Claim settlement case report processing method and device, electronic equipment and storage medium

Abstract

The invention belongs to the technical field of insurance claim settlement, and relates to a claim settlement report processing method and device, electronic equipment and a storage medium. A claim settlement document image uploaded by a user is acquired, identification and analysis are performed based on a multi-mode large model, an optical character identification model and a reasoning large model, document type information, accident information, accident type information, doctor seeing type information and accident reason information are acquired, a target insurance policy is matched according to the information, and a claim settlement report application form is automatically generated. On the basis, high-precision classification and key field extraction of various claim settlement documents are achieved through multi-model cooperation, the semantic understanding ability is achieved, the method can adapt to materials of different formats and insurance types, and the recognition accuracy and classification precision are remarkably improved. Intelligent analysis and form automatic filling can be completed in the claim application link, user experience is optimized, manual intervention is reduced, processing time is shortened, meanwhile, insurance responsibility intelligent association judgment is achieved, and the accuracy and automation level of claim application are improved.

Description

Technical Field

[0001] This application belongs to the field of insurance claims technology, specifically relating to a claim reporting processing method, device, electronic device, and storage medium. Background Technology

[0002] With the intelligent development of insurance claims processing, more and more insurance institutions are using OCR technology to automatically parse and classify claims documents to improve the efficiency of claims data processing. Existing technical solutions typically use OCR models to recognize document content and classify it based on keywords, achieving automatic filling of some fields.

[0003] However, in the process of developing this application, the inventors discovered at least the following problems with the existing technology: First, traditional OCR models rely heavily on keyword matching and lack the ability to understand the semantics of documents, resulting in low recognition accuracy when faced with non-standardized medical documents. Second, existing solutions are typically designed for specific types of documents (such as invoices, car insurance photos, etc.), making it difficult to adapt to the diverse claim materials under different insurance types. Third, some technologies are mainly applied to the claims review stage, resulting in insufficient automation for users in the claims application process. Therefore, the existing technology still has shortcomings in terms of recognition accuracy, document adaptability, and user experience. Summary of the Invention

[0004] To address the aforementioned issues, this application provides a claims reporting processing method that, through multi-model collaborative recognition and semantic reasoning, achieves high-precision classification and automatic extraction of key information for different claims documents, thereby significantly improving recognition accuracy and the efficiency of automated form filling.

[0005] To address the aforementioned technical problems, one technical solution adopted in this application is: providing a claims reporting processing method, the method comprising: acquiring a claims document image uploaded by a user; recognizing the claims document image based on a preset multimodal big data model and an inference big data model to obtain document type information of the claims document image; and executing a claims reporting application process when the claims document image corresponding to the document type information meets a preset first verification condition, the claims reporting application process comprising: parsing the claims document image corresponding to the document type information based on the multimodal big data model to obtain the user's claims document type information. The system analyzes the claim document images corresponding to the document type information, including claim information and claim type information. Based on a pre-defined optical character recognition (OCR) model and a multimodal inference model, it analyzes the claim document images corresponding to the document type information to obtain the user's claim cause information. Based on the claim type information and claim cause information, it matches the target policy. Based on the claim type information, claim cause information, claim information, and claim type information, it populates the target policy to obtain a claim report application form.

[0006] In some embodiments, based on the preset multi-modal large model and the inference large model, the claim document image is recognized to obtain the document type information of the claim document image, including: based on the multi-modal large model, the claim document image is detected to obtain the text feature information corresponding to the claim document image; the text feature information is verified; if the text feature information satisfies the preset second verification condition, the document type information of the claim document image whose text feature information satisfies the second verification condition is obtained according to the text feature information; if the text feature information does not satisfy the second verification condition, the text feature information is inferred based on the inference large model to obtain the document type information of the claim document image whose text feature information does not satisfy the second verification condition.

[0007] In some embodiments, the document type information includes medical record document information and discharge summary document information, and based on the multi-modal large model, the claim document image corresponding to the document type information is analyzed to obtain the user's risk information and risk type information, including: based on the multi-modal large model, the claim document image corresponding to the document type information is recognized to obtain the user's risk information; when the risk information satisfies the preset third verification condition, the inference large model is used to infer the risk information to obtain the user's risk type information.

[0008] In some embodiments, the document type information includes invoice document information, and based on the preset optical character recognition model and the inference large model, the claim document image corresponding to the document type information is analyzed to obtain the user's treatment type information, including: based on the optical character recognition model, the claim document image corresponding to the invoice document information is recognized to obtain the invoice text information of the claim document image corresponding to the invoice document information; when the invoice text information satisfies the preset fourth verification condition, the inference large model is used to infer the invoice text information to obtain the user's treatment type information.

[0009] In some embodiments, the document type information includes hospitalization expense item document information and examination report document information, and based on the optical character recognition model, the multi-modal large model and the inference large model, the claim document image corresponding to the document type information is analyzed to obtain the user's risk reason information, including: based on the optical character recognition model, the claim document image corresponding to the hospitalization expense item document information is recognized to obtain the hospitalization expense item information; based on the multi-modal large model, the claim document image corresponding to the examination report document information is recognized to obtain the examination report information; based on the inference large model, the hospitalization expense item information and the examination report information are inferred to obtain the user's risk reason information.

[0010] In some embodiments, according to the clinic type information and the accident cause information, the target insurance policy is matched, including: based on a preset mapping relationship strategy, the clinic type information and the accident cause information, at least one of the outpatient insurance policy, the hospitalization insurance policy and the drug purchase insurance policy is screened as the target insurance policy.

[0011] In some embodiments, according to the accident type information, the accident cause information, the accident information and the clinic type information, the target insurance policy is filled in to obtain a claim report application form, including: based on the multi-modal large model, the claim document image is recognized to obtain the identity information of the user; based on the preset template mapping model, the identity information, the accident type information, the accident cause information, the accident information and the clinic type information, the target insurance policy is filled in to obtain the claim report application form.

[0012] To solve the above technical problems, another technical solution adopted by the embodiments of the present application is to provide a claim report processing device, the device comprising: an image acquisition module, the image acquisition module being configured to acquire a claim document image uploaded by a user; a document type identification module, the document type identification module being configured to identify the claim document image based on a preset multi-modal large model and an inference large model to obtain document type information of the claim document image; and a claim report application module, the claim report application module being configured to execute a claim report application process when the claim document image corresponding to the document type information meets a preset first inspection condition, the claim report application process comprising: based on the multi-modal large model, the claim document image corresponding to the document type information is analyzed to obtain accident information and accident type information of the user; based on a preset optical character recognition model and the inference large model, the claim document image corresponding to the document type information is analyzed to obtain clinic type information of the user; based on the optical character recognition model, the multi-modal large model and the inference large model, the claim document image corresponding to the document type information is analyzed to obtain accident cause information of the user; according to the clinic type information and the accident cause information, a target insurance policy is matched; and according to the accident type information, the accident cause information, the accident information and the clinic type information, the target insurance policy is filled in to obtain a claim report application form.

[0013] To solve the above technical problems, still another technical solution adopted by the embodiments of the present application is to provide an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the above-mentioned method.

[0014] To solve the above technical problems, the application further provides a non-volatile computer readable storage medium storing computer executable instructions, when the computer executable instructions are executed by an electronic device, the electronic device executes the method described above.

