Insurance information processing method and device, computer program product and electronic equipment

Abstract

The embodiment of the invention relates to an insurance information processing method and device, a computer program product and electronic equipment, and relates to the technical field of computers, and the system comprises the following steps: obtaining input information on an insurance platform page; analyzing the input information through an underwriting evaluation intelligent agent used for insurance underwriting so as to identify disease information, and determining underwriting evaluation knowledge of the disease information; and decoding the underwriting evaluation point knowledge, the input information, the context information and the cue word based on a large language model to perform underwriting evaluation, and generating an underwriting mode of an insurance product for the input information and insurance product operation information corresponding to the underwriting mode. According to the invention, the underwriting efficiency and accuracy can be improved.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and more specifically, to an insurance information processing method, an insurance information processing device, a computer program product, and an electronic device. Background Technology

[0002] The number of users with abnormal health conditions such as pre-existing medical history, abnormal physical examination results, or chronic diseases is gradually increasing, so it is necessary to accurately recommend suitable insurance products to users.

[0003] Among related technologies, intelligent insurance platforms can recommend insurance products to users based on their health conditions. However, these methods are limited in their ability to answer questions beyond the scope of formatted Q&As, and the accuracy of recommended insurance products is relatively low. Furthermore, when encountering users with complex medical histories, human customer service is required to make insurance recommendations, and underwriting conclusions cannot be reached in real time, resulting in low efficiency. Summary of the Invention

[0004] The purpose of this disclosure is to provide an insurance information processing method, an insurance information processing device, a computer program product, and an electronic device, thereby overcoming, to at least a certain extent, the problems of low processing efficiency and poor reliability caused by the limitations and defects of related technologies.

[0005] According to one aspect of this disclosure, an insurance information processing method is provided, comprising: acquiring input information from an insurance platform page; parsing the input information using an underwriting assessment agent for insurance underwriting to identify disease information and determine underwriting assessment knowledge of the disease information; decoding the underwriting assessment knowledge, the input information, context information, and prompt words based on a large language model to perform underwriting assessment, and generating an underwriting method for an insurance product based on the input information and corresponding insurance product operation information.

[0006] In one exemplary embodiment of this disclosure, the step of decoding the underwriting assessment knowledge, the input information, the context information, and the prompt words based on a large language model to perform underwriting assessment, and generating the underwriting method of the insurance product for the input information and the corresponding insurance product operation information, includes:

[0007] If the input information is complete, the underwriting assessment knowledge, the input information, the context information, and the prompt words are input into the large language model;

[0008] The large language model is controlled to perform underwriting risk reasoning on the input information and context information according to the thought chain in the prompt words and the underwriting assessment knowledge, and to determine the underwriting level.

[0009] Based on the underwriting level, determine the underwriting method for the insurance product with the input information and the corresponding insurance product operation information.

[0010] In one exemplary embodiment of this disclosure, controlling the large language model to perform underwriting risk reasoning on the input information and context information according to the thought chain in the prompt words and the underwriting assessment knowledge, and determining the underwriting level, includes:

[0011] Based on underwriting assessment knowledge, the input information and context information are used to perform underwriting risk reasoning, and the overall assessment value of the input information in multiple overlapping diseases is calculated.

[0012] Determine the underwriting level that matches the overall assessment value.

[0013] In one exemplary embodiment of this disclosure, determining the underwriting method of the insurance product for the input information and the corresponding insurance product operation information based on the underwriting level includes at least one of the following:

[0014] The underwriting level is Level 1, the underwriting method is determined to be standard underwriting, and the insurance product operation information is a recommended insurance product.

[0015] The underwriting level is Level 2, the underwriting method is determined to be premium-increased underwriting, and the insurance product operation information is to determine the premium-increased coefficient of the insurance product based on the overall assessment value; or, the underwriting method is determined to be exclusion-based underwriting, and the insurance product operation information is to determine the exclusion scope of the insurance product;

[0016] The underwriting level is Level 3, the underwriting method is determined to be rejection, and the insurance product operation information is an operation suggestion to determine the insurance product for re-underwriting.

[0017] In one exemplary embodiment of this disclosure, the step of parsing the input information to identify disease information and determining the underwriting assessment knowledge of the disease information by an underwriting assessment agent for insurance underwriting includes:

[0018] The underwriting assessment agent parses the input information to determine the analysis result.

[0019] Based on the disease information present in the analysis results, knowledge retrieval is performed in the evaluation knowledge graph starting from the starting node to obtain the nodes directly associated with the starting node as the initial retrieval results;

[0020] The initial search results are then used to retrieve nodes indirectly associated with the starting node, which serve as underwriting assessment points for the disease information.

[0021] In one exemplary embodiment of this disclosure, the method further includes:

[0022] In response to the input information containing textual question information, the input information and historical question-and-answer information are input to the main agent, and the intent information is determined by using a large language model for intent recognition, and the sub-agent corresponding to the intent information is determined.

[0023] If the input information contains non-textual auxiliary information, and the type of the non-textual auxiliary information is medical information, the underwriting assessment agent is directly invoked.

[0024] If the input information contains non-textual auxiliary information, and the type of the non-textual auxiliary information is non-medical information, key information is extracted from the non-textual auxiliary information. The historical dialogue, question information, and the key information are input to the main agent. The large language model is used to perform intent recognition to determine the intent information, and a sub-agent matching the intent information is determined.

[0025] In one exemplary embodiment of this disclosure, the method further includes:

[0026] The responding sub-agent is a health service agent, which performs entity recognition on the input information to determine the entity;

[0027] The entity is analyzed using a large language model to determine the service type;

[0028] In response to the service type being medical service, the medical recommendation machine learning model service is invoked to determine recommended medical information based on the entities in the input information.

[0029] In response to the service type being a health service knowledge Q&A service, the input information is retrieved from the health service knowledge base to determine candidate knowledge. The candidate knowledge is then integrated with the input information to determine recommended answer information.

[0030] In one exemplary embodiment of this disclosure, the method further includes:

[0031] The responding sub-agent is an insurance knowledge question-answering agent that performs word segmentation on the input information and identifies insurance product names, insurance terms, and medical terms in the input information based on the word segmentation results;

[0032] By combining the recognition results with the attributes in the metadata, the inverted index of the attributes is used to perform a synonym search, resulting in multiple similar segments;

[0033] The similarity between the input information and the multiple similar segments is calculated to determine the similarity between the input information and the similar segments;

[0034] Based on multiple similar segments with a similarity greater than the similarity threshold, context information, and the input information, language parsing is performed to generate response information for the input information.

