Insurance application intelligent generation method and device, electronic equipment and storage medium
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA PING AN PROPERTY INSURANCE CO LTD
- Filing Date
- 2026-03-06
- Publication Date
- 2026-06-09
Smart Images
Figure CN122175709A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of artificial intelligence technology, and is applicable to the financial field, particularly to a method and device for intelligently generating insurance policies, electronic devices, and storage media. Background Technology
[0002] An insurance application is a formal document in the insurance business, used to record information about the policyholder's application to purchase insurance, as well as the insurance company's underwriting and insurance terms. In the financial sector, large language models can be used to generate insurance application forms (such as car insurance, life insurance, property insurance, group insurance, etc.) based on customer information.
[0003] When providing insurance services, related technologies typically first collect customer information (including insurance policies, communication records, images, voice recordings, etc.) and then fill in an application form template based on this information to obtain the application. However, the amount of customer information collected by these technologies is often limited, frequently requiring multiple requests for relevant information from the customer or simple inference of missing information using large language models to complete the application. However, due to the complexity of the insurance field, these methods are prone to errors in policy template filling. Summary of the Invention
[0004] The main objective of this application is to propose an intelligent method and device for generating insurance application forms, an electronic device, and a storage medium, which can solve the problem of errors caused by missing information during the application form filling process and improve the accuracy of insurance application form generation.
[0005] To achieve the above objectives, a first aspect of this application proposes an intelligent method for generating insurance application forms, the method comprising: Obtain multimodal information of the target object; Template matching is performed based on the multimodal information to obtain an insurance application template, and the type of insurance product in the insurance application template is determined. Data is collected from at least two preset data sources based on the insurance product type and the multimodal information to obtain object multi-source data; A risk profile of the object is constructed based on the multi-source data of the object. Based on the risk profile of the object, insurance rules are mapped to obtain an insurance plan; The target insurance application is obtained by filling in the application form template according to the insurance plan.
[0006] Optionally, the step of collecting data from at least two preset data sources based on the insurance product type and the multimodal information to obtain multi-source data of the object includes: Activate a pre-deployed intelligent information collector; wherein the intelligent information collector is bound to at least two of the data sources; The information intelligent collector collects data from each of the data sources based on the insurance product type and the multimodal information to obtain the object's multi-source data.
[0007] Optionally, the target object is a target enterprise, and at least two of the data sources include a business information database, a public opinion system, and a geographic location information platform. The information intelligent collector collects data from each of the data sources based on the insurance product type and the multimodal information to obtain multi-source data of the target object, including: The intelligent information collector reads information from the business registration information database based on the multimodal information to obtain the business registration information of the target enterprise. The information intelligent collector reads information from the public opinion system based on the insurance product type and the multimodal information to obtain public opinion information; wherein, the public opinion information includes insurance public opinion information and corporate public opinion information; The information intelligent collector performs environmental analysis on the geographic location information platform based on the multimodal information to obtain the target enterprise's corporate environmental risk information. The multi-source data of the object is obtained by fusing information from the enterprise's business registration information, public opinion information, and environmental risk information.
[0008] Optionally, after constructing the object risk profile based on the multi-source data of the object, the method further includes: Based on the requirement fields of the insurance application template, gap detection is performed on the multimodal information and the multi-source data of the object to obtain the gap fields; The small language model is invoked to infer the gap field based on the object risk profile, and the predicted field content is obtained; The risk profile of the object is updated based on the content of the predicted field.
[0009] Optionally, the target object is a target enterprise, and the step of mapping insurance professional rules based on the risk profile of the target enterprise to obtain an insurance plan includes: The insurance premium is calculated based on the enterprise environmental risk information, total number of employees, job risk category, and the proportion of employees in the job risk category in the risk profile of the target. The insured amount is calculated based on the different job risk categories to obtain the insurance insured amount for each job risk category; Personalized terms are generated based on the enterprise environmental risk information and / or the job risk category; The insurance plan is obtained by integrating the insurance premium, the insurance coverage amount, and the personalized terms.
[0010] Optionally, the step of performing template matching based on the multimodal information to obtain an insurance application template includes: The multimodal information is feature-encoded using a multimodal pre-trained model to obtain initial multimodal features; wherein, the initial multimodal features include initial text features of text, initial image features of images, and initial speech features of speech. When the text, the image, and the speech are in the same time window, cross-attention calculation is performed on the initial text features, the initial image features, and the initial speech features to obtain the target multimodal features; The candidate templates are filtered based on the target multimodal features to obtain the insurance policy template.
[0011] Optionally, the step of encoding the multimodal information using a multimodal pre-trained model to obtain initial multimodal features includes: The multimodal information is used to identify intents through the multimodal pre-trained model to obtain intents corresponding to different modalities; The multimodal information is extracted using the multimodal pre-trained model to obtain key information corresponding to different modalities; The initial multimodal features are obtained by performing feature mapping on all the intents and all the key information through the multimodal pre-trained model.
