Intelligent medical insurance auditing method and system based on multi-modal large model

Abstract

The invention discloses a medical insurance intelligent auditing method and system based on a multi-modal large model, and relates to the technical field of big data processing, and the method comprises the steps: recognizing a material name and a personal name, and auditing the compliance of a material needed for business handling according to the integrity and correctness of the material; identifying first expense detail data according to the expense detail list material and the expense identification cue word; matching the first expense detail data with a medical insurance catalog vector database to obtain a closest catalog entity, and combining the closest catalog entity with the first expense detail data to form second expense detail data; obtaining medicine knowledge from the medicine specification, matching the medicine knowledge with the medical insurance directory vector database, and associating the directory code of the optimal directory entity into the corresponding medicine knowledge; and for each record of the second expense detail data, performing expense rationality auditing according to the corresponding intelligent auditing rule and medicine knowledge. Content acquisition, medical project matching and intelligent auditing of the electronic material are realized, and the auditing speed and accuracy are improved.

Description

Technical Field

[0001] This invention relates to the field of big data processing technology, and in particular to a medical insurance intelligent review method and system based on a multimodal big model. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] When medical insurance participants are unable to directly claim reimbursement online due to various external factors, they need to bring relevant medical insurance expense materials to the medical insurance agency for offline manual reimbursement. Medical insurance staff need to enter and review the detailed expense information materials provided by the participants. Due to the large number of medical expense details and the difficulty in entering them, and the need for accurate matching of medical items with the medical insurance catalog, this places high demands on medical insurance staff.

[0004] In the traditional model, to improve the efficiency of medical insurance reimbursement, OCR recognition technology is usually used to identify medical insurance-related electronic materials. Medical insurance experts then correct any errors in the identification, obtain the relevant reimbursement information, and then manually match the medical insurance catalog and apply relevant review rules to review the reimbursement content.

[0005] This method has several drawbacks: first, it requires advanced OCR technology, which may not be accurate for materials with blurry handwriting or skewed charts; second, it requires manual association with the medical insurance catalog, which is time-consuming and tests the professional skills of medical insurance personnel, and different personnel may cause inconsistencies in standards; and third, it requires manual mapping of relevant review rules. Summary of the Invention

[0006] To address the aforementioned issues, this invention proposes a medical insurance intelligent review method and system based on a multimodal large model. Leveraging the image recognition and automatic reasoning capabilities of the multimodal large model, it enables functions such as content acquisition from electronic materials, medical item matching, and intelligent review, thereby improving review speed and accuracy.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a medical insurance intelligent review method based on a multimodal large model, comprising: Obtain the materials required for business processing and identify the names of the materials and the names of the personnel to whom they belong; After converting the material names and required materials for business processing into sentence vectors, text similarity is used for mapping and matching, and the names of the relevant personnel are matched character by character to verify the compliance of the materials required for business processing based on the completeness and correctness of the materials. Expense identification prompts are constructed, and based on the expense details list materials and expense identification prompts, a multimodal large model is used to identify the first expense details data; The first expense detail data is matched with the constructed medical insurance catalog vector database to obtain several similar catalog entities. Based on the constructed catalog matching prompt words, the closest catalog entity is selected from the several similar catalog entities and combined with the first expense detail data to form the second expense detail data. Construct drug knowledge recognition prompts to obtain drug knowledge from drug instructions using a multimodal large model, match the drug knowledge with the medical insurance catalog vector database, and combine the catalog matching prompts to associate the catalog code of the best matching catalog entity with the corresponding drug knowledge. A knowledge base of intelligent review rules is constructed. For each record of the second expense details data, the corresponding intelligent review rules and drug knowledge are obtained based on the medical insurance code of each record, so as to conduct a reasonableness review of the expenses.

[0008] As an alternative implementation method, the process of identifying the material name and the name of the person to whom it belongs includes: The pre-trained OCR recognition model is used to recognize the text content of the material. The text size is determined based on the Y-coordinate distance between the top and bottom vertices of the text box, and the text with the largest font size is used as the material name. The text following a specific character representing a name is extracted to form the name of the person to whom the material belongs.

