standard multi-source heterogeneous data to obtain pre-processed multi-source heterogeneous data, including: S131. Extract features from standard multi-source heterogeneous data to obtain standard multi-source heterogeneous features; S132. Perform a feature alignment operation on the standard multi-source heterogeneous features to align the features of the same entity in the standard multi-source heterogeneous data to obtain preprocessed multi-source heterogeneous data.

[0024] Furthermore, the preprocessed multi-source heterogeneous data includes structured data, unstructured data, time series data and graph data.

[0025] S2. Performing local dimensionality reduction processing on the preprocessed multi-source heterogeneous data at each data source node to obtain reduced-dimensionality multi-source heterogeneous data, and inputting the reduced-dimensionality multi-source heterogeneous data into a pre-built federal credit local model at each data source node for training to obtain a trained federal credit local model; The multi-source heterogeneous data after dimensionality reduction includes dimensionality-reduced structured data, dimensionality-reduced unstructured data, 32-dimensional time features and graph data features.

[0026] Specifically, in step S2, local dimensionality reduction processing is performed on the pre-processed multi-source heterogeneous data at each data source node to obtain the reduced dimensionality multi-source heterogeneous data, including: S21. Perform local dimensionality reduction on the structured data using the principal component analysis (PCA) method to obtain reduced-dimensional structured data. S22. Perform dimensionality reduction on unstructured data using the BERT model (Bidirectional Encoder Representations from Transformers, a pre-trained language model) and pooling operations to obtain reduced-dimensional unstructured data. S23. Use LSTM encoder to reduce the dimensionality of time series data to obtain 32-dimensional time features; S24. Use a semi-supervised algorithm to reduce the dimension of graph data and obtain graph data features.

[0027] Among them, the PCA principal component analysis method transforms the linear relationship in the original feature space into a new feature space through orthogonal transformation to achieve dimensionality reduction processing; optionally, the semi-supervised algorithm includes but is not limited to semi-supervised principal component analysis, semi-supervised linear discriminant analysis and semi-supervised autoencoder. The above algorithms are all existing technologies and will not be elaborated in detail here.

[0028] In one possible design, the federal credit local model is built based on a machine learning model.

[0029] S3. Obtain a user's loan request, the loan request including the applicant's authorization information, send a federated query request to relevant data source nodes based on the loan request, select a trained federal credit local model from each relevant data source node to process the loan request, and obtain multiple local feature results. Each relevant data source node uploads the multiple local feature results to a central processor; Specifically, in step S3, a federated query request is sent to the relevant data source node based on the loan request, including: S31. Obtaining preset dynamic triggering rules, the node type of each data source node, and a data quality assessment result, wherein the preset dynamic triggering rules include a time triggering rule, an event triggering rule, and a data volume triggering rule, and the data quality assessment result is an assessment result of multi-source heterogeneous data in each data source node, so that the central processing unit assigns a dynamic weight to the federal credit local model of each data source node based on the node type of each data source node, the preset dynamic triggering rules, and the data quality assessment result; Furthermore, time triggering rules include but are not limited to setting preset time, such as triggering every year or every month; event triggering rules include but are not limited to setting preset events, triggering when preset events occur; data volume triggering rules include but are not limited to triggering when the data volume reaches a preset data volume.

[0030] Preferably, when the preset dynamic triggering rule is triggered, the central processor updates the dynamic weight of the federal credit local model of each data source node.

[0031] S32. Select the relevant data source node based on the loan request and the data source nodes with different dynamic weights; S33. Generate a federated query request based on the loan request, and send the federated query request to the relevant data source node.

[0032] Specifically, in step S3, the trained federal credit local model of each relevant data source node is selected to process the loan request, and multiple local feature results are obtained, including: S34. Extracting features from the applicant's authorization information in the loan request to obtain authorization features; S35. Matching multi-source heterogeneous data of each relevant data source node based on the authorization feature to obtain multiple local data, extracting features from the multiple local data to obtain multiple local features; S36. Performing feature alignment on the multiple local features to obtain multiple aligned local features; S37. Input the authorization feature and multiple aligned local features into the trained federal credit local model of each relevant data source node respectively to obtain multiple local feature results.

