A medical and health question-answering method, device and medium based on intelligent agents

Abstract

This invention discloses a medical and health question-answering method, device, and medium based on intelligent agents, relating to the field of artificial intelligence medical question-answering technology. The method includes: utilizing a master intelligent agent to receive an intent parsing report and access a dynamic dialogue context pool; generating intelligent agent collaboration instructions based on the intent parsing report and the state of the dynamic dialogue context pool; using a credibility calibration intelligent agent to assess the credibility of a preliminary medical answer report, generating a confidence score and evidence conflict level; and using a dynamic risk assessment strategy to perform multi-dimensional weighted fusion and risk pattern recognition to generate diversified treatment instructions; when the diversified treatment instructions are permitted for release, performing a security and compliance check on the preliminary medical answer report to generate a compliant medical answer. This invention achieves multi-dimensional control over the credibility and security of medical answers, ultimately achieving the beneficial effect of providing personalized medical question-answering while ensuring the authenticity, reliability, and compliance of information.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence medical question-answering technology, and in particular to a medical and health question-answering method, device and medium based on intelligent agents. Background Technology

[0002] With the deepening application of artificial intelligence technology in the healthcare field, question-answering systems based on natural language processing (NLP) are gradually becoming important tools for assisting medical information retrieval and health management. Traditional healthcare question-answering systems often employ end-to-end deep learning models, such as pre-trained language models based on the Transformer architecture, which achieve question parsing and answer generation through large-scale medical text training. These systems typically rely on a single model to complete tasks such as intent recognition, entity extraction, and answer generation, with training data mostly derived from structured medical knowledge bases or unstructured electronic health records. In recent years, to improve answer accuracy, some systems have introduced Retrieval Augmentation (RAG) technology, which enhances contextual information through external knowledge base retrieval, reducing model illusions. Furthermore, some systems attempt to process professional medical terminology through rule engines or semantic template matching to improve the accuracy of entity recognition and intent classification. Existing technologies have demonstrated efficient information processing capabilities in specific scenarios, particularly in standardized areas such as common disease consultations and drug information queries.

[0003] Current technologies still face challenges in ensuring the credibility of medical answers and controlling dynamic risks. When user queries involve complex symptoms or multiple diseases, answers generated by a single model may ignore evidence conflicts between different knowledge sources or fail to fully consider key information from the user's historical conversations (such as allergy history and medication records), leading to insufficient confidence assessment of the answer. Furthermore, existing systems often employ static risk assessment strategies, making it difficult to dynamically adjust answer generation strategies based on real-time conversation status, potentially resulting in misleading suggestions. These problems are particularly prominent in high-risk scenarios such as chronic disease management and emergency triage, necessitating a solution that integrates multi-agent collaboration and dynamic risk perception. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an agent-based medical and health question-answering method to solve the problems of ignoring evidence conflicts and insufficient confidence assessment caused by static risk strategies when processing complex medical queries.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a medical and health question-answering method based on an intelligent agent, which includes receiving a medical and health question input by a user, parsing the medical and health question using a query understanding intelligent agent, identifying medical entities and determining the user's query intent, and generating an intent parsing report;

[0008] The master agent receives the intent parsing report and accesses the dynamic dialogue context pool. Based on the intent parsing report and the state of the dynamic dialogue context pool, it generates agent collaboration instructions.

[0009] The system utilizes collaborative instructions to drive various professional agents, which retrieve information from the corresponding medical knowledge base and generate preliminary answer fragments. The answer integration agent then merges all the preliminary answer fragments to generate a preliminary medical answer report.

[0010] The credibility of the preliminary medical answer report is assessed by a credibility calibration agent, which generates a confidence score and evidence conflict level. A dynamic risk assessment strategy is used to perform multi-dimensional weighted fusion and risk pattern recognition to generate diversified treatment instructions. When the diversified treatment instructions are allowed to be issued, the preliminary medical answer report is checked for safety and compliance to generate a compliant medical answer.

[0011] As a preferred embodiment of the agent-based medical and health question-answering method of the present invention, the medical and health questions include disease-related questions, symptom consultation, examination and medication questions, and health management and preventive healthcare questions.

[0012] As a preferred embodiment of the agent-based medical and health question-answering method of the present invention, the steps of using a query understanding agent to analyze medical and health questions, identify medical entities, determine the user's query intent, and generate an intent analysis report are as follows.

[0013] The system performs noise reduction, word segmentation, and syntactic analysis on medical and health issues, removing irrelevant information, punctuation marks, and redundant words, and outputs standardized text data.

[0014] The query understanding agent is used to identify medical entities and classify intents in standardized text data, obtain medical entities, and label the categories of medical entities; based on the user's context and the categories of medical entities, the user's query intent is determined by the classification algorithm, and structured medical entity data and user query intent are obtained.

[0015] The system performs semantic matching between structured medical entity data and user query intent, and infers the user's specific needs by combining the context of the query, generating an intent analysis report.

[0016] As a preferred embodiment of the agent-based medical health question-answering method of the present invention, the steps of receiving an intent parsing report by a master agent and accessing a dynamic dialogue context pool, and generating agent collaboration instructions based on the intent parsing report and the state of the dynamic dialogue context pool, are as follows:

[0017] The main control agent receives the intent parsing report and accesses the dynamic dialogue context pool through the interface to obtain the dialogue history and user health data related to the current query, thereby obtaining dynamic dialogue context information.

[0018] By combining and analyzing the structured medical entity data in the intent parsing report with dynamic dialogue context information, we can determine the user's current specific needs and obtain comprehensive needs analysis results.

[0019] Based on the comprehensive demand analysis results, the core user query needs are identified, the query processing task objectives are determined, and the query processing task is decomposed into multiple sub-tasks to generate intelligent agent collaboration instructions.