[0009] As an alternative implementation method, the process of verifying the compliance of the materials required for business processing includes: The material names and the materials required for business processing are mapped to sentence vectors using a pre-trained SimBERT model, and the cosine similarity between the two is calculated. Based on cosine similarity, all material names are matched with the materials required for business processing. If the similarity is greater than a set threshold, the material name is mapped to the materials required for business processing. If all the materials required for business processing have corresponding mapping objects, the materials are considered complete. Match the names of the personnel to whom the materials belong with the names of the personnel in charge of the business; if the names match, the materials are considered to be correct. If all the required materials for the business transaction are complete and correct, the materials are considered compliant.

[0010] As an alternative implementation method, the cost recognition prompt includes role / task + output method + additional knowledge; Directory matching suggestions include directory matching roles, matching processing steps, directory matching requirements, and similar directory entities; The drug knowledge identification prompts include the drug identification role, identification steps, and drug identification requirements.

[0011] As an alternative implementation, the matching process for similar directory entities includes: Concatenate the field name, specification, and unit fields of each first expense detail data to form the original expense information; The drug name, registered dosage form, and specifications information of the drug knowledge are combined into a drug information field; The original cost information or drug information fields are matched with the medical insurance catalog vector database using an approximate nearest neighbor search to obtain several similar catalog entities with a similarity greater than a set threshold.

[0012] As an alternative implementation method, the cost reasonableness review process includes: Intelligent review rules include review rule classification information, review rule information, and review element information; Iterate through each review rule information. For review rules with the attribute "rule", combine the review code of the review rule classification information and the corresponding review element information to generate review logic rules and execute them. If the rule fails, the review fails. For review rules with the attribute of knowledge, the project rule knowledge is concatenated according to the text knowledge concatenation rules of the corresponding review rule classification information, and the review result is obtained by the multimodal big model in combination with drug knowledge; among them, the intelligent review prompt words of the multimodal big model include intelligent review role + review steps + rule knowledge + drug knowledge + review requirements.

[0013] Secondly, the present invention provides a medical insurance intelligent review system based on a multimodal large model, comprising: The text recognition module is configured to retrieve the materials required for business processing and identify the material names and the names of the personnel to whom they belong. The compliance review module is configured to convert the material names and the materials required for business processing into sentence vectors, then perform mapping and matching through text similarity, and perform character matching of the names of the relevant personnel, so as to review the compliance of the materials required for business processing based on the completeness and correctness of the materials. The cost extraction module is configured to construct cost identification prompts. Based on the cost details list materials and the cost identification prompts, a multimodal large model is used to identify and obtain the first cost details data. The directory mapping module is configured to match the first expense detail data with the constructed medical insurance directory vector database to obtain several similar directory entities. Based on the constructed directory matching prompt words, the module selects the closest directory entity from the several similar directory entities and combines it with the first expense detail data to form the second expense detail data. The rule management module is configured to construct drug knowledge recognition prompts, which use a multimodal large model to obtain drug knowledge from drug instructions, match the drug knowledge with the medical insurance catalog vector database, and combine the catalog matching prompts to associate the catalog code of the best matching catalog entity with the corresponding drug knowledge. The review module is configured to build an intelligent review rule knowledge base. For each record of the second expense details data, it retrieves the corresponding intelligent review rules and drug knowledge based on the medical insurance code of each record, thereby reviewing the reasonableness of the expenses.

[0014] Thirdly, the present invention provides an electronic device including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the method described in the first aspect.

[0015] Fourthly, the present invention provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in the first aspect.

[0016] Fifthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention, based on traditional intelligent medical insurance review, leverages the image recognition and automatic reasoning capabilities of a multimodal large model to achieve content acquisition, medical item matching, and intelligent review functions for electronic materials. It solves the difficulties of entering and organizing multiple medical items, manual catalog matching, and manual review and verification in the traditional model, thereby improving review speed and reducing errors caused by manual operation.

[0018] In terms of compliance review, this invention firstly leverages the capabilities of OCR text recognition to use a pre-trained OCR recognition model to extract information from the original image material, thereby improving the ability to collect image information; secondly, it uses a pre-trained SimBERT model to generate text vectors and combines cosine similarity to compare similar texts, further completing the mapping of materials, solving the shortcomings of rule matching and keyword matching in traditional models, improving the ability of text semantic matching, and making it more reasonable.

