Loan interest rate pricing method based on large language model

Abstract

The invention discloses a loan interest rate pricing method and system based on a large language model. The loan interest rate pricing method comprises the steps that structured financial data and unstructured behavior data of a borrower are acquired and preprocessed; processing the data through the loan risk assessment model subjected to combined fine tuning, and generating a comprehensive credit assessment result; inputting a reinforcement learning dynamic pricing model to output a differentiated interest rate scheme; credit and market interest rate changes are monitored in real time, and re-pricing is triggered when threshold values are exceeded. The system comprises a data acquisition module, a risk assessment module, a dynamic pricing module and a monitoring and re-pricing module. According to the method, risk assessment precision is improved through multi-source data fusion, personalized dynamic pricing is realized through reinforcement learning, real-time monitoring guarantees interest rate and risk matching, data security is ensured through privacy calculation and encryption technologies, and the problems that traditional pricing data is insufficient in utilization and insufficient in personalization and real-time performance are solved.

Description

Technical Field

[0001] This invention relates to the intersection of artificial intelligence and financial technology, specifically to a loan interest rate pricing method based on a large language model. It is particularly applicable to risk assessment, dynamic interest rate generation, and real-time pricing adjustment technologies that integrate structured financial data and unstructured behavioral data, and belongs to the fields of natural language processing, deep learning, and financial risk control technology. Background Technology

[0002] In banking lending, loan interest rate pricing is a core element in achieving risk control and return optimization. Currently, the mainstream pricing methods mainly include cost-plus pricing, market competition pricing, and risk-based pricing. Cost-plus pricing is based on the bank's funding costs, operating costs, and risk costs, plus a fixed profit margin. Market competition pricing tracks the interest rate adjustment strategies of competitors. Risk-based pricing assesses risk levels based on factors such as borrower credit scores. These three methods together constitute the technical framework of traditional loan interest rate pricing.

[0003] However, existing technologies have significant limitations. Traditional models are weak in processing unstructured data and cannot effectively extract sentiment and repayment intention characteristics from texts such as customer service dialogues and loan purpose statements, resulting in a single dimension of risk assessment. Pricing strategies mostly use standardized formulas or rules, making it difficult to generate differentiated solutions based on borrowers' credit dynamics and behavioral preferences, affecting customer experience and market competitiveness. At the same time, dynamic factors such as market interest rate fluctuations and changes in borrowers' credit status are not monitored in real time and fed back to the pricing model, which may lead to a mismatch between interest rates and actual risks, exacerbating credit risk or loss of returns. Furthermore, traditional statistical models have limited ability to capture complex feature interactions and lack generalization performance in big data scenarios.

[0004] With the development of financial technology, single data types and static pricing logic can no longer meet the risk pricing needs in complex economic environments. Therefore, there is an urgent need for an intelligent loan interest rate pricing method that integrates multi-source heterogeneous data, dynamically responds to market and customer changes, has deep learning capabilities, and incorporates artificial intelligence technology. This method can solve problems such as insufficient data utilization, lack of personalization and real-time performance in existing technologies, and improve the accuracy and market adaptability of bank risk pricing. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention proposes a loan interest rate pricing method and system based on a large language model. The technical solution adopted in this invention includes:

[0006] In a first aspect, the present invention provides a loan interest rate pricing method based on a large language model, the method comprising:

[0007] S1. Obtain the borrower's structured financial data and unstructured behavioral data, and perform standardization processing on the data. The standardization processing includes sensitive data desensitization, outlier and missing value processing, unstructured data augmentation, and data standardization.

[0008] S2. The structured financial data and unstructured behavioral data are processed through a loan risk assessment model to generate a comprehensive credit assessment result. The loan risk assessment model adopts a multi-model collaborative processing approach, including feature extraction and multi-model fusion assessment.

[0009] S3. Input the comprehensive credit assessment results and the current market benchmark interest rate into the dynamic pricing model, and output a differentiated interest rate quotation scheme. The dynamic pricing model is based on a reinforcement learning architecture and includes state space, action space and reward function configuration.