[0015] Different from the related art, the application provides a claim settlement report case processing method and device, an electronic device, and a storage medium. By obtaining claim settlement document image uploaded by a user, identifying the claim settlement document image based on a preset multi-modal large model and an inference large model to obtain document type information of the claim settlement document image, when the claim settlement document image corresponding to the document type information meets a preset first inspection condition, a claim settlement report case application process is executed, which includes: based on the multi-modal large model, analyzing the claim settlement document image corresponding to the document type information to obtain accident information and accident type information of the user; based on a preset optical character recognition model and the inference large model, analyzing the claim settlement document image corresponding to the document type information to obtain medical treatment type information of the user; based on the optical character recognition model, the multi-modal large model, and the inference large model, analyzing the claim settlement document image corresponding to the document type information to obtain accident cause information of the user; matching a target insurance policy according to the medical treatment type information and the accident cause information; and filling the target insurance policy according to the accident type information, the accident cause information, the accident information, and the medical treatment type information to obtain a claim settlement report case application form. Based on this, the application realizes high-precision automatic classification and key field information extraction of various claim settlement documents through the collaborative application of the multi-modal large model, the optical character recognition model, and the inference large model. Compared with the prior art which only relies on keyword matching or single OCR recognition, the application can not only understand the semantic information of the document, but also adapt to claim settlement materials under different formats and different insurance types, thereby significantly improving the recognition accuracy and classification precision. At the same time, the intelligent analysis of the user-uploaded document and the automatic filling of the form can be completed at the claim settlement application link, which optimizes the user operation experience, reduces manual intervention, shortens the claim settlement application processing time, and improves the overall claim settlement efficiency. In addition, the combination of multi-model parallel cooperation and inference capability can also automatically match appropriate insurance policies according to the accident information, the medical treatment type information, and the accident cause information, realize intelligent association judgment of insurance liability, and further enhance the accuracy and automation level of the claim settlement application, thereby effectively solving the deficiencies of the prior art in recognition accuracy, bill adaptability, and user experience. BRIEF DESCRIPTION OF DRAWINGS

[0016] One or more embodiments are illustrated by way of example with reference to the accompanying drawings, which are schematic and not intended to be limiting of the embodiments, and in which like numerals refer to like elements throughout the drawings. The drawings are not limiting as to scale of proportions, as the dimensions of the various layers, regions, and elements have been exaggerated for clarity.

[0017] Figure 1 is a schematic diagram of an application scenario of a claim settlement report processing method provided by an embodiment of the present application; Figure 2 is a flowchart of a claim settlement report processing method provided by an embodiment of the present application; Figure 3 is a flowchart of a claim settlement report application process provided by an embodiment of the present application; Figure 4 is a schematic diagram of an interaction process of a claim settlement report processing method provided by an embodiment of the present application; Figure 5 is a schematic diagram of a claim settlement report application page provided by an embodiment of the present application; Figure 6 is a schematic diagram of a claim settlement report application page provided by another embodiment of the present application; Figure 7 is a structural schematic diagram of a claim settlement report processing apparatus provided by an embodiment of the present application; Figure 8 is a hardware structural schematic diagram of an electronic device for executing a claim settlement report processing method provided by an embodiment of the present application. DETAILED DESCRIPTION

[0018] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in detail with reference to the accompanying drawings. Obviously, the described embodiments are some but not all of the embodiments of the present application. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application.

[0019] It should be noted that, if there is no conflict, each feature in the embodiments of the present application can be combined with each other, and all within the protection scope of the present application. In addition, although the functional modules are divided in the device schematic diagram, and the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order from the module division in the device schematic diagram or the order in the flowchart.

[0020] The terms "first", "second", and the like in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are usually a class, and do not limit the number of objects, for example, the first object can be one or more.

[0021] Unless otherwise defined, all technical and scientific terms used in the present disclosure have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the description of the application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0022] Traditional insurance claim and reporting processes highly rely on user filling in claim reporting application forms manually and uploading paper or electronic documents one by one. The existing technology usually uses a single OCR model or keyword-based rules to parse and classify bills, which can obtain good results on structured invoice documents, but when facing diversified and non-standard medical documents such as hospital records, expense details, test reports, and discharge summaries, the recognition rate and classification accuracy decrease significantly. At the same time, single OCR or keyword rules lack understanding of the semantics of the documents and the cause-effect relationship of the accident, making it difficult to automatically determine the accident type, match the liability clause, or complete the policy liability verification, resulting in heavy burden on the user to fill in the forms and poor experience. A large amount of manual review is costly and inefficient, and is prone to misjudgment or omission, affecting the speed and compliance of claim settlement. Further, existing solutions focus on automatic review at the enterprise end, and it is difficult to achieve end-to-end automatic document classification, field extraction, and policy matching capabilities at the user end.

[0023] To solve the above problems, the present application proposes an overall technical architecture in which multi-modal large models, reasoning large models, and self-developed / third-party OCR models are complementary and parallelly scheduled, which runs through the end-to-end process from the user reporting entry to the generation of the claim application form. The specific idea includes: first, when the user uploads the documents, the preset multi-modal large model is used to perform overall semantic perception and document type judgment on the image (for example, invoice, expense list, test report, medical record, discharge summary, etc.); then, the system triggers parallel model combination calling according to the document type - for standardized bills, high-precision OCR models are used for field extraction, and for complex and unstructured medical documents, multi-modal understanding models and reasoning large models are used in parallel to make up for the shortcomings of OCR recognition (for example, reasoning about the type of visit, standardizing the names of diseases with abnormal writing, and inferring the cause and timeline of the accident); next, the information obtained by extraction and reasoning (visit time, hospital, diagnosis name, visit type, etc.) is input into the test judgment module based on the policy logic, and the responsibility matching, waiting period verification, policy validity period, and mandatory document determination are automatically completed according to the policy clauses, so as to automatically select or recommend the target policy that meets the conditions for the user and fill in the corresponding fields in the claim application form; finally, in a multi-terminal environment (WeChat mini-program, APP, background control system), the capability is provided uniformly with parallel interfaces and fault-tolerant strategies, supporting batch document classification, policy information detection and matching, and supplementary prompts.

[0024] It can be understood that the core of the present concept is: first, the multi-modal large model (for semantic understanding and reasoning) and the high-precision OCR model (for accurate field identification) are intelligently arranged in parallel according to task ability, which makes up for the short board of single model in identifying diversified medical documents; second, the closed-loop document inspection judgment logic covering the front-end user experience to the back-end compliance verification is designed, which automatically judges whether the document is necessary, the time and diagnosis meet the insurance clauses under the responsibility framework, so as to realize the end-to-end closed loop from "recognition" to "responsibility matching" to "automatic reporting"; third, the parallel application interface scheme of multiple types and sizes of models is proposed, so that the ability can be seamlessly deployed in WeChat mini program, APP and background system, meeting the delay, robustness and security requirements of production environment.

[0025] Please refer to Figure 1 , Figure 1 is an application scenario diagram of a claim handling method provided by an embodiment of the present application. The application scenario includes a user end and a server end. The user end can access the online claim system 100 through various forms of mobile terminal interfaces such as WeChat mini program, APP or web page end. In the server end, the online claim system 100 is deployed in a cloud server or an enterprise internal server, which is used to realize the online automatic processing of insurance claim.