[0033] S4. Use the central processing unit's federal credit global model to fuse multiple local feature results to obtain fused feature results, obtain real-time macroeconomic indicators, obtain dynamic risk preference coefficients based on the real-time macroeconomic indicators, obtain a final score based on the fused feature results and the dynamic risk preference coefficient, and output a loan decision based on the final score.

[0034] Among them, real-time macroeconomic indicators include but are not limited to the PMI index (Purchasing Managers' Index) and the unemployment rate.

[0035] Specifically, in step S4, a dynamic risk preference coefficient is obtained based on real-time macroeconomic indicators, including: S41. Construct a dynamic risk preference regulator and obtain historical macroeconomic indicators, wherein the historical macroeconomic indicators include a historical PMI index and a historical unemployment rate; Among them, the dynamic risk preference regulator is constructed based on the time series feature extraction algorithm.

[0036] S42. Inputting historical macroeconomic indicators into the dynamic risk appetite regulator for training, thereby obtaining a trained dynamic risk appetite regulator; In one possible design, in step S42, inputting historical macroeconomic indicators into the dynamic risk preference regulator for training includes: S421. Use wavelet transform to process historical macroeconomic indicators to obtain trend item data, cycle item data, and event impact residual data; S422. Use the Prophet prediction model to predict the trend item data to obtain the first processed data; S423. Processing the periodic term data using a Fourier fitting model to obtain second processed data; S424. Using the XGBoost classifier to classify the event impact residual data to obtain third processed data; S425. Fuse the first processed data, the second processed data, and the third processed data to obtain fused data, and input the fused data into the dynamic risk preference regulator for training.

[0037] S43. Input the real-time macroeconomic indicators into the trained dynamic risk preference regulator to obtain the dynamic risk preference coefficient.

[0038] Among them, the Prophet prediction model, Fourier fitting model and XGBoost classifier are all existing technologies and will not be elaborated here.

[0039] This embodiment discloses a bank loan decision-making method based on big data analysis. By building a federalized dynamic processing mechanism, it can dynamically adapt to changes in the economic environment and individual risks and has the ability to dynamically adjust according to changes in risks. At the same time, it can efficiently process multi-source heterogeneous data, make the decision-making process transparent, meet customer needs, improve bank risk control personnel's understanding of risk logic, and improve the effectiveness of credit risk assessment for customers who lack traditional credit records, such as small and micro enterprises, new employees, or freelancers.

[0040] like Figure 2 As shown, the second aspect of this embodiment provides a bank loan decision-making system based on big data analysis, including: A data acquisition module is used to acquire multi-source heterogeneous data from different data source nodes, and locally pre-process the multi-source heterogeneous data of each data source node to obtain pre-processed multi-source heterogeneous data, wherein the multi-source heterogeneous data includes enterprise data and internal bank data; A model training module is used to perform local dimensionality reduction processing on the pre-processed multi-source heterogeneous data at each data source node to obtain the reduced-dimensional multi-source heterogeneous data, and input the reduced-dimensional multi-source heterogeneous data into a pre-built federal credit local model at each data source node for training to obtain a trained federal credit local model; a local computing module configured to obtain a user's loan request, the loan request including information authorized by the applicant, send a federated query request to a relevant data source node based on the loan request, select a trained federal credit local model from each relevant data source node to process the loan request, and obtain a plurality of local feature results. Each relevant data source node then uploads the plurality of local feature results to a central processing unit; The fusion decision module is used to fuse multiple local feature results using the federal credit global model of the central processor to obtain fusion feature results, obtain real-time macroeconomic indicators, obtain dynamic risk preference coefficients based on the real-time macroeconomic indicators, obtain a final score based on the fusion feature results and the dynamic risk preference coefficient, and output a loan decision based on the final score.

[0041] The working process, working details and technical effects of the bank loan decision-making system based on big data analysis provided in the second aspect of this embodiment can be found in the bank loan decision-making method based on big data analysis described in the first aspect, and will not be repeated here.

