Life insurance company underwriting robot and method based on large language model

Abstract

The invention relates to the technical field of insurance underwriting, and provides a life insurance company underwriting robot and method based on a large language model, and the robot comprises a data input module which is used for receiving and processing multi-channel data needed by underwriting; the natural language processing module is used for deeply analyzing the input text through a natural language model and an emotion recognition algorithm; the decision engine is used for carrying out preliminary decision making on the underwriting request of the client according to the result of the natural language processing module and the underwriting rule in combination with a self-adaptive risk scoring system; the interaction module is used for enabling underwriting personnel to interact with the underwriting robot through the interaction module, and auditing and feeding back the preliminary decision provided by the robot; and the learning module continuously optimizes model parameters by continuously monitoring feedback data of underwriting personnel. According to the invention, the underwriting efficiency can be improved, manual errors are reduced, the underwriting process is optimized, and the user experience is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of insurance underwriting, and in particular to an underwriting robot and method for a life insurance company based on a large language model. Background Art

[0002] The underwriting process of traditional life insurance companies mainly relies on manual review. Underwriters manually review customers' medical records, financial information, and risk assessment data to make underwriting decisions. This method has problems such as high manual error rate, low processing efficiency, and lack of decision consistency. It is difficult to meet the market's demand for efficient and accurate underwriting processes. In addition, as the amount and complexity of data increase, traditional manual review methods are difficult to fully utilize these data resources. Summary of the invention

[0003] The present invention provides a life insurance company underwriting robot and method based on a large language model, which can solve the problems of inefficiency and inaccuracy in the existing life insurance underwriting process.

[0004] According to the present invention, a life insurance company underwriting robot based on a large language model includes:

[0005] Data input module: used to receive and process multi-channel data required for underwriting, supporting real-time dynamic update and integrity verification of data;

[0006] Natural language processing module: Through natural language models and sentiment recognition algorithms, it deeply analyzes the input text, extracts key information, identifies potential risk factors, and forms core data for decision-making;

[0007] Decision engine: Makes preliminary decisions on customers’ underwriting requests based on the results of the natural language processing module and underwriting rules, combined with the adaptive risk scoring system;

[0008] Interaction module: Designed as a multimodal interactive interface, it supports text, chart and voice query methods. Underwriters can interact with the underwriting robot through the interaction module and review and provide feedback on the preliminary decisions provided by the robot;

[0009] Learning module: Continuously optimize model parameters by continuously monitoring the feedback data from underwriters.

[0010] Preferably, the multi-channel data includes medical records, financial information, and risk assessment questionnaires.

[0011] Preferably, the natural language models include BERT, RoBERTa, and T5.

[0012] Preferably, the data input module comprises:

[0013] Data collection module: Users submit underwriting applications through the online platform, and the system automatically obtains relevant data;

[0014] Data verification module: The system verifies data integrity, identifies and marks abnormal data or format errors;

[0015] Data encryption and storage module: De-identify and encrypt data to ensure data security and store it in a secure database.

[0016] Preferably, the natural language processing module includes:

[0017] Data cleaning and preprocessing module: denoising the input text data, removing redundant characters, and standardizing the format;

[0018] Semantic analysis module: identifies text semantics through language models and extracts key information such as medical history and income level;

[0019] Emotion recognition module: identifies the emotional tendencies in user texts for contextual adaptation of subsequent interaction modules;

[0020] Data labeling and classification module: classifies and labels key information for use by the decision engine.

[0021] Preferably, the decision engine includes:

[0022] Rule matching and personalized assessment module: perform risk assessment based on preset underwriting rules and user information;

[0023] Adaptive scoring module: uses adaptive algorithms to dynamically score the user's overall risk;

[0024] Decision generation and output module: outputs underwriting, premium underwriting or rejection recommendations and passes them to the interactive module for review by underwriters.

[0025] Preferably, the interaction module includes:

[0026] Query interface module: underwriters input queries through natural language or voice, and the robot responds in real time;

[0027] Data display module: displays decision results in the form of charts and text to help underwriters quickly understand the basis for decision-making;

[0028] Feedback mechanism module: underwriters confirm or modify decisions and send feedback results to the learning module;

[0029] Interaction recording and analysis module: records all interaction data for analysis and model optimization by the learning module.

[0030] Preferably, the learning modules include:

[0031] Data collection and annotation module: collects underwriters’ feedback and final decisions and generates training samples;

[0032] Reinforcement learning and model training module: Retrain the model through reinforcement learning and dynamically adjust the model parameters based on the labeled data;

[0033] Performance evaluation and update module: Regularly evaluate the performance of the model in different scenarios and update the model based on the evaluation results.

