Medical industry intelligent question and answer method and system based on hybrid expert large model

Through the integration of hybrid expert big model analysis and subdivided field processing models, the problem of intelligent question-and-answer system understanding professional knowledge in the medical field is solved, providing more accurate and detailed medical answers, and improving the quality of medical consultation.

CN118919059APending Publication Date: 2024-11-08SHANDONG LANGCHAO YUNTOU INFORMATION TECH CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411091691.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-09
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

Existing smart Q&A systems have difficulty understanding expertise and terminology when dealing with complex or professional medical issues, resulting in the inability to provide accurate or detailed answers.

Method used

Using the intelligent question-and-answer method of the medical industry based on the hybrid expert model, the target subdivision field is determined by analyzing the matching of medical problem information with the information database and medical knowledge graphs of multiple subdivision fields, and calling the corresponding subdivision field processing model. , integrate answers from various segments to provide medical answers.

Benefits of technology

It achieves more accurate and detailed answers to complex medical problems in multiple sub-fields, improves the accuracy and quality of medical consultation, and can understand and integrate professional knowledge in multiple sub-fields.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118919059A_ABST
    Figure CN118919059A_ABST
Patent Text Reader

Abstract

The invention discloses a medical industry intelligent question answering method and system based on a hybrid expert large model, and the method comprises the steps: obtaining to-be-processed medical problem information; analyzing a matching condition of the medical problem information and an information base of a plurality of subdivided fields and a matching condition of the medical problem information and a medical knowledge graph through a hybrid expert large model so as to determine a target subdivided field associated with the medical problem information; a subdivision field processing model corresponding to each target subdivision field is called to process the medical problem information, a subdivision field answer corresponding to the medical problem information in each target subdivision field is obtained, and the subdivision field processing model is constructed according to an information base of the corresponding subdivision field; and integrating the subdivided domain answers according to the hybrid expert large model to obtain medical answer information of the medical question information.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of intelligent question answering technology, and in particular to an intelligent question answering method and system for the medical industry based on a hybrid expert large model. Background Art

[0002] With the rapid development of artificial intelligence technology, intelligent question-answering systems have become one of the important tools in the information age and are widely used in various fields, including the medical field. In the medical field, intelligent question-answering systems can provide fast and accurate medical consultation services, help patients or doctors obtain the required medical information, and improve the efficiency and quality of medical services.

[0003] The medical field is highly specialized and often involves a large amount of professional knowledge and terminology, which are generally difficult for existing intelligent question-answering systems to understand. As a result, existing intelligent question-answering systems are unable to provide accurate or detailed answers when dealing with some complex or professional medical issues. Summary of the invention

[0004] To this end, this application discloses the following technical solutions:

[0005] The first aspect of the present application provides a medical industry intelligent question answering method based on a hybrid expert large model, comprising:

[0006] Obtain information on pending medical issues;

[0007] Analyze the matching of the medical problem information with the information bases of multiple sub-fields and the matching of the medical problem information with the medical knowledge graph through a hybrid expert big model to determine the target sub-field associated with the medical problem information;

[0008] Calling a subdivision field processing model corresponding to each of the target subdivision fields to process the medical problem information, and obtaining a subdivision field answer corresponding to the medical problem information in each of the target subdivision fields;

[0009] The answers to the subdivided fields are integrated to obtain medical answer information for the medical problem information.

[0010] Optionally, obtaining the information of the medical problem to be processed includes:

[0011] Get the current question entered by the user;

[0012] Obtaining the historical questions input by the user before the current question and the historical answers corresponding to the historical questions;

[0013] The current question, the historical question and the historical answer are integrated to obtain the medical question information to be processed.

[0014] Optionally, also include:

[0015] identifying user emotion information representing an emotional state of a corresponding user according to the medical problem information;

[0016] The step of integrating the answers to the subdivided fields to obtain the medical answer information of the medical question information includes:

[0017] The answers to the subdivided fields are integrated according to the user emotion information to obtain medical answer information for the medical question information.

[0018] Optionally, also include:

[0019] Obtaining user feedback information corresponding to the medical answer information;

[0020] The hybrid expert large model is optimized according to the user feedback information.

[0021] Optionally, a method for constructing a segmented field processing model corresponding to any segmented field includes:

[0022] Obtaining an information database corresponding to the subdivision field;

[0023] According to a plurality of dialogue samples corresponding to the subdivided fields, each of the dialogue samples includes at least one sample question and at least one sample answer corresponding to the sample question;

[0024] Based on the information base corresponding to the subdivided field, an initial small parameter model is called to process the conversation sample to obtain a model prediction answer corresponding to the conversation sample;

[0025] The model loss of the initial small-parameter model is determined according to the model prediction answer and the sample answer, and the model parameters of the initial small-parameter model are adjusted according to the model loss until the model loss converges; wherein, when the model loss converges, the initial small-parameter model serves as the segmentation field processing model corresponding to the segmentation field.

[0026] Optionally, also include:

[0027] Regularly update the information database corresponding to the subdivision field;

[0028] The subdivision field processing model corresponding to the subdivision field is optimized according to the updated information base corresponding to the subdivision field.

[0029] Optionally, the step of integrating the answers to the subdivided fields to obtain medical answer information for the medical question information includes:

[0030] Obtaining auxiliary integration information, wherein the auxiliary integration information includes at least one of medical object status information, medical risk information, and expected effect information;

[0031] The answers to the subdivided fields are integrated according to the auxiliary integration information to obtain medical answer information for the medical question information.

[0032] Optionally, also include:

[0033] Determining language expression features representing the user's language expression habits according to the medical problem information;

[0034] Optimize each of the subdivided field answers according to the language expression features to obtain a corresponding optimized subdivided field answer;

[0035] The step of integrating the subdivided field answers according to the auxiliary integration information to obtain medical answer information for the medical question information includes:

[0036] The plurality of optimized answers to the subdivided fields are integrated according to the auxiliary integration information to obtain medical answer information for the medical problem information.

[0037] The second aspect of the present application provides a medical industry intelligent question-answering system based on a hybrid expert large model, comprising:

[0038] An acquisition unit, used for acquiring information of medical problems to be processed;

[0039] An analysis unit, configured to analyze the matching of the medical problem information with information bases of multiple subdivisions and the matching of the medical problem information with a medical knowledge graph through a hybrid expert big model, so as to determine a target subdivision associated with the medical problem information;

[0040] A processing unit, configured to call a subdivision field processing model corresponding to each of the target subdivision fields to process the medical problem information, and obtain a subdivision field answer corresponding to the medical problem information in each of the target subdivision fields;

[0041] The integration unit is used to integrate the answers to the subdivided fields to obtain medical answer information for the medical problem information.

