AI-driven intelligent guiding method and system for gynecological medical resources

Abstract

The invention provides an AI-driven intelligent guiding method and system for gynecological medical resources, and relates to the technical field of resource guiding, and the method comprises the steps: receiving an initial symptom description, and then carrying out the matching of a correlation problem; performing semantic recombination, and outputting a plurality of fuzzy variant problems; progressive polling is executed, and a polling reply pair sequence is obtained; performing symptom mode membership mapping to obtain an initial membership degree vector sequence; carrying out confidence correction, and outputting a confidence membership degree vector sequence; performing question semantic clustering grouping to obtain K groups of symptom semantic question clusters of the K core symptom modes; mapping and aggregating to obtain K confidence membership degree vector sequences; matching a real-time shunting strategy; and the information is sent to a hospital guide terminal as a reference basis for dynamic scheduling of gynecological medical resources. According to the invention, the technical problem that patients are easily allocated to unsuitable diagnosis and treatment channels due to improper communication because the diagnosis and treatment diversion for patients often depends on simple communication and judgment based on static empirical rules in the prior art is solved.

Description

Technical Field

[0001] This invention relates to the field of resource guidance technology, specifically to an AI-driven intelligent guidance method and system for gynecological medical resources. Background Technology

[0002] In modern medicine, the diagnosis and treatment of gynecological diseases is a complex and diverse process, involving the identification of a large number of symptoms, the derivation of diagnostic pathways, and the rational allocation of medical resources. Due to the high degree of overlap between disease symptoms and pathological patterns, patients often face repetitive and complex consultation processes, and medical resources are often misallocated due to poor communication and inaccurate symptom descriptions.

[0003] Manual consultations often rely on the personal experience of the triage staff and are usually brief and quick, with rapid and rather crude decision-making. It is difficult to accurately triage patients based on their conditions, which means that the treatment path assigned to patients is often not based on comprehensive and objective analysis, but rather on the triage staff's intuition and experience.

[0004] This triage method can easily lead to patients being incorrectly assigned to unsuitable treatment pathways, resulting in delayed emergency response, non-emergency cases occupying scarce resources, and overall low efficiency in medical care. For example, patients with mild symptoms may be incorrectly assigned to emergency departments, while some critically ill patients requiring urgent treatment may be assigned to general outpatient clinics, leading to a misallocation of medical resources and exacerbating hospital overcrowding. Summary of the Invention

[0005] This application provides an AI-driven intelligent guidance method and system for gynecological medical resources, aiming to solve the technical problems that existing technologies for patient triage often rely on triage personnel to make simple communication judgments based on static experience rules. This often leads to patients being assigned to unsuitable treatment channels due to improper communication, resulting in delayed emergency response, non-emergency patients occupying scarce resources, and low overall treatment efficiency.

[0006] The first aspect disclosed in this application provides an AI-driven intelligent guidance method for gynecological medical resources. The method includes: receiving an initial symptom description input by a real-time user; extracting multiple symptom keywords and matching them with related questions in a gynecological symptom association graph; semantically reorganizing the related questions to output multiple fuzzy variant questions; using the multiple fuzzy variant questions as the starting point for polling interaction, performing progressive polling on the real-time user to obtain a polling question-answer pair sequence; performing symptom pattern membership mapping on the question-answer pair sequence to obtain an initial membership vector sequence; and responding according to the polling question-answer pair sequence. The time-delay sequence is subjected to confidence correction of the initial membership vector sequence, and a confidence membership vector sequence is output; the question-answer pair sequence is subjected to question semantic clustering to obtain K groups of symptom semantic question clusters with K core symptom patterns; the confidence membership vector sequence is mapped and aggregated according to the K groups of symptom semantic question clusters to obtain K confidence membership vector sequences; the K core symptom patterns and K confidence membership vector sequences are used to match a real-time triage strategy in the hierarchical triage decision rule base; the real-time triage strategy is sent to the triage terminal as a reference for the dynamic scheduling of gynecological medical resources.

[0007] The second aspect of this application discloses an AI-driven intelligent guidance system for gynecological medical resources. This system is used in the aforementioned AI-driven intelligent guidance method for gynecological medical resources. The system includes: a problem-matching module, used to receive an initial symptom description input by a real-time user and extract multiple symptom keywords to match related questions in a gynecological symptom association graph; a semantic recombination module, used to perform semantic recombination on the related questions and output multiple fuzzy variant questions; a progressive polling module, used to perform progressive polling on the real-time user, starting with the multiple fuzzy variant questions, to obtain a polling-answer pair sequence; a membership mapping module, used to perform symptom pattern membership mapping on the question-answer pair sequence to obtain an initial membership vector sequence; and a confidence correction module. The system comprises the following modules: a module for performing confidence correction on the initial membership vector sequence based on the response delay sequence of the polling-answer pair sequence, and outputting a confidence membership vector sequence; a clustering and grouping module for performing question semantic clustering on the polling-answer pair sequence to obtain K groups of symptom semantic question clusters with K core symptom patterns; a mapping and aggregation module for mapping and aggregating the confidence membership vector sequence based on the K groups of symptom semantic question clusters to obtain K confidence membership vector sequences; a strategy matching module for matching real-time triage strategies in a hierarchical triage decision rule base using the K core symptom patterns and the K confidence membership vector sequences; and a strategy sending module for sending the real-time triage strategy to the triage terminal as a reference for dynamic scheduling of gynecological medical resources.