[0010] S4. Monitor and trigger the interest rate repricing mechanism in real time. The interest rate repricing mechanism is based on the credit score drift indicator and the preset threshold of the market interest rate fluctuation range. It starts the repricing process and records the adjustment log.

[0011] Furthermore, the sensitive data processing in step S1 includes:

[0012] The ID card number and mobile phone number are encrypted using SHA-256 hash encryption;

[0013] After extracting key entities from text data using a named entity recognition model, they are desensitized using alternative symbols;

[0014] In the summary statistics, a Laplace mechanism with differential privacy is used to add random perturbations, and the privacy budget is controlled within epsilon=1.0.

[0015] Furthermore, the feature extraction in step S2 includes:

[0016] Semantic feature extraction: A BERT-based multi-task learning model is used to complete the sentiment identification, semantic slot extraction and text summary generation of unstructured text, and the TextRank algorithm is combined to mine risk keywords;

[0017] Topological relationship feature extraction: Construct a borrower relationship network and use a Graph Attention Network to generate node topological feature vectors;

[0018] Structured data feature extraction: The XGBoost model is used to process structured financial data and output numerical feature vectors.

[0019] Furthermore, the multi-model fusion evaluation in step S2, employing the Stacking-Ensemble fusion framework, specifically includes:

[0020] Level 1 models include BERT, Graph Attention Network, and XGBoost;

[0021] The secondary model uses LightGBM, which takes the predicted probabilities of the primary model and intermediate layer features as input, and outputs a comprehensive credit assessment result, including risk level and default probability value, through a multi-source feature fusion network.

[0022] Furthermore, the dynamic pricing model configuration in step S3 includes:

[0023] The model structure adopts a combined architecture of DQN + Double Q-learning + Prioritized Experience Replay;

[0024] The state space includes changes in the borrower's credit rating, probability of default, market benchmark interest rate, transaction frequency, and repayment history;

[0025] The action space is a set of discrete interest rate adjustment values;

[0026] The reward function is defined as: actual revenue - default penalty - customer churn penalty, with a segmented incentive mechanism set for customers with different risk levels.

[0027] Furthermore, the calculation of monitoring indicators and repricing triggers in step S4 include:

[0028] Credit Score Drift Monitoring: Calculates the distribution drift index of the borrower's credit score, and triggers an assessment update when PSI > 0.2;

[0029] Market interest rate fluctuation monitoring: Track the fluctuation range of the market benchmark interest rate in real time, and initiate the repricing process when the fluctuation exceeds ±0.5%.

[0030] Optional features also include data security and compliance mechanisms:

[0031] A federated learning framework is used to achieve distributed model training across institutions, with the original data remaining locally.

[0032] Utilize multi-party secure computation to handle cross-agency risk assessment tasks;

[0033] Embed the SHAP value analysis tool in the risk assessment output to visualize the scoring results by feature contribution.

[0034] Secondly, the present invention also provides a loan interest rate pricing system based on a large language model, applying the method described in the first aspect.

[0035] The system includes: a data acquisition module configured to perform standardization processing on the data, the standardization processing including sensitive data desensitization, outlier and missing value processing, unstructured data augmentation, and data standardization;

[0036] The risk assessment module is configured to process the structured financial data and unstructured behavioral data through a loan risk assessment model to generate a comprehensive credit assessment result. The loan risk assessment model adopts a multi-model collaborative processing approach, including feature extraction and multi-model fusion assessment.

[0037] The dynamic pricing module is configured to input the comprehensive credit assessment results and the current market benchmark interest rate into the dynamic pricing model and output a differentiated interest rate quotation scheme. The dynamic pricing model is based on a reinforcement learning architecture and includes state space, action space and reward function configuration.

[0038] The monitoring repricing module is configured to monitor and trigger the interest rate repricing mechanism in real time. The interest rate repricing mechanism starts the repricing process and records the adjustment log based on the credit score drift indicator and the preset threshold of market interest rate fluctuation.

[0039] Thirdly, the present invention also provides an electronic device, including a processor and a memory;

[0040] The memory is used to store operation instructions;

[0041] The processor is configured to implement the steps of the method described in the first aspect by invoking the operation instructions, wherein the processor executes the program.