[0042] Optionally, when the obtaining unit obtains the medical problem information to be processed, it is specifically used to:

[0043] Get the current question entered by the user;

[0044] Obtaining the historical questions input by the user before the current question and the historical answers corresponding to the historical questions;

[0045] The current question, the historical question and the historical answer are integrated to obtain the medical question information to be processed.

[0046] The beneficial effects of this solution are:

[0047] There are a large number of different sub-fields in the medical field (for example, sub-fields corresponding to different departments such as dermatology, cardiology, respiratory medicine, otolaryngology, etc.). Each sub-field involves a large amount of professional knowledge and terminology. A single question-answering model cannot understand the knowledge of multiple sub-fields, resulting in inaccurate and inadequate answers to questions in the medical field. This solution uses the information base of each sub-field to build a sub-field processing model specifically for understanding the knowledge in that field. When dealing with medical problems, the target sub-fields related to the medical problem information are first analyzed through the hybrid expert large model, and then the sub-field processing model corresponding to the target sub-field is called to give the answer to each sub-field. Finally, the answers to multiple sub-fields are integrated to obtain medical answer information. In summary, this solution uses multiple sub-field processing models to solve the problem that a single model cannot understand the knowledge of each sub-field, so that the large model can give more accurate and detailed medical answers to complex medical problems in various fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0049] Figure 1 It is a flow chart of an intelligent question-answering method for the medical industry based on a hybrid expert large model provided in an embodiment of the present application;

[0050] Figure 2 It is a structural diagram of a medical industry intelligent question-answering system based on a hybrid expert large model provided in an embodiment of the present application. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0052] This application embodiment provides a medical industry intelligent question answering method based on a hybrid expert large model. Figure 1 , is a flow chart of the method, and the method may include the following steps.

[0053] S101, obtaining medical problem information to be processed.

[0054] This embodiment can be executed by an intelligent question-answering computer system (hereinafter referred to as the system) that is pre-configured with a large hybrid expert model and multiple segmented domain processing models.

[0055] In step S101, the system may receive a medical-related question input by any user, and determine the input question as medical question information to be processed. The user may input the question by voice or text.

[0056] The users of this embodiment may include patients who need to consult medical issues, and may also include medical personnel such as doctors and nurses.

[0057] The question input by the user may include any medical-related information. For example, the question input may be a question about the cause of a specific symptom, in which case the question may include a description of the symptom; or the question input may be a question about whether a specific medical plan is recommended, in which case the question may include one or more medical plans for a specific disease.

[0058] In some optional embodiments, the system may also obtain medical problem information in the following manner:

[0059] Get the current question entered by the user;

[0060] Obtain the historical questions entered by the user before the current question and the historical answers corresponding to the historical questions;

[0061] Current questions, historical questions and historical answers are integrated to obtain the medical problem information to be processed.

[0062] Among them, the historical questions may include each question that the same user has asked before the current question after logging into the system this time, and the historical answers may include the medical answer information output by the system for each historical question. For example, after user A logs into the system, he asks 5 questions successively. For each question, the system processes and outputs the corresponding medical answer information according to the method of this embodiment. When user A asks the 6th question, the system can merge the 6th question, the previous 5 questions and the 5 medical answer information corresponding to the 5 questions to obtain the medical question information to be processed.

[0063] When performing fusion processing, the system can use cutting-edge natural language processing technology to capture the keywords, professional terms and concepts in each historical question and the corresponding historical answer, and splice the extracted keywords, professional terms and concepts into the current question to form the medical question information to be processed, thereby achieving a deep understanding of the context.

[0064] For example, assuming that the user described some symptoms in the previous question and described the examination results or treatment experience in the current question, the system can extract keywords describing the symptoms from previous historical questions, and extract professional terms and concepts corresponding to the symptoms described by the user from the historical answers to the historical questions, and combine these symptom-related keywords, professional terms and concepts with the currently input examination results or treatment experience to form medical problem information to be processed. This will help to integrate this information for in-depth analysis in the subsequent processing process, thereby giving more accurate diagnosis or treatment recommendations.

[0065] The above fusion processing system can call a pre-built hybrid expert large model for execution.

[0066] In order to enhance the ability of the hybrid expert model to integrate context in the above manner, when training the hybrid expert model, some consecutive rounds of medical question-and-answer conversations between patients and doctors can be selected as training data sets, so as to train a model that can support the above-mentioned integrated continuous conversations.

[0067] The benefit of obtaining medical problem information in the above manner is that in the field of medical consultation, the communication between patients and doctors often develops dynamically, involving multiple rounds of dialogue and the exchange of complex information. In order to simulate multiple rounds of dialogue between patients and doctors, the system integrates historical questions and answers into the current medical problem information to be processed, which can not only analyze the patient's current problem in real time, but also review and integrate the content of previous dialogues, thereby providing coherent, accurate and personalized medical advice, significantly improving the user's consultation experience, and ensuring that users can obtain more professional and meticulous medical services.

[0068] S102, using a hybrid expert big model to analyze the matching of medical problem information with information bases of multiple sub-fields, and the matching of medical problem information with the medical knowledge graph, to determine the target sub-field associated with the medical problem information.

[0069] The medical problem information to be processed may be associated with one or more target segments.

[0070] Before executing step S102, an information base corresponding to each subdivision field may be constructed. The information base of each subdivision field may include relevant knowledge of the field, for example, basic theories, latest research results, clinical practice experience and other knowledge of the field.

[0071] In this embodiment, multiple sub-fields can be divided according to the division of departments in the hospital, for example, internal medicine, surgery, obstetrics and gynecology, pediatrics, dermatology and other sub-fields can be divided. Correspondingly, an internal medicine information database, a surgical information database, an obstetrics and gynecology information database, etc. can be constructed.

[0072] In order to build a detailed information database, for each sub-field, the system can collect medical literature, clinical guidelines, case reports and other materials in the sub-field through various methods such as manual entry and document scanning, and then use a large language model (such as Ghat-GPT) to conduct in-depth analysis and organization of the collected materials, and enter the organized information into the information database corresponding to the sub-field to form an information database corresponding to the sub-field.

[0073] Among them, medical literature and clinical guidelines can be collected from pre-designated literature databases, and case reports can be collected from hospital case databases.

[0074] By analyzing and integrating medical literature, clinical guidelines, case reports and other materials, we can ensure that the constructed information database has sufficient breadth (covering both theoretical research and clinical practice) and depth in the corresponding sub-fields, thereby ensuring the accuracy and authority of the sub-field processing model established based on the information database when providing medical advice.