[0042] A third aspect of this embodiment provides a computer device comprising a memory, a processor, and a transceiver communicatively connected in sequence. The memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program to execute the bank loan decision-making method based on big data analysis as described in the first aspect. Specifically, the memory may include, but is not limited to, random-access memory (RAM), read-only memory (ROM), flash memory, first-input first-output (FIFO), and / or first-input last-output (FILO) memory; the processor may include, but is not limited to, a microprocessor from the STM32F105 series. Furthermore, the computer device may include, but is not limited to, a power module, a display screen, and other necessary components.

[0043] The working process, working details and technical effects of the aforementioned computer equipment provided in the third aspect of this embodiment can be found in the bank loan decision-making method based on big data analysis described in the first aspect, and will not be repeated here.

[0044] A fourth aspect of this embodiment provides a computer program product, including a computer program or instructions, which, when executed by a computer, are used to implement the bank loan decision-making method based on big data analysis as described in the first aspect.

[0045] The working process, working details and technical effects of the aforementioned computer program product provided in the fourth aspect of this embodiment can be found in the bank loan decision-making method based on big data analysis as described in the first aspect, and will not be repeated here.

[0046] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included in the scope of protection of the present invention.

Claims

1. A bank loan decision-making method based on big data analysis, characterized in that: include: Acquire multi-source heterogeneous data at different data source nodes, and locally pre-process the multi-source heterogeneous data at each data source node to obtain pre-processed multi-source heterogeneous data, wherein the multi-source heterogeneous data includes enterprise data and internal bank data; Performing local dimensionality reduction processing on the pre-processed multi-source heterogeneous data at each data source node to obtain dimensionality-reduced multi-source heterogeneous data, inputting the dimensionality-reduced multi-source heterogeneous data into a pre-built federal credit local model at each data source node for training, and obtaining a trained federal credit local model; Obtaining a user's loan request, the loan request including information authorized by the applicant, sending a federated query request to relevant data source nodes based on the loan request, selecting a trained federal credit local model from each relevant data source node to process the loan request, obtaining a plurality of local feature results, and each relevant data source node uploading the plurality of local feature results to a central processor; The federal credit global model of the central processing unit is used to fuse multiple local feature results to obtain fused feature results, obtain real-time macroeconomic indicators, obtain dynamic risk preference coefficients based on the real-time macroeconomic indicators, obtain the final score based on the fused feature results and the dynamic risk preference coefficient, and output the loan decision based on the final score.

2. A bank loan decision-making method based on big data analysis according to claim 1, characterized in that: Perform local preprocessing on the multi-source heterogeneous data of each data source node to obtain preprocessed multi-source heterogeneous data, including: Perform data cleaning on the multi-source heterogeneous data of each data source node to obtain cleaned multi-source heterogeneous data; Normalize the cleaned multi-source heterogeneous data to obtain standard multi-source heterogeneous data; Feature engineering is performed on standard multi-source heterogeneous data to obtain preprocessed multi-source heterogeneous data.

3. A bank loan decision-making method based on big data analysis according to claim 2, characterized in that: Perform feature engineering on standard multi-source heterogeneous data to obtain pre-processed multi-source heterogeneous data, including: Perform feature extraction on standard multi-source heterogeneous data to obtain standard multi-source heterogeneous features; A feature alignment operation is performed on the standard multi-source heterogeneous features to align the features of the same entity in the standard multi-source heterogeneous data to obtain preprocessed multi-source heterogeneous data.

4. The bank loan decision-making method based on big data analysis according to claim 1, characterized in that: The pre-processed multi-source heterogeneous data includes structured data, unstructured data, time series data and graph data, and the multi-source heterogeneous data after dimensionality reduction includes reduced-dimensional structured data, reduced-dimensional unstructured data, 32-dimensional time features and graph data features; The method of performing local dimensionality reduction processing on the pre-processed multi-source heterogeneous data at each data source node to obtain the reduced dimensionality multi-source heterogeneous data includes: Based on the PCA principal component analysis method, the structured data is locally reduced in dimension to obtain the reduced-dimensional structured data; Based on the BERT model and pooling operation, the unstructured data is reduced in dimension to obtain reduced-dimensional unstructured data; The LSTM encoder is used to reduce the dimensionality of the time series data to obtain 32-dimensional time features; A semi-supervised algorithm is used to reduce the dimension of graph data and obtain graph data features.