[0020] As a preferred embodiment of the agent-based medical and health question-answering method of the present invention, the specific steps of using agent collaboration instructions to drive various professional agents to retrieve information from the corresponding medical knowledge base and generate preliminary answer fragments are as follows.

[0021] Intelligent agent collaboration instructions drive various professional intelligent agents to retrieve information from the corresponding medical knowledge base for different types of medical and health problems, and obtain medical retrieval results data.

[0022] Organize and filter the medical search results to obtain a preliminary medical information set;

[0023] The system supplements and integrates the preliminary medical information set using structured medical entity data, and matches the structured medical entity data with the user's specific needs based on the user's query intent to generate initial answer fragments.

[0024] The initial answer fragments are optimized using a semantic understanding model, and further enhanced with deep semantics to obtain structured answer fragments.

[0025] The structured answer fragments are validated for relevance, completeness, and accuracy to generate preliminary answer fragments.

[0026] As a preferred embodiment of the agent-based medical and health question-answering method of the present invention, the step of fusing all preliminary answer fragments by the answer-integrating agent to generate a preliminary medical answer report includes the following specific steps.

[0027] The answer synthesis agent receives all preliminary answer fragments and performs unified syntactic and semantic processing on them. Through weighted fusion and structural adjustment, it generates an optimized set of answer fragments.

[0028] The optimized answer fragment set is structured to generate a preliminary medical answer report.

[0029] As a preferred embodiment of the agent-based medical and health question-answering method of the present invention, the following steps are described: The agent is used to perform a credibility assessment on the preliminary medical answer report, generating a confidence score and an evidence conflict level. A dynamic risk assessment strategy is then used for multi-dimensional weighted fusion and risk pattern recognition to generate diversified treatment instructions.

[0030] The credibility calibration agent receives preliminary medical answer reports, performs evidence source matching on each preliminary answer fragment, verifies the data consistency and accuracy of the preliminary answer fragments, and outputs the verified preliminary answer fragments and evidence source information;

[0031] Extract the degree of conflict between evidence sources from the evidence source information, and generate an evidence conflict level for each preliminary answer fragment based on the conflict analysis;

[0032] Calculate the confidence score for each verified preliminary answer fragment based on evidence quality, data consistency, and the level of evidence conflict.

[0033] A comprehensive credibility score is generated for each preliminary answer fragment by weighted fusion of the confidence score and the level of evidence conflict.

[0034] By employing a dynamic risk assessment strategy, the overall credibility score is weighted and integrated across multiple dimensions, and risk score is generated for each preliminary answer fragment through risk pattern recognition.

[0035] Based on the risk score, different diversified disposal instructions are generated for each initial answer fragment.

[0036] As a preferred embodiment of the intelligent agent-based medical and health question-answering method described in this invention, the diversified treatment instructions are divided into pending instructions, prohibited issuance instructions, and permitted issuance instructions according to risk scores;

[0037] When a diversified treatment instruction is pending review, the preliminary medical response report will not be released and will be marked as pending review.

[0038] When the diversified treatment instruction prohibits the release of information, the preliminary medical response report shall be prohibited from being released, and the reason shall be recorded or a risk warning shall be generated.

[0039] When the diversified treatment instruction is approved for release, the legality, medical accuracy and potential risks of the content in the preliminary medical answer report are verified item by item. After the verification is passed, a disclaimer is automatically added to the preliminary medical answer report to generate a compliant medical answer.

[0040] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the agent-based medical and health question-answering method as described in the first aspect of the present invention.

[0041] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the agent-based medical and health question-answering method as described in the first aspect of the present invention.

[0042] The beneficial effects of this invention are as follows: By generating collaborative instructions for the main control agent in conjunction with the dynamic dialogue context, precise task decomposition and answer generation based on individual differences are achieved, ensuring the continuity and personalization of the question-and-answer results; at the same time, by using a credibility calibration agent to perform evidence source matching and conflict analysis on the preliminary answers, and combining it with a dynamic risk assessment strategy to generate diversified treatment instructions, multi-dimensional control over the credibility and security of medical answers is achieved, ultimately achieving the beneficial effect of providing personalized medical questions and answers while ensuring the authenticity, reliability, compliance, and security of information. Attached Figure Description

[0043] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a flowchart of an agent-based medical and health question-answering method.

[0045] Figure 2 To query and understand the workflow diagram of the intelligent agent.

[0046] Figure 3 The flowchart of the integrated intelligent agent for answering questions.

[0047] Figure 4 Workflow diagram for calibrating the agent for credibility. Detailed Implementation

[0048] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0049] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0050] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0051] Reference Figures 1-4 This is one embodiment of the present invention, which provides a medical health question-answering method based on an intelligent agent, including the following steps:

[0052] S1. Receive medical and health questions input by the user, use the query understanding agent to analyze the medical and health questions, identify medical entities and determine the user's query intent, and generate an intent analysis report.

[0053] It should be noted that healthcare issues include disease-related issues, symptom consultation, examination and medication issues, and health management and preventive healthcare issues.

[0054] It should be noted that the "Query Understanding Agent" is responsible for transforming the user's original query into a structured representation that can be processed by the machine. This includes denoising / word segmentation / syntactic analysis, medical entity recognition (NER), entity type labeling and intent classification, and outputting structured medical entities and intent labels for subsequent processing.

[0055] By collecting and cleaning medical corpora (de-sensitized paragraphs of electronic medical records, question-and-answer pairs, clinical guidelines, and glossaries), and labeling them by entity and intent to build a training set, a text preprocessing pipeline (noise filtering, word segmentation / sub-wordization, syntactic analysis, standardization, and medical terminology mapping) is built. A pre-trained model based on a bidirectional encoder (such as Clinical-BERT / general BERT fine-tuning) is selected and fine-tuned for named entity recognition (NER) and intent classification. At the same time, a rule / dictionary layer (ICD, SNOMED mapping table, thesaurus) is built in parallel to improve recall. Hybrid inference (model output + rule post-processing) is used to output structured entity-attribute pairs and intent vectors. Cross-validation and F1 / precision / recall are used for evaluation and deployment as a microservice that can be called via API. Through active learning and online error correction closed loop, the labeled set is continuously expanded and biases are corrected to obtain a query understanding agent.