[0008] One or more technical solutions provided in this application have at least the following beneficial effects: By receiving symptom descriptions input by users, AI can extract key symptom words and match them according to a gynecological symptom association map to identify potential clinical problems. This ensures that the starting point of diagnosis is based on the user's specific symptoms, thereby ensuring an accurate understanding of the user's symptoms. Through semantic reorganization of related questions, multiple fuzzy variant questions are generated to adapt to different users' language expressions, allowing the system to interact with users more flexibly and naturally. This enhances the system's user-friendliness and usability, ensuring users can more easily understand and respond accurately. Through progressive polling, various aspects of the symptoms are gradually verified, ensuring that the user's answers cover all important features of the symptoms. This progressively deeper process avoids user confusion caused by asking for too much information at once, thus improving the effectiveness of the consultation. Through symptom pattern membership mapping, the user's symptom description is transformed into a numerical membership vector, effectively quantifying the relationship between symptoms and different pathological patterns. This allows the AI ​​system to process symptom data mathematically, improving the objectivity of the diagnostic process. By combining the user's response latency to assess their understanding of the questions and the quality of their answers, the membership vector can be dynamically adjusted, improving the confidence of diagnostic inference. The diagnostic path can be optimized based on the user's response behavior, avoiding errors caused by misunderstandings during the diagnostic process. This approach addresses errors caused by discrepancies in the diagnostic process. By semantically clustering question-answer pairs, multiple symptom information can be categorized into core symptom patterns, generating corresponding symptom semantic question clusters. This process helps extract the most important diagnostic information from massive amounts of data, improving questioning efficiency. Mapping multiple symptom semantic question clusters to confidence membership vectors allows for comprehensive analysis of diagnostic results for each symptom pattern, generating a final confidence membership vector sequence from multiple dimensions, further improving diagnostic accuracy. Finally, combining core symptom patterns with confidence vector sequences automatically matches appropriate triage strategies, providing data support for real-time patient triage and optimizing the process. The allocation of medical resources enables the system to intelligently triage patients to appropriate treatment pathways, such as emergency, outpatient, or home care, based on the severity of their symptoms and diagnostic results, thus avoiding resource waste. By sending real-time triage strategies to the triage terminals, the allocation of medical resources can be dynamically adjusted according to the severity of patients' symptoms and diagnostic results. This serves as the basis for triage personnel to implement physical space guidance and dynamically allocate medical staff. This not only effectively avoids the waste of medical resources but also improves the efficiency of hospital resource utilization, thereby alleviating hospital congestion pressure, optimizing the overall medical service process, and achieving real-time optimized matching of supply and demand for gynecological medical resources.

[0009] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0010] Figure 1 A schematic diagram of the AI-driven intelligent guidance method for gynecological medical resources provided in this application embodiment.

[0011] Figure 2 A schematic diagram of the structure of an AI-driven intelligent guidance system for gynecological medical resources provided in an embodiment of this application.

[0012] Figure labeling: 10 for association question matching module, 20 for semantic reorganization module, 30 for progressive polling module, 40 for membership mapping module, 50 for confidence correction module, 60 for clustering and grouping module, 70 for mapping aggregation module, 80 for policy matching module, and 90 for policy sending module. Detailed Implementation

[0013] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0014] Example 1, as Figure 1 As shown in the embodiments of this application, an AI-driven intelligent guidance method for gynecological medical resources is provided, the method comprising: A100: After receiving the initial symptom description from the user in real time, extract multiple symptom keywords and match them with the gynecological symptom association map to solve the association problem.

[0015] Users input initial symptom descriptions into the system, presented in natural language, such as "abdominal pain" or "menstrual irregularities." Natural language processing (NLP) techniques are used to process these descriptions, including: word segmentation, breaking down the user-input description into individual words or phrases (e.g., "abdominal pain" is broken down into "abdomen" and "pain"); and keyword extraction, using predefined rules or machine learning models to extract symptom-related keywords. A gynecological symptom association graph, which can be understood as a structured dataset containing different symptoms and their relationships, is used to match symptom-related questions with these keywords, serving as the basis for subsequent question-and-answer sessions.

[0016] A200: Semantically reorganize the associated problem to output multiple fuzzy variant problems.

[0017] The problem is broken down into multiple semantic units, which are the basic semantic units within the problem. Based on a thesaurus or terminology database, each semantic unit is expanded into different expressions; for example, "abdominal pain" is expanded to "stomach ache" or "abdominal discomfort." By combining these expanded semantic units, multiple versions of the fuzzy problem are formed. For instance, the question "Do you have abdominal pain?" can be varied into "Do you have a stomach ache?" or "Do you feel abdominal discomfort?" These variant questions are then recombined according to certain grammatical rules to ensure they are linguistically sound and understandable to users.

[0018] A300: Using the multiple fuzzy variant questions as the starting point for polling interaction, perform progressive polling on the real-time user to obtain a polling question-answer sequence.

[0019] Multiple ambiguous variant questions were chosen as the starting point for the initial interaction. These questions were semantically restructured and presented in various ways to ensure coverage of a wider range of user understanding and expression. The polling process involved progressively asking questions to the user to obtain more precise symptom information. This was a progressive dialogue; in each round, based on the user's response, it was determined whether to continue asking questions or further explore the details of a particular symptom. Through continuously progressive questioning, the scope of the questions was gradually narrowed to accurately understand the user's symptoms until all core symptom elements were confirmed.

[0020] A400: Perform symptom pattern membership mapping on the question-and-answer pair sequence to obtain an initial membership vector sequence.