[0042] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of the method described in the first aspect.

[0043] The beneficial effects of the technical solution provided by this invention are:

[0044] The technical solution provided by this invention trains a loan risk assessment model through a joint fine-tuning strategy, integrating multimodal features of structured financial data and unstructured behavioral data. Combined with semantic modeling and borrower network topology analysis, it achieves accurate credit risk assessment, breaking through the dependence of traditional models on a single data type. The comprehensive credit assessment results and market benchmark interest rates are input into a reinforcement learning dynamic pricing model. A state space is constructed using dynamic credit variables, market indicators, and repayment behavior. Decisions are optimized through a multi-objective reward function to generate differentiated interest rate schemes. Real-time monitoring of credit distribution drift and market fluctuations triggers a repricing mechanism to ensure risk adaptability. Simultaneously, federated learning and privacy-preserving computation technologies ensure data security and compliance. The end-to-end intelligent process significantly improves the accuracy of risk assessment, the level of personalized pricing, and the system's real-time response capability. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in this invention, the accompanying drawings used in the description of this invention will be briefly introduced below.

[0046] Figure 1 This is a schematic diagram of the loan interest rate pricing method based on a large language model as described in an embodiment of the present invention;

[0047] Figure 2 This is a schematic diagram of the loan interest rate pricing system based on a large language model as described in an embodiment of the present invention;

[0048] Figure 3 A schematic diagram of the structure of an exemplary electronic device provided by the present invention. Detailed Implementation

[0049] This application provides a loan interest rate pricing method and system based on a large language model, effectively solving the problems of insufficient utilization of existing loan interest rate data and inadequate personalization and real-time performance. It significantly improves the accuracy of loan interest rate risk assessment, ensures data-driven loan interest rate decision-making, and achieves personalized loan interest rate setting. The following will describe exemplary embodiments of this application in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0050] Example 1

[0051] This embodiment provides a loan interest rate pricing method based on a large language model. (See also...) Figure 1 Specifically, it includes the following steps S100 to S400.

[0052] S100. Obtaining borrowers' structured financial data and unstructured behavioral data.

[0053] This step involves collecting comprehensive data on borrowers and standardizing it to provide high-quality input for subsequent credit assessment and pricing. It includes the following sub-steps:

[0054] S110. Determine the data type and source.

[0055] Specifically, structured financial data includes borrower credit scores (such as FICO), income statements, pay stubs, tax records, balance sheets (net assets, total liabilities), historical loan amounts and repayment records, etc., sourced from the bank's internal database and credit reporting agency API interfaces (such as Baihang Credit and Sesame Credit).

[0056] Unstructured behavioral data includes customer service conversation records, loan purpose statements, loan application text descriptions, and image data (such as scanned copies of ID cards and pay slip photos). After recognizing the text in the images using OCR tools, the data is incorporated into the dataset.

[0057] S120, Data Preprocessing Operation

[0058] Sensitive data processing: Sensitive fields such as ID card numbers and mobile phone numbers are encrypted using SHA-256 hashing; after extracting key entities (such as names and company names) from text data using the Named Entity Recognition (NER) model, they are desensitized using substitution symbols (such as "[USER]");

[0059] In the summary statistics, differential privacy (Laplace mechanism) is used to add random perturbations to control the privacy budget within epsilon=1.0.

[0060] Outlier and missing value handling: For numerical fields (such as income), use the Random Forest algorithm (IsolationForest) to detect outliers (such as values ​​exceeding 4 times the standard deviation); for structured fields, missing values ​​are imputed using Random Forest, and text fields are categorized by whether they contain key information.

[0061] Unstructured data augmentation: Regular expressions are used to clean text data; DocFormer or LayoutLMv3 multimodal document understanding models are introduced for image data to jointly embed visual and linguistic information and improve the structured effect of OCR.

[0062] Standardization processing: Continuous fields (such as credit scores) are normalized using Z-score, while monetary fields are normalized using Min-Max (0-1) standardization.