[0056] S1.1 Perform noise reduction, word segmentation, and syntactic analysis on medical and health issues to remove irrelevant information, punctuation marks, and redundant words, and output standardized text data.

[0057] It should be noted that web page tags, emoticons, repeated spaces, abnormally formatted characters, and advertising or polite phrases unrelated to the meaning of medical and health issues should be removed and standardized, and character encoding should be unified, full-width and half-width characters should be unified, punctuation marks should be removed, and clean text data should be generated.

[0058] The clean text data is segmented according to common Chinese word segmentation methods, keeping the boundaries of words related to medical entities intact and merging obviously broken proper noun phrases, and the segmented and annotated text is output.

[0059] Based on dependency relations and component boundary identification, the text is segmented and annotated with the position and association of the subject, predicate, object, modifier, negation word, and time and quantity expressions. The grammatical relationships between words are established and ambiguous expressions are eliminated to generate standardized text data.

[0060] S1.2 Utilize query understanding agents to perform medical entity recognition and intent classification on standardized text data, obtain medical entities, and label the categories of medical entities.

[0061] It should be noted that the standardized text data generated after denoising, word segmentation and syntactic analysis is input into the query understanding agent. The query understanding agent performs sub-word-level feature encoding on the standardized text data through a general BERT fine-tuning model to generate context-aware semantic vectors for each word in the text.

[0062] By utilizing the context-aware semantic vectors of each word, named entity recognition is performed on potential medical entities in the text, identifying medical entities such as disease names, symptom descriptions, examination items, medication information, and health management and preventive healthcare related terms;

[0063] Each identified medical entity is labeled with a corresponding category using entity type labeling rules. For example, “hypertension” is labeled as a disease category, “headache” as a symptom category, “blood routine” as an examination category, “aspirin” as a medication category, and “health checkup plan” as a health management category, ensuring that the identified medical entities are consistent with the original semantics in the standardized text data.

[0064] It should also be noted that the specific training steps of the general BERT fine-tuning model are as follows: using the labeled medical entity categories as supervision signals, inputting standardized text data into the general BERT fine-tuning model for fine-tuning training, adjusting the parameters of the general BERT fine-tuning model to enable it to make accurate predictions on medical entity recognition and intent classification tasks, using cross-validation to evaluate the performance of the general BERT fine-tuning model, and optimizing the general BERT fine-tuning model based on the F1 score, precision, and recall.

[0065] S1.3. Based on the user's context and the category of medical entities, determine the user's query intent through a classification algorithm, and obtain structured medical entity data and the user's query intent.

[0066] It should be noted that the medical entities and the labeled medical entity categories are integrated with the user's contextual information, and the user's historical questions and answers in the current session, as well as relevant disease, symptom, examination and medication information in the health record, are associated with the identified medical entities to construct a comprehensive feature vector that includes contextual semantic features and entity features.

[0067] The comprehensive feature vector is input into the classification algorithm. Based on the word context information, medical entity category, entity location, dependency relationships, and user's preceding and following context information contained in the comprehensive feature vector, the algorithm calculates a matching score for each intent category. The expression is as follows:

[0068] ;

[0069] In the formula, Intent category The matching score, a dimensionless real number, is used to represent the degree of matching between the category and the comprehensive feature vector. Is related to the category of intent The corresponding weight vector, dimensions and To ensure that the dimensions of the inner product calculation are consistent, It is a transpose operation; Intent category The bias term, which is a dimensionless real number, is used to adjust the baseline of the matching score. It is a comprehensive feature vector, and its dimension is a dimensionless real number vector. It is an index of the intent category;

[0070] The scores of all intent categories are compared, the intent category with the highest score is selected as the user query intent, and it is matched with the medical entity and the labeled medical entity category to output the corresponding structured medical entity data;

[0071] It should also be noted that user context refers to semantic information related to medical and health issues generated by the user in the current session and in historical interactions, including questions previously asked by the user, answers received, disease records in health records, symptom descriptions, test results and medication information, etc.

[0072] Semantic information is obtained through a dynamic dialogue context pool, which collects and updates the user's historical dialogue records, health profile, and query content and interaction status in the current session in real time, and integrates all relevant information.

[0073] It should be noted that the dynamic dialogue context pool is formed by collecting and updating users' historical dialogue records, health records (such as disease, symptoms, examination and medication information) and query content and interaction status in the current session in real time, and integrating them into a structured representation that includes contextual semantic features and entity features.

[0074] S1.4 Semantically match structured medical entity data with user query intent, and infer the user's specific needs based on the context of the query, generating an intent analysis report.

[0075] It should be noted that the disease names, symptoms, examination items, medication information, and health management-related terms in the structured medical entity data are semantically aligned with the user's query intent to form a preliminary matching relationship;

[0076] By leveraging the user's context, the initial matching relationship is enhanced to infer the user's specific needs in the current conversation and generate a comprehensive matching table;

[0077] The comprehensive matching table is structured according to intent category and medical entity category to generate an intent parsing report that can be used for subsequent processing.

[0078] S2. Utilize the master agent to receive the intent parsing report and access the dynamic dialogue context pool. Based on the intent parsing report and the state of the dynamic dialogue context pool, generate agent collaboration instructions.

[0079] It should be noted that the "master control agent" is the decision-making layer responsible for synthesizing the intent parsing results and dialogue context into task planning and scheduling instructions, and is responsible for task decomposition, collaborative instruction generation and context management.