[0012] To achieve the above objectives, a second aspect of this application provides an intelligent insurance application generation device, the device comprising: The information acquisition module is used to acquire multimodal information about the target object; The template matching module is used to perform template matching based on the multimodal information to obtain an insurance application template and determine the insurance product type of the insurance application template. The data collection module is used to collect data from at least two preset data sources based on the insurance product type and the multimodal information to obtain multi-source data of the object. The profile building module is used to build a profile based on the multi-source data of the object to obtain a risk profile of the object. The rule mapping module is used to map insurance rules based on the risk profile of the object to obtain an insurance plan; The policy filling module is used to fill the application form template according to the insurance plan to obtain the target application form.
[0013] To achieve the above objectives, a third aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the intelligent insurance policy generation method described in the first aspect.
[0014] To achieve the above objectives, a fourth aspect of the present application provides a storage medium, which is a computer-readable storage medium storing a computer program that, when executed by a processor, implements the intelligent insurance application generation method described in the first aspect.
[0015] The intelligent insurance application generation method, device, electronic equipment, and storage medium proposed in this application, after obtaining the multimodal information of the target object, do not directly fill in the insurance application based on the multimodal information. Instead, it determines the insurance product type of the insurance application template, and then collects data from at least two preset data sources based on the insurance product type and the multimodal information to obtain multi-source data of the object. This multi-source data of the object contains information related to both the insurance product type and the target object, and its information content is richer than that of the multimodal information. Next, a profile is constructed based on the multi-source data of the object to obtain a risk profile of the object, which can represent the risk information of the target enterprise in insurance application. Then, insurance rules are mapped based on the risk profile of the object to obtain an insurance plan, which can represent an insurance plan suitable for the target object. Finally, the insurance application template is filled in according to the insurance plan to obtain the target insurance application. In summary, this application can solve the problem of errors caused by missing information in the insurance application filling process and improve the accuracy of insurance application generation.
[0016] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description
[0017] Figure 1 This is a flowchart of the intelligent insurance application generation method provided in the embodiments of this application; Figure 2 yes Figure 1 The flowchart for step 102 in the document; Figure 3 yes Figure 2 The flowchart for step 201 in the document; Figure 4 yes Figure 1 The flowchart for step 103 in the document; Figure 5 yes Figure 4 The flowchart for step 402 in the document; Figure 6 yes Figure 1The flowchart for step 105 in the document; Figure 7 This is a block diagram of the module structure of the intelligent insurance application generation device provided in the embodiments of this application; Figure 8 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0019] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.
[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.
[0021] First, let's analyze some of the terms used in this application: Artificial intelligence (AI) is a new branch of computer science that studies, develops, and applies theories, methods, technologies, and systems to simulate, extend, and expand human intelligence. It aims to understand the essence of intelligence and produce intelligent machines that can react in a way similar to human intelligence. Research in this field includes robotics, speech recognition, intelligent insurance policy generation, natural language processing, and expert systems. AI can simulate the information processes of human consciousness and thought. Furthermore, AI utilizes digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceiving the environment, acquiring knowledge, and using that knowledge to achieve optimal results.
[0022] Natural Language Processing (NLP): NLP uses computers to process, understand, and utilize human language (such as Chinese and English). It is a branch of artificial intelligence and an interdisciplinary field of computer science and linguistics, often referred to as computational linguistics. NLP includes syntactic analysis, semantic analysis, and discourse understanding. It is commonly used in machine translation, handwritten and printed character recognition, speech recognition and text-to-speech conversion, information and image processing, information extraction and filtering, text classification and clustering, sentiment analysis, and opinion mining. It involves data mining, machine learning, knowledge acquisition, knowledge engineering, artificial intelligence research, and linguistic research related to language computation.
[0023] Large Language Model (LLM): refers to a deep learning model trained on large-scale text data that can understand, generate, and process natural language text. They typically possess powerful semantic understanding, reasoning, dialogue, and text generation capabilities, and are widely used in tasks such as question answering, writing assistance, translation, summarization, and code generation. The core features of large language models include: (1) Scalable parameters: containing hundreds of millions to trillions of parameters, trained on massive amounts of text. (2) End-to-end learning: directly mapping from input text to output text, with little or no explicit manual feature engineering. (3) Common architectures: encoder-decoder or decoder / bidirectional self-attention structures with Transformer as the core (such as variants of GPT and BERT).