[0063] S200. Process the structured financial data and unstructured behavioral data through a loan risk assessment model to generate a comprehensive credit assessment result;

[0064] This step uses a multi-model collaborative process to preprocess the data and outputs the borrower's risk level (e.g., A to D) and probability of default. Specifically, it includes the following sub-steps:

[0065] S210, Feature Extraction and Modeling

[0066] Semantic feature extraction: A BERT-based multi-task learning model is adopted to simultaneously identify the sentiment of unstructured text (positive / neutral / negative), extract semantic slots (such as "loan purpose = working capital"), and generate text summaries. The TextRank algorithm is combined to mine risk keywords (such as "tight cash flow") and construct a high-dimensional semantic feature vector.

[0067] Topological relationship feature extraction: Construct a borrower association network (nodes are borrowers, edges are guarantee / joint loan relationships), and use Graph Attention Network (GAT) instead of traditional GraphSAGE to strengthen semantic weights through edge features (such as guarantee amount, joint loan amount) to generate node topological feature vectors.

[0068] Structured data feature extraction: The XGBoost model is used to process structured financial data (such as revenue and debt ratio) and output numerical feature vectors.

[0069] S220, Multi-model Fusion Evaluation

[0070] The Stacking-Ensemble framework is adopted: the first-level model includes BERT (semantic features), GAT (topological features), and XGBoost (structured features); the second-level model adopts LightGBM, which takes the predicted probability of the first-level model and the intermediate layer features as input, and outputs a comprehensive credit assessment result, including risk level (A to D) and default probability value, through a multi-source feature fusion network (introducing a gating mechanism to allocate feature attention weights).

[0071] S300. Input the comprehensive credit assessment results and the current market benchmark interest rate into the dynamic pricing model, and output a differentiated interest rate quotation scheme;

[0072] This step combines the comprehensive credit assessment results with the market benchmark interest rate to generate a personalized interest rate through a reinforcement learning model. Specifically, it includes the following sub-steps:

[0073] S310, Dynamic Pricing Model Configuration

[0074] Model structure: The DQN+Double Q-learning+Prioritized Experience Replay (PER) combined architecture is adopted to improve the stability of policy learning and sample efficiency;

[0075] State space: includes dynamic features such as changes in borrower credit rating, probability of default, market benchmark interest rate, transaction frequency, and repayment history;

[0076] Action space: a set of discrete interest rate adjustment values, such as {-0.3%, 0, +0.3%, +0.5%};

[0077] Reward function: A multi-objective weighted strategy is adopted, defined as "actual revenue - default penalty - customer churn penalty", where actual revenue is positively correlated with loan amount and interest rate, default penalty is calculated based on expected default loss rate, and customer churn penalty is set based on customer's historical contribution; a segmented incentive mechanism is set for customers with different risk levels.

[0078] S320, Interest Rate Calculation and Constraints

[0079] Final interest rate calculation formula: Final interest rate = Market LPR + Model output interest rate adjustment value.

[0080] Adjustment constraints: Introduce interest rate adjustment range limits (adjustment range ≤ ±0.3% when credit rating changes slightly); adjust action space bias based on customer historical acceptance rate data.

[0081] S400, real-time monitoring and triggering of interest rate repricing mechanisms

[0082] This step involves dynamically updating the interest rate plan by monitoring market conditions and changes in borrower credit, and specifically includes the following sub-steps:

[0083] S410, Calculation of Monitoring Indicators

[0084] Credit Score Drift Monitoring: Calculate the distribution drift index (PSI>0.2) of a borrower's credit score to assess the stability of their creditworthiness;

[0085] Market interest rate fluctuation monitoring: Real-time tracking of the fluctuation range of the market benchmark interest rate (LPR) and setting a fluctuation threshold (±0.5%).

[0086] S420, Repricing Triggering and Execution

[0087] When the credit score drift indicator or the fluctuation of market interest rates exceeds the preset threshold, the repricing process is initiated: the loan risk assessment model is called to update the credit assessment results, a new interest rate plan is generated through the dynamic pricing model, and the adjustment log is recorded for audit tracking.