[0034] The present invention provides a life insurance company underwriting method based on a large language model, which adopts the above-mentioned life insurance company underwriting robot based on a large language model and includes the following steps:

[0035] 1) Data input;

[0036] 1.1) The customer submits an underwriting application through the online platform, and the system collects the customer's medical records, financial information, risk assessment questionnaire and other data;

[0037] 1.2) The data input module verifies the integrity of the data, ensuring that the format is correct and meets the system requirements. The system automatically de-identifies and encrypts the data and stores it securely in the database;

[0038] 2) Natural language processing;

[0039] 2.1) The system calls the natural language processing module to perform semantic analysis on health records and extract key information;

[0040] 2.2) The module analyzes the customer's emotional tendencies through the emotion recognition function, providing a reference for the communication style of the subsequent interaction module;

[0041] 3) Decision engine processing;

[0042] 3.1) The decision engine provides personalized risk scores for customers’ health status, financial situation, and occupational risks based on natural language processing results and set underwriting rules;

[0043] 3.2) The engine uses adaptive algorithms to adjust underwriting rules, generate preliminary underwriting recommendations, and output them to the interactive module;

[0044] 4) Interaction and feedback;

[0045] 4.1) Underwriters review the preliminary decisions generated by the system through the interactive module, including underwriting recommendations and risk scores;

[0046] 4.2) The underwriter further inquires about the details, and the system displays the information in the form of charts and text;

[0047] 4.3) The underwriter confirms or proposes changes to the robot’s decision, and the system collects and records the feedback;

[0048] 5) Model learning and optimization;

[0049] 5.1) The learning module labels the feedback data from underwriters and generates new training samples;

[0050] 5.2) The system retrains the model through reinforcement learning and continuously optimizes the model parameters.

[0051] The beneficial effects of the present invention are as follows:

[0052] The present invention automatically processes large amounts of data and makes accurate decisions, thereby improving underwriting efficiency, reducing manual errors, optimizing the underwriting process, and enhancing user experience, as follows:

[0053] A) Improved underwriting efficiency: The system automatically processes complex data, enables rapid underwriting, and greatly improves processing efficiency.

[0054] B) Improved decision accuracy: Reduce human errors and improve the consistency of underwriting decisions through advanced natural language processing and deep learning technologies.

[0055] C) User experience optimization: The friendly multimodal interaction design enables underwriters to easily query and provide feedback, improving the overall user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 Schematic diagram of a life insurance company underwriting robot based on a large language model in an embodiment. DETAILED DESCRIPTION

[0057] In order to further understand the content of the present invention, the present invention is described in detail in conjunction with the accompanying drawings and embodiments. It should be understood that the embodiments are only for explaining the present invention and are not intended to limit it.

[0058] Example

[0059] like Figure 1 As shown, this embodiment provides a life insurance company underwriting robot based on a large language model, which includes:

[0060] Data input module: used to receive and process multi-channel data required for underwriting, supporting real-time dynamic update and integrity verification of data; multi-channel data includes medical records, financial information, and risk assessment questionnaires. This module also supports real-time dynamic update and integrity verification of data to ensure data accuracy and timeliness.

[0061] The data input module includes:

[0062] Data collection module: Users submit underwriting applications through the online platform, and the system automatically obtains relevant data;

[0063] Data verification module: The system verifies data integrity, identifies and marks abnormal data or format errors;

[0064] Data encryption and storage module: De-identify and encrypt data to ensure data security and store it in a secure database.

[0065] Natural language processing module: Through natural language models and sentiment recognition algorithms, the input text is deeply analyzed to extract key information, identify potential risk factors, and form core data for decision-making; natural language models include BERT, RoBERTa, and T5. This module specifically adds the function of identifying customer sentiment tendencies to optimize the user interaction experience.

[0066] The natural language processing module includes:

[0067] Data cleaning and preprocessing module: denoising the input text data, removing redundant characters, and standardizing the format;

[0068] Semantic analysis module: identifies text semantics through language models and extracts key information such as medical history and income level;

[0069] Emotion recognition module: Identify emotional tendencies (such as anxiety and worry) in user texts for context adaptation of subsequent interaction modules;

[0070] Data labeling and classification module: classifies and labels key information for use by the decision engine.

[0071] Decision engine: Based on the results of the natural language processing module and the underwriting rules, combined with the adaptive risk scoring system, it makes a preliminary decision on the customer's underwriting request. The engine has learning capabilities and can dynamically optimize the decision logic based on historical data.