[0026] As shown in Figure 1 , the online claim system 100 includes a claim application form page module 110, a document identification module 120 and an information reasoning module 130.

[0027] The claim application form page 110 is used to provide a visual interactive interface for the user to apply for a claim, which belongs to the front-end interactive layer of the online claim system 100. The page can run in WeChat mini program, mobile application or web page, which is used to guide the user to upload claim materials, select insurance policies and confirm the claim, etc.

[0028] The claim application form page 110 specifically includes: a material uploading sub-page for receiving the claim document images uploaded by the user, including medical invoices, medical records, discharge summaries, expense detail lists and other image files; a policy selection sub-page for displaying the insurance policy information held by the user and supporting automatic matching of target claimable policies; a form filling sub-page for automatically presenting the claim application form generated by the system after the uploaded materials are analyzed by the document identification module 120 and the information reasoning module 130, and allowing the user to confirm or supplement the input if necessary.

[0029] Through the above structural design, the claim application form page 110 can realize efficient linkage between the user and the background intelligent identification module in the front end, so that the user can complete the automatic process of claim application without manually entering complex information such as medical treatment, accident, insurance policy, etc.

[0030] The single document recognition module 120 is the basic perception layer of the online claim settlement system 100, and is used for recognizing and classifying the claim settlement document image uploaded by the user. The module can call a multi-modal large model, an optical character recognition (OCR) model and an inference large model to work cooperatively, so as to realize the structured understanding of multi-source information.

[0031] The single document recognition module 120 specifically includes the following functional units: a single document classification unit, which jointly analyzes the visual features and text features of the claim settlement document image based on a multi-modal large model, and judges that the image belongs to an invoice, a medical record, a fee list, a discharge summary or the like; a text extraction unit, which extracts the text information on the face of a bill (such as a public hospital invoice, a value-added tax invoice or the like) having a relatively fixed format by using an OCR model, including the key fields of the time of visit, the amount, the name of the hospital and the like; a field verification unit, which verifies the correctness of the format, the time and the amount of the extracted fields, so as to ensure that the recognition result meets the preset first verification condition; and a semantic analysis unit, which inputs the OCR result into an inference large model (such as an LLM) by using a prompt engineering technology, so as to generate semantic structured data and provide an input basis for the subsequent information inference module 130.

[0032] Through the above design, the single document recognition module 120 can realize high-precision classification and content extraction of different types of claim settlement documents under the multi-model cooperative mechanism, and can make up for the low recognition accuracy of non-standard medical documents by the traditional OCR scheme.

[0033] The information inference module 130 is the core intelligent layer of the online claim settlement system 100, and is used for automatically inferring the type of visit, the type of accident and the cause of accident, and matching the insurance policy responsibility, on the basis of the structured information output by the single document recognition module 120, by using an inference large model.

[0034] The information reasoning module 130 comprises the following functional units: a treatment type reasoning unit: according to the bill content (including invoices, medical records, examination sheets, prescriptions, etc.) parsed by the document identification module 120, the treatment behavior type of the user is judged, and scenarios such as outpatient, hospitalization and drug purchase in a pharmacy are distinguished; an accident type identification unit: based on the semantic understanding ability of the multi-modal large model, combined with the drug names, examination items and diagnosis contents involved in the bill, the accident type (disease class or accident class) of the user is reasoned out; an accident reason reasoning unit: through the context semantic association reasoning ability of the large model, the specific accident reason (such as pneumonia, cold and fever, falling injury, animal scratch and bite, etc.) is inferred; a policy matching unit: according to the reasoned treatment type information and accident reason information, the target policy held by the user is automatically searched and matched, and the policy clause that meets the current claim liability scope is determined; a form filling unit: the obtained treatment type information, accident information, accident type information, accident reason information and matched policy information are integrated to generate a claim report application form, and are returned to the claim report application form page 110 for display and confirmation.

[0035] Through the above structural design, the information reasoning module 130 can convert multi-source bill information into structured data with insurance claim semantics, realizing intelligent decision and automatic information filling in the claim application link.

[0036] In the online claim system 100 provided by the present application, the claim report application form page 110, the document identification module 120 and the information reasoning module 130 cooperate with each other to form a complete closed loop from user image uploading to automatic generation of a claim form. The implementation of the system effectively reduces the complexity of the claim report operation of the insurance customer, shortens the claim application time, and significantly improves the accuracy of policy matching and field recognition.

[0037] The specific implementation process of the claim report processing method provided by the present application will be described in detail below through specific embodiments.

[0038] Please refer to Figure 2 , Figure 2 is a flowchart of a claim report processing method provided by an embodiment of the present application. The method is applied to the online claim system 100 described above. As shown in Figure 2 , the method comprises steps S11-S13: S11: Obtain the claim document image uploaded by the user.

[0039] First, users choose a claim method. Once the user enters the claim application process, the claim report application page 110 will display multiple claim methods for the user to choose from. One is the traditional policy field filling mode, where the user enters fields such as the time of medical treatment, cause of the accident, name of the medical institution, and invoice amount according to a fixed template. The other is the intelligent claim AI assistant mode, where the user authorizes the system to call the document recognition module 120 and the information reasoning module 130 to automatically collect information and report the claim by uploading images of claim documents. Through this design, users can choose the interaction method according to their actual needs, and the intelligent claim AI assistant mode can complete the subsequent automated claim reporting process without manual input.

[0040] Secondly, the system identifies and matches policy requirements with documentation. When a user selects to enable the intelligent claims AI assistant mode, the system calls the backend policy management service to automatically retrieve and display the user's current insurance policy information based on the information provided by the user (such as WeChat ID, mobile phone number, or policy number). After obtaining the insurance policy information, the system automatically matches the required documentation for the selected policy type (such as medical insurance, accident insurance, hospitalization allowance insurance, etc.). For example, if the selected policy is medical insurance, the required documentation includes: medical invoices, expense details, medical record cover page, discharge summary, etc.; if the selected policy is accident insurance, the required documentation includes: accident certificate, diagnosis certificate, invoice, etc. The system sends the matching results to the claims application page 110, dynamically generating a "Required Documents Upload List," guiding the user to upload the corresponding claims documents in one go according to the list.

[0041] Finally, the document images are uploaded. After selecting all the required documents on the claim report application page 110, the user submits all claim document images at once through the image upload interface. Optionally, after receiving the claim document images, the document recognition module 120 can perform image quality detection and layout recognition to remove blurry, duplicate, or non-document types.

[0042] S12: Based on the preset multimodal big model and inference big model, the claim document image is identified to obtain the document type information of the claim document image.

[0043] The claim document image is identified based on the preset multi-modal large model and the inference large model to obtain claim document image information, including: detecting the claim document image based on the multi-modal large model to obtain text feature information corresponding to the claim document image; checking the text feature information; if the text feature information meets the preset second checking condition, obtaining the claim document image information of the claim document image whose text feature information meets the second checking condition according to the text feature information; and if the text feature information does not meet the second checking condition, inferring the text feature information based on the inference large model to obtain the claim document image information of the claim document image whose text feature information does not meet the second checking condition.