[0024] The relevant technologies exhibit the following characteristics when generating insurance policies: 1. Isolated modal processing and lack of associative reasoning: Existing solutions can only handle single modalities. For example, OCR tools only recognize text in images and cannot associate it with instructions like "Please enter this form" in chat text; the accuracy of general speech recognition (ASR) transcription drops significantly in insurance-related scenarios (such as medical terminology and product names), and the transcribed text still requires manual semantic analysis. 2. Shallow semantic understanding and lack of domain knowledge: Traditional NLP entity recognition is trained on general corpora and cannot accurately understand the complex semantics of the insurance domain. For example, it cannot distinguish the referential relationship between "Policyholder: Zhang San" and "Insured: Zhang San" in the same dialogue, or understand that "Exemption from Clause C" specifically refers to a certain insurance liability. 3. Non-end-to-end process and automation breaks: Existing technologies are mostly isolated tools, and the recognized information still needs to be manually copied and pasted into the business system. When conflicts arise between different modal information (such as the ID number spoken in the voice not matching the image recognition), there is no automatic decision-making mechanism, relying entirely on manual judgment, leading to process interruptions.
[0025] Based on this, embodiments of this application propose an intelligent insurance application generation method, an intelligent insurance application generation device, an electronic device, and a computer-readable storage medium, which can solve the problem of errors caused by missing information during the insurance application filling process and improve the accuracy of insurance application generation.
[0026] The intelligent insurance policy generation method provided in this application relates to the field of deep learning technology in artificial intelligence. This method can be applied to terminals and servers, or it can be software running on the server. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The software can be an application that implements the intelligent insurance policy generation method, but it is not limited to the above forms.
[0027] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include server computers, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.
[0028] This application provides a method for intelligently generating insurance application forms, an apparatus for intelligently generating insurance application forms, an electronic device, and a computer-readable storage medium. The specific details are illustrated in the following embodiments. First, the method for intelligently generating insurance application forms in this application is described.
[0029] It should be noted that in each specific embodiment of this application, when it is necessary to process data related to the user's identity or characteristics, such as image data or dialogue data, the user's permission or consent will be obtained first. Moreover, the collection, use and processing of this data will comply with relevant laws, regulations and standards.
[0030] Reference Figure 1 , Figure 1This is an optional flowchart of the intelligent insurance policy generation method provided in the embodiments of this application, which may include, but is not limited to, steps 101 to 106.
[0031] Step 101: Obtain the multimodal information of the target object; Step 102: Perform template matching based on multimodal information to obtain the insurance application template and determine the insurance product type of the insurance application template; Step 103: Collect data from at least two preset data sources based on the insurance product type and multimodal information to obtain multi-source data of the object; Step 104: Construct a profile based on multi-source data of the object to obtain an object risk profile; Step 105: Map insurance rules based on the risk profile of the target to obtain an insurance plan; Step 106: Fill in the insurance application template according to the insurance plan to obtain the target insurance application.
[0032] In steps 101 to 106 of this embodiment, after obtaining the multimodal information of the target object, the application form is not directly filled based on the multimodal information. Instead, the insurance product type of the application form template is determined. Then, data is collected from at least two preset data sources based on the insurance product type and the multimodal information to obtain multi-source data of the object. This multi-source data contains information related to both the insurance product type and the target object, and its information content is richer than that of the multimodal information. Next, a profile is constructed based on the multi-source data of the object to obtain a risk profile of the object, which can represent the risk information of the target enterprise in insurance application. Then, insurance rules are mapped based on the risk profile of the object to obtain an insurance plan, which can represent an insurance plan suitable for the target object. Finally, the application form template is filled based on the insurance plan to obtain the target insurance application. In summary, this application can solve the problem of errors caused by missing information during the policy filling process and improve the accuracy of insurance application generation.
[0033] It should be noted that the intelligent insurance policy generation method provided in this application is applicable to insurance scenarios in the financial field, and it significantly improves the accuracy of generating insurance policies for enterprises, especially when the target is an enterprise.
[0034] In step 101 of some embodiments, multimodal information of the target object is obtained. The target object refers to the entity applying for insurance, such as a company applying for employer's liability insurance. Multimodal information includes information belonging to different modalities. For example, the modalities involved in multimodal information include: voice, image, text, tables, etc. The company's multimodal information includes: company insurance policies, voice conversations, text conversations, and documents containing employee lists.
[0035] In one example, the multimodal information of the target object may include: (1) voice / text: "I want to buy public liability insurance for my shopping mall, which is located at No. XX, XX Road. The daily foot traffic is about 5,000 people." (2) pictures: business license photo, shopping mall floor plan.
[0036] In one example, the multimodal information of the target may include: (1) a text dialogue: “Insure our company’s 50 office staff and 20 warehouse loaders with employer’s liability insurance.” (2) a file: an employee list in Excel format (including job type and monthly salary).
[0037] In one example, the multimodal information of the target object may include: (1) Voice: "I want to renew the insurance for my 3-year-old BMW 5 Series to extend the warranty." Images: vehicle registration certificate photo, last maintenance receipt.
[0038] In one example, the multimodal information of the target object may include: (1) text: “Insure my newly started factory with comprehensive property insurance, mainly for the factory buildings and equipment.” (2) documents: a list of fixed assets (including purchase price and installation date) and a floor plan of the factory buildings.