[0035] According to one aspect of this disclosure, an insurance information processing apparatus is provided, comprising:

[0036] The input information acquisition module is used to acquire input information from the insurance platform page;

[0037] The assessment knowledge retrieval module is used to parse the input information through an underwriting assessment intelligence agent used for insurance underwriting to identify disease information and determine the underwriting assessment knowledge of the disease information;

[0038] The underwriting module is used to decode the underwriting assessment knowledge, the input information, the context information, and the prompt words based on a large language model to conduct underwriting assessment, and generate the underwriting method of the insurance product for the input information and the corresponding insurance product operation information.

[0039] According to one aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the insurance information processing method described in any of the preceding claims.

[0040] According to one aspect of this disclosure, an electronic device is provided, comprising:

[0041] Processor; and

[0042] Memory for storing the executable instructions of the processor;

[0043] The processor is configured to execute the insurance information processing method described above by executing the executable instructions.

[0044] In the technical solution provided in this disclosure, since underwriting assessment knowledge is obtained based on the disease information in the input information on the insurance platform page, and then the input information is underwritten based on the underwriting assessment knowledge to generate the underwriting method of the insurance product and the underwriting plan represented by the insurance product operation information, it can assist users in making timely judgments or decisions at the front end of insurance sales, and provide timely feedback on the underwriting plan, thereby realizing automated underwriting and improving underwriting efficiency.

[0045] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0046] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0047] Figure 1 The illustration shows a flowchart of an insurance information processing method according to an embodiment of the present disclosure.

[0048] Figure 2 The diagram illustrates the overall process of underwriting assessment in an embodiment of this disclosure.

[0049] Figure 3 The diagram illustrates the specific process of underwriting assessment in an embodiment of this disclosure.

[0050] Figure 4 The schematic diagram illustrates the process of allocating sub-intelligent agents in an embodiment of this disclosure.

[0051] Figure 5 The illustration shows a flowchart of an insurance knowledge Q&A session in an embodiment of this disclosure.

[0052] Figure 6 The schematic diagram illustrates a process flow diagram of health services in an embodiment of this disclosure.

[0053] Figure 7 The schematic diagram illustrates the interface diagram of the underwriting assessment in an embodiment of this disclosure.

[0054] Figure 8 A block diagram schematically illustrates an insurance information processing apparatus according to an embodiment of the present disclosure.

[0055] Figure 9 A schematic block diagram of an electronic device according to an embodiment of the present disclosure is shown. Detailed Implementation

[0056] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this disclosure more comprehensive and complete, and to fully convey the concept of the example embodiments to those skilled in the art. The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a full understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced with one or more of the specific details omitted, or other methods, components, apparatus, steps, etc., can be employed. In other instances, well-known technical solutions are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0057] Furthermore, the accompanying drawings are merely illustrative of this disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0058] Commercial health insurance is closely related to medicine and has complex terms: With the development of medical technology, various innovative drugs and medical devices, advanced treatment technologies are constantly emerging, and health management and services are constantly innovating, making health insurance terms increasingly complex. In order to manage the risk of claims after underwriting and balance mortality, health insurance has very complex rules for underwriting pre-existing conditions. These rules not only cover multiple stages such as health disclosure, underwriting review, and risk investigation, but also need to comprehensively consider factors such as different types of diseases, stages of disease, treatment, and prognosis.

[0059] User inquiries may involve medical expertise and detailed insurance terms. The health conditions of people with pre-existing conditions are diverse, and their insurance needs are more specific and complex, requiring comprehensive consideration of various medical factors and insurance policy details. The technical terminology and complex provisions of insurance policies themselves make it difficult for agents, even those with some insurance knowledge, to accurately understand and explain them in a short time, thus affecting their ability to make accurate recommendations to clients. Therefore, it is difficult for agents to quickly and accurately recommend suitable products based on a client's health condition.

[0060] To address the aforementioned technical challenges, insurance product recommendations can be implemented through intelligent insurance application platforms. Most existing intelligent insurance application platforms rely on fixed-format questionnaires for knowledge-based question-and-answer sessions, lacking in-depth reasoning. Key problems include: significant differences in health conditions among customers, making it difficult to accurately identify intent and responses to questions outside the scope of the formatted questionnaire; a lack of in-depth reasoning for complex cases with comorbidities or complications, resulting in coarse recommendations and poor judgment; and the need for manual processing for users with complex medical histories, requiring underwriting experts to consider the client's medical records for judgment and decision-making, thus hindering real-time underwriting feedback and creating gaps in the customer experience.

[0061] For customers who may be subject to increased premiums, the agent cannot provide a quantified risk premium coefficient. The agent can only consult underwriting experts, resulting in a break in the underwriting process. Therefore, the performance of the intelligent insurance product recommendation is poor.

[0062] The existing technical solution is based on a three-layer question-answering architecture: FAQ question answering, rule-based question answering, entity recognition after intent recognition, matching the identified intent and entity to the pre-configured questionnaire format, performing question normalization, and retrieving answers from the graph. Document retrieval and answer generation are based on a general large model + RAG approach.

[0063] Among the methods described above, while fixed, formatted questionnaires yield accurate conclusions, they suffer from limitations in understanding intent, have unfriendly interaction methods, and multi-turn dialogues are ineffective. Question-and-answer systems based on general large models and RAG lack deep reasoning capabilities, have coarse judgment scales for assessments, cannot quantify risk, and complex questions can only be assessed by human experts, resulting in 40% invalid responses. Furthermore, user risk is not quantified using EM values ​​(extra mortality rates), making it impossible to measure surcharge coefficients for users who may require additional premiums; agents can only consult underwriting experts, creating gaps in the underwriting process.

[0064] To address the aforementioned technical problems, this disclosure provides an insurance information processing method, with reference to... Figure 1 As shown, the main steps include:

[0065] Step S110: Obtain input information from the insurance platform page;

[0066] Step S120: The input information is parsed by an underwriting assessment agent used for insurance underwriting to identify disease information and determine the underwriting assessment knowledge of the disease information;

[0067] Step S130: Based on the large language model, decode the underwriting assessment knowledge, the input information, the context information, and the prompt words to perform underwriting assessment, and generate the underwriting method of the insurance product for the input information and the corresponding insurance product operation information.

[0068] Next, refer to Figure 1 The following diagram illustrates the insurance information processing method in the embodiments of this disclosure.

[0069] In step S110, the input information on the insurance platform page is obtained.