[0021] After completing the question-and-answer interaction, the obtained question-and-answer data is converted into information that can be used for further decision-making. At this point, the symptom information is mapped using the PID fuzzy membership degree technique. PID fuzzy membership degree is a mathematical method combined with fuzzy logic control to describe the membership relationship between different symptoms and pathological patterns. This technique quantifies the relationship between each symptom and a potential pathological pattern, generating a membership degree vector. Fuzzy membership degree represents the degree of adaptation of a symptom to a specific pathological pattern. For example, a membership degree of 0.8 for a symptom means that the symptom is highly relevant to a certain pathological pattern, but not entirely certain. Multiple symptoms form a complete initial membership degree vector sequence, constituting a holistic description of the user's symptom information.

[0022] A500: Based on the response delay sequence of the polling response pair sequence, perform confidence correction on the initial membership vector sequence and output a confidence membership vector sequence.

[0023] During the question-and-answer interaction, the time interval between each user's response is recorded, i.e., response latency. In medical diagnosis, the user's response speed reflects their understanding of certain questions or the certainty of their symptoms. For example, if the response latency to a question is long, it means the user's description of the symptom is uncertain or the symptom is not obvious. Based on individual differences among users, including their physiological characteristics and cognitive baseline, the expected baseline latency for users answering questions is determined. The deviation between the user's response latency and the expected baseline latency is calculated. Based on the percentage of latency deviation, the initial membership vector sequence is adjusted. A longer latency reduces the confidence of the answer, i.e., reduces the membership degree; conversely, a shorter latency indicates that the user is more certain, and the confidence degree is relatively high. After latency correction, the final confidence membership vector sequence is generated, providing a more reliable basis for subsequent decision-making.

[0024] A600: Perform question semantic clustering on the question-answer sequence to obtain K groups of symptom semantic question clusters with K core symptom patterns.

[0025] After multiple rounds of question-and-answer sessions, semantic clustering is performed on all questions. The core objective of this process is to group questions based on their semantic relationship with symptoms, identifying core question clusters that represent different core symptom patterns. Each cluster represents a potential symptom pattern; for example, one cluster might be associated with "abdominal pain," another with "menstrual irregularities," and so on. Ultimately, through clustering, K groups of symptom semantic question clusters are obtained, with each symptom pattern corresponding to a specific question cluster.

[0026] A700: Aggregate the confidence membership vector sequence according to the K groups of symptom semantic question cluster mapping to obtain K confidence membership vector sequences.

[0027] The generated K sets of symptom semantic question clusters are mapped to user confidence membership vector sequences. Each question cluster represents a set of related symptoms, and each confidence membership vector represents the strength, probability, etc., of these symptoms. Based on the questions contained in each question cluster, the confidence membership information related to these questions is summarized and aggregated. That is, based on the question-and-answer process and the relevance of each symptom, new K confidence membership vector sequences are generated. Each vector corresponds to a core symptom pattern, and these vectors provide accurate confidence and relevance for each symptom pattern.

[0028] A800: The K core symptom patterns and K confidence membership vector sequences are used to match real-time triage strategies in the hierarchical triage decision rule base.

[0029] K core symptom patterns, validated through progressive polling depth, and their confidence membership vector sequences, which are fused with latency correction and medical logic verification, are used for three-dimensional matching in a hierarchical triage decision rule base that dynamically correlates the real-time load status of hospital resources.

[0030] Specifically, firstly, confidence weights are loaded based on the degree of consistency between symptom patterns and gynecological emergency and critical illness pathology templates. For example, if the membership degree of an ectopic pregnancy pattern is ≥0.9, a T1-level response is activated. Secondly, the saturation of the department is combined. For example, if the gynecological emergency room load rate is >80%, a load balancing strategy is triggered. Finally, a dynamic triage strategy is generated that includes physical space guidance paths, such as "3rd floor emergency room 3", priority codes and time-sensitive instructions, and "receive patients within 15 minutes". This achieves precise and coordinated adaptation of symptom characteristics, resource capacity and treatment timeliness.

[0031] A900: Sends the real-time triage strategy to the triage terminal as a reference for the dynamic scheduling of gynecological medical resources.

[0032] The real-time triage strategy is pushed to the physical devices of the hospital's triage terminal. Based on the dynamic analysis of the membership degree of the patient's symptom pattern and the real-time load status of medical resources, the strategy generates a structured scheduling instruction that includes departmental orientation, such as gynecological emergency / specialist outpatient clinic, priority code, and expected waiting time. This serves as the decision-making basis for triage personnel to implement physical space guidance and dynamic allocation of medical and nursing staff, thereby realizing the real-time optimized matching of the supply and demand relationship of gynecological medical resources.

[0033] Furthermore, after receiving the initial symptom description from real-time user input, the method extracts multiple symptom keywords and matches them against a gynecological symptom association map. The method includes: A110: Using a gynecological-specific stop word list to traverse the initial symptom descriptions after text cleaning, perform non-symptom semantic filtering to obtain multiple imprecise descriptive keywords; A120: Perform medical semantic standardization mapping on the multiple imprecise descriptive keywords to output the multiple symptom keywords; A130: Using the multiple symptom keywords as source nodes, perform a minimum correlation connected subgraph search on the gynecological symptom association graph to obtain a symptom association connected subgraph; A140: Match the association problem according to the pathological node composition of the symptom association connected subgraph.

[0034] The initial symptom descriptions input by users are in natural language form. These texts are first cleaned, including basic processing such as removing irrelevant characters, punctuation, and redundant spaces. In addition, word segmentation (breaking sentences into words) and part-of-speech tagging (marking the part of speech of each word) are performed to facilitate subsequent processing. In the field of gynecology, some words appear in the symptom descriptions but do not directly represent the symptoms themselves. Therefore, these words need to be filtered using a specially designed gynecological stop word list. By matching against the gynecological stop word list, these irrelevant words are filtered out, leaving only keywords or phrases related to the symptoms. For example, the input text "I feel a bit of a headache" will, after stop word filtering, only retain "headache" as a keyword.