[0039] In one example, the multimodal information of the target object may include: (1) text: "Insure our logistics company's nationwide transportation business with logistics liability insurance, especially compensation for cargo damage and delays." (2) documents: description of business operation scope, freight volume data in previous years.
[0040] In step 102 of some embodiments, template matching is performed based on multimodal information to obtain an insurance application template, and the insurance product type of the application template is determined. The application template can be obtained and the insurance product type determined using a large language model based on multimodal information.
[0041] For example, when the multimodal information of the target object includes a text dialogue that reads "Insure our company's 50 office staff and 20 warehouse loaders with employer's liability insurance", an application form template related to the office staff and loaders can be identified, and the type of insurance product for the application form template can be identified as employer's liability insurance.
[0042] In one embodiment, reference is made to Figure 2 Step 102 may include: Step 201: Encode the multimodal information using a multimodal pre-trained model to obtain initial multimodal features; wherein, the initial multimodal features include initial text features of text, initial image features of images, and initial speech features of speech. Step 202: When text, image, and speech are in the same time window, perform cross-attention calculation on the initial text features, initial image features, and initial speech features to obtain the target multimodal features; Step 203: Filter the candidate templates according to the target multimodal features to obtain the insurance policy template.
[0043] Specifically, cross-attention is an attention mechanism used to establish informational connections between two different sequences or modalities, enabling one sequence to "pay attention" to relevant information in the other sequence. This embodiment utilizes the cross-attention mechanism to perform cross-attention calculations on initial text features, initial image features, and initial speech features, resulting in target multimodal features rich in feature information.
[0044] In one example, when processing a voice message, attention weights are automatically assigned to text and image information within the same time window to determine which information is relevant and credible. Inconsistencies are automatically resolved using preset confidence thresholds and conflict resolution algorithms (such as prioritizing the latest news or higher-resolution images).
[0045] The advantage of the above embodiments is that they realize intelligent association and collaborative reasoning between modalities, rather than simple splicing, and have a certain "judgment" ability.
[0046] In one embodiment, prior to step 201, an intelligent insurance policy generation method may further include: training a multimodal pre-trained model, specifically including: constructing a multimodal pre-trained model using massive amounts of insurance-related text, images (such as insurance policies and certificates), and speech data for pre-training. During the pre-training stage, tasks such as masked language models and image patch masking are introduced, enabling the model to learn the semantics of insurance terminology and its correspondence with visual information (such as the visual features of the text "ID card" and the image of an ID card). In this way, the model outputs no longer isolated text and visual features, but a unified semantic representation vector that integrates insurance-related knowledge, laying the foundation for deep semantic understanding.
[0047] In one example, the multimodal pre-trained model is trained on an existing model. There are multiple models to be trained, and after training, they are manually tested to select the model with the highest accuracy and correctness. The training process can be simplified as follows: (1) Multimodal data collection: insurance policies, communication records, pictures, voice; (2) Data security and compliance: de-identification, access control, auditing; (3) Data preprocessing and labeling: cleaning, de-identification, structuring; (4) Selecting existing basic models, such as DeepSeek, LLM, ViT, Whisper, etc.; (5) Multimodal pre-training with domain knowledge enhancement; (6) Multi-task instruction fine-tuning (MTL); (7) Human feedback reinforcement learning (RLHF); (8) Model evaluation and deployment; (9) Continuous learning and updating.
[0048] The multimodal pre-trained model includes: (1) a dedicated modality processing module (speech / image / text), which is responsible for converting the raw unstructured data into components of DeepSeek's understandable prompts. For example, the image module not only recognizes text with OCR, but also generates a short description of the image content, which together serve as the context for subsequent inputs. The output of this dedicated modality processing module includes preliminarily parsed text, image descriptions, and speech-to-text, which can serve as components of the prompt. (2) a core model (adapted to the insurance domain), which receives prompts from all modalities and uses its powerful semantic understanding and logical reasoning capabilities to perform referential resolution, information association, and conflict detection. For example, it understands that "this table" refers to the image that was just uploaded and determines whether the ID number is consistent. The output of this core model includes: a deep understanding of the user's intent, a fused structured information draft, and a report of potential conflict points. (3) an insurance domain knowledge inference engine, which works in collaboration with the model. The model's preliminary inference results are matched and decided upon again with the insurance knowledge base (such as terms and underwriting rules) and business rules (such as field validation). Ensure the output conforms to business specifications and resolve the potential "illusion" problem of DeepSeek. The output of this insurance domain knowledge inference engine may include: underwriting decision suggestions, mapping relationship between structured data and templates, and conflict alerts requiring manual review. (4) Structured data generator & API adapter: The natural language results generated by DeepSeek or the structured data produced by the inference engine are transformed into API call requests that meet the requirements of the core business system through a dynamic template matching engine. The output of this structured data generator & API adapter may include: the final structured insurance application form (such as JSON / XML) and request messages that can directly call the entry API.