[0044] The system identifies the claim document image uploaded by the user based on the preset multi-modal large model and the inference large model to obtain the claim document image corresponding to the claim document image. The "claim document image information" refers to the category label of each type of document required for claim reporting, such as public hospital medical invoice, value-added tax medical invoice, medical record, discharge summary, hospitalization expense detailed list, and other types. The system automatically classifies the content of the claim document image into the corresponding category to facilitate subsequent structured information extraction and automatic filling of the claim application form. The specific implementation process is as follows: The system identifies the claim document image based on the preset multi-modal large model. The multi-modal large model refers to an artificial intelligence model that can understand both visual information (such as image content, layout structure) and language information (such as ticket text, header semantics). This model can jointly analyze the text and semantic features in the image under conditions such as no fixed format, complex background, or low clarity. In this embodiment, the multi-modal large model can be a model based on the fusion architecture of VisionTransformer (ViT) and large language model (LLM), such as MiniGPT-4, Qwen-VL, InternVL, GPT-4V, or a custom multi-modal recognition network trained for insurance document scenarios, which is not limited here.

[0045] To guide the model to extract more accurate target information, the system designs a targeted text prompt (Prompt) before identification, which is used to indicate that the model should focus on key elements related to claims, such as "patient name", "hospital name", "ticket header", "monitoring unit", etc. Through the identification process of the model, the text feature information corresponding to the claim document image is obtained, which is all the text and its structural semantics extracted from the image.

[0046] The system performs feature checking on the extracted text feature information, that is, judges whether the text feature meets the preset second checking condition. The "second checking condition" refers to a condition set for determining whether the text feature is sufficient to directly determine the type of the single document. For example, when there are explicit feature words such as "Treasury Department supervision", "outpatient charge ticket", "invoice code", "charge ticket" in the text feature, it can be directly judged that the image belongs to the "public hospital medical invoice" type; when there are fields such as "value-added tax special invoice", "tax amount", "buyer name", etc., it can be judged as "value-added tax medical invoice". If the text feature information meets the second checking condition, the system directly determines the corresponding single document type information according to the text feature information.

[0047] If the text feature information does not meet the second checking condition, that is, the image lacks obvious keywords or the ticket format is blurred, the system calls the reasoning large model for further judgment. The reasoning large model refers to a large language model with semantic understanding and logical inference ability, which can complete type reasoning based on context semantic features. Unlike the multi-modal large model, which focuses on "image understanding", the reasoning large model focuses more on mining implicit semantic relationships from the extracted text information. In this embodiment, the reasoning large model can use language models with strong semantic reasoning ability such as GPT-4, Qwen2, GLM-4, ChatGLM3, etc., or industry reasoning models fine-tuned in the insurance claim field, without limitation.

[0048] For example, when the image contains a large number of drug names, charge items and amount lists, but does not explicitly mark the "invoice" word, the reasoning large model can determine that the single document belongs to "inpatient expense itemized list" through semantic inference; for example, if the image contains text elements such as "diagnosis result", "complaint", "medical advice", etc., it can be judged as "medical record" or "discharge summary". The system finally obtains the single document type information of the claim single document image whose text feature information does not meet the second checking condition through the auxiliary inference of the reasoning large model.

[0049] After multi-modal recognition and reasoning inference, the system outputs and structures the single document type information corresponding to all image documents. Each claim single document image is labeled with a specific category label, which is used to automatically extract the data content of the corresponding column in the subsequent steps, and realizes the automatic filling of the claim application form.

[0050] In this embodiment, through the cooperative recognition mechanism of the multi-modal large model and the reasoning large model, high-precision automatic classification and intelligent type determination of the claim document image are realized, and the automation degree and recognition robustness of the claim process are significantly improved. Specifically, the multi-modal large model can accurately extract the bill text and its semantic features from complex document images under the condition of no fixed format and different image quality, realize the rapid recognition and direct classification of significant feature samples; and the reasoning large model further makes up for the shortcomings of traditional OCR or shallow classification algorithms in the context of ambiguous semantics and missing text, and realizes intelligent classification of image pieces lacking obvious features through context semantic understanding and logical inference. The combination of the two enables the system to maintain high stability and high accuracy when facing claim image pieces of different sources and formats, effectively reducing manual intervention, reducing the labor cost and error rate of the claim audit link, and providing accurate and structured input basis for subsequent automatic field extraction and form filling, thereby improving the intelligent level and user experience of the claim business as a whole.

[0051] S13: When the claim document image corresponding to the document type information meets the preset first verification condition, a claim report application process is performed.

[0052] When the claim document image corresponding to the document type information is classified, the system determines whether the claim document image meets the preset first verification condition to determine whether to start the claim report application process. The "first verification condition" refers to a set of preliminary verification rules defined for the document type information, including but not limited to: whether the document type matches the material type required by the claim application (for example, invoice documents should be outpatient or inpatient medical invoices, medical records should be outpatient medical records or discharge summaries); whether the key field information of the document is complete (such as whether the patient's name, treatment time, and hospital name are complete); whether the time involved in the document falls within the effective period of the policy; and whether the document meets the requirements of the insurance disease waiting period or pre-appointment.

[0053] After confirming that the claim document image meets the first verification condition, the system starts the claim report application process to further analyze and extract information from the document. Specifically, the system combines optical character recognition (OCR) models, multi-modal large models, and reasoning large models to identify and reason about the content of the document, extract key data such as the user's risk information, risk type, treatment type, and risk cause. At the same time, the system will match the responsibility according to the user's policy responsibility, and determine whether the document is mandatory and whether the fields meet the requirements of the policy, to ensure that the classification, content, and time information of each document strictly comply with the requirements of the electronic policy.

[0054] Please refer to Figure 3 , Figure 3is a process diagram of a claim reporting application process provided by an embodiment of the present application. As shown in Figure 3 the claim reporting application process includes steps S131-S135: S131: based on a multi-modal large model, analyzing the claim document image corresponding to the document type information to obtain the user's risk information and risk type information.

[0055] Among them, the document type information includes medical record type information and discharge summary type information, based on the multi-modal large model, the claim document image corresponding to the document type information is analyzed to obtain the user's risk information and risk type information, including: based on the multi-modal large model, the claim document image corresponding to the document type information is identified to obtain the user's risk information; when the risk information meets the preset third verification condition, the risk information is inferred based on the inference large model to obtain the user's risk type information.

[0056] After classifying the claim document image through step S11, for the claim document image classified as medical record or discharge summary, the key information related to the user's risk is extracted, and the automatic inference of the risk type is completed based on the inference large model.

[0057] The multi-modal large model is called by the document recognition module 120 to identify the claim document image. For medical record or discharge summary image, the model extracts key field information through multi-modal feature fusion, including but not limited to: user's hospital name; hospitalization time, discharge time; disease diagnosis result or main symptom description; medical process or treatment type, etc. The extracted text information will be stored in a unified structure to form "risk information".

[0058] To ensure the reliability of the identification result, the system performs consistency and validity check on the extracted risk information. The check content includes: whether the time field exists and meets the logic (for example, the hospitalization time is earlier than the discharge time); whether the hospital name belongs to the list of identifiable medical institutions; whether the disease diagnosis field conforms to the disease code or term format specified in the insurance liability definition. If the extracted information passes the above check, it is determined that it meets the preset third verification condition.