[0035] Because different descriptions may refer to the same symptom—for example, "stomach pain," "abdominal pain," and "upper abdominal discomfort" all refer to the same symptom, namely stomach pain—medical semantic standardization is necessary. This process maps imprecise descriptive keywords (such as "abdominal pain" in different descriptions) to a standard medical term. In this way, multiple different symptom expressions can be unified into the corresponding standard medical terms. After standardization, the final output symptom keywords are officially recognized symptom terms in medicine.

[0036] A pre-constructed gynecological symptom association graph is used to analyze the relationships between symptom keywords. The graph is a graph structure of nodes (symptoms or symptoms) and edges (relationships between symptoms). Each node represents a symptom or symptom, and each edge represents the association between the two. The input symptom keywords are used as source nodes, and the association between these symptoms and other symptoms is searched. Through the search, other symptoms connected to these symptom keywords are found, forming the minimum connected subgraph. This subgraph represents the minimum symptom network associated with the user input symptoms. Here, "minimum" means that the nodes in the subgraph only contain those symptom nodes that are directly or indirectly related to the input symptoms, and contain as few irrelevant nodes as possible.

[0037] In addition to symptom keywords, the association graph also contains some related words that help describe the relationships between symptoms. For example, "pain" may be a connector between "abdominal pain" and "headache", or "severe" can describe the intensity of abdominal pain. These related words also serve as bridges when searching for the least connected subgraph.

[0038] Finally, the output is a symptom association connected subgraph containing multiple symptoms and their relationships, representing the association between the user-input symptoms and other symptoms. This subgraph contains both direct and indirect associations of symptoms, providing a basis for further diagnosis.

[0039] Based on the obtained symptom-related connected subgraph, matching related questions is performed according to pathological nodes. That is, standard questions associated with these symptoms or combinations of symptoms are found. These questions can help further confirm the possible causes of the symptoms. Finally, a series of related questions are generated as the basis for subsequent question-and-answer sessions. These questions are not only about single symptoms, but also about the relationships between multiple symptoms and the pathological background, helping to gain a deeper understanding of the user's health status.

[0040] Furthermore, the method involves semantically reorganizing the associated question to output multiple fuzzy variant questions, and includes: A210: Decompose the association question into multiple medical semantic elements; A220: Based on the multiple attribute dimensions of the multiple medical semantic elements, perform synonym matching and expansion in a multi-level synonym database for gynecological terms to obtain multiple sets of expanded semantic elements; A230: After combining and enumerating the multiple sets of expanded semantic elements, use a clinical consultation template for grammatical compliance reorganization to construct multiple intermediate semantic expressions; A240: Perform clinical diagnosis-oriented variant operations based on the sentence type of the multiple intermediate semantic expressions to generate the multiple fuzzy variant questions, wherein the fuzzy variant questions are used to perform intent equivalence verification using the association question.

[0041] To perform more precise language processing, the association question is broken down into multiple medical semantic elements. Semantic elements are the most basic semantic units extracted from the question, representing core medical concepts. By breaking them down, the core elements in the association question can be identified, and their meanings can be further analyzed.

[0042] Multiple medical semantic elements contain multiple attribute dimensions. For example, the anatomical dimension describes the location of the symptom, the degree dimension describes the intensity of the symptom, the time dimension describes the duration or timing of the symptom, and the feature dimension describes the characteristics of the symptom. Based on these dimensions, these medical semantic elements are expanded by querying a multi-level thesaurus of gynecological terms. The thesaurus contains a large number of synonyms for medical terms; for example, "severe abdominal pain" is synonymous with "acute abdominal pain" or "intense abdominal pain." These expansions generate multiple sets of expanded semantic elements for subsequent grammatical compliance and semantic variant operations.

[0043] All extended semantic elements are combined to form different expressions. To ensure that the syntax and structure of the questions conform to the standards of clinical diagnosis, these combined semantic elements are reorganized according to the clinical consultation template. The clinical consultation template is a predefined sentence structure that conforms to the norms of medical consultation. For example, combining the core symptom of "abdominal pain", multiple possible expressions are generated by combining different attributes such as degree and time: "How severe is the abdominal pain?" "How long has the abdominal pain lasted?" Each combination represents a possible way of asking questions, and these ways of asking questions serve as intermediate semantic expressions.

[0044] In clinical consultations, the same question can be asked in different ways to guide patients to provide more accurate information. Different variant operations are performed based on the sentence type of multiple intermediate semantic expressions. Exemplary variants are as follows: switch to severity rating, changing "Is the pain severe?" to "How severe is the pain?"; switch to related symptoms, changing "Do you have abdominal pain?" to "Do you have abdominal pain during menstruation?"; switch to temporal comparison, changing "How long have you had abdominal pain?" to "Is the abdominal pain more painful now than it was an hour ago?"; switch to open guidance, changing "When was your last menstrual period?" to "Please describe your most recent menstrual period?"

[0045] Each generated fuzzy variant question must maintain intent equivalence to the original associated question. This means that, despite different phrasing, the core purpose of the question and the information sought to be obtained are consistent. Ultimately, multiple fuzzy variant questions are generated based on the above variant operations. These questions use different expressions to ask users about the same symptom, ensuring coverage of a wider range of user understanding and descriptions.