[0019] In terms of obtaining expense details and matching with the medical insurance catalog, this invention employs two main methods: First, it uses a multimodal large model to obtain specific expense details, achieving significantly higher accuracy than traditional manual entry or OCR recognition. It also enables mapping and matching of similar meanings across different fields, resulting in expense detail data with a fixed structure. Second, it stores the three major catalogs in a vector database. First, it retrieves similar entities from the vector database based on the expense details obtained from electronic materials. Then, it uses a large model to obtain the most matching entity from the approximate entities, improving matching accuracy and avoiding errors caused by manual or fuzzy comparisons.

[0020] In terms of intelligent review, this invention retains the original rule matching and adds a large-scale intelligent review module. It transmits the necessary rule knowledge and necessary drug knowledge as additional prompts to the large model, so that the review that originally required manual processing by offline experts can be completed by the large model. This ensures the correct execution of the rules for rule matching review and also completes the review of content that traditional rule-based review cannot handle, thus improving the efficiency of review.

[0021] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0023] Figure 1 The flowchart of the intelligent medical insurance review method based on a multimodal large model provided in Embodiment 1 of the present invention is shown. Detailed Implementation

[0024] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0025] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0026] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form as well. Furthermore, it should be understood that the terms “comprising” and “including”, and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0028] Example 1 Multimodal large language models are a class of models that combine the natural language processing capabilities of large language models with the ability to understand and generate data from other modalities. By integrating various types of input and output, such as text, images, and sound, they provide a richer and more natural interactive experience.

[0029] Therefore, this embodiment provides a medical insurance intelligent review method based on a multimodal large model. The multimodal large language model used includes, but is not limited to, commercial multimodal large language models and open-source multimodal large language models. Leveraging the image recognition and automatic reasoning capabilities of the multimodal large model, it achieves functions such as content acquisition from electronic materials, medical item matching, intelligent review, and automatic calculation.

[0030] like Figure 1 As shown, it includes: Obtain the materials required for business processing and identify the names of the materials and the names of the personnel to whom they belong; After converting the material names and required materials for business processing into sentence vectors, text similarity is used for mapping and matching, and the names of the relevant personnel are matched character by character to verify the compliance of the materials required for business processing based on the completeness and correctness of the materials. Expense identification prompts are constructed, and based on the expense details list materials and expense identification prompts, a multimodal large model is used to identify the first expense details data; The first expense detail data is matched with the constructed medical insurance catalog vector database to obtain several similar catalog entities. Based on the constructed catalog matching prompt words, the closest catalog entity is selected from the several similar catalog entities and combined with the first expense detail data to form the second expense detail data. Construct drug knowledge recognition prompts to obtain drug knowledge from drug instructions using a multimodal large model, match the drug knowledge with the medical insurance catalog vector database, and combine the catalog matching prompts to associate the catalog code of the best matching catalog entity with the corresponding drug knowledge. A knowledge base of intelligent review rules is constructed. For each record of the second expense details data, the corresponding intelligent review rules and drug knowledge are obtained based on the medical insurance code of each record, so as to conduct a reasonableness review of the expenses.

[0031] The method of this embodiment will be described in detail below.

[0032] Step 1: Obtain the materials required for business processing, and use a pre-trained OCR recognition model to obtain the material names and the names of the personnel to whom they belong. Specifically: Step 1-1: Obtain basic information of the business handler and information on the electronic materials used for reimbursement and perform preprocessing; including: global threshold binarization of the image, noise removal of the data using linear or nonlinear filtering, and tilt correction of the image using Hough transform.

[0033] Steps 1-2: Use a pre-trained OCR recognition model to obtain the text information of the input electronic material, and obtain the title of the electronic material and the name of the person to whom the material belongs based on the text box size and rule matching technology.

[0034] The electronic materials required for OCR model recognition model pre-training include, but are not limited to, the insured person's valid ID, electronic medical insurance voucher, social security card, hospital billing invoice, inpatient expense list, and medical record copy. Among them, the hospital-related electronic materials should cover medical institutions nationwide and include materials from primary, secondary, tertiary, and unclassified medical institutions in a proportional manner. The pre-training electronic material data should be labeled and used for model training.

[0035] The OCR recognition model consists of a convolutional neural network (CNN), a recurrent neural network (RNN), and a connectionist temporal classification. The model's final output is the recognition results, coordinates, and confidence scores of multiple text boxes.