[0070] In this embodiment, the insurance platform page can be the display page of an insurance platform installed on a terminal. The insurance platform can be a standalone application deployed on the client, a mini-program embedded in a third party, a public account, or a separate webpage. The insurance platform can be used by agents, administrators, or operators, and can also be used by users after authorization.

[0071] For insurance platforms, a separate AI-powered underwriting entry point has been added, focusing on supporting underwriting assessment Q&A. The platform can recognize intent, providing social interaction, identifying intent to apply for disease-related underwriting, and guiding users away from invalid intents. It supports uploading images and documents, as well as various formats of files such as medical examination reports and medical records, or plain text input, from which entity recognition of information such as disease risks and products is performed. It supports continuous Q&A, allowing multiple uploads of files or supplementary information for subsequent Q&A sessions. During the Q&A process, the large language model performs logical reasoning based on customer input and underwriting assessment knowledge. If the information is insufficient to issue an underwriting conclusion, the large language model will guide the customer to supplement the necessary information. After the supplementary information is completed, the large language model will make an underwriting decision, issuing an underwriting conclusion, EM value (Extra Mortality Rate), surcharge factor, and recommended insurable products.

[0072] Therefore, the input information can include questions, which can be user-entered inquiries or consultations. These questions can be in text format. The input information can also include non-text auxiliary information, such as images and documents. This non-text auxiliary information can be medical or non-medical. Users can directly enter text in the input box to ask questions, or they can click on the "Add Control" button at the bottom of the insurance platform page to upload medical examination reports or medical records to submit their questions. Medical records can include one or more of the following: ultrasound, CT scans, biochemical tests, admission diagnoses, and discharge summaries.

[0073] In step S120, the input information is parsed by an underwriting assessment agent used for insurance underwriting to identify disease information and determine the underwriting assessment knowledge of the disease information.

[0074] After obtaining the input information from the insurance platform page, if there is only one underwriting assessment agent for insurance underwriting, the underwriting assessment knowledge corresponding to the disease information of the input information can be directly determined based on the underwriting assessment agent. If there are multiple sub-agents, the main agent can perform intent recognition on the input information to determine the intent information, which represents the true purpose of the input information. After determining the intent information, the input information can be assigned to the sub-agent corresponding to the intent information for execution. The main agent has multimodal interaction capabilities, can have natural and fluent dialogue, can understand the user's intent and provide accurate responses, and can be made easier to operate by taking photos, uploading files, etc. Sub-agents can include insurance knowledge Q&A agents, underwriting review agents, and health service agents.

[0075] First, we will explain the specific situation where there is only one underwriting assessment agent used for insurance underwriting.

[0076] The underwriting assessment agent is implemented using a task-based workflow, which orchestrates task nodes. The underwriting assessment agent is used for health insurance service recommendations, specifically outputting underwriting conclusions, suitable insurance products, the underwriting method of the insurance products, and one or more of the following: surcharge coefficient and adjustment direction.

[0077] For example, the input information is parsed by the underwriting assessment agent to determine the parsing result; further, it is determined whether disease information exists in the parsing result. If so, underwriting assessment knowledge is obtained by searching the assessment knowledge graph based on the disease information in the parsing result. For example, knowledge retrieval is performed in the assessment knowledge graph starting from the starting node to obtain the nodes directly associated with the starting node as the initial retrieval result; the initial retrieval result is then searched to obtain the underwriting assessment knowledge of the disease information from the nodes indirectly associated with the starting node.

[0078] In this embodiment of the disclosure, the input information is first parsed. For example, it is first determined whether the input information contains non-textual auxiliary information, which can be an image or a document. If non-textual auxiliary information exists, the text within it is identified using an image recognition model. The image recognition model can be an OCR model.

[0079] Furthermore, abnormal indicators in the input and auxiliary information can be identified and integrated to obtain analysis results. These results can indicate whether disease information is included. Disease information may include the disease name, abnormal indicators, and information about potential insurance products. Specifically, this can be obtained as the disease name, abnormal indicators from the disease / test, and whether there are potential insurance products.

[0080] In this embodiment, if the parsing results contain disease names or abnormal information, the underwriting assessment knowledge can be retrieved from the assessment knowledge graph based on the existing disease names or abnormal information to obtain the underwriting assessment knowledge corresponding to the disease names or abnormal information. Underwriting assessment knowledge can be underwriting rules. Specifically, an assessment knowledge graph can be established centered on disease information, linking diseases to each other and diseases to underwriting assessment knowledge. The triplets stored in the assessment knowledge graph mainly include: the relationship between a disease and its associated diseases (including hierarchical relationships, comorbidities, complications, etc., such as the "three highs" including hypertension, hyperlipidemia, and hyperglycemia); the relationship between a disease and assessment knowledge of different types of insurance products (such as "hypertension" and "hypertension: general inpatient medical assessment"); all diseases and the assessments they need to refer to (such as "hypertension" and "metabolic syndrome assessment"). The description of the assessment knowledge for each disease is unstructured text knowledge. In the underwriting knowledge graph, specific risk quantification data governance has been performed on the assessment scales of complex past cases, significantly improving the refinement of underwriting.

[0081] For example, a second-degree search can be performed in the review knowledge graph based on disease names or abnormal information. The second-degree search leverages the semantic relationships between entities in the knowledge graph, performing logical operations to retrieve entity attributes or relationships based on the initial search results. In this embodiment, the second-degree search refers to starting from the initial node, traversing at most two edges, and retrieving disease nodes indirectly related to the initial node to uncover deeper levels of association information. For example, the initial node can be determined based on the disease name or abnormal information, and the search can begin in the review knowledge graph from the initial node to obtain nodes directly related to the initial node as the initial search results. Further, the nodes directly related to the initial search results can be searched to obtain nodes indirectly related to the initial node as underwriting review knowledge corresponding to the disease names or abnormal information involved in the input information. For example, for the "three highs" (hypertension, hyperglycemia, and hyperlipidemia), a first-degree search would recall hypertension, hyperglycemia, and hyperlipidemia, while a second-degree search would recall metabolic syndrome as underwriting review knowledge.

[0082] During the retrieval process, entity information such as diseases and symptoms is first identified and converted into standard terminology. Using this standard terminology, disease-related underwriting assessment knowledge is retrieved from the assessment knowledge graph. Combining the underwriting assessment thought chain processing logic set in the intelligent agent system's prompts with the user's diseases, a large language model analyzes and understands the underwriting assessment knowledge, determining the reasoning and calculation for cases with multiple overlapping diseases, thus enabling underwriting judgments under such conditions.

[0083] If no matching underwriting assessment knowledge is retrieved, the entire process ends, and a message indicating that the disease is not included is displayed.