[0088] Data security and compliance mechanisms

[0089] To ensure that the data processing process complies with privacy protection requirements, this method also includes the following security mechanisms:

[0090] Cross-institutional collaborative modeling: Distributed model training is achieved using a federated learning framework (FedAvg + Differential Privacy), with the original data remaining locally.

[0091] Joint Risk Computation: Utilizes multi-party secure computation (MPC) or homomorphic encryption to handle cross-agency risk assessment tasks, avoiding data privacy breaches.

[0092] Explainability assurance: Embedding SHAP value analysis tools into the risk assessment output, visualizing borrower scores by feature contribution, and ensuring that the pricing logic is transparent and traceable.

[0093] This embodiment, through the aforementioned steps, trains a loan risk assessment model using a joint fine-tuning strategy. It integrates multimodal features from structured financial data and unstructured behavioral data, combining semantic modeling and borrower network topology analysis to achieve accurate credit risk assessment, breaking through the dependence of traditional models on a single data type. The comprehensive credit assessment results and market benchmark interest rates are input into a reinforcement learning dynamic pricing model. A state space is constructed using dynamic credit variables, market indicators, and repayment behavior. Decisions are optimized through a multi-objective reward function to generate differentiated interest rate schemes. Real-time monitoring of credit distribution drift and market fluctuations triggers a repricing mechanism to ensure risk adaptability. Simultaneously, federated learning and privacy-preserving computation technologies ensure data security and compliance. This end-to-end intelligent process significantly improves the accuracy of risk assessment, the personalization of pricing, and the system's real-time response capabilities.

[0094] Example 2

[0095] like Figure 2 As shown, this embodiment provides a loan interest rate pricing system based on a large language model. The system includes:

[0096] The data acquisition module 1 is configured to perform standardization processing on the data, including sensitive data desensitization, outlier and missing value processing, unstructured data augmentation, and data standardization.

[0097] Risk assessment module 2 is configured to process the structured financial data and unstructured behavioral data through a loan risk assessment model to generate a comprehensive credit assessment result. The loan risk assessment model adopts a multi-model collaborative processing approach, including feature extraction and multi-model fusion assessment.

[0098] The dynamic pricing module 3 is configured to input the comprehensive credit assessment results and the current market benchmark interest rate into the dynamic pricing model and output a differentiated interest rate quotation scheme. The dynamic pricing model is based on a reinforcement learning architecture and includes state space, action space and reward function configuration.

[0099] The monitoring repricing module 4 is configured to monitor and trigger the interest rate repricing mechanism in real time. The interest rate repricing mechanism starts the repricing process and records the adjustment log based on the credit score drift indicator and the preset threshold of the market interest rate fluctuation range.

[0100] It should be noted that the specific manner in which each module performs its operation in the apparatus described in the above embodiments has been described in detail in the embodiments of the method, and will not be elaborated here.

[0101] Exemplary electronic devices and computer-readable storage media

[0102] Figure 3 A schematic diagram of the structure of an electronic device according to an embodiment of this application is shown.

[0103] The electronic device 300 includes a processor 301 and a memory 303. The processor 301 and the memory 303 are connected, for example, via a bus 302. Optionally, the electronic device 300 may also include a transceiver 304. It should be noted that in practical applications, the transceiver 304 is not limited to one. It is understood that the structure illustrated in the embodiments of the present invention does not constitute a specific limitation on the specific structure of the electronic device 300. In other embodiments of this application, the electronic device 300 may include more or fewer components than illustrated, or combine some components, or split some components, or arrange different components. The illustrated components may be implemented in hardware, software, or a combination of software and hardware. Optionally, the electronic device may also include a display screen 305 for displaying images or receiving user operation commands when needed.

[0104] In this embodiment, processor 301 is used to implement the method shown in the above method embodiment. Transceiver 304 may include a receiver and a transmitter. Transceiver 304 is used in this embodiment to enable the electronic device of this embodiment to communicate with other devices during execution.