[0036] The pre-trained OCR recognition model is used to recognize the text content of electronic materials. The text size is determined based on the Y-coordinate distance between the top and bottom vertices of the text box, and the text with the largest font size is selected as the title of the electronic material. The text rule matching method is used to extract the text after the specific characters representing the name as the name of the person to whom the material belongs. The specific characters representing the name include, but are not limited to: name, insured person, payer, patient name, etc.

[0037] Step 2: Use the pre-trained SimBERT model to convert the OCR-recognized material names and the materials required for business processing into sentence vectors, perform mapping and matching through text similarity, and perform character matching of the relevant personnel names to review the compliance of the materials based on their completeness and correctness.

[0038] Specifically: Step 2-1: Construct a set of similar material sentences and pre-train the SimBERT model.

[0039] Step 211: Compile a training dataset and generate statements with names similar to the materials required for business processing, resulting in a set of several similar material statements.

[0040] Taking the inpatient expense list as an example, the set of similar material statements includes: {inpatient expense lists of various hospitals, summary lists of inpatient expenses, patient expense lists, hospital inpatient expense lists, inpatient expense details, and outpatient expense lists}.

[0041] Step 212: In the same set of similar material sentences, perform arbitrary pairwise matching on the material sentences to generate multiple similar sentence pairs, and use the similar sentences to train the SimBERT model.

[0042] In the specific implementation, the BERT word segmenter is used to convert sentences into token IDs, adding [CLS] and [SEP] tags and padding to a fixed length. During training, some tokens are randomly masked (15% probability), with 80% replaced with [MASK], 10% replaced with random words, and 10% kept as original words, enhancing the model's robustness. An adaptive learning rate optimization algorithm is used, with the learning rate and batch size set accordingly, to train the model.

[0043] Step 2-2: The electronic document title and the required materials for business processing are mapped to sentence vectors using the pre-trained SimBERT model. The cosine similarity matching method is then used to perform text similarity matching between the electronic document title and the required materials for business processing.

[0044] The formula for text similarity matching is as follows: ; ; in, is the sentence vector obtained by using the pre-trained SimBERT to compute text a, and similarity is the cosine similarity between the two sentence vectors.

[0045] Steps 2-3: Based on text similarity, match all material names with the materials required for business processing. If the similarity is greater than a set threshold, map the material names to the materials required for business processing. If all the materials required for business processing have corresponding mapping objects, then the materials required for business processing are considered complete.

[0046] Steps 2-4 involve using OCR to identify the name of the person to whom the material belongs, and then matching the names with the name of the person in charge of the business. If the names match, the material is considered to be correct and required for the business processing.

[0047] If all the required materials for the business processing are complete and correct, then the materials are considered compliant. (Steps 2-5)

[0048] Step 3: Construct a multimodal large model for cost identification prompts. Based on the mapping and matching results, input the electronic materials corresponding to the cost details list materials required for business processing and the cost identification prompts into the multimodal large model. The large model will judge the materials and obtain relevant material information to generate the first cost details data.

[0049] Specifically: Step 3-1: Construct multimodal large model cost recognition prompts to guide the large model in recognizing cost list data. Multimodal large model cost recognition prompts include: {role / task + output method + additional knowledge}; Role / Task: Assign a specific role to a large model and specify the job of that role. For example: You are a data analyst who needs to generate detailed expense data from input images according to the output format; Output method: This guides the large model to output results in a fixed format after recognizing the content of electronic materials. For example, the large model may be required to output each line of data according to the project name, specifications, unit price, quantity, unit, and amount. Additional knowledge involves enumerating synonyms for each column, using these synonyms as supplementary knowledge to the larger model, and mapping fixed words to the same field for output.

[0050] In practical implementation, different fields have the following synonyms: Project Name: {Name, Project Name, Fee Item}; Specifications: {Specifications, Drug Specifications, Preparation Specifications}; Unit price: {unit price, unit price, selling price}; Unit: {unit, pricing unit}; Amount: {Amount, Amount Due, Total Price}. Step 3-2: Transmit the detailed cost list materials and cost identification prompts to the multimodal big language model. The multimodal big model will identify the first detailed cost data, including: project name, specifications, unit price, quantity, unit, and amount, and output it in a fixed format.