[0075] For example, in the field of obstetrics and gynecology, the corresponding information database can collect case reports of pregnant women at different stages of pregnancy and relevant medical literature at different stages of pregnancy. This can ensure that the obstetrics and gynecology treatment model built based on the information database can not only understand the symptoms and complications that may be encountered at different stages of pregnancy, but also provide personalized medical guidance and health advice based on the latest medical research.

[0076] In the field of dermatology, the corresponding information database can include case reports related to various skin diseases, so that the dermatology treatment model built based on the information database can identify and distinguish various skin lesions and provide patients with scientific treatment plans.

[0077] In step S102, the system can use the hybrid expert model to determine the target sub-fields associated with the medical problem information from the following two aspects.

[0078] First, the hybrid expert model analyzes the matching of medical problem information with information bases of multiple segments to determine the target segment associated with the medical problem information.

[0079] Specifically, the hybrid expert model can extract professional terms and concepts describing symptoms, professional terms and concepts describing causes of disease, and professional terms and concepts describing treatment methods (including drugs and surgery) from medical problem information, and match these extracted professional data and concepts with the information database of each sub-field. If the match is successful, it can be considered that the medical problem information to be processed is related to this sub-field, and this sub-field is identified as the target sub-field.

[0080] For example, if the professional terms and concepts describing the cause of disease extracted by the hybrid expert model appear in the cardiology information database, then it can be considered that the medical problem information to be processed is related to the cardiology sub-field, and cardiology is therefore determined as a target sub-field associated with the medical problem information.

[0081] Secondly, the hybrid expert big model analyzes the matching between medical problem information and the medical knowledge graph to determine the target sub-segment areas associated with the medical problem information.

[0082] In order to determine the target sub-field according to the second aspect, the system of this embodiment can construct a medical knowledge graph in advance based on the information bases of multiple sub-fields.

[0083] The medical knowledge graph may include the following nodes:

[0084] Symptom nodes: Each symptom node is used to describe a symptom. Different symptom nodes describe different symptoms. For example, one symptom node can describe a high fever, while another symptom node can describe abdominal pain.

[0085] Cause node: Each cause node can describe a cause. Different cause nodes describe different causes. For example, the cause described by one cause node may be appendicitis, while the cause described by another cause node may be influenza A virus infection.

[0086] Drug nodes. Each drug node can correspond to a drug. Different drug nodes correspond to different drugs. For example, one drug node corresponds to drug X, and another drug node corresponds to drug Y.

[0087] Surgical node: Each surgical node can correspond to a type of surgery. Different surgical nodes correspond to different surgeries. For example, one surgical node corresponds to surgery A, and another surgical node corresponds to surgery B.

[0088] Any two nodes may be connected to each other through an edge or not. If there is an edge connecting any two nodes, it means that the two nodes are correlated. Each edge can correspond to a weight value. The larger the weight value, the stronger the correlation between the two nodes connected by this edge, and the smaller the weight value, the weaker the correlation.

[0089] For example, if there is an edge connection between two symptom nodes, it means that the two symptoms may occur at the same time, and the stronger the weight value, the greater the probability of the two symptoms occurring simultaneously;

[0090] If there is an edge connection between two drug nodes, it means that the two drugs are generally taken at the same time. The stronger the weight value, the greater the probability of taking them at the same time.

[0091] If there is an edge connection between the drug node and the symptom node, it means that patients with the symptom generally take the drug. The larger the weight value, the greater the probability that the patient with the symptom takes the drug.

[0092] If there is an edge connection between the cause node and the symptom node, it means that the symptom may be caused by the cause. The larger the weight value, the greater the probability that the symptom is caused by the cause.

[0093] The system can use any existing knowledge graph construction technology to process the information base of each sub-field to obtain the above-mentioned medical knowledge graph.

[0094] After constructing the above knowledge graph, the system can match the medical problem information to be processed with the nodes in the medical knowledge graph to find multiple matching nodes.

[0095] For example, if a symptom C is described in the medical problem information, then the node in the medical knowledge graph that also describes the symptom C is a matching node; if a surgery A is described in the medical problem information, then the node in the medical knowledge graph that also describes the surgery A is a matching node.

[0096] For each matching node, the system may determine as an associated node a node that has an edge connection with the matching node and a weight value corresponding to the edge that is greater than a screening threshold. After traversing each matching node in this manner, a number of associated nodes may be obtained.

[0097] Exemplarily, if there is an edge connection between the node of symptom C and the node of cause D, and the weight value of the edge is greater than the screening threshold, it can be determined that cause D is an associated node.

[0098] The screening threshold can be set by the user, or it can be a fixed value preset by the system, or it can be determined by the system based on the number of edges connected to the matching node. If a matching node is connected to more edges, then a larger screening threshold can be determined when determining the associated node corresponding to the matching node. If a matching node is connected to fewer edges, then a smaller screening threshold can be determined when determining the associated node corresponding to the matching node.

[0099] That is to say, for each matching node, the associated nodes of the matching node may be screened according to the screening threshold corresponding to the matching node, and the screening threshold is proportional to the number of edges connected to the matching node.

[0100] After multiple associated nodes are identified, the subdivided fields to which the symptoms, causes, drugs, and surgeries described by these associated nodes belong can be determined and identified as target subdivided fields.

[0101] By determining the target sub-segment in the manner of the second aspect, the system can not only identify the various sub-segments clearly stated in the user's question, but also analyze other sub-segments that the user failed to point out due to the limitations of his or her own experience and knowledge, thereby helping to provide the user with more accurate medical answer information.

[0102] By constructing the above medical knowledge graph, we can effectively connect and integrate the knowledge points and concepts of different departments and improve the ability to integrate cross-departmental knowledge. This integration not only improves the system's ability to handle complex problems involving professional knowledge from multiple departments at the same time, but also enables patients to obtain more comprehensive and in-depth medical consulting services.

[0103] In addition, the integration of cross-department knowledge can also help improve the coherence and consistency of medical consultation. When faced with complex cases involving multiple departments, the system can comprehensively consider various factors, avoid one-sided and misleading information, and provide more scientific and reasonable medical advice. This cross-domain knowledge integration not only improves the accuracy of consultation, but also greatly improves user satisfaction and trust, which is conducive to bringing users a more humane and high-quality medical service experience.

[0104] S103, calling the segment field processing model corresponding to each target segment field to process the medical problem information, and obtaining the segment field answer corresponding to the medical problem information in each target segment field.

[0105] Each sub-field can correspond to a sub-field processing model built using the information base of the sub-field. For example, obstetrics and gynecology can correspond to the obstetrics and gynecology processing model, and dermatology can correspond to the dermatology processing model.