[0084] In step S130, the underwriting assessment knowledge, the input information, the context information, and the prompt words are decoded based on the large language model to perform underwriting assessment, and the underwriting method of the insurance product for the input information and the corresponding insurance product operation information are generated.

[0085] In this embodiment, after obtaining the underwriting assessment knowledge, the underwriting assessment knowledge, input information, context information, and prompt words can be input into the large language model. The prompt words can include thought chains for reasoning, representing the specific process of reasoning. Based on this, the large language model is controlled to perform underwriting assessments on the input information and context information according to the thought chains in the prompt words and the underwriting assessment knowledge, generating the underwriting method for the insurance product based on the input information and the corresponding insurance product operation information. The underwriting method can be any one of standard underwriting, premium-increased underwriting, exclusion underwriting, or rejection underwriting. The insurance product operation information corresponding to the underwriting method can be determining the premium increase coefficient of the insurance product, determining the exclusion scope of the insurance product, or determining operational suggestions for re-underwriting the insurance product, etc.

[0086] Figure 2 The flowchart of the underwriting assessment agent is illustrated in the figure. (Refer to...) Figure 2 As shown, the system retrieves problem information, performs in-depth analysis on the input information to obtain analysis results, searches the knowledge graph based on the analysis results, and conducts underwriting risk assessment and product recommendations based on the search results. If no further questions are needed, the system outputs the result; if further questions are needed, the system updates the input information and continues the above steps until the input information is complete and no further questions are needed, at which point the overall underwriting conclusion is output.

[0087] For example, after determining the underwriting assessment knowledge of the disease information, the underwriting assessment knowledge, input information, context information of the input information, and prompt words can be used as input. A large language model is used to parse and process the underwriting assessment knowledge, input information, context information of the input information, and prompt words. The large language model is controlled to evaluate the input information and context information according to the underwriting assessment knowledge and the reasoning chain of the underwriting assessment agent, assessing whether the input information meets the insurance underwriting conditions, outputting the underwriting level, and then generating an underwriting plan that conforms to the output format specified by the prompt words based on the underwriting level. The underwriting plan may include the underwriting method of the insurance product of the input information and the corresponding insurance product operation information. The context information can be historical dialogue information, such as the historical dialogue information of the previous 5 rounds of input information.

[0088] Specifically, when the underwriting assessment agent performs underwriting evaluation on input and contextual information based on underwriting assessment knowledge, it can calculate the assessment value of a single disease from multiple analytical dimensions. In addition, it can determine the assessment values ​​of complications and comorbidities. Multiple analytical dimensions can be determined based on underwriting assessment knowledge. For example, multiple analytical dimensions could include disease severity, treatment effectiveness, recurrence risk, complications, and underwriting standard knowledge. Furthermore, the assessment values ​​of a single disease, complications, and comorbidities can be superimposed to obtain an overall assessment value. The overall assessment value can be a risk value, specifically represented by an EM value. Finally, the matching underwriting level is determined based on the comparison results of the overall assessment value with a first threshold and a second threshold.

[0089] When the overall assessment value does not exceed the first threshold of the insurance product, the underwriting level can be considered Level 1. When the overall assessment value is greater than the first threshold but less than the second threshold, the underwriting level can be considered Level 2. Based on this, it can be determined whether the insurance product has any additional premiums or exclusions. If there are additional premiums, the additional premium coefficient can be calculated. If there are exclusions, the scope of the exclusions can be determined. If the overall assessment value is greater than the second threshold, the underwriting level is Level 3. Level 1 can be low risk, Level 2 can be high risk, and Level 3 can be excessively high risk. The first threshold is less than the second threshold.

[0090] After determining the overall assessment value corresponding to the input information, the underwriting level can be determined. Based on the underwriting level, the corresponding underwriting method and the corresponding insurance product operation information can be determined. The underwriting level can be Level 1, Level 2, Level 3, etc. The underwriting method can be any of the following: standard underwriting, underwriting with added premium, underwriting with exclusions, or rejection. The insurance product operation information corresponding to the underwriting method can be the insurance product's premium factor, exclusions, or reassessment recommendations, etc. When the underwriting level is Level 1, the underwriting method is standard underwriting, and the corresponding insurance product operation information can be a recommended matching insurance product. When the underwriting level is Level 2, and the underwriting method is underwriting with added premium, the corresponding insurance product operation information can include calculating the premium factor and recommending a matching insurance product. The premium factor can be determined based on the calculated overall assessment value; for example, the premium factor can be positively correlated with the calculated overall assessment value, meaning the higher the overall assessment value, the larger the premium factor. When the underwriting level is Level 2 and the coverage method is exclusion coverage, the corresponding insurance product operation information can include clearly specifying the scope of exclusions (e.g., excluding treatment related to thyroid disease) and recommending matching insurance products. When the underwriting level is Level 3 and the coverage method is rejection, the corresponding insurance product operation information can include analyzing the reasons for rejection and providing operational suggestions for re-underwriting the insurance product. Operational suggestions can include health management recommendations or areas for improvement. For example, a health management recommendation could be to reapply after achieving target blood sugar levels for 6 months.

[0091] In this embodiment, a comprehensive assessment and insurance product recommendation can be made for users with complications and comorbidities based on their overall evaluation values. An algorithm for premium increases based on disease EM values ​​and underwriting factors is introduced to quantify disease risk, make the reasoning and decision-making process transparent, and improve the accuracy of the underwriting plan.

[0092] Furthermore, the completeness of input information can be determined from multiple assessment dimensions. For example, after obtaining underwriting assessment knowledge of the disease name or abnormal information, the necessary fields for the disease can be determined based on this knowledge. The input information is then compared with these necessary fields; if they are fully covered, the input information is considered complete; otherwise, it is considered incomplete. If the input information is complete, the underwriting method of the insurance product and the corresponding insurance product operation information can be directly output as the underwriting plan output by the underwriting assessment agent, ending the underwriting assessment process. If the input information is incomplete, the user is instructed on the data types that need to be supplemented, and the process returns to the input information collection step. Multiple assessment dimensions may include disease name, diagnosis time, treatment status, examination indicators, and complications. For example, the user may be instructed to supplement three months of blood pressure monitoring records. Specifically, the necessary fields for each disease can be dynamically determined based on the underwriting assessment knowledge corresponding to the disease. When it is determined that the provided input information does not cover the necessary fields for the disease, a prompt message is generated based on the missing fields. The generated prompt message could be, for example, "I need to further determine detailed information about blood pressure." After obtaining the supplementary information, the underwriting assessment can be carried out by combining the input information and the supplementary information, which will not be elaborated here.