[0059] When the risk information meets the third checking condition, the system calls the reasoning large model to perform reasoning analysis of the risk type. The reasoning process includes: reading the extracted disease diagnosis result or main symptom description; matching based on medical semantic rules and insurance liability knowledge base; combining keywords and context logic to determine whether the current risk belongs to the disease class, the accident class, or other special types (such as pregnancy, slow disease recheck, etc.). For example, when the diagnosis field contains "fall injury", "fracture" and other accident-related semantics, the model will reason that the risk type is an accident; when the diagnosis result is "pneumonia", "high blood pressure" and other disease names, it is reasoned as a disease.

[0060] After the above processing, the system obtains structured risk information (including hospital, hospital time, hospital disease, etc.) and risk type information (disease or accident), and outputs both.

[0061] In this embodiment, through intelligent semantic analysis and automatic inference of risk type of medical record and discharge summary type of claim documents, the automation and accuracy of claim data extraction are significantly improved. On the one hand, the multi-modal large model can accurately identify and extract risk information in the scene of various medical record formats and unstable bill formats, breaking through the format dependence and text omission problem of traditional OCR in complex medical document recognition; on the other hand, the reasoning large model realizes intelligent judgment of disease and accident scene by combining medical semantics, insurance liability rules and context logic, reducing the repeated confirmation of risk nature in manual review link. Not only improves the structured degree of medical record information in the claim process, but also lays an accurate data foundation for subsequent automatic matching of insurance policies and automatic generation of claim report application forms, thereby improving the intelligent level and business efficiency of the claim system as a whole.

[0062] S132: Based on the preset optical character recognition model and the reasoning large model, the claim document image corresponding to the document type information is analyzed to obtain the user's hospital type information.

[0063] Among them, the document type information includes invoice type document information, based on the preset optical character recognition model and the reasoning large model, the claim document image corresponding to the document type information is analyzed to obtain the user's hospital type information, including: based on the optical character recognition model, the claim document image corresponding to the invoice type document information is recognized to obtain the invoice text information corresponding to the claim document image corresponding to the invoice type document information; when the invoice text information meets the fourth preset checking condition, the reasoning large model is used to reason the invoice text information to obtain the user's hospital type information.

[0064] The system uses an optical character recognition model (OCR) combined with a large inference model to automatically analyze and infer the user's visit type information (such as outpatient, inpatient, and drug purchase) for claim documents images classified as invoices in step S11.

[0065] First, based on the preset OCR model, the corresponding invoice document image is recognized and text is extracted. Since the format of public hospital invoices is relatively standardized and the field position is stable, the OCR model can accurately identify the core element information in the invoice, including the hospital name, invoice code, invoice date, item details, and cost amount, thereby generating structured invoice text information.

[0066] Subsequently, the system performs a series of preset verification steps on the extracted invoice text information to ensure data integrity and validity. The fourth verification condition includes: invoice time and visit time check: the system compares the invoice date on the invoice with the previously analyzed visit date. If the time interval between the two exceeds a reasonable range (such as exceeding a preset threshold), it may indicate a time anomaly, thereby avoiding the misclassification of non-associated invoices into the claim range. Total amount and item amount check: the system verifies the consistency of the total amount and item amount fields on the invoice using a rule algorithm to prevent errors caused by OCR misidentification or abnormal invoices.

[0067] When the invoice text information meets the above-mentioned fourth preset verification condition, the system inputs the structured text information into the large inference model for high-level semantic inference analysis. The large inference model can automatically determine the visit type corresponding to the invoice based on context semantics, field association features, and a medical scene knowledge graph.

[0068] In terms of specific rules, the system has both fast determination rules based on explicit keywords (e.g., invoices containing "bed fee," "surgery fee," and "nursing fee" usually indicate inpatient visits; invoices containing "registration fee," "outpatient treatment fee," and "outpatient infusion" indicate outpatient visits) and inference capabilities based on patterns and context (e.g., if multiple invoices for the same case appear within the same time window and the total amount is much higher than the average outpatient consumption, the system may determine that the visit type is inpatient). The large inference model can integrate these explicit / implicit clues to give a more robust judgment, especially when a single field is not sufficient to determine the visit type.

[0069] In this embodiment, through the three-layer link of "high-quality OCR extraction + strict field verification (the fourth verification condition) + inference model based on business knowledge", the recognition of medical bills from the image level to the semantic level of the type of visit is realized. Specifically, the OCR model ensures the accuracy of the basic data in structured information extraction, and the large inference model further realizes the intelligent judgment of the medical scene semantics through semantic understanding and knowledge inference, thereby providing reliable data support for subsequent claim settlement and matching of compensation rules.

[0070] S133: Based on the optical character recognition model, the multi-modal large model, and the inference large model, the claim document image corresponding to the single document type information is analyzed to obtain the user's cause of loss information.

[0071] Among them, the single document type information includes hospitalization expense item single document information and examination report single document information, and based on the optical character recognition model, the multi-modal large model and the inference large model, the claim document image corresponding to the single document type information is analyzed to obtain the user's cause of loss information, including: based on the optical character recognition model, the claim document image corresponding to the hospitalization expense item single document information is identified to obtain the hospitalization expense item information; based on the multi-modal large model, the claim document image corresponding to the examination report single document information is identified to obtain the examination report information; based on the inference large model, the hospitalization expense item information and the examination report information are inferred to obtain the user's cause of loss information.

[0072] For two types of claim materials of hospitalization expense item single document and examination report single document, the key information reflecting the user's "cause of loss" is extracted and summarized, thereby providing interpretable and standardized medical semantic basis for subsequent claim settlement and policy matching. Since this type of single document contains a large number of medical terms, complex structures and cross-document semantic associations, it is difficult to accurately identify and understand by relying on traditional OCR analysis alone, therefore, the embodiment realizes structured extraction and semantic inference through the hierarchical collaborative mechanism of "optical character recognition model + multi-modal large model + inference large model". The specific implementation process is as follows: First, the system performs structured recognition on the hospitalization expense item document image. The hospitalization expense item document usually comes from the hospital expense list or detailed table, which has a relatively standard format but many fields and different project names. The system performs layout structure analysis on the image based on an optical character recognition model (OCR) to identify the project name, unit price, quantity, amount, department, and keywords of drugs or treatment items, and reconstructs the table structure based on the recognition results to obtain the hospitalization expense item information. The hospitalization expense item information records the confidence level and position information at the field level to ensure traceability for subsequent semantic processing. To ensure accuracy, the system performs several preprocessing steps after the OCR output, including: synonym standardization: through word vector matching or medical terminology dictionary, equivalent names such as "tetanus human immunoglobulin for injection" and "human immunoglobulin (tetanus)" are standardized; expense classification aggregation: according to the project category (drug, treatment, examination, and surgery), the information is aggregated to facilitate subsequent semantic reasoning to focus on key fields; abnormal item filtering: remove the amount of money abnormal or non-medical meaning items (such as table decoration symbols, header and footer text) misrecognized by OCR.

[0073] Next, the system calls a multi-modal large model for semantic-level understanding of the examination report document image. The examination report document has large differences in format and contains multi-modal information such as images, tables, and text, making it difficult for traditional OCR to directly analyze its core indicators. The multi-modal large model uses a visual-linguistic joint encoding method to fuse image features (such as image scans and pathology pictures) and text information (such as diagnostic conclusions, detection indicators, and reference intervals) in the report to extract structured examination report information. For example, when the model analyzes a chest CT report, it automatically extracts key indicators ("thickening of lung texture", "a small amount of exudation", and "suggests pneumonia") and normalizes them into standard medical entities (such as disease names ICD codes), while identifying report dates, departments, and detection methods. Through this way, the multi-modal large model can identify structured clues such as report titles, image page numbers, and conclusion paragraph boundaries at the visual level, ensuring the accuracy and completeness of information extraction.