[0106] In this embodiment, each subdivision field processing model may be a small parameter model.

[0107] For each subdivision, the system can use the information library of the subdivision to conduct targeted training on an initial small parameter model, thereby obtaining a subdivision processing model that can accurately grasp the medical knowledge of its specific subdivision. By constructing and training a series of such small models corresponding to specific subdivisions, the system of this embodiment can comprehensively cover all areas of the medical industry and provide patients with more professional and accurate medical consulting services.

[0108] Specifically, for any sub-field, the sub-field processing model corresponding to the sub-field can be constructed according to the following construction method.

[0109] A1, obtain the information database corresponding to the subdivision field;

[0110] A2, based on multiple dialogue samples corresponding to the subdivided fields, each dialogue sample includes at least one sample question and at least one sample answer corresponding to the sample question;

[0111] A3, based on the information database corresponding to the subdivided field, calls the initial small parameter model to process the dialogue sample and obtains the model prediction answer corresponding to the dialogue sample;

[0112] A4. Determine the model loss of the initial small parameter model according to the model prediction answer and the sample answer, and adjust the model parameters of the initial small parameter model according to the model loss until the model loss converges; wherein, when the model loss converges, the initial small parameter model serves as the segmentation field processing model corresponding to the segmentation field.

[0113] In step A2, the dialogue sample can be obtained as follows:

[0114] Collect case reports of departments corresponding to the subdivided fields, as well as conversations between patients and doctors, and between doctors in the corresponding departments. Extract questions raised by patients and doctors from the conversations as sample questions, parse the corresponding answers from the case reports as sample answers, and obtain conversation samples. Conversation samples can be differentiated by patient, that is, several questions and answers extracted from the conversations and case reports of one patient are used as one conversation sample, and several questions and answers extracted from the conversations and case reports of another patient are used as another conversation sample.

[0115] In step A3, the sample questions in the dialogue sample and the information base of the segmented field can be input into the initial small parameter model, and the small parameter model is used to process and generate a model to predict the answer.

[0116] In step A4, for each dialogue sample, the model predicted answer corresponding to the dialogue sample and the sample answer can be compared to determine the degree of difference between the model predicted answer and the sample answer corresponding to each dialogue sample, and the model loss is determined in combination with the degree of difference of multiple dialogue samples. The smaller the overall degree of difference of the multiple dialogue samples, for example, the smaller the average value of the degree of difference, the smaller the model loss; the greater the overall degree of difference of the multiple dialogue samples, for example, the greater the average value of the degree of difference, the greater the model loss.

[0117] Steps A3 and A4 can be iterated multiple times. After adjusting the model parameters once according to step A4, the process returns to A3.

[0118] Model loss convergence can be defined as the model loss being less than or equal to a preset convergence threshold.

[0119] The specific method of adjusting model parameters according to model loss can be found in the relevant prior art and will not be described in detail.

[0120] The advantages of using a small parameter model as the segment processing model for each segment are:

[0121] In the design of medical question-answering systems, although traditional large-parameter models are powerful, they often consume huge computing resources when dealing with complex medical problems and are accompanied by high training and maintenance costs.

[0122] A significant advantage of small parameter models is their high efficiency in training and inference while maintaining high accuracy. Such models are particularly suitable for dealing with specific medical problems that require expertise, such as accurate diagnosis of specific diseases or appropriateness assessment of specific drug use.

[0123] In addition, models with small parameters have obvious advantages in terms of interpretability. In the medical field, the transparency of the model's decision-making process is crucial to building trust between doctors and patients. Due to the simplicity of its structure, models with small parameters are easier to interpret and understand, allowing doctors and patients to clearly grasp the model's reasoning logic and decision-making basis. This high degree of interpretability not only helps to increase patients' trust in medical advice, but also provides doctors with a powerful auxiliary tool, enabling them to better understand the model's recommendations and make appropriate adjustments or supplements when necessary.

[0124] In some optional embodiments, the system may also periodically optimize the subdivision field processing model corresponding to each subdivision field in the following manner:

[0125] Regularly update the information database corresponding to the sub-sectors;

[0126] The segment processing model corresponding to the segment is optimized according to the updated information base corresponding to the segment.

[0127] The update method can be:

[0128] For each sub-field, every update cycle (for example, 1 day, 7 days, etc.), the system checks whether there are new documents in the sub-field in the detection literature database and case database. The new documents include at least one of new medical literature, new clinical guidelines and new case reports. If so, the system collects the new documents in the sub-field, and then extracts the knowledge of the sub-field in the new documents and stores it in the information database of the sub-field in the manner of constructing the information database in the aforementioned embodiment, thereby completing the update.

[0129] After updating the information base, the system can optimize the segment processing model corresponding to the segment as follows:

[0130] Based on the updated information library corresponding to the segmented field, the segmented field processing model of the segmented field is called again to process the dialogue sample in step A3 to obtain the model prediction answer corresponding to the dialogue sample, and then continue to determine the model loss of the segmented field processing model in the manner of step A4, and adjust the parameters based on the model loss until the model loss converges.

[0131] The segmentation domain processing model when the model loss converges is the new segmentation domain processing model optimized according to the updated information base.

[0132] The system can realize a continuous learning mechanism through the above optimization method. With the continuous updating of medical knowledge and the development of clinical practice, the system can regularly update the information base of each sub-field processing model through a continuous learning mechanism and update the sub-field processing model accordingly, so as to ensure that the latest medical advice provided always meets the current medical standards. This continuous learning and updating, combined with the efficiency and interpretability of the small parameter model, provides the system of the present invention with strong vitality and lasting competitiveness.

[0133] Optionally, when optimizing the segmented domain processing model, the optimization method can also be simplified. Specifically, the segmented domain processing model can be optimized only based on the newly added information in the updated information base, rather than based on the entire updated information base. This method not only greatly reduces the time and resources required for the update, but also ensures the continuous operation of the system during the update process, avoids service interruptions, and ensures the continuity and reliability of medical consultation services. It allows the system to respond quickly to changes in medical knowledge, significantly improves the update efficiency and flexibility compared to traditional full system retraining, and ensures the timeliness and accuracy of information.

[0134] In addition, the above simplified method significantly improves the maintainability of the system. System administrators can flexibly adjust and optimize the processing models of various sub-fields through a simplified update process based on user feedback and needs. This flexibility enables the system to quickly adapt to changes in user needs and improve personalized service capabilities. It not only improves the adaptability of the system, but also enhances user satisfaction. Through this continuous optimization and updating, the medical question and answer system of the present invention can maintain its leading position in the field of medical consultation, provide users with high-quality services, and provide medical professionals with a reliable knowledge update platform to ensure that they can obtain the latest medical information.