[0093] Figure 3 The diagram illustrates the flowchart of the underwriting assessment process. To enhance the professionalism and rigor of the underwriting assessment agent, an underwriting assessment is implemented through a technical architecture that incorporates process orchestration, a large language model, RAG (Retrieval-augmented Generation), a rule engine, and limited output. (Reference) Figure 3 As shown, the main steps include:

[0094] Step S301: Obtain input information.

[0095] Step S302: Determine whether the input information contains an image or a file. If yes, proceed to step S303; otherwise, proceed to step S304.

[0096] Step S303: OCR recognition identifies and determines the text in the image or file.

[0097] Step S304: Integrate the problem information, physical examination report, and medical record information from the input information.

[0098] Step S305: Scope of Responsibility Determination. If the intention is for underwriting assessment, proceed to step S306; otherwise, proceed to step S310.

[0099] Step S306: Underwriting assessment is performed through the thought chain of the underwriting assessment intelligence agent. For example, disease risk identification can be performed first. Based on the identified disease name or abnormal information, underwriting assessment knowledge for that disease name or abnormal information is retrieved from the assessment knowledge graph. The completeness of the input information is determined. If complete, information extraction and summarization are performed, and the risk values ​​of individual diseases, complications, and comorbidities in the input information are queried based on the disease knowledge graph. The risk values ​​of individual diseases, complications, and comorbidities are superimposed to obtain the overall risk value, which is then compared with a risk value threshold. The underwriting level is determined based on the comparison result. Based on the underwriting level and the input information, insurance product information is queried in the insurance graph to obtain matching insurance products for recommendation.

[0100] Specifically, for the knowledge graph assessment, the underwriting risk of diseases is quantified, and EM (Extra Mortality Rate) is introduced to calculate the premium surcharge factor. A new knowledge management feature for overlapping risk diseases has been added, enabling comprehensive assessment and product recommendations for customers with complications and comorbidities. Calculators for risks such as hypertension and physical condition have also been added.

[0101] Step S307: Determine whether the input information is complete and whether further questions are needed. If yes, proceed to step S308; otherwise, proceed to step S309.

[0102] Step S308, follow-up questioning guidance.

[0103] Step S309: Output the underwriting plan and recommended insurance products.

[0104] Step S310: Guide students to ask the correct questions.

[0105] Step S311: Display the results.

[0106] In this embodiment, a balance between the degree of freedom in generation and the accuracy of facts is achieved through the synergy of structured constraints and flexible generation, guiding the creativity of the large model onto a compliant and controllable track. This not only significantly reduces the risk of illusion but also improves professionalism in the insurance field. Process orchestration breaks down tasks into manageable atomic steps, improving processing efficiency. The large model focuses on complex reasoning, while the rule engine handles deterministic processing. The large model + RAG uses retrieved credible content as the basis for generation, introducing a confidence threshold mechanism: when a key entity (such as a disease name) does not match the knowledge base, it explicitly outputs that the disease is not included, limiting the large model's fictitiousness and improving reliability. For explicit rules and mathematical formulas, the rule engine performs hard constraints and logical verification. For example, the calculation of disease risk values ​​and premium surcharge schemes are all implemented by the rule engine, ensuring the professionalism and reliability of the output insurance product's underwriting method and the corresponding insurance product operation information, thus achieving differentiated insurance. By outputting underwriting methods and insurance product operation information at the front end of insurance sales, the convenience of choosing insurance is improved, a reference is provided for insurance sales, automated underwriting is achieved, underwriting efficiency is improved, and accurate insurance product recommendations are made for users, thus improving accuracy.

[0107] In this embodiment, the sub-agents, in addition to the underwriting assessment agent, may also include an insurance knowledge question-and-answer agent and a health service agent. If multiple sub-agents are included, the sub-agent to be invoked can be determined based on the intent information. When determining the intent information, it can be based on the content of the input information. If the input information contains text-based questions, the question information, context information, and prompts can be input to the main agent. A large language model is used for intent recognition to determine the intent information, and the corresponding sub-agent is identified. The input information is then forwarded to the corresponding sub-agent, achieving end-to-end intent recognition. The context information can be historical question-and-answer information. Prompts can include task type, task execution order, and rules for task return results. For example, the task execution order can prioritize matching direct reply-type intents; if a match is found, the intent name and corresponding preset reply are returned directly. If no match is found, the closest business consultation-type intent is matched, and entity recognition is performed to extract entities such as disease, insurance product name, occupation name, age, and gender. If no category can be defined, an unknown intent is returned. The entity recognition specifications are as follows: For diseases, the identification should include the disease diagnosis or abbreviation / alias, recognizing disease symptoms and location information. For insurance product names, the identification should include the official name and commonly used abbreviations / aliases. Output should be in JSON format. Historical question-and-answer information can include multi-turn dialogue information preceding the question information, such as five or ten rounds, etc.

[0108] If the input information includes text-based questions as well as non-textual auxiliary information such as images or files, it can be determined whether the non-textual auxiliary information is medical information. If the non-textual auxiliary information is medical information, the underwriting assessment agent is directly used as a sub-agent. Here, medical information can be medical reports, such as physical examination reports, medical diagnoses, imaging reports, etc.

[0109] If the question information contains non-textual auxiliary information, and the type of non-textual auxiliary information is non-medical, key information is extracted from the non-textual auxiliary information to determine key information. The historical dialogue, question information, and key information are input into the main agent, and a large language model is used for intent recognition to determine the intent information. A sub-agent matching the intent information is then identified. For example, if the input information contains textual question information and non-textual auxiliary information, and the type of non-textual auxiliary information is not medical information such as medical reports, physical examination reports, or imaging reports, but rather a regular document, key information can be extracted from the non-textual auxiliary information. Key information can be keywords, etc. Furthermore, the historical dialogue, question information, and key information from the non-textual auxiliary information can be integrated to obtain integrated information. This integrated information, along with prompts, is input into a large language model, which encodes and decodes the information to output the intent information corresponding to the input information. The agent corresponding to this intent information can then be designated as a sub-agent. Intent information can include direct response intents, insurance knowledge Q&A, disease-related products, and health services. Direct response intents can be, for example, social interactions.