[0074] Subsequently, the system standardizes and semantically aligns the above hospitalization expense item information and examination report information. In this process, the hospitalization expense item information and the examination report information are mapped into the same medical knowledge graph space. The knowledge graph pre-constructs the association relationship between drugs, symptoms, diseases, and treatment methods. For example, "tetanus human immunoglobulin" has a high correlation with "animal bite" and "trauma infection prevention" nodes, and "azithromycin", "white blood cell elevation", and "chest inflammation" have a semantic convergence relationship with the "pneumonia" node. The system projects the drug names recognized by the OCR and the examination conclusions extracted by the multi-modal model into the knowledge graph to form a joint semantic representation. This standardization process can significantly reduce semantic deviations caused by different hospital terminologies or abbreviations, ensuring the generalization and consistency of downstream reasoning.

[0075] After semantic alignment, the system enters the reasoning phase of the cause of the accident. This phase is dominated by the reasoning large model, which reasons out the main cause of the user's medical treatment or claim through the context logical chain. The reasoning process includes: first, retrieving the disease entity with the highest co-occurrence probability in the knowledge graph with the hospitalization expense item information and the examination report information; combining the expense type weight (drug > treatment > examination) and the report conclusion confidence to calculate the confidence score of the candidate disease or the cause of the accident; according to the reasoning rules (for example, if the drug is used for preventing diseases and the examination result shows normal, then the reasoning is accidental or exposure type cause of the accident), the most likely cause of the accident is screened. For example, when the hospitalization expense item information includes "tetanus human immunoglobulin" and "rabies vaccine", and the examination report shows "skin damage" and "normal serum antibody", the reasoning large model will output the conclusion: "the cause of the accident is animal bite causing trauma"; for example, when the hospitalization expense item information includes "azithromycin" and "infusion fee", and the examination report extracts "pulmonary infection" and "white blood cell elevation", the model reasons that the "cause of the accident is pneumonia infection".

[0076] In this embodiment, by hierarchically coordinating the optical character recognition model, the multi-modal large model, and the reasoning large model, a closed loop can be formed at three levels of structured extraction, semantic understanding, and logical reasoning, thereby accurately capturing the implicit cause of the accident information in complex medical materials. Not only does it improve the accuracy and consistency of the cause of the accident recognition, but also through the knowledge graph support and confidence fusion mechanism, the reasoning process has explainability and traceability, significantly reducing the misjudgment rate. At the same time, the system can maintain stable parsing ability in the scene of unstructured single certificate and mixed formats of multiple types, realizing the intelligent transition of the cause of the accident from "keyword matching" to "causal understanding".

[0077] S134: According to the medical treatment type information and the cause of the accident information, match the target insurance policy.

[0078] The target policy is matched according to the treatment type information and the accident cause information, including: based on a preset mapping relationship strategy, the treatment type information and the accident cause information, screening at least one of the outpatient policy, the hospitalization policy and the drug purchase policy as the target policy.

[0079] In the claim report application process, the system needs to match the target policy held by the user according to the treatment type information (such as outpatient, hospitalization, and drug purchase) and the accident cause information (such as cold, fall injury, and pneumonia) obtained in the previous step. To achieve this goal, a preset mapping relationship strategy is introduced, which defines the corresponding rules between various accident causes, treatment types and insurance policy coverage responsibilities. The system first standardizes the user's treatment type information and accident cause information, converting them into internal unified representations to ensure the accuracy and consistency of subsequent mapping logic.

[0080] Subsequently, the system calls the policy selector module to screen the medical insurance policy held by the user in the medical insurance policy library. Medical insurance policies usually cover three types of responsibilities: outpatient, hospitalization, and drug purchase. Different accident causes and treatment types are mapped to corresponding coverage responsibilities. For example: if the user's accident cause is "cold" and the treatment type is "outpatient", the system can match the outpatient-only coverage policy or the outpatient + hospitalization combination policy held by the user as the target policy; if the accident cause is "accidental fall injury" and the treatment type includes outpatient + hospitalization, the system can match multiple policies covering outpatient, hospitalization and outpatient + hospitalization combination as the target policy; if the accident cause is "pneumonia" and the treatment type is outpatient + hospitalization, the system matches the policy covering outpatient + hospitalization as the target policy.

[0081] In this embodiment, through the combination of mapping rules and policy screening, the system can automatically select the target policy covering the user's current treatment and accident scene in a short time, thereby providing accurate policy information basis for subsequent automatic filling of claim report application forms, reducing the risk of manual selection and mis-matching, and improving claim processing efficiency and accuracy.

[0082] S135: Fill in the target policy according to the accident type information, the accident cause information, the accident information and the treatment type information, and obtain the claim report application form.

[0083] The target policy is filled in according to the accident type information, the accident cause information, the accident information and the treatment type information, and the claim report application form is obtained, including: based on a multi-modal large model, identifying the claim document image to obtain the user's identity information; based on a preset template mapping model, the identity information, the accident type information, the accident cause information, the accident information and the treatment type information, filling in the target policy to obtain the claim report application form.

[0084] In the final stage of the claim report application process, the system needs to automatically fill in the risk type information (such as illness, accident, etc.), risk cause information (such as cold, fall injury, pneumonia, etc.), risk information (such as hospital, treatment time, etc.), and treatment type information (such as outpatient, hospitalization, drug purchase) obtained from the previous steps into the target policy, thereby generating a complete claim report application form. The specific implementation process is as follows: First, the system uses a multi-modal large model to identify the claim document images corresponding to the target policy and obtain the user's identity information. The identity information here includes the user's name, ID number, contact information, and other key fields. The multi-modal large model can process both image and text information, comprehensively understand the visual elements (such as head portrait, seal, table position) and text information (such as text content) in the claim document, and ensure the completeness and accuracy of the identity information.

[0085] Subsequently, the system inputs the aforementioned parsed risk type information, risk cause information, risk information, treatment type information, and user identity information into a template mapping model. The template mapping model can use a structured information filling system, a rule engine, a differentiable form generation model, or a specialized mapping model fine-tuned in the insurance claim field, without limitation.

[0086] The template mapping model maps each type of information to the corresponding field position based on a pre-set claim application form template. For example, it maps the user's name to the "insured person's name" field, the treatment time to the "treatment time" field, and the outpatient type to the "medical behavior type" field. At the same time, the template mapping model automatically selects the corresponding policy field according to the risk type and risk cause, and performs necessary merging or splitting processing on complex fields.

[0087] In this embodiment, the system can generate a complete and accurate claim report application form without requiring the user to manually input any document information or field content, thereby significantly reducing human error and improving claim application efficiency and user experience. In addition, this automatic filling process has high adaptability and expandability, and can be compatible with multiple policy types, multiple types of claim documents, and different formats of claim application templates, achieving the intelligentization and standardization of online claim processes.