[0049] In one embodiment, reference is made to Figure 3 Step 201 may include: Step 301: Perform intent recognition on multimodal information using a multimodal pre-trained model to obtain the intent corresponding to different modalities; Step 302: Extract information from multimodal information using a multimodal pre-trained model to obtain key information corresponding to different modalities; Step 303: Perform feature mapping on all intentions and all key information using a multimodal pre-trained model to obtain initial multimodal features.
[0050] In step 301, for example, if the multimodal information of the target object includes: (1) speech / text: "I want to buy public liability insurance for my shopping mall, which is located at No. XX, XX Road, and the daily foot traffic is about 5,000 people." (2) images: business license photo, shopping mall floor plan, then the multimodal pre-trained model can perform intent recognition on speech / text and obtain the intent to apply for public liability insurance. It can also perform intent recognition on images and obtain the intent to be related to business and shopping mall insurance (such as public liability insurance, comprehensive property insurance, etc.).
[0051] In step 302, for example, for the above multimodal information, the multimodal pre-trained model extracts information from the speech / text and can obtain key information including: business location: "shopping mall", address: "XX Road XX No.", business nature: "retail", daily customer flow: "5000", company name and registered capital (recognized from image OCR).
[0052] In step 303, initial multimodal features can be obtained by feature mapping of all intentions and all key information through the encoding layer in the multimodal pre-trained model.
[0053] The advantage of the above embodiments is that by utilizing the deep semantic understanding capabilities of the multimodal pre-trained model, the information content of the initial multimodal features is increased, which in turn improves the accuracy of feature encoding.
[0054] In step 103 of some embodiments, data is collected from at least two preset data sources based on the insurance product type and multimodal information to obtain object multi-source data.
[0055] In one embodiment, reference is made to Figure 4 Step 103 may include: Step 401: Activate the pre-deployed information intelligent collector; wherein the information intelligent collector is bound to at least two data sources; Step 402: Using an intelligent information collector, data is collected from various data sources based on insurance product type and multimodal information to obtain multi-source data of the object.
[0056] In step 401, the information intelligent collector can be an intelligent agent / processing module for information collection.
[0057] In step 402, when activating the intelligent information collector to collect information, it relies not only on the type of insurance product but also on multimodal information. When the target is a target enterprise, at least two data sources include business registration databases, public opinion systems, and geolocation information platforms.
[0058] In one embodiment, step 402 may include: acquiring object entity information (including enterprise entity information, such as "insured for XX Technology Co., Ltd.") from multimodal information; and collecting data from various data sources based on insurance product type and object entity information through an information intelligent collector to obtain multi-source object data.
[0059] The advantage of the above embodiments is that data collection is carried out by a specially designed intelligent information collector, which improves the efficiency and accuracy of data collection.
[0060] In one embodiment, the target object is the target enterprise, referring to... Figure 5 Step 402 may include: Step 501: Using an intelligent information collector, information is read from the business registration information database based on multimodal information to obtain the business registration information of the target company. Step 502: Using an intelligent information collector, information is read from the public opinion system based on insurance product type and multimodal information to obtain public opinion information; among which, public opinion information includes insurance public opinion information and corporate public opinion information; Step 503: Using an intelligent information collector, conduct environmental analysis on the geographic location information platform based on multimodal information to obtain the target company's corporate environmental risk information; Step 504: Information fusion is performed based on enterprise business registration information, public opinion information, and enterprise environmental risk information to obtain multi-source data of the object.
[0061] Specifically, the information intelligent collector invokes a built-in multi-source data collector to initiate collaborative queries to various authorized external databases and API interfaces. For example, it obtains registered capital, years of establishment, and industry classification from business registration databases; obtains clues about operational stability from public financial reports or public opinion systems; analyzes the natural environmental risks (such as whether it is located in a flood-prone area) and surrounding safety conditions from geolocation information platforms; and even analyzes the size and job composition (such as the proportion of manual labor positions) from recruitment websites.
[0062] The advantage of the above embodiments is that by acquiring multi-source data of the object from multiple dimensions, the adverse effects of insufficient multimodal information on the generation of insurance policies are reduced, thereby improving the accuracy of insurance policy generation.
[0063] In step 104 of some embodiments, a profile is constructed based on multi-source object data to obtain an object risk profile. Specifically, the collected multi-source object data is fed into a "risk profile construction engine," which outputs an object risk profile containing a structured enterprise risk label system through predefined rules and models. For example, the object risk profile is: "A high-tech enterprise, established for 5 years, with 200 employees, of which 40% are production line workers, located in the High-tech Zone of City A (a low-risk earthquake zone), with no major labor disputes reported last year."
[0064] In one embodiment, after step 104, the intelligent insurance application generation method may further include: performing gap detection on multimodal information and multi-source data of the object based on the required fields of the insurance application template to obtain gap fields; calling a small language model to infer the gap fields based on the object risk profile to obtain predicted field content; and updating the object risk profile based on the predicted field content.