[0046] Furthermore, based on multiple attribute dimensions of the aforementioned multiple medical semantic elements, synonym matching and expansion are performed in a multi-level synonym database for gynecological terminology to obtain multiple sets of expanded semantic elements. The method includes: A221: Perform four-dimensional classification and annotation on the first medical semantic element to generate a first attribute dimension, wherein the first attribute dimension covers one or more of the anatomical dimension, degree dimension, time dimension, and feature dimension; A222: Activate the corresponding sub-library of the gynecological terminology multi-level synonym library according to the first attribute dimension, and perform three-level term matching expansion using the first medical semantic element to obtain a first set of candidate semantic elements, wherein the gynecological terminology multi-level synonym library includes an anatomical location mapping library, a symptom intensity grading library, a symptom temporal pattern library, and a symptom feature description library; A223: Progressively perform cross-dimensional conflict detection on the first set of candidate semantic elements to generate a first set of extended semantic elements.

[0047] In medical consultation and symptom analysis, each symptom involves multiple attributes. These attributes are not limited to the symptom name but also include other characteristics of the symptom. To accurately describe symptoms, each medical semantic element is classified into four dimensions, divided into four main attribute dimensions: Anatomical dimension describes the location of the symptom, such as "abdominal pain" which can be divided into "upper abdominal pain" or "lower abdominal pain," or more specific locations such as "right lower abdominal pain"; Intensity dimension describes the strength or severity of the symptom, such as "mild abdominal pain" or "severe abdominal pain"; Temporal dimension describes the temporal characteristics of the symptom, involving the time of onset and duration of the symptom, such as "persistent abdominal pain," "occasional abdominal pain," or "abdominal pain since last month"; Feature dimension describes the specific characteristics of the symptom, such as "stabbing pain," "dull pain," or "distending pain." By analyzing the first medical semantic element and mapping it to one or more of the above four dimensions, this classification allows for the creation of a multi-dimensional description for each symptom semantic element, thus comprehensively capturing different aspects of the symptom.

[0048] The Gynecological Terminology Multilevel Synonym Database is a knowledge base containing multiple terminology sub-databases. Each sub-database provides terminology extensions for symptom descriptions across different dimensions. For example, terms related to anatomical location, symptom intensity, and temporal patterns can all be found in dedicated sub-databases. Specifically, the Anatomical Location Mapping Database contains terms for different anatomical locations, the Symptom Intensity Grading Database includes descriptions of symptoms at different intensities, the Symptom Temporal Pattern Database includes the temporal characteristics of symptoms, and the Symptom Feature Description Database contains specific feature descriptions of symptoms.

[0049] Based on the previous four-dimensional classification of the first medical semantic elements, a three-level term matching expansion is performed using the corresponding sub-database for targeted activation. This process is hierarchical and consists of the following stages: Level 1 expansion matches medical semantic elements with basic terms; Level 2 expansion further expands based on other symptom attributes, such as intensity or temporal patterns; and Level 3 expansion provides a more detailed classification and description of the symptoms, generating more precise expanded terms. Through this three-level matching expansion, multi-level and multi-dimensional candidate terms are provided for each symptom. Finally, based on these expansions, the first set of candidate semantic elements is obtained. These semantic elements contain different ways of describing the symptoms and provide more options for subsequent semantic reasoning.

[0050] In medical semantic analysis, conflicts can arise between different dimensions. For example, a symptom might be described as "persistent" in the time dimension and "mild" in the severity dimension. This creates a semantic conflict because "persistent" implies a significant symptom, while "mild" suggests a mild symptom. Therefore, cross-dimensional conflict detection is necessary. This involves analyzing whether there are logical inconsistencies between the dimensions of each candidate semantic element. When conflicts occur, conflicting candidates are automatically adjusted or removed to ensure that the generated semantic elements are compatible across different dimensions. Through progressive conflict detection, a first set of conflict-free extended semantic elements is ultimately determined. These extended semantic elements satisfy the requirements of medical logic and comprehensively cover all dimensions of the symptom.

[0051] Furthermore, using the multiple fuzzy variant questions as the starting point for polling interaction, progressive polling is performed on the real-time user to obtain a polling response sequence. The method includes: A310: Randomly extract a first fuzzy variant question without replacement from the plurality of fuzzy variant questions; A320: Receive the first round-robin response from the real-time user to the first fuzzy variant question; A330: Compare the first fuzzy variant question and the first round-robin response to generate a first core element coverage; A340: If the first core element coverage is higher than a preset medical credibility threshold, generate a first round-robin baseline question based on the uncovered core elements of the first fuzzy variant question according to the first round-robin response, wherein the first round-robin response and the first fuzzy variant question constitute a first round-robin question-answer pair; A350: Perform semantic recombination on the first round-robin baseline question to obtain a first fuzzy round-robin question; A360: Generate a second round-robin baseline question based on the uncovered core elements of the first fuzzy round-robin question according to the second round-robin response from the real-time user, wherein the second round-robin response and the first fuzzy round-robin question constitute a second round-robin question-answer pair; A370: Perform progressive round-robin on the real-time user in analogy until the uncovered core elements are an empty set, and output the round-robin question-answer pair sequence.

[0052] Among multiple fuzzy variant questions, one is randomly selected as the first fuzzy variant question. This random selection process, not in a specific order, is designed to avoid user bias based on question type and to increase the diversity of interaction. Once a fuzzy variant question is selected, it will not be drawn again, thus ensuring that each question is used only once and avoiding duplicate questioning.