[0110] Figure 4 The flowchart for determining the sub-agent is illustrated in the diagram. (Refer to...) Figure 4 As shown, the main steps include:

[0111] The system acquires the input question information and, through conversation management and historical question-and-answer information, determines the intent of the input information through the main intelligence agent. The routing and forwarding module then forwards the input information to different sub-intelligent agents. For example, questions about insurance product knowledge are forwarded to the insurance knowledge question-and-answer agent, questions about eligible products for medical conditions are forwarded to the underwriting assessment agent, and questions about health services are forwarded to the health service agent. For intent information that cannot be recognized, a general response is output to guide the user to ask the question again.

[0112] For example, if the input information is "details of a specific insurance product" or "waiting period, limit, and scope of coverage for a specific product," the intended information is insurance knowledge Q&A. If the input information is "which type of insurance can a diabetic patient buy," the intended information is products available for the disease. If the input information is "which hospital to go to for high blood pressure treatment," the intended information is health services.

[0113] This embodiment employs an agent routing architecture combining a master agent and sub-agents. The master agent, responsible for intent recognition, understands the user's input question and routes it to the corresponding sub-agent for processing. During multi-turn interactions, a session management module enhances the contextual judgment capabilities of the large language model, preventing errors in sub-agent judgment. The session management module records each user's historical question-and-answer information. The master agent's session is composed of all its own and all sub-agents' sessions; concatenating these sessions allows the large language model to more accurately judge the context, avoiding forwarding to the wrong sub-agent.

[0114] When the intent of the input information is insurance knowledge Q&A, the invoked sub-agent can be an insurance knowledge Q&A agent. In some embodiments, when the sub-agent is an insurance knowledge Q&A agent, the agent segments the input information into words, identifies the insurance product name, insurance description, and medical description in the input information based on the segmentation results, and performs a synonym search on the identification results to determine the answer information for the input information. For example, attributes in the metadata can be combined to retrieve the inverted index of the attributes, obtaining multiple similar segments; similarity calculations are performed between the input information and multiple similar segments to determine the similarity between the input information and the similar segments; and the answer information for the input information is generated based on the k similar segments with similarity greater than a similarity threshold, the context information of the input information, and the input information itself.

[0115] For an insurance knowledge question-answering agent, during the document storage stage, the unstructured knowledge of insurance products can be vectorized using an embedding model and stored in a vector database. This database stores vectors, metadata, and tags, and an inverted index is built for the attributes in the metadata. The unstructured knowledge can include insurance terms and conditions, underwriting policy descriptions, etc. The metadata can include terms and conditions names, effective dates, etc. Building an inverted index for the attributes in the metadata yields the following form:

[0116] {"text":"A 60-day waiting period begins from the effective date of this contract...","metadata":{"doc_type":"Terms","product":"XXXX 2024PLUS Medical Insurance (Adjustable Rate)","section":"Mandatory Coverage","keywords":["Waiting Period","Responsibility","XX Agreement"]}}.

[0117] Further, a thesaurus can be established, such as creating abbreviations or aliases for product terms: Health Appointment -> XXX Health 2024 PLUS Medical Insurance (adjustable rate); and creating insurance terminology synonyms: Scope of Coverage -> Insurance Liability / Mandatory Liability, Optional Liability, Sum Insured -> Product Limit, Remaining Sum Insured -> Claim Limit, Underwriting Conditions -> Underwriting Policy.

[0118] In the knowledge-based question-answering stage of the insurance knowledge-answering agent, the user's input information is first segmented into words to obtain segmentation results. NLP (Natural Language Semantics Recognition) is then applied to the insurance product names, insurance terms, and medical terms in the segmentation results. The identified terms are retrieved from synonyms, and the keywords in the query are expanded using synonyms. A hybrid retrieval scheme is adopted, combining attributes from metadata for matching to obtain initial screening results. Further refinement is performed on these initial screening results to obtain final screening results. When matching attributes from metadata to obtain initial screening results, synonym retrieval can be performed on the inverted index of the attributes. The retrieval algorithm uses the term frequency-inverse document frequency algorithm to recall multiple similar segments, which can be the top k similar segments. Further refinement is performed on the initial screening results. During the refinement process, the similarity between the input information and the multiple segments recalled in the initial screening is calculated to determine the similarity between the text vector of the input information and the text vectors of the multiple similar segments. For example, cosine similarity can be used to calculate the similarity between the two. Based on this, the most similar segments, contextual information, input information, and prompts can be fed into a large language model to output the response information. Contextual information can be, for example, the most recent five rounds of dialogue.

[0119] Figure 5 The diagram illustrates a flowchart of an insurance knowledge question-and-answer process. The intent is to identify the intelligent agent that, upon determining input information such as insurance product terms and conditions or underwriting policies, will invoke the insurance knowledge question-and-answer intelligent agent to perform the task. This agent is built upon a large model and knowledge base-based retrieval enhancement system, focusing on retrieving document knowledge and generating answers. It is suitable for question-and-answer scenarios involving unstructured documents such as FAQs, insurance terms and conditions, and underwriting policies. (Reference) Figure 5 As shown, the main steps include:

[0120] The insurance knowledge question-answering AI sends the input information to a knowledge base for retrieval. It then determines if the retrieved knowledge exists; if so, it selects the correct knowledge and summarizes the answer based on the input information and the knowledge, then displays it. If no retrieved knowledge exists, it returns to the original question-and-answer section.

[0121] In other embodiments, when the service type of the input information is health service, a health service intelligent agent can be invoked. The responding intelligent agent, also a health service intelligent agent, performs entity recognition on the input information to determine the entities. For example, it extracts information such as diseases, products, hospitals, regions, and symptoms as entities. A large language model is used to analyze the entities and determine the service type. Using prompt words, the input information and built-in prompt words are used by the large language model to determine whether the service type is a consultation on health service knowledge or a request for medical treatment. If the service type is medical treatment, a medical treatment recommendation machine learning model service is invoked. Based on the entities in the input information, recommended medical treatment information is determined; based on information such as diseases, symptoms, products, and regions, recommended medical treatment information is directly output, which can include recommended matching hospitals, departments, and specialists. If the service type is health service knowledge question-and-answer service, the input information is searched in the health service knowledge base to determine candidate knowledge. This candidate knowledge is then integrated with the input information to determine the recommended answer. Candidate knowledge is the knowledge most relevant to the input information.

[0122] For example, when the intent recognition agent determines that the input information relates to health service issues, it invokes a health service agent to perform the task. This agent integrates a large model, a knowledge base RAG scheme, and a medical recommendation machine learning model. For health service knowledge-based questions, it uses a large model combined with RAG to retrieve health service documents and generate recommended answer information, suitable for unstructured document question-and-answer formats such as medical network hospitals, hospital introductions, and high-end medical case studies. If the input information is related to disease treatment, it invokes a medical recommendation model to determine recommended medical information based on entities such as diseases, symptoms, products, and regions extracted from the input information.