[0065] Specifically, for key information that cannot be directly obtained (such as the exact number of employees in a specific job category or past records of minor work-related injuries), a small language model will be activated for deep reasoning. This small language model is trained on a massive amount of insurance cases and data, and can perform collaborative filtering and probabilistic inference based on the obtained enterprise risk profile. For example, if the "percentage of workers working at heights" cannot be directly obtained, the model can perform Bayesian inference based on the industry (such as "building exterior cleaning") and company size included in the enterprise risk profile, and refer to the average data of similar enterprises to provide a high-confidence estimate, thus completing the enterprise risk profile.
[0066] Step 105: Map insurance rules based on the object's risk profile to obtain an insurance plan. Specifically, the object's risk profile contains much deeper information than traditional methods, and this information can be mapped to insurance professional rules through a pre-deployed plan generation engine to obtain an insurance plan.
[0067] In one embodiment, reference is made to Figure 6 Step 105 may include: Step 601: Calculate the insurance premium based on the enterprise environmental risk information, total number of employees, job risk category, and the proportion of employees in each job risk category in the target risk profile. Step 602: Calculate the insured amount based on different job risk categories to obtain the insurance insured amount for each job risk category; Step 603: Generate personalized clauses based on the enterprise's environmental risk information and / or job risk categories to obtain personalized clauses; Step 604: Combine the insurance premium, the sum insured, and the personalized terms to obtain the insurance plan.
[0068] In step 601, premium calculation: Based on the enterprise's environmental risk information and total number of employees, the premium can be calculated in a differentiated and refined manner by using a large language model, combined with the job risk category (such as office staff, general workers, and high-altitude workers) and the inferred number of people.
[0069] In step 602, the insured amount is set: different death / disability liability limits can be programmed for employees with different job risk categories through a large language model.
[0070] In step 603, the terms stipulate that personalized special provisions can be automatically generated through a large language model. For example, for companies that are identified as having personnel engaged in "high-altitude operations," an "additional high-altitude operation liability clause" is automatically added; for companies located in specific areas, an "additional specific natural disaster liability clause" is added.
[0071] Finally, in step 604, the insurance premium, the sum insured, and the personalized terms are all treated as elements of the personalized plan, and the insurance plan is obtained by merging all the personalized plan elements.
[0072] The advantage of the above embodiments is that they enable the mapping from profiles to insurance rules, which helps improve the accuracy of insurance policy generation.
[0073] Step 106: Fill in the application form template according to the insurance plan to obtain the target application form. All personalized plan elements of the insurance plan will be automatically and accurately filled into the corresponding fields of the pre-selected application form template, completing the entire closed loop.
[0074] In one example, a dynamic template matching engine is designed on the backend. Based on key information such as the "type of insurance" identified from multimodal information, this engine automatically loads the corresponding JSON Schema or XML application template from the template library. Subsequently, the extracted structured data can be precisely mapped to the template fields through the data population module, and an API call request message (such as the body of an HTTP POST request) that meets the requirements of the core system can be automatically generated to directly complete the application entry.
[0075] In one example, a public liability insurance application includes the following: json { "policyHolder":{ "name":" / *Extracted from business license OCR* / ", "idNumber":" / *...* / " }, "insuredPremises":{ "address":"XX Road, XX Number", "nature":"retail", "dailyFootfall":5000 }, "coverage":{ "limitPerOccurrence":2000000, / / Preset based on risk characteristics "limitAnnualAggregate":5000000 }, "specialClauses":["Additional: Elevator Liability Clauses"] / / Automatically added when elevators are detected in the mall } .
[0076] In one example, an employer's liability insurance application includes the following: json { "enterpriseInfo":{ / *...* / }, "employeeDetails":{ "totalNumber":70, "byCategory":[ { "jobType":"Office worker", "count":50, "riskLevel":"Low", "sumInsuredPerPerson":600000, / / Liability limit for death / disability[7](@ref) "rate": 0.8‰ / / Example rate }, { "jobType":"Warehouse Loading and Unloading Worker", "count":20, "riskLevel":"High", "sumInsuredPerPerson":600000, "rate": 2.5‰, / / A higher rate is applied based on the risk level. "additionalMedicalCover":true / / Automatically add accidental medical coverage [7](@ref) } ] } } .
[0077] In one example, the car insurance extended policy application includes the following: json { "vehicleDetails":{ "vin":" / * Extracted from vehicle registration certificate OCR * / ", "brand":"BMW", "model":"5 series", "age":3 }, "extendedWarranty":{ "planType":"Preferred Package", / / Matches packages based on vehicle value "durationMonths":24, "coveredComponents":[ "engine", "transmission", "Transmission System" "Main electrical equipment" ], "terms": "Based on maintenance records, the deductible for the engine portion is waived" / / Personalized Special Offer } } .