[0088] It should be noted that the above steps S131 to S133 involve multi-dimensional information extraction on the claim document image, including risk information and risk type information (S131), treatment type information (S132), and risk reason information (S133). In this embodiment, the above information acquisition process can be performed simultaneously in parallel, rather than in the traditional sequential processing mode. Specifically, based on the combination of the optical character recognition model, the preset multi-modal large model, and the inference large model, the system can simultaneously complete text recognition, visual feature analysis, and semantic reasoning on a single uploaded claim document image, thereby significantly shortening the processing time while ensuring the accuracy of information extraction, and improving the overall claim application efficiency. This parallel processing mode, combined with the ability of multi-model collaborative work, enables the claim reporting system to quickly and stably complete classification, field extraction, and policy matching operations in complex scenarios with multiple types of documents, multiple fields, and multiple rules, thereby significantly improving user experience and operational efficiency.

[0089] The specific implementation process of the claim reporting processing method provided by the present application will be introduced below with a specific example. Figures 4 to 6

[0090] (1) The user enters the claim reporting application process. It can be understood that the user enters the claim reporting application page 110 through the micro-insurance applet or APP. As shown in Figure 5 , the claim reporting application page 110 displays the basic information input items required for claim reporting (such as the insured person's identity information, medical information, etc.), and the user can choose to use the intelligent claim AI assistant. The user uploads all the claim document images at one time according to the prompt, including invoices, medical records, hospitalization expense details, examination reports, and discharge summaries.

[0091] (2) The system performs document type identification. It can be understood that the document identification module 120 analyzes the uploaded claim document images using a multi-modal large model to extract text information and visual features. For documents with obvious text features (such as public hospital invoices and value-added tax medical invoices), the system directly determines the corresponding document type information based on the text feature information. For documents with unclear text features (such as medical records, examination reports, and discharge summaries), the inference large model is used to infer the text feature information to obtain the document type information of the claim document image whose text feature information does not satisfy the second verification condition. Finally, the system generates "document type information" for each document, indicating its specific category (medical record type, discharge summary type, invoice type, expense list type, etc.).

[0092] (3) The system performs document verification and starts the claim reporting application process. It can be understood that the system performs a first verification condition judgment on the classification result of each document. If the document satisfies the first verification condition, the system starts the claim reporting application process. ​

[0093] (4) Perform the claim report application process. Understandably, first, the document recognition module 120 analyzes the medical record and discharge summary through the multi-modal large model, extracts the user's insurance information (hospital, treatment time, and disease). The information reasoning module 130 reasons the user's insurance type (such as accident, disease, etc.) according to the extracted insurance information through the reasoning large model. Second, the information reasoning module 130 identifies the invoice type of document through the OCR model, extracts the invoice text information (invoice time, amount, and treatment department). Then, the reasoning large model reasons the treatment type (outpatient / hospitalization) according to the invoice information. Third, the information reasoning module 130 extracts the medication and cost information by identifying the hospitalization expense item document information through OCR. The multi-modal large model identifies the inspection report information and extracts the index results. The reasoning large model comprehensively reasons the user's insurance reason (such as cat and dog scratch, pneumonia, etc.) based on the expense item and inspection report.

[0094] (5) Perform information verification and confirmation. Understandably, as shown in Figure 6 , the system automatically checks all extracted information, prompts the user to check the document status and recognition results (such as no error information prompt). The user can confirm the information, check the identity information and treatment information, and complete the online signature confirmation.

[0095] (6) Perform automatic matching and filling. Understandably, the system calls the policy matching logic according to the insurance information, insurance type, treatment type, and insurance reason information: filters the user's insurance policy that covers the current insurance situation (outpatient, hospitalization, and drug purchase). The system automatically fills the user information and claim information into the target policy based on the template mapping model, and generates a complete claim report application form. The user can submit the claim application online without manually filling in most of the information.

[0096] The embodiment of the application provides a claim processing method, which realizes high automation and intelligence of information extraction, document classification, policy matching and claim form filling in an insurance claim process. First, a multi-modal technology mode is adopted, a multi-modal large model, an OCR model and an inference large model are jointly applied, various claim documents are comprehensively recognized and analyzed, key data such as risk information, treatment type and risk cause are accurately extracted from image files uploaded by a user, and the limitation of a single model in recognition accuracy and applicable scene is compensated through the collaborative work of multiple models, so that the accuracy and efficiency of claim information processing are improved. Secondly, a claim application mode is proposed, that is, through functions such as automatic classification of documents by photographing and automatic filling of claim information, a new claim experience is provided for the user, the complexity and understanding threshold encountered by the user in the claim operation are greatly reduced, and the user can conveniently and efficiently complete the claim application. Thirdly, the acquisition process of key data such as risk information, treatment type and risk cause based on the claim processing method provided by the application can be carried out in parallel, and the claim efficiency and user experience are effectively improved. Based on this, the application realizes the integrated application of multi-model collaboration, intelligent information processing and accurate policy matching in the technical field, has significant advantages in operation efficiency, accuracy and user experience, and fully embodies the innovation and practicality of the application in the field of insurance claim.

[0097] Based on the claim processing method provided in the above embodiment, the embodiment of the application further provides a claim processing device. Please refer to Figure 7 , Figure 7 is a structural schematic diagram of the claim processing device. As Figure 7 shown, the claim processing device 200 includes an image acquisition module 210, a document type identification module 220 and a claim application module 230.

[0098] The image acquisition module 210 is configured to acquire an uploaded claim document image of a user; the document type identification module 220 is configured to identify the claim document image based on a preset multi-modal large model and an inference large model, to obtain document type information of the claim document image; the claim report application module 230 is configured to execute a claim report application process when the claim document image corresponding to the document type information meets a preset first inspection condition, the claim report application process including: analyzing the claim document image corresponding to the document type information based on the multi-modal large model, to obtain accident information and accident type information of the user; analyzing the claim document image corresponding to the document type information based on a preset optical character recognition model and the inference large model, to obtain treatment type information of the user; analyzing the claim document image corresponding to the document type information based on the optical character recognition model, the multi-modal large model and the inference large model, to obtain accident cause information of the user; matching a target insurance policy according to the treatment type information and the accident cause information; filling the target insurance policy according to the accident type information, the accident cause information, the accident information and the treatment type information, to obtain a claim report application form.

[0099] It should be noted that the claim report processing apparatus described above can execute the claim report processing method provided in the embodiments of the present application, and has the corresponding function modules and beneficial effects of the execution method. Technical details not described in detail in the claim report processing apparatus embodiments can be referred to the claim report processing method provided in the embodiments of the present application.

[0100] The embodiments of the present application also provide an electronic device 300, please refer to Figure 8 which shows a hardware structure schematic diagram of the electronic device 300 capable of executing the method described in the above embodiments. The electronic device 300 includes at least one processor 310, and a memory 320 connected with the at least one processor 310, Figure 8 The memory 320 stores instructions executable by the at least one processor 310, and the instructions are executed by the at least one processor 310 to enable the at least one processor 310 to execute the claim report processing method described in the above embodiments. The processor 310 and the memory 320 can be connected by a bus or other means, Figure 8 The bus connection is taken as an example.

[0101] The memory 320, as a non-volatile computer readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the claim handling method in the embodiments of the present application. The processor 310 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 320, that is, implements the claim handling method described in the above embodiments.