[0135] Optionally, the above process of optimizing the subdivided domain processing model can also be performed when it is determined that the current subdivided domain processing model is idle. For example, if the system is using the obstetrics and gynecology processing model to process the problem, the obstetrics and gynecology processing model is not optimized according to the above method, and the optimization is performed after the processing is completed.

[0136] On the basis of constructing the segmented field processing model for each segmented field, in step S103, the system can call the segmented field processing model corresponding to each target segmented field, and transmit the medical problem information and the information library of the target segmented field to the corresponding segmented field processing model, so that the segmented field processing model gives the corresponding segmented field answer based on the professional knowledge of the segmented field.

[0137] Exemplarily, the determined target sub-fields may include dermatology and internal medicine. In S103, the system may transmit the medical problem information and the dermatology information library to the dermatology processing model, so that the dermatology processing model gives answers to the dermatology sub-fields related to dermatology based on the knowledge of dermatology. The system may also transmit the medical problem information and the internal medicine information library to the internal medicine processing model, so that the internal medicine processing model gives answers to the internal medicine sub-fields related to internal medicine based on the knowledge of internal medicine.

[0138] The content of the answer to the subdivided field may be related to the question raised by the user and the corresponding subdivided field, which is not limited in this embodiment.

[0139] For example, when the question asked by the user is about the cause of specific symptoms, the answer to each segmented field may include: within the segmented field, the possible causes of the symptoms described by the user in the question, and the treatment plans recommended for these causes in the segmented field; when the question asked by the user is about whether a specific medical plan is recommended, the answer to each segmented field may include: what effects and side effects the specific medical plan described by the user will have in the corresponding segmented field, and whether it will cause some complications in the corresponding segmented field.

[0140] S104, integrating the answers in the subdivided fields to obtain medical answer information for the medical problem information.

[0141] Step S104 can be implemented through the aforementioned hybrid expert big model, that is, the system can input one or more answers obtained in the subdivided fields into the hybrid expert big model, and use the hybrid expert big model to integrate these answers in the subdivided fields to obtain the final medical answer information.

[0142] In the architecture of the hybrid expert model, a question may require the joint participation and answer of multiple segmented domain processing models. In order to ensure that the medical consultation provided to the patient is both accurate and consistent in content, the system of this embodiment can use the hybrid expert model to integrate the segmented domain answers output by one or more segmented domain processing models. This integration mechanism can carefully integrate and coordinate the responses from different segmented domain processing models to generate a comprehensive, coherent, and easy-to-understand final answer, namely the above-mentioned medical answer information.

[0143] The beneficial effects of this embodiment are:

[0144] There are a large number of different sub-fields in the medical field (for example, sub-fields corresponding to different departments such as dermatology, cardiology, respiratory medicine, otolaryngology, etc.). Each sub-field involves a large amount of professional knowledge and terminology. A single question-answering model cannot understand the knowledge of multiple sub-fields, resulting in inaccurate and inadequate answers to questions in the medical field. This solution uses the information base of each sub-field to build a sub-field processing model specifically for understanding the knowledge in that field. When dealing with medical problems, the target sub-fields related to the medical problem information are first analyzed through the hybrid expert large model, and then the sub-field processing model corresponding to the target sub-field is called to give the answer to each sub-field. Finally, the answers to multiple sub-fields are integrated to obtain medical answer information. In summary, this solution uses multiple sub-field processing models to solve the problem that a single model cannot understand the knowledge of each sub-field, so that the large model can give more accurate and detailed medical answers to complex medical problems in various fields.

[0145] On the other hand, through the above method, the system of this embodiment can use the hybrid expert model to integrate multiple specially trained segmented domain processing models to form a comprehensive intelligent system with extensive domain expertise. This design not only improves the flexibility of the system in handling problems, but also significantly enhances its ability to solve complex medical problems. The system can achieve the synergistic effect of domain expertise and knowledge complementarity. When a user raises a medical consultation, the system can intelligently identify the key areas involved in the problem and call the segmented domain processing model of the corresponding field for in-depth analysis and answer.

[0146] For example, when facing issues related to heart disease, the system can call up a sub-field of the heart disease field for processing, and use its corresponding information database in the heart disease field to provide professional and accurate medical advice. This on-demand calling mechanism ensures that the answers patients receive are not only professional, but also highly targeted.

[0147] In addition, the hybrid expert model shows unique advantages in dealing with cross-domain problems. In medical practice, many diseases and treatment strategies often involve knowledge and skills from multiple disciplines. By integrating sub-domain processing models from different fields, the hybrid expert model can achieve knowledge complementarity and integration, and provide patients with more comprehensive and in-depth medical consultation. For example, in the face of complex cases involving the heart and kidneys, the system can simultaneously call the sub-domain processing model corresponding to the field of heart disease and the sub-domain processing model corresponding to the field of kidney disease, thereby integrating the expertise of the two aspects and formulating a comprehensive treatment plan for the patient.

[0148] Furthermore, the system of this embodiment can continuously optimize the performance of the processing models in each segment based on patient feedback and new medical data, and improve the accuracy and reliability of the overall system. Through continuous learning and iteration, it ensures that the system can adapt to the rapidly changing medical environment and maintain its leading position in the field of medical consultation.

[0149] Furthermore, in the face of the high resource consumption problem of large parameter models in traditional medical question-and-answer systems, the system of this embodiment adopts a solution of training specific segmented field processing models for each segmented field separately and integrating these segmented field processing models through a hybrid expert large model. This solution not only simplifies the training process and reduces resource requirements, but also through an intelligent reasoning mechanism, the system can accurately call small models in related fields according to the nature of the problem, avoiding unnecessary calculations and significantly improving operating efficiency. In addition, this design also enhances the scalability of the system, so that as medical knowledge is updated, the addition of new segmented field processing models can be easily realized without interfering with the existing system, ensuring long-term system stability and maintainability.

[0150] In some optional embodiments, the system of this embodiment can also identify the user's emotions in the question-answering process in the following manner, and adjust the final output medical answer information according to the user's emotions. The specific adjustment method is as follows:

[0151] First, the system can identify user emotion information representing the emotional state of the corresponding user based on the medical problem information.

[0152] The system can use existing natural language processing technology to analyze the current and historical questions contained in the medical question information, extract emotional keywords related to the user's emotions, and then determine the user's emotional information based on these emotional keywords.

[0153] Optionally, the determined user emotion information may specifically include different emotions of the user towards issues in different aspects, or different emotions of the user towards different sub-fields.

[0154] On the one hand, the current and historical issues may include those related to the expected course of the disease, the expected therapeutic effects of treatment plans, the causes of symptoms, the economic costs of treatment plans, etc.