[0123] Figure 6 A flowchart illustrating health services is shown below. Figure 6 As shown, health services mainly include the following steps:

[0124] In step S601, the health service agent sends the input question to the large language model.

[0125] In step S602, the large language model determines the service type. If the service type is medical service, proceed to step S603; if the service type is health service knowledge query service, proceed to step S605.

[0126] In step S603, the medical recommendation model searches for departments based on diseases, symptoms, and locations.

[0127] In step S604, recommended medical information is determined based on information such as department, region, and insurance product.

[0128] In step S605, knowledge retrieval is performed based on the knowledge base.

[0129] In step S606, recall knowledge is determined.

[0130] In step S607, it is determined whether recall knowledge exists. If it exists, proceed to step S608; otherwise, proceed to step S610.

[0131] In step S608, select the correct knowledge.

[0132] In step S609, answer information is recommended based on the input information and knowledge summary.

[0133] In step S610, the question and answer are returned.

[0134] In step S611, the results are displayed.

[0135] By invoking the health service intelligent agent, the result corresponding to the input information can be accurately determined based on the service type.

[0136] In this embodiment, the insurance platform can be accessed via a WeChat official account. The insurance platform page can be displayed by clicking a trigger control. The insurance platform page may include input boxes and add controls. Users or agents can directly enter text in the input boxes, or click the add control or the photo control to directly upload medical examination reports or medical records to generate input information. Supported medical records include, but are not limited to, documents such as ultrasound, CT scans, biochemical tests, admission diagnoses, and discharge summaries.

[0137] When the intent recognition agent identifies the input information as indicating a disease-related product, it can directly invoke the underwriting assessment agent as a sub-agent to reason about the input information according to the steps in the reasoning chain represented by the prompt words, thus obtaining the underwriting plan for the input information. For example, the input information can be parsed to generate a parsing result, which may contain disease names or abnormal information. If the parsing result contains a disease name, a secondary search can be performed on the assessment knowledge graph based on the disease name to obtain knowledge indirectly related to the disease name or abnormal information as the underwriting assessment knowledge for that disease. Based on the underwriting assessment knowledge, multiple analysis dimensions are determined, and the assessment values ​​for individual diseases, complications, and comorbidities in the input information are determined from these dimensions. The assessment values ​​of different diseases are then superimposed to obtain an overall assessment value. An underwriting level matching the overall assessment value is determined, and the underwriting method of the insurance product and the corresponding insurance product operation information are determined based on the underwriting level.

[0138] For example, after the reasoning process, a risk assessment is first performed on each disease. Finally, the overall assessment value for the combined "three highs" (hypertension, hyperlipidemia, and hyperglycemia) is summarized, resulting in an EM risk value of 175 points. Based on this overall assessment value, the underwriting factor for insurance products is calculated to be 2.75. Based on the underwriting conclusion, an insurable product is recommended, offering either an additional premium or exclusion of coverage, with an additional premium coefficient of 2.75. The output underwriting plan based on the input information can be as follows: Figure 7 As shown in the image.

[0139] The technical solutions in this disclosure can provide comprehensive assessments and insurance product recommendations for users with complications and comorbidities. A large language model is introduced to address the difficulty of deep-level structured parsing of medical images and text, supporting the parsing of multimodal data such as plain text, medical images, and physical examination reports to extract key information and achieve a more accurate understanding of disease risk. The large model, combined with the underwriting knowledge graph RAG, provides a more comprehensive and accurate interpretation of medical records for underwriting. A premium addition algorithm based on the overall disease assessment value and underwriting factors is introduced to quantify disease risk and make the reasoning and decision-making process transparent, improving the accuracy of underwriting for users with complex diseases. The question-and-answer architecture, primarily based on a rule engine, is upgraded to a multi-agent architecture, supporting smooth multi-turn dialogues and more accurately understanding the customer's question-and-answer intent, improving user experience and product intelligence. The system resolves the breakpoint problem of relying on manual intervention for complex case underwriting, improving efficiency and convenience. It reduces the amount of manual consultation, significantly lowering company operating costs.

[0140] In this embodiment of the disclosure, an insurance information processing device is provided, with reference to... Figure 8 As shown, the insurance information processing device 800 mainly includes the following modules:

[0141] The input information acquisition module 801 is used to acquire input information from the insurance platform page;

[0142] The knowledge retrieval module 802 is used to parse the input information through an underwriting assessment agent used for insurance underwriting to identify disease information and determine the underwriting assessment knowledge of the disease information;

[0143] The underwriting module 803 is used to decode the underwriting assessment knowledge, the input information, the context information and the prompt words based on the large language model to perform underwriting assessment, and generate the underwriting method of the insurance product for the input information and the insurance product operation information corresponding to the underwriting method.

[0144] It should be noted that the specific details of each step in the above insurance information processing method have been described in detail in the corresponding insurance information processing method, so they will not be repeated here.

[0145] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0146] Furthermore, although the steps of the method in this disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0147] In an exemplary embodiment of this disclosure, an electronic device capable of implementing the above-described method is also provided.

[0148] Those skilled in the art will understand that various aspects of this disclosure can be implemented as a system, method, or program product. Therefore, various aspects of this disclosure can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software aspects, collectively referred to herein as a "circuit," "module," or "system."

[0149] The following reference Figure 9 To describe an electronic device 900 according to such an embodiment of the present disclosure. Figure 9 The electronic device 900 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments disclosed herein.

[0150] like Figure 9 As shown, the electronic device 900 is manifested in the form of a general-purpose computing device. The components of the electronic device 900 may include, but are not limited to: at least one processing unit 910, at least one storage unit 920, a bus 930 connecting different system components (including storage unit 920 and processing unit 910), and a display unit 940.

[0151] The storage unit stores program code that can be executed by the processing unit 910, causing the processing unit 910 to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure. For example, the processing unit 910 can perform actions such as... Figure 1 The steps are shown in the figure.

[0152] Storage unit 920 may include readable media in the form of volatile storage units, such as random access memory (RAM) 9201 and / or cache memory 9202, and may further include read-only memory (ROM) 9203.

[0153] Storage unit 920 may also include a program / utility 9204 having a set (at least one) program module 9205, such program module 9205 including but not limited to: operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.

[0154] Bus 930 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.