[0053] After posing the first fuzzy variant question to the user, the system receives the user's real-time response, which can be in various forms, such as text, multiple-choice options, or audio. Based on the user's response, the system analyzes the key elements of the symptoms, extracts valuable information, and generates the first round of responses.

[0054] The received responses to the first round of queries are compared with the first fuzzy variant question to check whether the user has effectively answered the key elements of the question. Core elements refer to key information about the symptoms or problem; for example, if the question is "How severe is the abdominal pain?", then the core element is "Intensity of abdominal pain". By comparing the user's answer with the question, the first core element coverage is calculated. This refers to whether the user's response covers the core elements of the question, as well as the quality and completeness of the answer. For example, if the user fully describes the intensity of the abdominal pain (e.g., "Severe abdominal pain") and its duration (e.g., "Three days"), then the core element coverage is high.

[0055] A medical credibility threshold is set for each question, which is the accuracy of the answer in providing the expected medical information. If the user's answer meets this threshold, it is considered sufficiently credible. For example, if the user describes the intensity, location, and duration of abdominal pain in detail, and this information can be effectively used for diagnosis, then the answer is considered highly credible. Based on elements not covered in the user's answer, a baseline question is generated for subsequent inquiries. The baseline question asks questions based on information missing from previous answers. For example, if the user did not explicitly mention the duration of the abdominal pain, a new question is generated: "How long has the abdominal pain lasted?" The first fuzzy variant question and the user's first round of responses are combined into a complete question-answer pair. This ensures that subsequent inquiries are based on the user's existing answers and that information already known is not repeatedly asked.

[0056] The first round of baseline questions is semantically reorganized to generate the first fuzzy round of questions. Semantic reorganization refers to modifying the way the questions are expressed to make them more diverse so as to better adapt to the user's answers. Through semantic reorganization, the way the questions are expressed can be adjusted to avoid users missing key information due to misunderstandings or incomplete answers.

[0057] After the first fuzzy round-robin question is posed, the user provides a second round-robin response. This response offers further information about the symptoms, but may still only cover some of the core elements. The second round-robin response is analyzed to determine if it covers all the core elements of the first fuzzy round-robin question. If any core elements remain unanswered, a new question—the second round baseline question—is generated based on this missing information. This question continues to guide the user to provide more symptom information. The second round-robin response is then combined with the first fuzzy round-robin question to form a complete second round of question-and-answer dialogue.

[0058] Continue the progressive polling process, that is, adjust the questions based on each user's answer. When all core elements have been effectively covered without omission, the uncovered core elements will become an empty set, indicating that enough information has been collected. At this point, output the complete polling answer sequence. This sequence contains all the polling questions and the user's answers, which is the basis for subsequent symptom analysis, diagnosis, or decision-making.

[0059] Furthermore, the method also includes: A371: If the coverage of the first core element is lower than the preset medical credibility threshold, then randomly extract a second fuzzy variant question without replacement from the plurality of fuzzy variant questions; A372: Receive the first update response from the real-time user to the second fuzzy variant question; A373: By comparing the second fuzzy variant question and the first update response, generate a calculation of the coverage of the second core element; A374: By analogy, iteratively extract questions from the plurality of fuzzy variant questions to conduct a comprehension test on the real-time user until a fuzzy variant question that matches the user's cognitive ability is obtained, and start progressive polling.

[0060] If the coverage of the first core element is lower than the preset medical credibility threshold, that is, if the user's answer does not fully cover the core information, the current answer is considered not detailed or clear enough. In order to supplement the insufficient answer, a new question is randomly selected from the multiple fuzzy variant questions generated earlier. This question is different from the previous question to avoid repeated difficulties in understanding for the user. By randomly selecting without replacement, it is ensured that each question is used only once to avoid question duplication.

[0061] After posing the second ambiguous variant question to the user, the system receives the user's first updated response. The user's answer to the question will provide new information, involving more details of the symptoms or different descriptions.

[0062] The user's first updated response is compared with the second ambiguous variant question to check whether the user's answer effectively covers the core elements of the question, thus determining the coverage of the second core elements. If the user answers most or all of the key information of the question, the coverage will be high; if the answer is incomplete or ambiguous, the coverage will be low.

[0063] Based on the user's answers, the questions are gradually adjusted and iteratively extracted. If the second round of questions still fails to fully cover the core elements, questions are randomly selected again from multiple fuzzy variant questions and adjusted. Each adjustment of the questions is a test of the user's comprehension. The user's comprehension ability is judged based on their answers, and questions suitable for their cognitive ability are selected. Through this iterative process, the question expression method most suitable for the user's comprehension ability is gradually found. Finally, when the user can answer the questions clearly and accurately, it is considered that the user's cognitive ability has been adapted, and the formal progressive polling stage begins.

[0064] Furthermore, the method involves semantically reorganizing the first polling baseline problem to obtain a first fuzzy polling problem, the method comprising: A351: Backtrack the first clinical diagnosis-oriented variant operation rule of the first fuzzy variant problem; A352: Use the first clinical diagnosis-oriented variant operation rule to perform semantic reorganization of the first polling benchmark problem to generate the first fuzzy polling problem.

[0065] In clinical diagnosis, to accurately identify symptoms and determine diseases, there is a set of clinical diagnosis-oriented variant operating rules. These rules are specifically designed to adjust and optimize the semantic structure of questions. Based on medical knowledge bases and clinical experience, these rules aim to make the questioning more precise and better guide users to provide symptom information that aids in diagnosis. For example, if the symptom description is not specific or detailed enough, the rules suggest changing the question to a more specific form, such as changing "Do you feel pain?" to "Is your abdominal pain dull or severe?" This revisits the first clinical diagnosis-oriented variant operating rule to ensure that the questioning better meets clinical needs, thereby collecting accurate symptom information that is helpful for diagnosis.