5. The bank loan decision-making method based on big data analysis according to claim 1, characterized in that: Send a federated query request to the relevant data source nodes based on the loan request, including: Obtaining preset dynamic triggering rules, a node type of each data source node, and a data quality assessment result, wherein the preset dynamic triggering rules include a time triggering rule, an event triggering rule, and a data volume triggering rule, and the data quality assessment result is an assessment result of multi-source heterogeneous data in each data source node, so that the central processing unit assigns a dynamic weight to the federal credit local model of each data source node based on the node type of each data source node, the preset dynamic triggering rules, and the data quality assessment result; Select relevant data source nodes based on loan requests and data source nodes with different dynamic weights; Generate a federated query request based on the loan request and send the federated query request to the relevant data source node.

6. The bank loan decision-making method based on big data analysis according to claim 1, characterized in that: The trained federal credit local model of each relevant data source node is selected to process the loan request, and multiple local feature results are obtained, including: Extract features from the applicant's authorization information in the loan request to obtain authorization features; Based on the authorization features, the multi-source heterogeneous data of each relevant data source node is matched to obtain multiple local data, and features are extracted from the multiple local data to obtain multiple local features; Performing feature alignment processing on multiple local features to obtain multiple aligned local features; The authorization feature and the multiple aligned local features are respectively input into the trained federal credit local model of each relevant data source node to obtain multiple local feature results.

7. The bank loan decision-making method based on big data analysis according to claim 1, characterized in that: The real-time macroeconomic indicators include the PMI index and the unemployment rate; the dynamic risk preference coefficient obtained based on the real-time macroeconomic indicators includes: Construct a dynamic risk preference regulator and obtain historical macroeconomic indicators, wherein the historical macroeconomic indicators include historical PMI index and historical unemployment rate; Input historical macroeconomic indicators into the dynamic risk preference regulator for training, and obtain the trained dynamic risk preference regulator; Input the real-time macroeconomic indicators into the trained dynamic risk preference regulator to obtain the dynamic risk preference coefficient.

8. A bank loan decision-making system based on big data analysis, used to implement the method according to any one of claims 1 to 7, characterized in that: include: A data acquisition module is used to acquire multi-source heterogeneous data from different data source nodes, and locally pre-process the multi-source heterogeneous data of each data source node to obtain pre-processed multi-source heterogeneous data, wherein the multi-source heterogeneous data includes enterprise data and internal bank data; A model training module is used to perform local dimensionality reduction processing on the pre-processed multi-source heterogeneous data at each data source node to obtain the reduced-dimensional multi-source heterogeneous data, and input the reduced-dimensional multi-source heterogeneous data into a pre-built federal credit local model at each data source node for training to obtain a trained federal credit local model; a local computing module configured to obtain a user's loan request, the loan request including information authorized by the applicant, send a federated query request to a relevant data source node based on the loan request, select a trained federal credit local model from each relevant data source node to process the loan request, and obtain a plurality of local feature results. Each relevant data source node then uploads the plurality of local feature results to a central processing unit; The fusion decision module is used to fuse multiple local feature results using the federal credit global model of the central processor to obtain fusion feature results, obtain real-time macroeconomic indicators, obtain dynamic risk preference coefficients based on the real-time macroeconomic indicators, obtain a final score based on the fusion feature results and the dynamic risk preference coefficient, and output a loan decision based on the final score.

9. A computer device, characterized in that: It includes a memory, a processor and a transceiver that are communicatively connected in sequence, wherein the memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the bank loan decision-making method based on big data analysis as described in any one of claims 1 to 7.

10. A computer program product comprising a computer program or instructions, characterized in that When executed by a computer, the computer program or the instruction implements the bank loan decision-making method based on big data analysis as described in any one of claims 1 to 7.

Images