[0105] Processor 301 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 301 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0106] Processor 301 may also include one or more processing units, such as an application processor (AP), a modem processor, a graphics processing unit (GPU), an image signal processor (ISP), a controller, a memory, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural network processing unit (NPU). Different processing units can be independent devices or integrated into one or more processors. The controller can be the central nervous system and command center of the electronic device 300. The controller can generate operation control signals based on instruction opcodes and timing signals to control instruction fetching and execution. Processor 301 may also include a memory for storing instructions and data. In some embodiments, the memory in processor 301 is a cache memory. This memory can store instructions or data that processor 301 has just used or is reusing. If processor 301 needs to reuse the instruction or data, it can directly retrieve it from the memory. This avoids repeated access, reduces the waiting time of processor 301, and thus improves system efficiency.

[0107] The processor 301 can run a loan interest rate pricing method based on a large language model provided in the embodiments of this application. The processor 301 may include different devices, such as when integrating a CPU and a GPU, the CPU and GPU can work together to execute the loan interest rate pricing method based on a large language model provided in the embodiments of this application. Some algorithms are executed by the CPU and other algorithms are executed by the GPU to obtain faster processing efficiency.

[0108] Bus 302 may include a pathway for transmitting information between the aforementioned components. Bus 302 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 302 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0109] The memory 303 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory), or high-speed random access memory. It may also include non-volatile memory, such as at least one disk storage device, flash memory device, universal flash storage (UFS), or other optical disc storage, optical disk storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0110] Optionally, the memory 303 is used to store application code that executes the scheme of this application, and the execution is controlled by the processor 301. The processor 301 is used to execute the application code stored in the memory 303 to implement the loan interest rate pricing method based on a large language model provided in any embodiment of this application.

[0111] The memory 303 can be used to store computer executable program code, which includes instructions. The processor 301 executes various functional applications and data processing of the electronic device 300 by running the instructions stored in the memory 303. The memory 303 may include a program storage area and a data storage area. The program storage area can store the operating system, application code, etc. The data storage area can store data created during the use of the electronic device 300 (such as images and videos captured by a camera application).

[0112] The memory 303 may also store one or more computer programs corresponding to the loan interest rate pricing method based on a large language model provided in the embodiments of this application. These one or more computer programs are stored in the memory 303 and configured to be executed by the one or more processors 301. The one or more computer programs include instructions that can be used to perform the various steps in the corresponding embodiments described above.

[0113] Of course, the code for the loan interest rate pricing method based on a large language model provided in this application embodiment can also be stored in external memory.

[0114] The display screen 305 includes a display panel. The display panel can be a liquid crystal display (LCD), an organic light-emitting diode (OLED), an active-matrix organic light-emitting diode (AMOLED), a flexible light-emitting diode (FLED), a minimized LED, a microLED, a quantum dot light-emitting diode (QLED), etc. In some embodiments, the electronic device 300 may include one or N displays 305, where N is a positive integer greater than 1. The display screen 305 can be used to display information input by the user or information provided to the user, as well as various graphical user interfaces (GUIs). For example, the display screen 305 can display photos, videos, web pages, or documents.

[0115] The electronic device provided in this application is applicable to any of the above-described methods. Therefore, the beneficial effects it can achieve can be referred to the beneficial effects in the corresponding methods provided above, and will not be repeated here.

[0116] Embodiments of the present invention also provide a computer-readable storage medium capable of implementing all the steps of the methods in the above embodiments, wherein the computer-readable storage medium stores a computer program that, when executed by a processor, implements all the steps of the methods in the above embodiments.

[0117] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0118] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.

[0119] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including an instruction set implemented in a process. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0120] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1The steps of the functions specified in one or more boxes. Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0121] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A loan interest rate pricing method based on a large language model, characterized in that, Includes the following steps: S1. Obtain the borrower's structured financial data and unstructured behavioral data, and perform standardization processing on the data. The standardization processing includes sensitive data desensitization, outlier and missing value processing, unstructured data augmentation, and data standardization. S2. The structured financial data and unstructured behavioral data are processed through a loan risk assessment model to generate a comprehensive credit assessment result. The loan risk assessment model adopts a multi-model collaborative processing approach, including feature extraction and multi-model fusion assessment. S3. Input the comprehensive credit assessment results and the current market benchmark interest rate into the dynamic pricing model, and output a differentiated interest rate quotation scheme. The dynamic pricing model is based on a reinforcement learning architecture and includes state space, action space and reward function configuration. S4. Monitor and trigger the interest rate repricing mechanism in real time. The interest rate repricing mechanism is based on the credit score drift indicator and the preset threshold of the market interest rate fluctuation range. It starts the repricing process and records the adjustment log.