[0102] The memory 320 can include a program storage area and a data storage area, wherein the program storage area can store an operating system and at least one application required by a function; the data storage area can store data created according to the use of the computing device, etc. In addition, the memory 320 can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state memory device. In some embodiments, the memory 320 can optionally include a memory remotely arranged with respect to the processor 310, which can be connected to the computing device through a network. Examples of the above network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0103] The one or more modules are stored in the memory 320, and when executed by the one or more processors 310, the claim handling method described in the above embodiments is executed.

[0104] It should be noted that the electronic device 300 can be an electronic device with data processing and artificial intelligence computing capability, such as a server, a personal computer, a smart terminal, a mobile device or an embedded device capable of running the multi-modal large model, the OCR model and the inference large model in the present application, or a self-defined intelligent hardware platform specially designed for the insurance claim business scenario, without limitation.

[0105] The above product can execute the method provided by the embodiments of the present application, and has the corresponding functional modules and beneficial effects of executing the method. Technical details not described in detail in the present embodiment can be referred to the claim handling method described in the embodiments of the present application.

[0106] The embodiment of the present application provides a nonvolatile computer readable storage medium, the nonvolatile computer readable storage medium stores computer executable instructions, the computer executable instructions are executed by one or more processors, so that the at least one processor can execute the claim processing method described in the above embodiment. For example, the nonvolatile computer readable storage medium can be a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a read-only compact disc (Compact Disc Read-Only Memory, CDROM), a magnetic tape, a floppy disk and an optical data storage device and the like.

[0107] The embodiment of the present application provides a computer program product, the computer program product includes a computer program stored on a nonvolatile computer readable storage medium, the computer program includes program instructions, when the program instructions are executed by an electronic device, the electronic device can execute the claim processing method in any method embodiment.

[0108] It should be noted that the above-described device embodiments are only schematic, wherein the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, that is, they can be located in one place, or distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment of the present application.

[0109] Through the description of the above embodiments, those skilled in the art can clearly understand that the above embodiment method can be realized by software plus the necessary general hardware platform, and of course it can also be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disc), and includes a plurality of instructions for making a terminal (which can be a mobile phone, computer, server, air conditioner or network equipment) execute the method described in each embodiment of the present application.

[0110] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application, but not to limit them; under the idea of the present application, the technical features in the above examples or different embodiments can also be combined, the steps can be implemented in any order, and there are many other changes of different aspects of the present application as described above, which are not provided in details for simplicity; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for processing claims reports, characterized in that, The method includes: Obtain images of claim documents uploaded by the user; Based on the preset multimodal big model and reasoning big model, the claim document image is identified to obtain the document type information of the claim document image; When the claim document image corresponding to the document type information meets the preset first inspection condition, the claim reporting application process is executed, which includes: Based on the multimodal large model, the claim document image corresponding to the document type information is parsed to obtain the user's accident information and accident type information; Based on the preset optical character recognition model and the inference model, the image of the claim document corresponding to the document type information is parsed to obtain the user's medical treatment type information; Based on the optical character recognition model, the multimodal big model, and the inference big model, the claim document image corresponding to the document type information is analyzed to obtain the user's accident cause information; Match the target insurance policy based on the described medical visit type information and the described cause of the claim information; Based on the information on the type of accident, the cause of the accident, the accident information, and the type of medical treatment, the target policy is filled in to obtain a claim report application form.

2. The claims reporting and processing method according to claim 1, characterized in that, The method, based on a preset multimodal large model and inference large model, identifies the claim document image to obtain the document type information of the claim document image, including: Based on the multimodal large model, the claim document image is detected to obtain text feature information corresponding to the claim document image; The text feature information is verified; If the text feature information meets the preset second verification condition, the document type information of the claim document image that meets the second verification condition is obtained based on the text feature information; If the text feature information does not meet the second test condition, the inference model is used to infer the text feature information to obtain the document type information of the claim document image where the text feature information does not meet the second test condition.

3. The claims reporting and processing method according to claim 1, characterized in that, The document type information includes medical record document information and discharge summary document information. The process involves parsing the claim document image corresponding to the document type information based on the multimodal large model to obtain the user's accident information and accident type information, including: Based on the multimodal large model, the claims document image corresponding to the document type information is identified to obtain the user's accident information; When the incident information meets the preset third verification condition, the incident information is inferred based on the inference big model to obtain the incident type information of the user.

4. The claims reporting and processing method according to claim 1, characterized in that, The document type information includes invoice-type document information. The method, based on a preset optical character recognition model and the inference model, parses the claim document image corresponding to the document type information to obtain the user's medical visit type information, including: Based on the optical character recognition model, the claim document image corresponding to the invoice document information is recognized to obtain the invoice text information of the claim document image corresponding to the invoice document information; When the invoice text information meets the preset fourth verification condition, the user's medical treatment type information is obtained by reasoning based on the inference big model.

5. The claims reporting and processing method according to claim 1, characterized in that, The document type information includes inpatient expense document information and examination report document information. The method, based on the optical character recognition model, the multimodal large model, and the inference large model, parses the claim document image corresponding to the document type information to obtain the user's accident cause information, including: Based on the optical character recognition model, the image of the claim document corresponding to the inpatient expense item document information is recognized to obtain the inpatient expense item information; Based on the multimodal large model, the claims document image corresponding to the inspection report document information is identified to obtain the inspection report information; Based on the aforementioned reasoning model, the inpatient expense information and the examination report information are used to deduce the user's cause of the accident information.

6. The claims reporting and processing method according to claim 1, characterized in that, The process of matching target insurance policies based on the medical visit type information and the cause of the claim information includes: Based on a preset mapping strategy, the medical visit type information, and the claim cause information, at least one of outpatient policies, inpatient policies, and medication purchase policies is selected as the target policy.

7. The claims reporting and processing method according to claim 1, characterized in that, The process involves filling in the target policy based on the incident type information, incident cause information, incident information, and medical treatment type information to obtain a claim application form, including: Based on the multimodal large model, the user's identity information is obtained by recognizing the claim document image. Based on the preset template mapping model, the identity information, the accident type information, the accident cause information, the accident information, and the medical treatment type information, the target policy is populated to obtain the claim report application form.

8. A claims reporting and processing device, characterized in that, The device includes: An image acquisition module is used to acquire images of claim documents uploaded by the user. The document type recognition module is used to recognize the claim document image based on a preset multimodal big model and inference big model to obtain the document type information of the claim document image. The claims reporting application module is used to execute a claims reporting application process when the claims document image corresponding to the document type information meets a preset first verification condition. The claims reporting application process includes: parsing the claims document image corresponding to the document type information based on the multimodal big data model to obtain the user's accident information and accident type information; parsing the claims document image corresponding to the document type information based on a preset optical character recognition model and the inference big data model to obtain the user's medical visit type information; parsing the claims document image corresponding to the document type information based on the optical character recognition model, the multimodal big data model, and the inference big data model to obtain the user's accident cause information; matching a target policy based on the medical visit type information and the accident cause information; and filling the target policy with the accident type information, the accident cause information, the accident information, and the medical visit type information to obtain a claims reporting application form.

9. An electronic device, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1-7.

10. A non-volatile computer-readable storage medium, characterized in that, The non-volatile computer-readable storage medium stores computer-executable instructions that, when executed by an electronic device, cause the electronic device to perform the method described in any one of claims 1-7.

Images