[0051] Step 3-3: Parse the first cost detail data output by the large model, and store the first cost detail data into the relational data and vector database according to the corresponding fields.

[0052] Step 4: Construct a vector database of the three major medical insurance catalogs, build a catalog matching intelligent agent, obtain similar catalog entities through near-nearest neighbor matching, select the closest similar catalog entity from the similar catalog entities through a large model, and combine it with the first expense detail data to form the second expense detail data.

[0053] Specifically: Step 4-1: Construct a vector database for the three major medical insurance catalogs. Store the information of the three major medical insurance catalogs into the vector database through embedding to form the vector data of the three major catalogs. Treat all data entries in the catalogs as a single entity.

[0054] The three major medical insurance catalogs refer to the medical insurance drug catalog, the medical treatment item catalog, and the medical service facility catalog. These three catalogs define the scope of basic medical insurance fund payments, and the specific information of the three catalogs may vary from city to city.

[0055] Step 4-2: Construct a directory matching agent, generate directory matching prompts by combining similar directory entities, and have the large model select the closest similar directory entity from several similar directory entities.

[0056] Step 421: Concatenate the field name, specification, and unit fields of each first expense detail data into a long text field to form the original expense information. After vectorizing the original expense information, perform an approximate nearest neighbor (ANN) search and match it with the three major medical insurance catalog vector databases to obtain and return several similar catalog entities with a similarity greater than a set threshold.

[0057] Step 422: Construct directory matching prompts to help the large model complete the mapping and matching between the cost details data and the three major directories. Directory matching prompts include {directory matching role + matching processing steps + directory matching requirements + similar directory entities}. Match roles to the directory; select a current role for the large model; the large model must complete the directory matching. The matching process guides how the large model works. It requires the large model to select the most similar directory entity to the cost details data from similar directory entities based on semantic similarity, and return the best matching directory entity and the confidence score of the match according to a fixed format. The directory matching requirement is that the large model should only output the most matching directory entity and must output the best matching directory entity and its confidence score. Similar directory entities are objects that are searched and matched from the three major medical insurance directory vector databases. They contain all the fields and field values ​​of the relevant directory entities.

[0058] Step 423: Transmit the prompt words to the large model, which then matches the closest similar directory entity for each first expense detail data.

[0059] Step 4-3: Obtain key information such as entity name, medical insurance code, co-payment ratio, and application restrictions from the similar directory entities fed back by the large model, and add them to the corresponding first expense detail data to form the second expense detail data.

[0060] Step 5: Construct an intelligent review rule knowledge base, store intelligent review rules according to a fixed storage logic, use a multimodal large model to obtain drug knowledge from drug instructions, complete the mapping of drug information with the corresponding three major directories, and perform structured processing of drug knowledge.

[0061] Step 5-1: Process the rules and knowledge points of the two intelligent supervision databases and the rules and knowledge accumulated by the medical insurance department in daily operations to generate review rule classification information, review rule information and review element information.

[0062] (1) Review rule classification information: used to classify different types of rules. For example, drugs are restricted to medical institutions of a certain level, drugs are restricted to certain genders, etc. Review rule classification information is formed according to different classifications.

[0063] The review rule classification information should include the review rule category ID, review rule category name, medical insurance catalog category, violation level, rule type, rule logic, violation example, review code, and text knowledge concatenation rules.

[0064] Among them, the categories of the medical insurance catalog are used to distinguish between drugs, consumables, medical service items, and others; The violation level indicates the severity of the violation, including clear violation, highly suspicious, and slightly suspicious. Rule characterization is used to characterize violations of rules, such as double charging for medical services. The rule logic indicates the processing logic of the rule. For example, if the medical information contains drugs that are restricted to certain medical institution levels, and a medical institution below the restricted level uses them and the drugs are covered by the medical insurance fund, then this rule is violated.

[0065] The violation examples are used to display cases of violations of this type of rule. For example, if "Zhongjiefeng Injection" is used and included in the medical insurance fund payment, and the medical institution is below the secondary level, the system will remind you of the violation. The review code is rule calculation code written in QL expressions; Text knowledge concatenation rules are the rules for concatenating specific fields in knowledge retrieval.

[0066] (2) Review rule information: Used to store specific rule clauses. For example, the restriction rules for a specific drug.