[0072] The decision engine includes:

[0073] Rule matching and personalized assessment module: perform risk assessment based on preset underwriting rules and user information (such as occupational risks, health status, etc.);

[0074] Adaptive scoring module: uses adaptive algorithms to dynamically score the user's overall risk;

[0075] Decision generation and output module: outputs underwriting, premium underwriting or rejection recommendations and passes them to the interactive module for review by underwriters.

[0076] Interaction module: Designed as a multi-modal interactive interface, it supports text, chart and voice query methods. Underwriters can interact with the underwriting robot through the interaction module and review and provide feedback on the initial decision made by the robot. The module has the ability to adapt to situations and can adjust the interaction mode according to the user's emotional tendencies.

[0077] The interactive modules include:

[0078] Query interface module: underwriters input queries through natural language or voice, and the robot responds in real time;

[0079] Data display module: displays decision results in the form of charts and text to help underwriters quickly understand the basis for decision-making;

[0080] Feedback mechanism module: underwriters confirm or modify decisions and send feedback results to the learning module;

[0081] Interaction recording and analysis module: records all interaction data for analysis and model optimization by the learning module.

[0082] Learning module: Continuously optimize model parameters by continuously monitoring the feedback data from underwriters. The module uses reinforcement learning and transfer learning techniques to enable the model to quickly adapt to new data scenarios and improve decision accuracy and processing efficiency through continuous learning.

[0083] Learning modules include:

[0084] Data collection and annotation module: collects underwriters’ feedback and final decisions and generates training samples;

[0085] Reinforcement learning and model training module: Retrain the model through reinforcement learning and dynamically adjust the model parameters based on the labeled data;

[0086] Performance evaluation and update module: Regularly evaluate the performance of the model in different scenarios and update the model based on the evaluation results.

[0087] This embodiment provides a life insurance company underwriting method based on a large language model, which adopts the above-mentioned life insurance company underwriting robot based on a large language model and includes the following steps:

[0088] 1) Data input;

[0089] 1.1) The customer submits an underwriting application through the online platform, and the system collects the customer's medical records, financial information, risk assessment questionnaire and other data;

[0090] 1.2) The data input module verifies the integrity of the data, ensuring that the format is correct and meets the system requirements. The system automatically de-identifies and encrypts the data and stores it securely in the database;

[0091] 2) Natural language processing;

[0092] 2.1) The system calls the natural language processing module to perform semantic analysis on health records and extract key information;

[0093] 2.2) The module analyzes the customer's emotional tendencies through the emotion recognition function, providing a reference for the communication style of the subsequent interaction module;

[0094] 3) Decision engine processing;

[0095] 3.1) The decision engine provides personalized risk scores for customers’ health status, financial situation, and occupational risks based on natural language processing results and set underwriting rules;

[0096] 3.2) The engine uses adaptive algorithms to adjust underwriting rules, generate preliminary underwriting recommendations, and output them to the interactive module;

[0097] 4) Interaction and feedback;

[0098] 4.1) Underwriters review the preliminary decisions generated by the system through the interactive module, including underwriting recommendations and risk scores;

[0099] 4.2) The underwriter further inquires about the details, and the system displays the information in the form of charts and text;

[0100] 4.3) The underwriter confirms or proposes changes to the robot’s decision, and the system collects and records the feedback;

[0101] 5) Model learning and optimization;

[0102] 5.1) The learning module labels the feedback data from underwriters and generates new training samples;

[0103] 5.2) The system retrains the model through reinforcement learning and continuously optimizes the model parameters.

[0104] Through the above implementation process, the intelligent underwriting robot of this embodiment can efficiently and accurately process life insurance underwriting requests, significantly reduce manual workload, and improve the automation level of the underwriting process. Compared with the traditional manual review method, the system shortens the underwriting time from several days to several minutes, and the consistency and accuracy of decision-making are also significantly improved.

[0105] The present invention and its embodiments are described schematically above, and the description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. Therefore, if a person skilled in the art is inspired by it and designs a structural method and an embodiment similar to the technical solution without creativity without departing from the purpose of the invention, they shall all fall within the protection scope of the present invention.