[0066] Guided by the operating rules of the first clinical diagnostic variant, the first round of baseline questions undergoes semantic reorganization. Semantic reorganization refers to adjusting the language of the questions according to the rules to make them more targeted, precise, and in line with users' understanding habits. The reorganized first fuzzy round question will better meet the needs of clinical diagnosis and guide users to provide more detailed and accurate symptom information; this is the first fuzzy round question.

[0067] Furthermore, the method involves performing symptom pattern membership mapping on the question-and-answer pair sequence to obtain an initial membership vector sequence, the method comprising: A410: Extract the first symptom entity from the first fuzzy variant question; A420: Extract the first attribute modification, the first spatiotemporal feature, and the first confirmation intensity from the first polling response, wherein the first symptom entity, the first attribute modification, the first spatiotemporal feature, and the first confirmation intensity constitute the first symptom feature quadruple; A430: Preconstruct the symptom-pathological pattern matrix; A440: Calculate the weighted cosine similarity between the first symptom feature quadruple and the symptom-pathological pattern matrix, and output the first initial membership vector.

[0068] From the first fuzzy variant problem, identify and extract the first symptom entity, which is the symptom content mentioned in the problem. The symptom entity refers to the specific description of the symptom, which is the name of the pathological symptom or the description of the symptom itself, and can reflect the user's pain. For example, abdominal pain, headache, nausea, etc. are all symptom entities.

[0069] The first round of responses is the user's answer to the first fuzzy variant question. Features are extracted from these responses: the first attribute modifier refers to the attribute description of the symptom entity, such as degree, nature, and intensity; the first spatiotemporal feature refers to the temporal and spatial characteristics of the symptom's occurrence, describing the duration, timing, or trend of the symptom; and the first confirmation strength refers to the degree of confirmation regarding the symptom in the symptom description or the user's level of confidence. By extracting these three features from the user's response, and adding them to the first symptom entity, a complete first symptom feature quadruple is formed, which can quantify and structure the symptom description provided by the user.

[0070] The symptom-pathology pattern matrix is ​​associated with known pathological patterns (such as ectopic pregnancy, pelvic inflammatory disease, etc.) to quantify the manifestation of symptoms under different pathological patterns. The row vectors represent predefined typical gynecological pathological patterns, such as ectopic pregnancy, ovarian cysts, pelvic inflammatory disease, etc.; the column vectors represent the core elements of the symptoms, including location, nature, severity, time, and accompanying symptoms; the matrix value represents the association weight between the symptom and the pathological pattern. This value is a numerical value between 0 and 1, indicating the theoretical association strength of the symptom under the pathological pattern. 0 means completely unrelated, and 1 means completely related.

[0071] The similarity between the first symptom feature quadruple and the pathological patterns in the symptom-pathological pattern matrix is ​​calculated. The weighted cosine similarity algorithm is used to measure the similarity between the symptom features and the pathological patterns. Based on the calculated weighted cosine similarity, a membership value is generated for each pathological pattern. This membership value represents the matching degree between the symptom and each pathological pattern. Finally, these membership values ​​form the first initial membership vector.

[0072] Furthermore, based on the response delay sequence of the polling response pair sequence, the initial membership vector sequence is confidence-corrected, and a confidence membership vector sequence is output. The method includes: A510: Match the response delay baseline interval sequence based on the real-time user's physiological characteristics and cognitive baseline; A520: Calculate the delay deviation percentage sequence between the response delay baseline interval sequence and the response delay sequence; A530: After correcting the initial membership vector sequence for medical logic conflicts, use the delay deviation percentage sequence for confidence correction, and output the confidence membership vector sequence.

[0073] The analysis is based on users' physiological characteristics and cognitive baselines. Physiological characteristics include age, gender, and health conditions such as chronic diseases like hypertension and diabetes, which affect users' thinking speed when answering questions. The cognitive baseline includes users' cognitive abilities, cultural background, and language habits, which affect their understanding of questions and their response speed. For example, older users react more slowly, or users with cognitive impairments require more time to answer questions. Based on these physiological characteristics and cognitive baselines, a pre-defined sequence of response delay benchmark intervals is established. These intervals represent the normal time range required for users to answer questions under different physiological characteristics and cognitive baselines.

[0074] As users answer each question, their actual response latency is recorded—the time taken from when the question is asked to when the user provides an answer. This actual response latency is compared to a preset latency baseline range, and a latency deviation percentage is calculated. The latency deviation percentage represents the difference between the actual latency and the baseline latency. For example, if the baseline latency is 20 seconds, and the user actually took 30 seconds, the deviation is 50%. A complete latency deviation percentage sequence is generated to describe the user's latency deviation across multiple questions.

[0075] The initial membership vector sequence undergoes medical logic conflict correction. This is achieved by detecting contradictory statements in users' responses across multiple rounds to identify and correct medical inconsistencies. For example, contradictory statements in symptom descriptions, such as "My abdominal pain is severe, but I can move around normally," violate common medical sense, as severe abdominal pain typically impairs mobility. All question-and-answer pairs are examined, particularly symptom descriptions, and validated against gynecological clinical guidelines. Responses containing medical inconsistencies or contradictions are downweighted or corrected. For instance, if a user's answer violates clinical guidelines, its impact on decision-making is reduced, or the answer is excluded.