[0155] In this case, emotional information may include emotions towards the above-mentioned different aspects of the problem. The system can determine the user's degree of concern about different aspects of the problem based on the user's emotions towards different aspects of the problem. For example, it is determined that the user has the highest degree of concern about issues related to the expected treatment effects of the treatment plan, and has a lower degree of concern about issues related to the expected development of the disease.

[0156] Secondly, the current and historical questions may include issues involving multiple sub-fields, such as issues involving pediatrics, issues involving obstetrics and gynecology, issues involving internal medicine, etc.

[0157] In this case, the emotional information may include emotions towards issues in the above-mentioned different sub-fields. The system can determine the user's degree of concern about different sub-fields based on the user's emotions towards issues in different sub-fields. For example, it can be determined that the user has the highest degree of concern about obstetrics and gynecology issues and a lower degree of concern about pediatric issues.

[0158] The system can identify the user's emotional information before integrating it into the medical answer information, and then perform the following steps based on the emotional information:

[0159] Integrate answers to segmented fields based on user emotional information to obtain medical answer information for medical problems.

[0160] After obtaining the user emotion information of the first aspect, the system can determine the user's attention to different aspects of the questions based on the user emotion information, and then adjust the order of the answers to the subdivided fields of different aspects from high to low according to the attention level. For example, if the question about the expected treatment effect of the treatment plan has the highest attention level, and the question about the expected disease progression has a lower attention level, the system will rank the subdivided field answers related to the expected treatment effect first, and rank the subdivided field answers related to the expected disease progression last;

[0161] After obtaining the user emotional information in the second aspect mentioned above, the system can determine the user's concern level for issues in different sub-fields based on the user emotional information, and then adjust the order of answers to different sub-fields from high to low according to the degree of concern. For example, if the user's concern level for obstetrics and gynecology is higher and the concern level for pediatrics is lower, the system will rank the answers to obstetrics and gynecology sub-field first and the answers to pediatrics sub-field last.

[0162] After the sorting is completed, the system integrates the answers in these subdivided fields according to the sorting results to form medical answer information, thereby ensuring that answers that users pay more attention to are displayed first in the medical answer information, and answers that users pay less attention to are displayed in the subsequent part of the medical answer information.

[0163] Through the above method, the system can adjust its answer strategy according to the patient's mood and tone, providing more considerate and humane services. This comprehensive understanding of the context enables the system to cope with more complex medical consultation scenarios such as multi-symptom diagnosis and treatment plan discussion, ensuring the professionalism and pertinence of medical advice.

[0164] In some optional embodiments, the system can also optimize the hybrid expert large model as follows:

[0165] Obtain user feedback information corresponding to the medical answer information;

[0166] The hybrid expert model is optimized based on user feedback.

[0167] The user feedback information may include information on the user's satisfaction with the medical answer information output by the system.

[0168] The optimization process can be:

[0169] The feedback information and the medical answer information are input into the hybrid expert big model together. After the hybrid expert big model obtains the feedback information and the medical answer information, the medical answer information can be classified and stored according to the feedback information. For example, a type of medical answer information that satisfies the user can be stored in one database, and a type of medical answer information that the user is dissatisfied with can be stored in another database. When the medical answer information is requested again, the hybrid expert big model predicts the expected satisfaction level of the medical answer information to be outputted based on the above two types of medical answer information. If the expected satisfaction level is insufficient, the medical answer information to be outputted is adjusted to obtain medical answer information with a higher expected satisfaction level, so that the output medical answer information is more in line with the needs of users.

[0170] In the above-mentioned optimization method for the large hybrid expert model, this embodiment introduces the reinforcement learning technology in machine learning. Through actual interaction with patients, the system can continuously learn and self-optimize to form a more efficient dialogue strategy. This adaptive learning mechanism enables the system to quickly adapt to new medical problems and consulting needs, and adjust the answer strategy in time to provide more accurate and useful medical information. With the passage of time and the accumulation of data, the system will become more intelligent, better able to predict patients' needs, and provide more accurate medical consulting services. Through this continuous learning and optimization, the system will continuously improve its professional level and service quality in the field of medical question and answer.

[0171] Optionally, when integrating the answers of the sub-fields, the hybrid expert model can also refer to the auxiliary integration information to integrate the answers of each sub-field, as follows:

[0172] Obtaining auxiliary integration information, the auxiliary integration information including at least one of medical object status information, medical risk information, and expected effect information;

[0173] Answers to subdivided fields are integrated based on the auxiliary integrated information to obtain medical answer information for medical problem information.

[0174] Exemplarily, when the user asking the question is a patient, the system can obtain from the relevant database the patient's health status (i.e., medical object status information), the treatment risks corresponding to the treatment plans currently consulted by the patient and the treatment plans that have been applied (i.e., medical risk information), the treatment effects of the treatment plans currently consulted by the patient and the treatment plans that have been applied (i.e., expected effect information), and other information, and use this information as auxiliary integrated information.

[0175] After obtaining the auxiliary integration information, the auxiliary integration information and the answers to each segmented field can be input into the hybrid expert big model together, so that the hybrid expert big model can integrate the answers to each segmented field together under the guidance of the auxiliary integration information.

[0176] For example, when faced with different views of two small models on the effectiveness of a treatment plan, the mixed expert large model can make a balanced and comprehensive medical recommendation by comprehensively considering the above-mentioned auxiliary integrated information, including patient condition, treatment risks, expected effects, etc.

[0177] Combining auxiliary integrated information to integrate answers from different sub-fields is helpful to resolve information conflicts between answers from different sub-fields. In the medical field, experts from different fields may put forward different opinions and suggestions based on their respective professional knowledge. The present invention uses advanced information fusion technology and combines auxiliary integrated information to conduct in-depth analysis and comprehensive evaluation of these diverse opinions to determine the most reasonable and scientific medical answer information.

[0178] Optionally, the system can also integrate answers based on the user's language expression habits. Specifically, before integrating answers in segmented fields, the system can use the hybrid expert model to perform the following steps:

[0179] Determine language expression features representing the user's language expression habits based on the medical problem information;

[0180] Optimize the answer to each sub-field according to the language expression characteristics, and obtain the corresponding optimized answer to each sub-field;

[0181] After obtaining the optimized segment answers, the system can:

[0182] By integrating multiple optimized answers to subdivided fields based on the auxiliary integration information, medical answer information for medical problem information is obtained.

[0183] The user's language expression features can be extracted using existing natural language processing technology.