[0078] In one example, a comprehensive property insurance policy includes the following: json { "insuredProperty":[ { "item":"factory", "sumInsured":50000000, / / Based on the fixed asset list "rate": 0.1‰, / / Example rate "insuredValueBasis": "Reset Value" }, { "item":"machinery and equipment", "sumInsured":30000000, "rate": 0.15‰, "insuredValueBasis": "Reset Value" } ], "totalSumInsured":80000000, "specialClauses":["Automatic coverage of new location clauses"] / / Automatically add [2](@ref) for plant operation characteristics } .
[0079] In one example, a logistics liability insurance policy includes the following: json { "logisticsCompanyInfo":{ "name":" / *Extracted from business license OCR* / ", "operationScope":"National" }, "coverageDetails":{ "mainLiability":{ "description": "Goods damage and delays", "limitPerOccurrence":500000, "limitAnnualAggregate":5000000 } }, "riskCharacteristics":{ "cargoType":"electronic products, etc." "annualFreightVolume":" / * Parse from business data* / " }, "specialClauses":["Additional: Error and Omission Clauses"] / / Automatically add [2](@ref) for logistics business features } .
[0080] Based on the above embodiments, this application can achieve at least the following beneficial effects: (1) Based on the multimodal information provided by the customer, a set of insurance plans that best suit the user's current situation is generated, and then the target insurance policy is obtained. The target is not simply filling in known fields, but generating a highly personalized insurance plan by actively mining and inferring the enterprise risk profile. (2) Quantitative improvement in processing efficiency and accuracy: The processing time of a single insurance policy is reduced from 5-15 minutes manually to within 10 seconds fully automatically by the system, with an efficiency improvement of more than 98%. Through multimodal cross-validation and domain knowledge enhancement, the accuracy of key fields (ID number, insured amount) input is increased from about 95% manually to more than 99.9%, eliminating data errors from the source and reducing subsequent error correction costs. (3) System performance and resource saving: Based on the unified representation of the pre-trained model, compared with the use of multiple independent AI services in series (such as general OCR + general ASR), the consumption of computing resources is reduced by about 30%, and the response latency is lower. The fully automated process can save more than 70% of the manual cost of primary data entry, allowing human resources to focus on high-value underwriting and risk control tasks.
[0081] Please see Figure 7 This application also provides an intelligent insurance application generation device, which can implement the above-mentioned intelligent insurance application generation method. Figure 7 The present invention provides a module structure block diagram of an intelligent insurance policy generation device. The device includes: an information acquisition module 701 for acquiring multimodal information of a target object; a template matching module 702 for performing template matching based on the multimodal information to obtain an insurance policy template and determining the insurance product type of the insurance policy template; a data collection module 703 for collecting data from at least two preset data sources based on the insurance product type and multimodal information to obtain multi-source data of the object; a profile construction module 704 for constructing a profile based on the multi-source data of the object to obtain a risk profile of the object; a rule mapping module 705 for mapping insurance rules based on the risk profile of the object to obtain an insurance plan; and a policy filling module 706 for filling the insurance policy template according to the insurance plan to obtain the target insurance policy.
[0082] In one embodiment, the intelligent insurance application generation device may further include a profile update module, used for: performing gap detection on multimodal information and multi-source data of the object according to the required fields of the insurance application template to obtain gap fields; calling a small language model to infer the gap fields based on the object risk profile to obtain predicted field content; and updating the object risk profile according to the predicted field content.
[0083] It should be noted that the specific implementation method of the intelligent insurance application generation device is basically the same as the specific implementation method of the intelligent insurance application generation method described above, and will not be repeated here.
[0084] This application also provides an electronic device, which includes: a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for communication between the processor and the memory. When the program is executed by the processor, it implements the above-described intelligent insurance policy generation method. This electronic device can be any intelligent terminal, including tablet computers, in-vehicle computers, etc.
[0085] Please see Figure 8 , Figure 8 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 801 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 802 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 802 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 802 and is called and executed by the processor 801 using the intelligent insurance policy generation method of the embodiments of this application. The 803 input / output interface is used to implement information input and output. The communication interface 804 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 805 transmits information between various components of the device (e.g., processor 801, memory 802, input / output interface 803, and communication interface 804); The processor 801, memory 802, input / output interface 803, and communication interface 804 are connected to each other within the device via bus 805.
[0086] This application embodiment also provides a storage medium, which is a computer-readable storage medium for computer-readable storage. The storage medium stores one or more programs, which can be executed by one or more processors to implement the above-described intelligent insurance policy generation method.
[0087] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.
[0088] The intelligent insurance application generation method, device, electronic equipment, and storage medium provided in this application, after obtaining the multimodal information of the target object, do not directly fill in the insurance application based on the multimodal information. Instead, they determine the insurance product type of the insurance application template, and then collect data from at least two preset data sources based on the insurance product type and multimodal information to obtain multi-source data of the object. This multi-source data contains information related to both the insurance product type and the target object, and its information content is richer than that of the multimodal information. Next, a profile is constructed based on the multi-source data of the object to obtain a risk profile of the object, which can represent the risk information of the target enterprise in insurance application. Then, insurance rules are mapped based on the risk profile of the object to obtain an insurance plan, which can represent an insurance plan suitable for the target object. Finally, the insurance application template is filled in according to the insurance plan to obtain the target insurance application. In summary, this application can solve the problem of errors caused by missing information during the insurance application filling process and improve the accuracy of insurance application generation.