[0080] By defining a task decomposition semantic framework and sub-task templates (such as retrieval, verification, generation, and compliance checks), historical dialogues and user health profiles are aggregated as context storage and connected to the state machine / dialogue context pool interface. Rule-based judgment logic ensures security and compliance boundaries and priority constraints. A lightweight policy network (small Transformer or sequence-to-sequence model) is then trained on labeled dialogue-task decomposition pairs to learn how to decompose high-level intents into deliverable sub-tasks. The task delivery logic is responsible for instruction serialization and delivery, error rollback strategies, and timeout handling. After end-to-end evaluation based on simulation scenarios (task completion rate, instruction accuracy, and failure recovery capability), the master control logic is deployed as an auditable orchestration service. The master control agent is obtained by continuously refining the strategy through policy replay, offline / online testing, and security sandbox.

[0081] It should also be noted that the rule judgment logic performs logical judgment and screening of security compliance boundaries, priority constraints and policy constraints before task decomposition, in order to guide the generation and distribution of subsequent sub-tasks;

[0082] Task distribution logic refers to the rules and processes by which subtasks generated by the master intelligent agent are organized, serialized and distributed to each collaborating intelligent agent according to execution order, priority and security compliance constraints. It is obtained by making logical judgments on task decomposition results, context state and policy constraints.

[0083] The main control logic is to decompose and distribute executable tasks by integrating high-level intents, intent parsing reports, and dynamic dialogue contexts through rule judgment, policy networks, and scheduling strategies.

[0084] The process involves defining a task decomposition semantic framework, establishing rule-based judgment constraints, training a policy network on labeled dialogue-task decomposition pairs, combining it with a scheduler for task serialization, conducting end-to-end verification through simulation scenarios and human review, and deploying it as an auditable master control service to form the master control logic.

[0085] S2.1 Utilize the master control agent to receive the intent parsing report, and access the dynamic dialogue context pool through the interface to obtain the dialogue history and user health data related to the current query, thereby obtaining dynamic dialogue context information.

[0086] It should be noted that the intent parsing report is used as input to the master intelligent agent, which accesses the dynamic dialogue context pool through an interface.

[0087] Retrieve all relevant interaction records of the user in the current and historical sessions from the dynamic dialogue context pool, including historical questions and answers, disease records in health records, symptom descriptions, examination results and medication information, as well as query content and interaction status in the current session;

[0088] The retrieved dialogue history and health record information are integrated and formatted to form dynamic dialogue context information that can represent the user's context and state.

[0089] S2.2 Combine and analyze the structured medical entity data in the intent parsing report with the dynamic dialogue context information to determine the user's current specific needs and obtain comprehensive needs analysis results.

[0090] It should be noted that the structured medical entity data in the intent parsing report is compared semantically with the historical questions, historical answers, health record records and current conversation status contained in the dynamic dialogue context information. The semantic similarity is obtained by combining the meaning of words and the user's context, and the content is judged to be most relevant to the user's current expression, thus identifying the medical issues that the user is concerned about in this round of conversation.

[0091] Based on the intent category and medical entity category, the matching results are categorized, summarized, and prioritized to infer the user's specific needs in the current session. The comprehensive needs analysis results are output, including the intent category, user focus, semantic matching degree, and contextual information for each medical entity.

[0092] It should be noted that the meanings of the words are derived from the medical entities (diseases, symptoms, drugs, etc.) extracted from the intent analysis report, as well as the keywords and terms in the dialogue context.

[0093] S2.3 Based on the comprehensive demand analysis results, identify the core user query needs, determine the query processing task objectives, decompose the query processing task into multiple sub-tasks, and generate intelligent agent collaboration instructions.

[0094] It should be noted that, based on the intent category and semantic similarity of each medical entity in the comprehensive demand analysis results, the medical entity and intent category are combined for feature representation. For example, each combination includes disease name, symptoms, examination or medication information and corresponding intent weight.

[0095] The ranking algorithm or rules are used to prioritize and rank combinations of medical entities and intent categories according to user attention, historical interaction frequency and contextual relevance.

[0096] The ranking results identify the core medical issues that users are most concerned about in the current session, and thus clarify the query processing task objectives, such as obtaining disease diagnosis and treatment information, symptom relief suggestions, or medication guidance.

[0097] Based on the query processing task objectives, the query processing task is divided into independently executable sub-tasks according to functions and processing flow, including information retrieval, evidence verification, preliminary answer generation and answer integration, etc. Each sub-task has a clear input-output relationship and execution order.

[0098] The decomposed subtasks, their execution order, required medical entities, and intent information are integrated to generate intelligent agent collaboration instructions.

[0099] S3. Using agent collaboration instructions, drive various professional agents to retrieve information from the corresponding medical knowledge base and generate preliminary answer fragments. The answer integration agent then merges all the preliminary answer fragments to generate a preliminary medical answer report.

[0100] It should be noted that the "Answer Integration Agent" is an aggregation layer responsible for semantic fusion, conflict resolution, and unified expression of the preliminary answer fragments returned by various professional agents, outputting a structured and readable preliminary medical answer report;

[0101] By defining standardized representations of answer fragments (text, structured fields, source identifiers, timestamps, and initial confidence values), semantic vectorization is performed on preliminary answer fragments. A two-layer retrieval structure is constructed by combining keyword matching rules. The preliminary answer fragments are then relevance-scored and re-ranked (dense retriever + lexical re-ranker). Conflict detection and clustering algorithms are used to merge similar preliminary answer fragments and mark inconsistencies. Rule / template method is used to complete structured merging and source tracing. If necessary, a fine-tuned generative model (RAG or seq2seq) is used to perform natural language fusion and rewriting on verified evidence to improve readability. At the same time, fact completion and consistency verification are performed to obtain a comprehensive intelligent agent for the answer.

[0102] S3.1 The intelligent agent collaboration instructions drive various professional intelligent agents to retrieve information from the corresponding medical knowledge base for different types of medical and health problems, and obtain medical retrieval results data.