[0184] The user's language expression characteristics may include the following features:

[0185] The user's usual sentence patterns;

[0186] The order of the subject, predicate, object and other grammatical structures that the user usually uses when speaking;

[0187] The order in which the user is accustomed to describing different contents when speaking, for example, whether the user is accustomed to describing the symptoms first and then the condition, or describing the condition first and then the symptoms.

[0188] After obtaining the above-mentioned language expression features, the system can optimize each answer to a specific field based on the language expression features. For example, the core content involved in the answer to a specific field, including professional terms and concepts in the answer, can be extracted, and these core contents can be reintegrated according to the sentence patterns accustomed to the user to obtain the optimized answer to the specific field; the order of the subject, predicate, object and other grammatical structures in the answer to the specific field can be adjusted according to the order of the subject, predicate, object and other grammatical structures accustomed to the user, so that the latter adapts to the former to obtain the optimized answer to the specific field; the different contents of the answer to the specific field can also be reordered according to the order of different contents that the user is accustomed to using to obtain the optimized answer to the specific field.

[0189] In this embodiment, the system integrates the user's language expression characteristics to improve the readability and comprehensibility of the final output medical answer information, ensuring that the final output medical consultation answer is clear, logical, and well-organized. This optimization process not only improves the patient's understanding of medical advice, but also enhances the authority and persuasiveness of the answer. Through carefully designed language expressions, the system can transform complex medical information into a form that is easy for patients to accept and understand.

[0190] It can be seen from the above technical solution that the question-answering method of this embodiment also has the following beneficial effects.

[0191] First, by constructing a hybrid expert model, the present invention can use the segmented field processing models of different segments to deeply understand the medical expertise and terminology of the corresponding segments and provide more accurate and in-depth medical consulting services. This model structure enables the system to provide more accurate and comprehensive answers when answering complex or professional medical questions.

[0192] Secondly, the system can significantly improve the contextual understanding of the model by training a large hybrid expert model based on continuous multi-round question-answering in the medical field. This enables the system to better understand and utilize contextual information during the conversation, thereby providing coherent and consistent medical consultation and improving the accuracy of diagnosis and advice.

[0193] Thirdly, the system can provide personalized medical consultation based on the patient's specific situation, including the patient's medical object status information, medical risk information and expected effect information, the patient's emotions, the patient's language expression habits, etc., abandoning the limitations of traditional unified question and answer templates. This personalized service can better meet the unique needs of patients, improve patient satisfaction and the pertinence of medical services.

[0194] Fourthly, by optimizing the model structure and reasoning process, the present invention reduces the demand for computing resources and storage resources of the intelligent question-answering system during training and reasoning. During training, the corresponding initial small parameter model can be trained for each subdivision field, and during reasoning, a few subdivision field processing models associated with the medical problem information to be processed can be selected from multiple subdivision fields for reasoning, which not only reduces the operating cost of the system, but also makes the system more suitable for use in resource-constrained environments.

[0195] Fifth, when updating the sub-domain processing model, the model update process can be simplified, and only the new information in the information database can be updated, so that the system can more easily absorb new medical information. This update mechanism not only reduces the difficulty and cost of maintenance, but also ensures that the system can continue to provide the latest medical consulting services.

[0196] Sixth, the system of the present invention demonstrates the great potential of applying artificial intelligence in the medical field, and provides strong technical support for the digital transformation of the medical industry. Through intelligent question-and-answer tools, the efficiency of medical services can be improved, labor costs can be reduced, and patients can be provided with a more convenient service experience, thereby promoting the digital development of the medical industry.

[0197] Seventhly, through the above improvements, the system of the present invention can provide higher quality medical services. Whether in terms of accuracy, consistency, personalized services, or resource consumption and update maintenance, the present invention has brought significant overall quality improvement to the provision of medical services.

[0198] Eighth, through the intelligent question-and-answer system, the workload of doctors can be reduced, allowing medical staff to devote more time and energy to medical activities that require more professional judgment and operation. This helps to optimize the allocation of medical resources, improve the overall efficiency of medical services, and promote the rational allocation of medical resources.

[0199] Ninth, accurate, coherent and personalized intelligent question-answering systems can provide more humane medical services and enhance patients’ trust in medical services, which is one of the key factors for the success of medical services.

[0200] In the tenth aspect, the system of the present invention can be used not only for clinical consultation, but also as a tool for medical research and education. The large amount of medical question and answer data accumulated by the system can provide valuable information resources for medical research, and can also be used as an auxiliary tool for medical education to help medical students better understand and master medical knowledge, thereby providing support for medical research and education.

[0201] To sum up, the intelligent question-and-answer system for the medical industry of this patent has achieved a significant improvement in the efficiency and quality of medical services through a number of innovations, provided strong support for the digital transformation and intelligent upgrading of the medical industry, and has broad application prospects and practical value.

[0202] The present application also provides a medical industry intelligent question-answering system based on a hybrid expert large model. Figure 2 , the method may include the following steps.

[0203] An obtaining unit 201 is used to obtain medical problem information to be processed;

[0204] An analysis unit 202 is used to analyze the matching of the medical problem information with the information bases of multiple subdivision fields and the matching of the medical problem information with the medical knowledge graph through a hybrid expert big model, so as to determine the target subdivision field associated with the medical problem information;

[0205] The processing unit 203 is used to call the subdivision field processing model corresponding to each target subdivision field to process the medical problem information, and obtain the subdivision field answer corresponding to the medical problem information in each target subdivision field;

[0206] The integration unit 204 is used to integrate the answers in the subdivided fields to obtain medical answer information for the medical problem information.

[0207] Optionally, when the obtaining unit 201 obtains the medical problem information to be processed, it is specifically used to:

[0208] Get the current question entered by the user;

[0209] Obtain the historical questions entered by the user before the current question and the historical answers corresponding to the historical questions;

[0210] Current questions, historical questions and historical answers are integrated to obtain the medical problem information to be processed.

[0211] Optionally, the integration unit 204 is further configured to:

[0212] identifying user emotion information representing an emotional state of a corresponding user based on the medical problem information;

[0213] When the integration unit 204 integrates the subdivided field answers to obtain the medical answer information of the medical question information, it is specifically used for:

[0214] Integrate answers to segmented fields based on user emotional information to obtain medical answer information for medical problems.

[0215] Optionally, the system further includes an optimization unit 206, configured to:

[0216] Obtain user feedback information corresponding to the medical answer information;

[0217] The hybrid expert model is optimized based on user feedback.