[0089] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.
[0090] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.
[0091] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.
[0092] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.
[0093] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0094] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.
[0095] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.
[0096] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0097] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0098] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0099] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.
Claims
1. A method for intelligently generating insurance application forms, characterized in that, The method includes: Obtain multimodal information of the target object; Template matching is performed based on the multimodal information to obtain an insurance application template, and the type of insurance product in the insurance application template is determined. Data is collected from at least two preset data sources based on the insurance product type and the multimodal information to obtain object multi-source data; A risk profile of the object is constructed based on the multi-source data of the object. Based on the risk profile of the object, insurance rules are mapped to obtain an insurance plan; The target insurance application is obtained by filling in the application form template according to the insurance plan.
2. The method according to claim 1, characterized in that, The step involves collecting data from at least two preset data sources based on the insurance product type and the multimodal information to obtain multi-source data for the object, including: Activate a pre-deployed intelligent information collector; wherein the intelligent information collector is bound to at least two of the data sources; The information intelligent collector collects data from each of the data sources based on the insurance product type and the multimodal information to obtain the object's multi-source data.
3. The method according to claim 2, characterized in that, The target object is the target enterprise, and at least two of the data sources include a business registration information database, a public opinion system, and a geographic location information platform. The intelligent information collector collects data from each of the data sources based on the insurance product type and the multimodal information to obtain multi-source data of the target enterprise, including: The intelligent information collector reads information from the business registration information database based on the multimodal information to obtain the business registration information of the target enterprise. The information intelligent collector reads information from the public opinion system based on the insurance product type and the multimodal information to obtain public opinion information; wherein, the public opinion information includes insurance public opinion information and corporate public opinion information; The information intelligent collector performs environmental analysis on the geographic location information platform based on the multimodal information to obtain the target enterprise's corporate environmental risk information. The multi-source data of the object is obtained by fusing information from the enterprise's business registration information, public opinion information, and environmental risk information.
4. The method according to any one of claims 1 to 3, characterized in that, After constructing the object risk profile based on the multi-source data of the object, the method further includes: Based on the requirement fields of the insurance application template, gap detection is performed on the multimodal information and the multi-source data of the object to obtain the gap fields; The small language model is invoked to infer the gap field based on the object risk profile, and the predicted field content is obtained; The risk profile of the object is updated based on the content of the predicted field.
5. The method according to any one of claims 1 to 3, characterized in that, The target object is the target enterprise. The process of mapping insurance professional rules based on the risk profile of the target enterprise to obtain an insurance plan includes: The insurance premium is calculated based on the enterprise environmental risk information, total number of employees, job risk category, and the proportion of employees in the job risk category in the risk profile of the target. The insured amount is calculated based on the different job risk categories to obtain the insurance insured amount for each job risk category; Personalized terms are generated based on the enterprise environmental risk information and / or the job risk category; The insurance plan is obtained by integrating the insurance premium, the insurance coverage amount, and the personalized terms.
6. The method according to any one of claims 1 to 3, characterized in that, The step of performing template matching based on the multimodal information to obtain the insurance application template includes: The multimodal information is feature-encoded using a multimodal pre-trained model to obtain initial multimodal features; wherein, the initial multimodal features include initial text features of text, initial image features of images, and initial speech features of speech. When the text, the image, and the speech are in the same time window, cross-attention calculation is performed on the initial text features, the initial image features, and the initial speech features to obtain the target multimodal features; The candidate templates are filtered based on the target multimodal features to obtain the insurance policy template.
7. The method according to claim 6, characterized in that, The step of encoding the multimodal information using a multimodal pre-trained model to obtain initial multimodal features includes: The multimodal information is used to identify intents through the multimodal pre-trained model to obtain intents corresponding to different modalities; The multimodal information is extracted using the multimodal pre-trained model to obtain key information corresponding to different modalities; The initial multimodal features are obtained by performing feature mapping on all the intents and all the key information through the multimodal pre-trained model.
8. An intelligent insurance application generation device, characterized in that, The device includes: The information acquisition module is used to acquire multimodal information about the target object; The template matching module is used to perform template matching based on the multimodal information to obtain an insurance application template and determine the insurance product type of the insurance application template. The data collection module is used to collect data from at least two preset data sources based on the insurance product type and the multimodal information to obtain multi-source data of the object. The profile building module is used to build a profile based on the multi-source data of the object to obtain a risk profile of the object. The rule mapping module is used to map insurance rules based on the risk profile of the object to obtain an insurance plan; The policy filling module is used to fill the application form template according to the insurance plan to obtain the target application form.
9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 7.
10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.