[0103] It should be noted that the sub-task information, the medical entities involved, and the query processing task objectives explicitly stated in the intelligent agent collaboration instructions should be transmitted to each professional intelligent agent;

[0104] Each specialized intelligent agent determines its corresponding medical knowledge base based on the type of medical and health problems it is responsible for, such as a disease diagnosis and treatment knowledge base, a symptom analysis knowledge base, an examination and testing knowledge base, or a medication guidance knowledge base.

[0105] Each professional intelligent agent uses structured query methods or semantic retrieval methods to locate relevant information in the corresponding medical knowledge base. During the retrieval process, it uses medical entity names, intent categories and contextual keywords for precise matching, and combines information such as time and source to filter the retrieval results.

[0106] Each professional intelligent agent organizes and outputs medical search results data that includes text descriptions, source identifiers, timestamps, and preliminary confidence levels.

[0107] S3.2 Organize and filter the medical search results data to obtain a preliminary medical information set.

[0108] It should be noted that the medical search results data are filtered, deduplicated, and structured according to the degree of matching of medical entities, information completeness, source credibility, and time relevance to generate a preliminary medical information set.

[0109] S3.3. Use structured medical entity data to supplement and integrate the preliminary medical information set, and match the structured medical entity data with the user's specific needs according to the user's query intent to generate initial answer fragments.

[0110] It should be noted that the structured medical entity data is semantically aligned and fields are completed with the text content in the initial medical information set to ensure that each piece of information covers the relevant medical entity attributes, thus obtaining the supplemented medical information set;

[0111] Identify user concerns from user query intent, such as disease name, symptom type, examination items or medication requirements, and vectorize these user concerns.

[0112] For each piece of information in the supplemented medical information set, extract the corresponding medical entity, keywords, and contextual features, and calculate the semantic similarity with the user's points of interest. The expression is as follows:

[0113] ;

[0114] In the formula, This is the first in the supplemented medical information collection. message With the User Focus The semantic similarity between them, with values ​​ranging from [0,1]; This is the first in the supplemented medical information collection. message semantic vector; This is the first in the supplemented medical information collection. User concerns semantic vector; This is the first in the supplemented medical information collection. message The second norm of semantic vectors; This is the first in the supplemented medical information collection. User Focus The second norm of semantic vectors; It is an index of information; It is an index of user concerns;

[0115] Based on the semantic similarity, content highly relevant to the user's query intent is filtered out, and relevant medical entities and intent categories are labeled to obtain the filtering and labeling results;

[0116] The filtered and labeled results are organized into structured initial answer fragments according to relevant medical entities, intent categories, and semantic similarity.

[0117] S3.4 Optimize the expression of the initial answer fragment through a semantic understanding model, and further enhance the deep semantics to obtain a structured answer fragment.

[0118] It should be noted that the initial answer fragments are input into the semantic understanding model. The semantic understanding model performs word-by-word semantic vector analysis on each initial answer fragment, identifies the deep semantic information and contextual logical relationships in the initial answer fragments, and optimizes and reorganizes medical entities, keywords and modifiers. Through semantic similarity calculation and contextual association enhancement, unclear or repetitive parts are merged and adjusted, while maintaining the consistency between medical entity categories and user query intent, and outputting structured answer fragments.

[0119] It should also be noted that the training process of the semantic understanding model is as follows: by collecting medical question-and-answer pairs, clinical guideline texts and desensitized paragraphs of electronic medical records, the text is segmented, standardized and labeled with medical entities, and a mapping relationship between the input text and the target semantic labels is constructed. The loss between the input text features and the target labels is calculated using supervised learning methods. The semantic representation parameters are iteratively adjusted through optimization algorithms so that the semantic understanding model can capture medical entities, contextual logic and deep semantic information in the text.

[0120] S3.5. Verify the relevance, completeness, and accuracy of the structured answer fragments to generate preliminary answer fragments.

[0121] It should be noted that, based on the structured medical entity data and the user's query intent, the medical entities, keywords and semantic information contained in each structured answer fragment are compared with the preliminary medical information set, and structured answer fragments that do not match the user's query intent or have serious information deficiencies are filtered out.

[0122] For each retained structured answer fragment, medical knowledge verification is performed to check whether the medical entity attributes, medication recommendations, examination or symptom descriptions conform to validated clinical guidelines and medical literature, and to mark the verification results and suspicious items.

[0123] All validated structured answer fragments that cover the core needs of user queries are integrated to output preliminary answer fragments.

[0124] S3.6 The answer synthesis agent accepts all preliminary answer fragments and performs unified processing on the preliminary answer fragments at the grammatical and semantic levels. Through weighted fusion and structural adjustment, it generates an optimized set of answer fragments.

[0125] It should be noted that each preliminary answer fragment should be standardized at the grammatical level, including standardized terminology, sentence structure, and punctuation, to ensure consistent text expression;

[0126] Semantic unification is performed on each preliminary answer fragment. By analyzing medical entities, intent categories, and contextual information, the order of expression and logical relationships are adjusted to ensure semantic coherence within and between preliminary answer fragments.

[0127] The content of the preliminary answer fragments is integrated by weighted fusion method, taking into account the importance, coverage and matching degree with the user's query intent of each preliminary answer fragment, and generating the fused structured content;

[0128] The integrated structured content is restructured, including the classification and labeling of medical entities, keywords, and intent categories, to form an optimized set of answer fragments.

[0129] S3.7. Perform structured processing on the optimized answer fragment set to generate a preliminary medical answer report.

[0130] It should be noted that the optimized answer fragments in the optimized answer fragment set are uniformly organized into standardized fields. Optimized answer fragments from different sources but with similar or repeated content are merged, and conflict markers and consistency checks are performed on the parts with differences. Optimized answer fragments are grouped and sorted according to intent categories to ensure logical coherence of the answers. While maintaining the traceability information, the results are integrated into a highly readable and semantically complete structured document, which is output as a preliminary medical answer report.