[0155] Electronic device 900 can also communicate with one or more external devices 1000 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with electronic device 900, and / or with any device that enables electronic device 900 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 950. Furthermore, electronic device 900 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 960. As shown, network adapter 960 communicates with other modules of electronic device 900 via bus 930. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 900, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0156] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, terminal device, or electronic device, etc.) to execute the methods according to the embodiments of this disclosure.

[0157] In exemplary embodiments of this disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the methods described above is stored. In some possible implementations, various aspects of this disclosure may also be implemented as a program product including program code that, when the program product is run on a terminal device, causes the terminal device to perform the steps of the various exemplary embodiments of this disclosure described in the "Exemplary Methods" section above.

[0158] The program product for implementing the above-described method according to embodiments of the present disclosure may employ a portable compact disc read-only memory (CD-ROM) and include program code, and may run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto. In this document, the readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0159] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0160] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting programs for use by or in conjunction with an instruction execution system, apparatus, or device.

[0161] The program code contained on the readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0162] Program code for performing the operations of this disclosure can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0163] Furthermore, the above figures are merely illustrative of the processes included in the method according to exemplary embodiments of this disclosure and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0164] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention described herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not invented by this disclosure. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

Claims

1. A method for processing insurance information, characterized in that, include: Obtain the input information from the insurance platform page; The input information is parsed by an underwriting assessment agent used for insurance underwriting to identify disease information and determine the underwriting assessment knowledge of the disease information; Based on a large language model, the underwriting assessment knowledge, the input information, the context information, and the prompt words are decoded to perform underwriting assessment, and the underwriting method of the insurance product for the input information and the corresponding insurance product operation information are generated.

2. The insurance information processing method according to claim 1, characterized in that, The process involves decoding the underwriting assessment knowledge, input information, context information, and prompts based on a large language model to perform underwriting evaluation, generating an insurance product underwriting method for the input information and corresponding insurance product operation information, including: If the input information is complete, the underwriting assessment knowledge, the input information, the context information, and the prompt words are input into the large language model; The large language model is controlled to perform underwriting risk reasoning on the input information and context information according to the thought chain in the prompt words and the underwriting assessment knowledge, and to determine the underwriting level. Based on the underwriting level, determine the underwriting method for the insurance product with the input information and the corresponding insurance product operation information.

3. The insurance information processing method according to claim 2, characterized in that, The control system, based on the thought chain in the prompt words and the underwriting assessment knowledge, performs underwriting risk reasoning on the input information and contextual information to determine the underwriting level, including: Based on underwriting assessment knowledge, the input information and context information are used to perform underwriting risk reasoning, and the overall assessment value of the input information in multiple overlapping diseases is calculated. Determine the underwriting level that matches the overall assessment value.

4. The insurance information processing method according to claim 2, characterized in that, The step of determining the underwriting method for the insurance product based on the underwriting level and the corresponding insurance product operation information includes at least one of the following: The underwriting level is Level 1, the underwriting method is determined to be standard underwriting, and the insurance product operation information is a recommended insurance product. The underwriting level is Level 2, the underwriting method is determined to be premium-increased underwriting, and the insurance product operation information is to determine the premium-increased coefficient of the insurance product based on the overall assessment value; or, the underwriting method is determined to be exclusion-based underwriting, and the insurance product operation information is to determine the exclusion scope of the insurance product; The underwriting level is Level 3, the underwriting method is determined to be rejection, and the insurance product operation information is an operation suggestion to determine the insurance product for re-underwriting.

5. The insurance information processing method according to claim 1, characterized in that, The process of parsing the input information to identify disease information and determining the underwriting assessment knowledge of the disease information by an underwriting assessment intelligent agent used for insurance underwriting includes: The underwriting assessment agent parses the input information to determine the analysis result. Based on the disease information present in the analysis results, knowledge retrieval is performed in the evaluation knowledge graph starting from the starting node to obtain the nodes directly associated with the starting node as the initial retrieval results; The initial search results are then retrieved to obtain nodes indirectly associated with the starting node as underwriting assessment points for the disease information.

6. The insurance information processing method according to claim 1, characterized in that, The method further includes: In response to the input information containing textual question information, the input information and historical question-and-answer information are input to the main agent, and the intent information is determined by using a large language model for intent recognition, and the sub-agent corresponding to the intent information is determined. If the input information contains non-textual auxiliary information, and the type of the non-textual auxiliary information is medical information, the underwriting assessment agent is directly invoked. If the input information contains non-textual auxiliary information, and the type of the non-textual auxiliary information is non-medical information, key information is extracted from the non-textual auxiliary information. The historical dialogue, question information, and the key information are input to the main agent. The large language model is used to perform intent recognition to determine the intent information, and a sub-agent matching the intent information is determined.

7. The insurance information processing method according to claim 6, characterized in that, The method further includes: The responding sub-agent is a health service agent, which performs entity recognition on the input information to determine the entity; The entity is analyzed using a large language model to determine the service type; In response to the service type being medical service, the medical recommendation machine learning model service is invoked to determine recommended medical information based on the entities in the input information. In response to the service type being a health service knowledge Q&A service, the input information is retrieved from the health service knowledge base to determine candidate knowledge. The candidate knowledge is then integrated with the input information to determine recommended answer information.

8. The insurance information processing method according to claim 6, characterized in that, The method further includes: The responding sub-agent is an insurance knowledge question-answering agent that performs word segmentation on the input information and identifies insurance product names, insurance terms, and medical terms in the input information based on the word segmentation results. By combining the recognition results with the attributes in the metadata, the inverted index of the attributes is used to perform a synonym search, resulting in multiple similar segments; The similarity between the input information and the multiple similar segments is calculated to determine the similarity between the input information and the similar segments; Based on multiple similar segments with a similarity greater than the similarity threshold, context information, and the input information, language parsing is performed to generate response information for the input information.

9. An insurance information processing device, characterized in that, include: The input information acquisition module is used to acquire input information from the insurance platform page; The assessment knowledge retrieval module is used to parse the input information through an underwriting assessment intelligence agent used for insurance underwriting to identify disease information and determine the underwriting assessment knowledge of the disease information; The underwriting module is used to decode the underwriting assessment knowledge, the input information, the context information, and the prompt words based on a large language model to conduct underwriting assessment, and generate the underwriting method of the insurance product for the input information and the corresponding insurance product operation information.

10. A computer program product, characterized in that, The invention includes a computer program, characterized in that, when the computer program is executed by a processor, it implements the insurance information processing method according to any one of claims 1-8.

11. An electronic device, characterized in that, include: processor; as well as Memory for storing the executable instructions of the processor; The processor is configured to execute the insurance information processing method according to any one of claims 1-8 by executing the executable instructions.

Images