[0067] The audit rule information should include at least the audit rule ID, audit rule category ID, directory number, common name, rule usage, detection logic, and logical basis.

[0068] The catalog number refers to the three major catalog numbers corresponding to specific drugs, consumables, or services. The common name corresponds to the three major directory names; The usage of a rule indicates whether it belongs to a rule or knowledge. The detection logic is the logic for handling abnormal issues detected by the rule, such as the method of seeking medical treatment using medication not conforming to the restricted method of seeking medical treatment; The logical basis is the basis for the detection logic, such as limiting payment to patients with severe inflammation and infection, requiring nutritional risk screening, clarifying that they have nutritional risks, and only hospitalized patients who cannot obtain sufficient nutrition through diet or the use of "enteral nutrition agents".

[0069] (3) The audit elements are used to store the specific element values, element units and cycles when the rules are processed. The audit elements are then embedded into the QL expression to complete the rule matching.

[0070] Step 5-2: Use a multimodal large model to obtain drug knowledge from the drug instructions, complete the mapping of drug information to the corresponding three major directories, and perform structured processing of drug knowledge.

[0071] Step 521: Construct drug knowledge recognition prompts to assist the large model in acquiring drug knowledge. The prompts include {drug recognition role + recognition steps + drug recognition requirements}.

[0072] Drug identification role: Assign a role to the large model, requiring the large model to identify drug knowledge from the drug instructions in a fixed format; The drug identification process guides the data processing of large models. It involves segmenting the drug instruction manual into text blocks according to a fixed format, identifying specific content, and then converting it into a specific output format. Drug identification requirements: The large model must be output in a fixed format, including drug name, category, registered dosage form, specifications, ingredients, drug indications, contraindications, dosage and administration, adverse reactions, precautions, precautions for use in pregnant and lactating women, precautions for use in children, precautions for use in elderly patients, and drug incompatibilities. Step 522: Transmit the drug instructions and drug knowledge recognition prompts to the large model, which will generate relevant fields and organize them into preliminary drug knowledge information; Step 5-3: Use a large model to map drug knowledge to the medical insurance catalog, and associate the drug codes in the medical insurance catalog with the corresponding drug knowledge.

[0073] Step 531: Obtain the drug name, registered dosage form, and specification information from the drug knowledge information and concatenate them into a drug information field. Perform an approximate nearest neighbor (ANN) search and match the drug information field with the three major medical insurance catalog vector databases to obtain and return the several similar catalog entities with the highest similarity.

[0074] Step 532: Transmit the directory matching prompts and similar directory entities to the large model. The large model returns the best matching directory entity from the similar directory entities and associates the directory code of the best matching directory entity with the drug knowledge.

[0075] Step 6: Construct an intelligent auditing agent to audit the reasonableness of each expense in the second expense details data, output the audit results, and calculate the expense reimbursement status based on the audit results.

[0076] Step 6-1: Traverse each record in the second expense details data, and obtain the relevant audit rule classification information, audit rule information, audit element information, and drug knowledge information based on the medical insurance code of the record.

[0077] Step 6-2: Traverse each review rule information, analyze the rule category based on the rule usage attribute, and for review rules with the attribute "rule", generate review logic rules by combining the review code of the review rule category and the corresponding review elements, execute the review logic rules, and if the rule fails, the review fails.

[0078] Step 6-3: If the rule usage attribute is "knowledge", then the text knowledge of the corresponding review rule category is used to concatenate the project rule knowledge, and the review result is obtained by the big model intelligent review in combination with drug knowledge.

[0079] Step 631: Construct intelligent review prompts for the large model to assist the large model in completing intelligent review. The prompts for the large model during the intelligent review stage include: {intelligent review role + review steps + rule knowledge + drug knowledge + review requirements}. The review role is assigned to the large model, allowing the large model to determine whether the reimbursement item should be reimbursed using the medical insurance catalog based on the submitted second expense details data and relevant knowledge. The review process requires the large model to read the detailed data of the second-ranked drug and combine it with rule knowledge, drug knowledge and other business materials to determine whether the current medical insurance catalog can be used. Rule knowledge refers to rule knowledge generated by concatenating textual knowledge according to concatenation rules. Drug knowledge is the corresponding drug knowledge obtained from the catalog code in the drug knowledge associated with the medical insurance code in the second expense details data; The review requirements stipulate that the large model should concisely output the review process, review results, and confidence levels, and output the results in a fixed format.