[0131] S4. Utilize the credibility calibration agent to assess the credibility of the preliminary medical answer report, generate a confidence score and evidence conflict level, and use a dynamic risk assessment strategy to perform multi-dimensional weighted fusion and risk pattern recognition to generate diversified treatment instructions.

[0132] It should be noted that the "credibility calibration agent" is responsible for verifying the chain of evidence, calculating confidence and risk stratification for each segment of the preliminary medical answer, and outputting confidence scores, evidence conflict levels and diversified handling instructions to determine whether to issue or proceed to manual review.

[0133] By collecting source information, content completeness, and historical verification records of preliminary medical answer reports, a traceable evidence comparison table is established. Based on factors such as source reliability, information completeness, and consistency of each preliminary answer fragment, evaluation indicators are set to assign weights and conduct level assessments for each factor. The scores of each factor are summarized according to weighted rules to generate a confidence score. Contradictions and uncertainties between answer fragments are identified according to preset conflict level standards. Factors such as source reliability, semantic consistency, and contextual relevance are weighted and integrated in multiple dimensions to form a risk assessment result. The result is calibrated and optimized using simulation test labeled data to obtain a credibility calibration agent.

[0134] It should also be noted that the evaluation indicators were determined based on the reliability of the source, the completeness of the information, and the consistency of the information for each preliminary answer fragment;

[0135] The conflict level standard is set by comparing the consistency and contradictions of information between different initial answer fragments to classify conflict situations into different levels.

[0136] S4.1 The credibility calibration agent receives the preliminary medical answer report, performs evidence source matching on each preliminary answer fragment, verifies the data consistency and accuracy of the preliminary answer fragment, and outputs the verified preliminary answer fragment and evidence source information.

[0137] It should be noted that medical entities, keywords and related information are extracted from the initial answer fragments, and the corresponding original evidence is located in multi-source medical information through the retrieval tool, such as desensitized paragraphs of electronic medical records, clinical guidelines, drug instructions or consultation question-and-answer pairs. During the matching process, the evidence is classified and scored according to the source type, peer review / guideline level and publication date.

[0138] The matched evidence is compared with the medical entity information in the preliminary answer fragment to verify the accuracy and consistency of the diseases, symptoms, examination items, medication information and health management terminology described in the preliminary answer fragment, including checking the entity names, attributes, relationships and logical rationality;

[0139] The verification results are labeled, the evidence source and verification status corresponding to each preliminary answer fragment are clearly defined, and the verified preliminary answer fragment and evidence source information are output.

[0140] S4.2 Extract the degree of conflict between evidence sources from the evidence source information, and generate an evidence conflict level for each preliminary answer fragment based on the conflict analysis.

[0141] It should be noted that each of the evidence sources matched with each verified preliminary answer fragment is analyzed one by one, and the content of the disease name, symptom description, examination items, medication information and health management terminology involved in each evidence source is compared.

[0142] The comparison operation includes checking the attribute values, logical relationships, time information and hierarchical concepts of each medical entity to determine whether there are inconsistencies in description, attribute conflicts or logical contradictions between corresponding medical entities in different evidence sources.

[0143] The comparison results are converted into numerical values ​​that can represent the intensity of the conflict by using conflict quantification methods, and the conflict intensity values ​​are output, for example by calculating the proportion of inconsistent attributes or the degree of semantic mismatch.

[0144] The conflict intensity value is associated with each verified preliminary answer fragment, and the degree of conflict of evidence sources is divided into different levels by the conflict intensity threshold.

[0145] It should also be noted that the conflict intensity threshold is determined by comparing the attributes, logic, and semantic consistency of multiple evidence sources matched for each historical answer fragment, recording the conflict situation and number of conflicts that occur between different evidence sources for each answer fragment, compiling the conflict distribution of all historical answer fragments, and analyzing the relationship between conflict frequency and the reliability of answer fragments.

[0146] S4.3 Calculate the confidence score for each verified preliminary answer fragment based on evidence quality, data consistency, and level of evidence conflict.

[0147] It should be noted that the quality of evidence is determined by comprehensively evaluating the reliability of the source of the matched evidence, the completeness of the information, the relevance of the context, and the time of publication.

[0148] The quality of evidence, data consistency, and level of evidence conflict are integrated to form a feature vector for each preliminary answer fragment;

[0149] The feature vector set is input into the confidence calibrator to generate a quantified confidence score for each validated preliminary answer fragment, expressed as:

[0150] ;

[0151] In the formula, It is the first The confidence score of the preliminary answer fragment after verification is dimensionless and can be standardized to [0, 1]. It is the weighting coefficient for the overall evidence quality score; It is the first The overall evidence quality score of the preliminary answer fragments after verification is dimensionless and can be obtained by combining factors such as source type, peer review level, guideline authority, and publication date. It is the weighting coefficient for the data consistency score; It is the first The data consistency score of the preliminary answer fragment after verification is dimensionless and is calculated by checking the consistency between the preliminary answer fragment and the entities, symptoms, examination items and medication descriptions of the evidence source. It is the weighting coefficient for the level of conflict of evidence; It is the first The evidence conflict level of the preliminary answer fragment after verification is dimensionless and can be represented by numerical values ​​such as low, medium, and high, corresponding to 0, 0.5, and 1, respectively. It is an index of the preliminary answer fragments after verification.

[0152] S4.4 Generate a comprehensive credibility score for each preliminary answer fragment by weighted fusion of the confidence score and the numerical results of the evidence conflict level.

[0153] It should be noted that, based on the preset confidence weighting coefficient, the confidence score of each preliminary answer segment is weighted to ensure that the evidence quality and data consistency of different preliminary answer segments are reflected in the comprehensive evaluation; the level of evidence conflict is numerically weighted so that preliminary answer segments with a high degree of conflict automatically have their credibility reduced in the comprehensive evaluation.