2. The method according to claim 1, characterized in that, Sensitive data processing in step S1 includes: The ID card number and mobile phone number are encrypted using SHA-256 hash encryption; After extracting key entities from text data using a named entity recognition model, they are desensitized using alternative symbols; In the summary statistics, a Laplace mechanism with differential privacy is used to add random perturbations, and the privacy budget is controlled within epsilon=1.

0.

3. The method according to claim 1, characterized in that, The feature extraction in step S2 includes: Semantic feature extraction: A BERT-based multi-task learning model is used to complete the sentiment identification, semantic slot extraction and text summary generation of unstructured text, and the TextRank algorithm is combined to mine risk keywords; Topological relationship feature extraction: Construct a borrower relationship network and use a Graph Attention Network to generate node topological feature vectors; Structured data feature extraction: The XGBoost model is used to process structured financial data and output numerical feature vectors.

4. The method according to claim 1, characterized in that, The multi-model fusion evaluation in step S2, using the Stacking-Ensemble ensemble framework, specifically includes: Level 1 models include BERT, Graph Attention Network, and XGBoost; The secondary model uses LightGBM, which takes the predicted probabilities of the primary model and intermediate layer features as input, and outputs a comprehensive credit assessment result, including risk level and default probability value, through a multi-source feature fusion network.

5. The method according to claim 1, characterized in that, The dynamic pricing model configuration in step S3 includes: The model structure adopts a combined architecture of DQN + Double Q-learning + Prioritized Experience Replay; The state space includes changes in the borrower's credit rating, probability of default, market benchmark interest rate, transaction frequency, and repayment history; The action space is a set of discrete interest rate adjustment values; The reward function is defined as: actual revenue - default penalty - customer churn penalty, with a segmented incentive mechanism set for customers with different risk levels.

6. The method according to claim 1, characterized in that, Step S4 includes the calculation of monitoring indicators and repricing triggers, which include: Credit Score Drift Monitoring: Calculates the distribution drift index of the borrower's credit score, and triggers an assessment update when PSI > 0.2; Market interest rate fluctuation monitoring: Track the fluctuation range of the market benchmark interest rate in real time, and initiate the repricing process when the fluctuation exceeds ±0.5%.

7. The method according to claim 1, characterized in that, It also includes data security and compliance mechanisms: A federated learning framework is used to achieve distributed model training across institutions, with the original data remaining locally. Utilize multi-party secure computation to handle cross-agency risk assessment tasks; Embed the SHAP value analysis tool in the risk assessment output to visualize the scoring results by feature contribution.

8. A loan interest rate pricing system based on a large language model, using the method described in any one of claims 1-7, characterized in that, include: The data acquisition module is configured to perform standardization processing on the data, including sensitive data desensitization, outlier and missing value handling, unstructured data augmentation, and data standardization. The risk assessment module is configured to process the structured financial data and unstructured behavioral data through a loan risk assessment model to generate a comprehensive credit assessment result. The loan risk assessment model adopts a multi-model collaborative processing approach, including feature extraction and multi-model fusion assessment. The dynamic pricing module is configured to input the comprehensive credit assessment results and the current market benchmark interest rate into the dynamic pricing model and output a differentiated interest rate quotation scheme. The dynamic pricing model is based on a reinforcement learning architecture and includes state space, action space and reward function configuration. The monitoring repricing module is configured to monitor and trigger the interest rate repricing mechanism in real time. The interest rate repricing mechanism is based on a preset threshold of credit score drift indicator and market interest rate fluctuation range, and initiates the repricing process and records the adjustment log.

9. An electronic device, characterized in that, Including processor and memory; The memory is used to store operation instructions; The processor is configured to execute the method of any one of claims 1-7 by invoking the operation instructions.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the method of any one of claims 1-7.

Images