[0080] Step 632: Transmit the second expense details data, other electronic materials for business processing, and intelligent review prompts from the big model to the big model. The big model then outputs the review results for each record.

[0081] Step 6-4: Based on the review results of each item, include the approved items in the reimbursement scope, and calculate the patient's total consumption amount, out-of-pocket amount, and reimbursement amount according to medical insurance standards.

[0082] It should be noted that all data acquisition is conducted in accordance with laws and regulations and with user consent, and the data is used legally.

[0083] Example 2 This embodiment provides a medical insurance intelligent review system based on a multimodal large model, including: The text recognition module is configured to retrieve the materials required for business processing and identify the material names and the names of the personnel to whom they belong. The compliance review module is configured to convert the material names and the materials required for business processing into sentence vectors, then perform mapping and matching through text similarity, and perform character matching of the names of the relevant personnel, so as to review the compliance of the materials required for business processing based on the completeness and correctness of the materials. The cost extraction module is configured to construct cost identification prompts. Based on the cost details list materials and the cost identification prompts, a multimodal large model is used to identify and obtain the first cost details data. The directory mapping module is configured to match the first expense detail data with the constructed medical insurance directory vector database to obtain several similar directory entities. Based on the constructed directory matching prompt words, the module selects the closest directory entity from the several similar directory entities and combines it with the first expense detail data to form the second expense detail data. The rule management module is configured to construct drug knowledge recognition prompts, which use a multimodal large model to obtain drug knowledge from drug instructions, match the drug knowledge with the medical insurance catalog vector database, and combine the catalog matching prompts to associate the catalog code of the best matching catalog entity with the corresponding drug knowledge. The review module is configured to build an intelligent review rule knowledge base. For each record of the second expense details data, it retrieves the corresponding intelligent review rules and drug knowledge based on the medical insurance code of each record, thereby reviewing the reasonableness of the expenses.

[0084] It should be noted that the above modules correspond to the steps described in Embodiment 1, and the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1. It should also be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer-executable instructions.

[0085] In further embodiments, the following is also provided: An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the method described in Embodiment 1. For brevity, further details are omitted here.

[0086] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0087] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.

[0088] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in Embodiment 1.

[0089] The method in Example 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.

[0090] A computer program product includes a computer program that, when executed by a processor, implements the method described in Embodiment 1.

[0091] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions included in program modules, which execute in a device on a target real or virtual processor to perform the processes / methods described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided among program modules as needed. The machine-executable instructions for the program modules can execute within a local or distributed device. In a distributed device, the program modules can reside in both local and remote storage media.

[0092] The computer program code used to implement the methods of the present invention may be written in one or more programming languages. This computer program code may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the computer or other programmable data processing device, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a computer, partially on a computer, as a stand-alone software package, partially on a computer and partially on a remote computer, or entirely on a remote computer or server.

[0093] In the context of this invention, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, and the like. Examples of signals may include electrical, optical, radio, sound, or other forms of propagation signals, such as carrier waves, infrared signals, etc.

[0094] Those skilled in the art will recognize that the units and algorithm steps described in connection with the various examples of this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention.

[0095] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A medical insurance intelligent review method based on a multimodal large model, characterized in that, include: Obtain the materials required for business processing and identify the names of the materials and the names of the personnel to whom they belong; After converting the material names and required materials for business processing into sentence vectors, text similarity is used for mapping and matching, and the names of the relevant personnel are matched character by character to verify the compliance of the materials required for business processing based on the completeness and correctness of the materials. Expense identification prompts are constructed, and based on the expense details list materials and expense identification prompts, a multimodal large model is used to identify the first expense details data; The first expense detail data is matched with the constructed medical insurance catalog vector database to obtain several similar catalog entities. Based on the constructed catalog matching prompt words, the closest catalog entity is selected from the several similar catalog entities and combined with the first expense detail data to form the second expense detail data. Construct drug knowledge recognition prompts to obtain drug knowledge from drug instructions using a multimodal large model, match the drug knowledge with the medical insurance catalog vector database, and combine the catalog matching prompts to associate the catalog code of the best matching catalog entity with the corresponding drug knowledge. A knowledge base of intelligent review rules is constructed. For each record of the second expense details data, the corresponding intelligent review rules and drug knowledge are obtained based on the medical insurance code of each record, so as to conduct a reasonableness review of the expenses.