[0154] The confidence score, adjusted by the confidence weighting coefficient, is combined with the evidence conflict level in sequence so that each verified preliminary answer fragment receives a comprehensive credibility score that matches the quality of evidence, data consistency, and degree of conflict.

[0155] It should also be noted that the confidence weighting coefficient is obtained by analyzing the contribution of evidence quality, data consistency, and conflict level to the accuracy of the answer in historical verification data, and by calculating the relative importance ratio by statistically analyzing the proportion and influence of each feature in the successful answer segments.

[0156] S4.5 Utilize a dynamic risk assessment strategy to perform multi-dimensional weighted fusion of the comprehensive credibility score, and generate a risk score for each preliminary answer fragment through risk pattern recognition.

[0157] It should be noted that the overall credibility score of each preliminary answer fragment is received, and the overall credibility score is adjusted in the dynamic risk assessment strategy according to the weight relationship of different risk dimensions, so that the overall credibility score reflects the correlation between evidence consistency, evidence conflict level and overall credibility in each risk dimension.

[0158] The adjusted overall credibility score is classified using risk pattern recognition rules. Preliminary answer fragments with high overall credibility scores and good evidence consistency are marked as low risk, while preliminary answer fragments with low overall credibility scores or high evidence conflict levels are marked as high risk. The risk overall credibility score corresponding to each preliminary answer fragment is output.

[0159] S4.6. Based on the risk score, generate different diversified disposal instructions for each preliminary answer fragment.

[0160] It should be noted that, based on the risk score of each preliminary answer segment (high, medium, or low), different diversified handling instructions are assigned to each preliminary answer segment.

[0161] S5. When the diversified treatment instruction is permitted to be issued, conduct a safety and compliance check on the preliminary medical answer report and generate a compliant medical answer.

[0162] S5.1 Diversified disposal instructions are classified into pending instructions, prohibited instructions, and permitted instructions based on risk scores.

[0163] It should be noted that the risk scores of each preliminary answer fragment are sorted from low to high and compared with the upper and lower thresholds. The risk scores are divided into low risk scores, medium risk scores, and high risk scores. Preliminary answer fragments with low risk scores and high evidence consistency are classified as permitted to issue instructions, preliminary answer fragments with medium risk scores and some evidence conflict are classified as pending instructions, and preliminary answer fragments with high risk scores and high level of evidence conflict are classified as prohibited from issuing instructions.

[0164] It should also be noted that by analyzing the risk score distribution of historical preliminary answer fragments, the median or a certain percentile (e.g., the top 30 percentile) of the fragments with scores in the low-risk range is used as the lower limit threshold.

[0165] By statistically analyzing the distribution of high-risk scores for historical preliminary answer segments, the median or a certain percentile (e.g., the last 30 percentile) of the segments with scores in the high-risk range is used as the upper limit threshold.

[0166] S5.2 When a diversified treatment instruction is pending review, the preliminary medical response report will not be released and will be marked as pending review.

[0167] It should be noted that when a diversified treatment instruction is classified as pending review, the corresponding preliminary medical response report is marked as "pending review" in the internal records to prevent it from being directly published or displayed to users;

[0168] The risk score, evidence conflict level, and overall credibility information of each preliminary medical response segment in the preliminary response report are attached for subsequent automatic review.

[0169] The report to be reviewed will be added to the review queue, triggering the subsequent review process.

[0170] S5.3 When the diversified treatment instruction is to prohibit the release, the preliminary medical answer report shall be prohibited from being released, and the reason shall be recorded or a risk warning shall be generated.

[0171] It should be noted that when a diversified treatment instruction is classified as prohibited from publication, the preliminary medical answer report is immediately prohibited from being displayed or published to users. At the same time, the preliminary medical answer report is marked as "prohibited from publication" in the internal records, and the risk score, evidence conflict level and overall credibility information of each preliminary answer segment are attached.

[0172] The specific reasons for the ban should be recorded in detail, such as excessive conflict of evidence, low credibility, or insufficient data consistency, and risk warnings should be generated to trigger subsequent review and improvement processes.

[0173] S5.4 When the diversified treatment instruction is permitted to be issued, the legality, medical accuracy and potential risks of the content in the preliminary medical answer report are verified item by item, and after the verification is passed, a disclaimer is automatically added to the preliminary medical answer report to generate a compliant medical answer.

[0174] It should be noted that when the diversified treatment instruction is allowed to be issued, the credibility calibration agent will check each preliminary answer segment in the preliminary medical answer report item by item. After all checks and verifications are passed, the disclaimer text will be matched and attached to the content of the preliminary medical answer report to generate the final compliant medical answer, so that the legality, medical accuracy and risk information of each preliminary answer segment are clearly marked and guaranteed.

[0175] It should also be noted that the disclaimer text was obtained from the security requirements and potential risks of information use in medical Q&A. The content includes clauses on the reference nature of the information, usage restrictions and precautions, etc., and is attached after being matched with the structured content of the preliminary medical answer report, so that the final compliant medical answer can clearly inform users of the reference and potential risks of each answer when it is published.

[0176] This embodiment also provides a computer device applicable to the agent-based medical and health question-answering method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the agent-based medical and health question-answering method proposed in the above embodiment.

[0177] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0178] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the agent-based medical and health question-answering method proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0179] In summary, this invention achieves precise task decomposition and answer generation based on individual differences by: a master control agent generating collaborative instructions based on dynamic dialogue context, ensuring the continuity and personalization of question-and-answer results; simultaneously, a credibility calibration agent performs evidence source matching and conflict analysis on preliminary answers, and generates diversified treatment instructions in conjunction with a dynamic risk assessment strategy, achieving multi-dimensional control over the credibility and security of medical answers. Ultimately, this invention achieves the beneficial effect of providing personalized medical question-and-answer while ensuring the authenticity, reliability, compliance, and security of information.