Claims

1. A life insurance company underwriting robot based on a large language model, characterized by: Including sequential connection: Data input module: used to receive and process multi-channel data required for underwriting, supporting real-time dynamic update and integrity verification of data; Natural language processing module: Through natural language models and sentiment recognition algorithms, it deeply analyzes the input text, extracts key information, identifies potential risk factors, and forms core data for decision-making; Decision engine: Makes preliminary decisions on customers’ underwriting requests based on the results of the natural language processing module and underwriting rules, combined with the adaptive risk scoring system; Interaction module: Designed as a multimodal interactive interface, it supports text, chart and voice query methods. Underwriters can interact with the underwriting robot through the interaction module and review and provide feedback on the preliminary decisions provided by the robot; Learning module: Continuously optimize model parameters by continuously monitoring the feedback data from underwriters.

2. The life insurance company underwriting robot based on a large language model according to claim 1, characterized in that: Multi-channel data includes medical records, financial information, and risk assessment questionnaires.

3. The life insurance company underwriting robot based on a large language model according to claim 2, characterized in that: Natural language models include BERT, RoBERTa, and T5.

4. The life insurance company underwriting robot based on a large language model according to claim 3, characterized in that: The data input module includes: Data collection module: Users submit underwriting applications through the online platform, and the system automatically obtains relevant data; Data verification module: The system verifies data integrity, identifies and marks abnormal data or format errors; Data encryption and storage module: De-identify and encrypt data to ensure data security and store it in a secure database.

5. The life insurance company underwriting robot based on a large language model according to claim 4, characterized in that: The natural language processing module includes: Data cleaning and preprocessing module: denoising the input text data, removing redundant characters, and standardizing the format; Semantic analysis module: identifies text semantics through language models and extracts key information such as medical history and income level; Emotion recognition module: identifies the emotional tendencies in user texts for contextual adaptation of subsequent interaction modules; Data labeling and classification module: classifies and labels key information for use by the decision engine.

6. The life insurance company underwriting robot based on a large language model according to claim 5, characterized in that: The decision engine includes: Rule matching and personalized assessment module: perform risk assessment based on preset underwriting rules and user information; Adaptive scoring module: uses adaptive algorithms to dynamically score the user's overall risk; Decision generation and output module: outputs underwriting, premium underwriting or rejection recommendations and passes them to the interactive module for review by underwriters.

7. The life insurance company underwriting robot based on a large language model according to claim 6, characterized in that: The interactive modules include: Query interface module: underwriters input queries through natural language or voice, and the robot responds in real time; Data display module: displays decision results in the form of charts and text to help underwriters quickly understand the basis for decision-making; Feedback mechanism module: underwriters confirm or modify decisions and send feedback results to the learning module; Interaction recording and analysis module: records all interaction data for analysis and model optimization by the learning module.

8. The life insurance company underwriting robot based on a large language model according to claim 7, characterized in that: Learning modules include: Data collection and annotation module: collects underwriters’ feedback and final decisions and generates training samples; Reinforcement learning and model training module: Retrain the model through reinforcement learning and dynamically adjust the model parameters based on the labeled data; Performance evaluation and update module: Regularly evaluate the performance of the model in different scenarios and update the model based on the evaluation results.

9. A life insurance company underwriting method based on a large language model, characterized by: It adopts a life insurance company underwriting robot based on a large language model as described in any one of claims 1 to 8, and includes the following steps: 1) Data input; 1.1) The customer submits an underwriting application through the online platform, and the system collects the customer's medical records, financial information, risk assessment questionnaire and other data; 1.2) The data input module verifies the integrity of the data, ensuring that the format is correct and meets the system requirements. The system automatically de-identifies and encrypts the data and stores it securely in the database; 2) Natural language processing; 2.1) The system calls the natural language processing module to perform semantic analysis on health records and extract key information; 2.2) The module analyzes the customer's emotional tendencies through the emotion recognition function, providing a reference for the communication style of the subsequent interaction module; 3) Decision engine processing; 3.1) The decision engine provides personalized risk scores for customers’ health status, financial situation, and occupational risks based on natural language processing results and set underwriting rules; 3.2) The engine uses adaptive algorithms to adjust underwriting rules, generate preliminary underwriting recommendations, and output them to the interactive module; 4) Interaction and feedback; 4.1) Underwriters review the preliminary decisions generated by the system through the interactive module, including underwriting recommendations and risk scores; 4.2) The underwriter further inquires about the details, and the system displays the information in the form of charts and text; 4.3) The underwriter confirms or proposes changes to the robot’s decision, and the system collects and records the feedback; 5) Model learning and optimization; 5.1) The learning module labels the feedback data from underwriters and generates new training samples; 5.2) The system retrains the model through reinforcement learning and continuously optimizes the model parameters.

Images