2. The intelligent medical insurance review method based on a multimodal large model as described in claim 1, characterized in that, The process of identifying the material name and the name of the person to whom it belongs includes: The pre-trained OCR recognition model is used to recognize the text content of the material. The text size is determined based on the Y-coordinate distance between the top and bottom vertices of the text box, and the text with the largest font size is used as the material name. The text following a specific character representing a name is extracted to form the name of the person to whom the material belongs.

3. The intelligent medical insurance review method based on a multimodal large model as described in claim 1, characterized in that, The process of verifying the compliance of the materials required for business processing includes: The material names and the materials required for business processing are mapped to sentence vectors using a pre-trained SimBERT model, and the cosine similarity between the two is calculated. Based on cosine similarity, all material names are matched with the materials required for business processing. If the similarity is greater than a set threshold, the material name is mapped to the materials required for business processing. If all the materials required for business processing have corresponding mapping objects, the materials are considered complete. Match the names of the personnel to whom the materials belong with the names of the personnel in charge of the business; if the names match, the materials are considered to be correct. If all the required materials for the business transaction are complete and correct, the materials are considered compliant.

4. The intelligent medical insurance review method based on a multimodal large model as described in claim 1, characterized in that, The fee recognition prompt includes the role / task, output method, and additional knowledge. Directory matching suggestions include directory matching roles, matching processing steps, directory matching requirements, and similar directory entities; The drug knowledge identification prompts include the drug identification role, identification steps, and drug identification requirements.

5. The intelligent medical insurance review method based on a multimodal large model as described in claim 1, characterized in that, The matching process for similar directory entities includes: Concatenate the field name, specification, and unit fields of each first expense detail data to form the original expense information; The drug name, registered dosage form, and specifications information of the drug knowledge are combined into a drug information field; The original cost information or drug information fields are matched with the medical insurance catalog vector database using an approximate nearest neighbor search to obtain several similar catalog entities with a similarity greater than a set threshold.

6. The intelligent medical insurance review method based on a multimodal large model as described in claim 1, characterized in that, The process of reviewing the reasonableness of fees includes: Intelligent review rules include review rule classification information, review rule information, and review element information; Iterate through each review rule information. For review rules with the attribute "rule", combine the review code of the review rule classification information and the corresponding review element information to generate review logic rules and execute them. If the rule fails, the review fails. For review rules with the attribute of knowledge, the project rule knowledge is concatenated according to the text knowledge concatenation rules of the corresponding review rule classification information, and the review result is obtained by the multimodal big model in combination with drug knowledge; among them, the intelligent review prompt words of the multimodal big model include intelligent review role + review steps + rule knowledge + drug knowledge + review requirements.

7. A medical insurance intelligent review system based on a multimodal large model, characterized in that, include: The text recognition module is configured to retrieve the materials required for business processing and identify the material names and the names of the personnel to whom they belong. The compliance review module is configured to convert the material names and the materials required for business processing into sentence vectors, then perform mapping and matching through text similarity, and perform character matching of the names of the relevant personnel, so as to review the compliance of the materials required for business processing based on the completeness and correctness of the materials. The cost extraction module is configured to construct cost identification prompts. Based on the cost details list materials and the cost identification prompts, a multimodal large model is used to identify and obtain the first cost details data. The directory mapping module is configured to match the first expense detail data with the constructed medical insurance directory vector database to obtain several similar directory entities. Based on the constructed directory matching prompt words, the module selects the closest directory entity from the several similar directory entities and combines it with the first expense detail data to form the second expense detail data. The rule management module is configured to construct drug knowledge recognition prompts, which use a multimodal large model to obtain drug knowledge from drug instructions, match the drug knowledge with the medical insurance catalog vector database, and combine the catalog matching prompts to associate the catalog code of the best matching catalog entity with the corresponding drug knowledge. The review module is configured to build an intelligent review rule knowledge base. For each record of the second expense details data, it retrieves the corresponding intelligent review rules and drug knowledge based on the medical insurance code of each record, thereby reviewing the reasonableness of the expenses.

8. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, perform the method described in any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, implements the method described in any one of claims 1-6.

Images