[0180] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A medical health question-answering method based on intelligent agents, characterized in that: include, It receives medical and health questions input by users, uses a query understanding agent to analyze the medical and health questions, identifies medical entities and determines the user's query intent, and generates an intent analysis report; The master agent receives the intent parsing report and accesses the dynamic dialogue context pool. Based on the intent parsing report and the state of the dynamic dialogue context pool, it generates agent collaboration instructions. The system utilizes collaborative instructions to drive various professional agents, which retrieve information from the corresponding medical knowledge base and generate preliminary answer fragments. The answer integration agent then merges all the preliminary answer fragments to generate a preliminary medical answer report. The credibility of the preliminary medical answer report is assessed by a credibility calibration agent, which generates a confidence score and evidence conflict level. A dynamic risk assessment strategy is used to perform multi-dimensional weighted fusion and risk pattern recognition to generate diversified treatment instructions. When the diversified treatment instructions are allowed to be issued, the preliminary medical answer report is checked for safety and compliance to generate a compliant medical answer. The specific steps for generating diversified processing instructions are as follows: The credibility calibration agent receives preliminary medical answer reports, performs evidence source matching on each preliminary answer fragment, verifies the data consistency and accuracy of the preliminary answer fragments, and outputs the verified preliminary answer fragments and evidence source information; Extract the degree of conflict between evidence sources from the evidence source information, and generate an evidence conflict level for each preliminary answer fragment based on the conflict analysis; Calculate the confidence score for each verified preliminary answer fragment based on evidence quality, data consistency, and the level of evidence conflict. A comprehensive credibility score is generated for each preliminary answer fragment by weighted fusion of the confidence score and the level of evidence conflict. By employing a dynamic risk assessment strategy, the overall credibility score is weighted and integrated across multiple dimensions, and risk score is generated for each preliminary answer fragment through risk pattern recognition. Based on the risk score, different diversified disposal instructions are generated for each initial answer fragment.

2. The agent-based medical health question-answering method as described in claim 1, characterized in that: The healthcare issues mentioned include disease-related issues, symptom consultation, examination and medication issues, and health management and preventive healthcare issues.

3. The agent-based medical health question-answering method as described in claim 2, characterized in that: The process of using a query understanding agent to analyze healthcare questions, identify medical entities, determine the user's query intent, and generate an intent analysis report involves the following steps: The system performs noise reduction, word segmentation, and syntactic analysis on medical and health issues, removing irrelevant information, punctuation marks, and redundant words, and outputs standardized text data. The query understanding agent is used to identify medical entities and classify intents in standardized text data, obtain medical entities, and label the categories of medical entities; based on the user's context and the categories of medical entities, the user's query intent is determined by the classification algorithm, and structured medical entity data and user query intent are obtained. The system performs semantic matching between structured medical entity data and user query intent, and infers the user's specific needs by combining the context of the query, generating an intent analysis report.

4. The agent-based medical and health question-answering method as described in claim 3, characterized in that: The process of utilizing a master intelligent agent to receive intent parsing reports and access a dynamic dialogue context pool, and generating agent collaboration instructions based on the intent parsing reports and the state of the dynamic dialogue context pool, involves the following specific steps. The main control agent receives the intent parsing report and accesses the dynamic dialogue context pool through the interface to obtain the dialogue history and user health data related to the current query, thereby obtaining dynamic dialogue context information. By combining and analyzing the structured medical entity data in the intent parsing report with dynamic dialogue context information, we can determine the user's current specific needs and obtain comprehensive needs analysis results. Based on the comprehensive demand analysis results, the core user query needs are identified, the query processing task objectives are determined, and the query processing task is decomposed into multiple sub-tasks to generate intelligent agent collaboration instructions.

5. The agent-based medical and health question-answering method as described in claim 4, characterized in that: The process of using intelligent agent collaboration instructions to drive various specialized intelligent agents to retrieve information from the corresponding medical knowledge base and generate preliminary answer fragments involves the following specific steps. Intelligent agent collaboration instructions drive various professional intelligent agents to retrieve information from the corresponding medical knowledge base for different types of medical and health problems, and obtain medical retrieval results data. Organize and filter the medical search results to obtain a preliminary medical information set; The system supplements and integrates the preliminary medical information set using structured medical entity data, and matches the structured medical entity data with the user's specific needs based on the user's query intent to generate initial answer fragments. The initial answer fragments are optimized using a semantic understanding model, and further enhanced with deep semantics to obtain structured answer fragments. The structured answer fragments are validated for relevance, completeness, and accuracy to generate preliminary answer fragments.

6. The agent-based medical health question-answering method as described in claim 5, characterized in that: The process involves an intelligent agent that integrates all preliminary answer fragments to generate a preliminary medical answer report. The specific steps are as follows: The answer synthesis agent receives all preliminary answer fragments and performs unified syntactic and semantic processing on them. Through weighted fusion and structural adjustment, it generates an optimized set of answer fragments. The optimized answer fragment set is structured to generate a preliminary medical answer report.

7. The agent-based medical health question-answering method as described in claim 1, characterized in that: When the diversified treatment instruction allows for issuance, a security and compliance check is performed on the preliminary medical answer report to generate a compliant medical answer. The specific steps are as follows. Diversified disposal instructions are categorized into pending instructions, prohibited instructions, and permitted instructions based on risk scores. When a diversified treatment instruction is pending review, the preliminary medical response report will not be released and will be marked as pending review. When the diversified treatment instruction prohibits the release of information, the preliminary medical response report is prohibited from being released, and a risk warning is generated. When the diversified treatment instruction is approved for release, the legality, medical accuracy and potential risks of the content in the preliminary medical answer report are verified item by item. After the verification is passed, a disclaimer is automatically added to the preliminary medical answer report to generate a compliant medical answer.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the agent-based medical and health question-answering method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the agent-based medical and health question-answering method according to any one of claims 1 to 7.

Images