[0218] Optionally, the system further includes a construction unit 205, which is used to construct a subdivision field processing model corresponding to each subdivision field, wherein the construction method of the subdivision field processing model corresponding to any subdivision field includes:

[0219] Obtain the information database corresponding to the subdivision field;

[0220] According to a plurality of dialogue samples corresponding to the subdivided fields, each dialogue sample includes at least one sample question and at least one sample answer corresponding to the sample question;

[0221] Based on the information database corresponding to the subdivided field, call the initial small parameter model to process the dialogue sample and obtain the model prediction answer corresponding to the dialogue sample;

[0222] The model loss of the initial small parameter model is determined according to the model prediction answer and the sample answer, and the model parameters of the initial small parameter model are adjusted according to the model loss until the model loss converges; wherein, when the model loss converges, the initial small parameter model serves as the segmentation field processing model corresponding to the segmentation field.

[0223] Optionally, the optimization unit 206 is further configured to:

[0224] Regularly update the information database corresponding to the sub-sectors;

[0225] The segment processing model corresponding to the segment is optimized according to the updated information base corresponding to the segment.

[0226] Optionally, when the integration unit 204 integrates the subdivided field answers to obtain the medical answer information of the medical question information, it is specifically used for:

[0227] Obtaining auxiliary integration information, the auxiliary integration information including at least one of medical object status information, medical risk information, and expected effect information;

[0228] Answers to subdivided fields are integrated based on the auxiliary integrated information to obtain medical answer information for medical problem information.

[0229] Optionally, the integration unit 204 is further configured to:

[0230] Determine language expression features representing the user's language expression habits based on the medical problem information;

[0231] Optimize the answer to each sub-field according to the language expression characteristics, and obtain the corresponding optimized answer to each sub-field;

[0232] When the integration unit 204 integrates the subdivided field answers according to the auxiliary integration information to obtain the medical answer information of the medical question information, it is specifically used for:

[0233] By integrating multiple optimized answers to subdivided fields based on the auxiliary integration information, medical answer information for medical problem information is obtained.

[0234] The working principle of the intelligent question-answering system for the medical industry based on the hybrid expert big model provided in this embodiment can be found in the relevant steps of the intelligent question-answering method for the medical industry based on the hybrid expert big model provided in any embodiment of the present application, and will not be repeated here.

[0235] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0236] For the convenience of description, the above system or device is described by dividing it into various modules or units according to its functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0237] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or partly contributed to the prior art in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present application or certain parts of the embodiments.

[0238] Finally, it should be noted that, in this article, relational terms such as first, second, third and fourth are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0239] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. An intelligent question-answering method for the medical industry based on a hybrid expert large model, characterized in that: include: Obtain information on pending medical issues; Analyze the matching of the medical problem information with the information bases of multiple sub-fields and the matching of the medical problem information with the medical knowledge graph through a hybrid expert big model to determine the target sub-field associated with the medical problem information; Calling a segmented field processing model corresponding to each of the target segmented fields to process the medical problem information, and obtaining a segmented field answer corresponding to the medical problem information in each of the target segmented fields, wherein the segmented field processing model is constructed according to an information base corresponding to the segmented field; The answers to the subdivided fields are integrated to obtain medical answer information for the medical problem information.

2. The method according to claim 1, characterized in that The obtaining of information on medical issues to be processed includes: Get the current question entered by the user; Obtaining the historical questions input by the user before the current question and the historical answers corresponding to the historical questions; The current question, the historical question and the historical answer are integrated to obtain the medical question information to be processed.

3. The method according to claim 1, characterized in that Also includes: identifying user emotion information representing an emotional state of a corresponding user according to the medical problem information; The step of integrating the answers to the subdivided fields to obtain the medical answer information of the medical question information includes: The answers to the subdivided fields are integrated according to the user emotion information to obtain medical answer information for the medical question information.

4. The method according to claim 1, characterized in that: Also includes: Obtaining user feedback information corresponding to the medical answer information; The hybrid expert large model is optimized according to the user feedback information.

5. The method according to claim 1, characterized in that The method for constructing a segmented field processing model corresponding to any segmented field includes: Obtaining an information database corresponding to the subdivision field; According to a plurality of dialogue samples corresponding to the subdivided fields, each of the dialogue samples includes at least one sample question and at least one sample answer corresponding to the sample question; Based on the information base corresponding to the subdivided field, an initial small parameter model is called to process the conversation sample to obtain a model prediction answer corresponding to the conversation sample; The model loss of the initial small-parameter model is determined according to the model prediction answer and the sample answer, and the model parameters of the initial small-parameter model are adjusted according to the model loss until the model loss converges; wherein, when the model loss converges, the initial small-parameter model serves as the segmentation field processing model corresponding to the segmentation field.

6. The method according to claim 5, characterized in that Also includes: Regularly update the information database corresponding to the subdivision field; The subdivision field processing model corresponding to the subdivision field is optimized according to the updated information base corresponding to the subdivision field.

7. The method according to claim 1, characterized in that The step of integrating the answers to the subdivided fields to obtain the medical answer information of the medical question information includes: Obtaining auxiliary integration information, wherein the auxiliary integration information includes at least one of medical object status information, medical risk information, and expected effect information; The answers to the subdivided fields are integrated according to the auxiliary integration information to obtain medical answer information for the medical question information.

8. The method according to claim 2, characterized in that Also includes: Determining language expression features representing the user's language expression habits according to the medical problem information; Optimize each of the subdivided field answers according to the language expression features to obtain a corresponding optimized subdivided field answer; The step of integrating the subdivided field answers according to the auxiliary integration information to obtain medical answer information for the medical question information includes: The plurality of optimized answers to the subdivided fields are integrated according to the auxiliary integration information to obtain medical answer information for the medical problem information.

9. An intelligent question-answering system for the medical industry based on a hybrid expert large model, characterized in that: include: An acquisition unit, used for acquiring information of medical problems to be processed; An analysis unit, configured to analyze the matching of the medical problem information with information bases of multiple subdivisions and the matching of the medical problem information with a medical knowledge graph through a hybrid expert big model, so as to determine a target subdivision associated with the medical problem information; A processing unit, configured to call a segmented field processing model corresponding to each of the target segmented fields to process the medical problem information, and obtain a segmented field answer corresponding to the medical problem information in each of the target segmented fields, wherein the segmented field processing model is constructed according to an information base corresponding to the segmented field; The integration unit is used to integrate the answers to the subdivided fields to obtain medical answer information for the medical problem information.

10. The system according to claim 9, characterized in that When the obtaining unit obtains the medical problem information to be processed, it is specifically used to: Get the current question entered by the user; Obtaining the historical questions input by the user before the current question and the historical answers corresponding to the historical questions; The current question, the historical question and the historical answer are integrated to obtain the medical question information to be processed.

Citation Information

Cited By

  • Method for providing medical question and answer service, related device and computer program product

    CN119560183A