[0076] After correcting for medical logic conflicts, a confidence adjustment is performed using a time delay bias percentage sequence. Time delay bias reflects the user's comprehension and the accuracy of their answer. If a user spends too much time on a question, they perceive the answer as highly uncertain, thus lowering the confidence level of that question; conversely, if the response time is short and meets expectations, the answer has higher confidence. Based on time delay bias and medical logic correction, a confidence membership vector sequence is finally calculated, reflecting the credibility of each pathological pattern in the user's symptom description.

[0077] Example 2, based on the same inventive concept as the AI-driven intelligent guidance method for gynecological medical resources in the foregoing examples, such as... Figure 2 As shown in the figure, this application embodiment provides an AI-driven intelligent guidance system for gynecological medical resources, the system comprising: The system comprises the following modules: a question matching module 10, which receives an initial symptom description from a real-time user and extracts multiple symptom keywords to match related questions in a gynecological symptom association graph; a semantic reorganization module 20, which performs semantic reorganization on the related questions and outputs multiple fuzzy variant questions; a progressive polling module 30, which uses the multiple fuzzy variant questions as the starting point for polling interaction and performs progressive polling on the real-time user to obtain a polling question-answer pair sequence; a membership mapping module 40, which performs symptom pattern membership mapping on the question-answer pair sequence to obtain an initial membership degree vector sequence; and a confidence correction module 50, which adjusts the initial membership degree vector based on the response delay sequence of the polling question-answer pair sequence. The system performs confidence correction on the quantity sequence and outputs a confidence membership vector sequence; a clustering grouping module 60 is used to perform question semantic clustering on the round-robin question-answer pair sequence to obtain K groups of symptom semantic question clusters with K core symptom patterns; a mapping aggregation module 70 is used to map and aggregate the confidence membership vector sequence according to the K groups of symptom semantic question clusters to obtain K confidence membership vector sequences; a strategy matching module 80 is used to match real-time triage strategies in the hierarchical triage decision rule base using the K core symptom patterns and K confidence membership vector sequences; and a strategy sending module 90 is used to send the real-time triage strategy to the triage terminal as a reference for the dynamic scheduling of gynecological medical resources.

[0078] Furthermore, the associated question matching module 10 is used to perform the following operation steps: The initial symptom descriptions after text cleaning and processing are traversed using a gynecological stop word list to perform non-symptom semantic filtering, resulting in multiple imprecise descriptive keywords. These imprecise keywords are then subjected to medical semantic standardization mapping to output multiple symptom keywords. Using these symptom keywords as source nodes, a minimum connectivity subgraph search is performed on the gynecological symptom association graph to obtain a symptom association connected subgraph. Based on the pathological node composition of the symptom association connected subgraph, the association problem is matched.

[0079] Furthermore, the semantic recombination module 20 is used to perform the following operational steps: The associated question is broken down into multiple medical semantic elements; based on multiple attribute dimensions of the multiple medical semantic elements, synonym matching and expansion are performed in a multi-level synonym database for gynecological terms to obtain multiple sets of expanded semantic elements; after combining and enumerating the multiple sets of expanded semantic elements, grammatical compliance reorganization is performed using a clinical consultation template to construct multiple intermediate semantic expressions; based on the sentence type of the multiple intermediate semantic expressions, a clinical diagnosis-oriented variant operation is performed to generate the multiple fuzzy variant questions, wherein the fuzzy variant questions are used to perform intent equivalence verification using the associated question.

[0080] Furthermore, the semantic recombination module 20 is used to perform the following operational steps: The first medical semantic element is classified and labeled in four dimensions to generate a first attribute dimension, wherein the first attribute dimension covers one or more of the anatomical dimension, degree dimension, time dimension, and feature dimension; the corresponding sub-library of the gynecological terminology multi-level synonym library is activated according to the first attribute dimension, and the first medical semantic element is used to perform three-level term matching expansion to obtain a first set of candidate semantic elements, wherein the gynecological terminology multi-level synonym library includes an anatomical location mapping library, a symptom intensity grading library, a symptom temporal pattern library, and a symptom feature description library; cross-dimensional conflict detection is progressively performed on the first set of candidate semantic elements to generate a first set of extended semantic elements.

[0081] Furthermore, the progressive polling module 30 is used to perform the following operation steps: Randomly and without replacement, a first fuzzy variant question is extracted from the plurality of fuzzy variant questions; a first round-robin response from the real-time user to the first fuzzy variant question is received; the first fuzzy variant question and the first round-robin response are compared to generate a first core element coverage; if the first core element coverage is higher than a preset medical credibility threshold, a first round-robin baseline question is generated based on the first round-robin response for the uncovered core elements of the first fuzzy variant question, wherein the first round-robin response and the first fuzzy variant question constitute a first round-robin question-answer pair; the first round-robin baseline question is semantically reorganized to obtain a first fuzzy round-robin question; a second round-robin baseline question is generated based on the second round-robin response from the real-time user for the uncovered core elements of the first fuzzy round-robin question, wherein the second round-robin response and the first fuzzy round-robin question constitute a second round-robin question-answer pair; progressive round-robin is performed on the real-time user in a similar manner until the uncovered core elements are an empty set, and the round-robin question-answer pair sequence is output.

[0082] Furthermore, the progressive polling module 30 is used to perform the following operation steps: If the coverage of the first core element is lower than a preset medical credibility threshold, a second fuzzy variant question is randomly extracted from the plurality of fuzzy variant questions without replacement; the first update response from the real-time user to the second fuzzy variant question is received; by comparing the second fuzzy variant question and the first update response, a second core element coverage calculation is generated; similarly, questions are iteratively extracted from the plurality of fuzzy variant questions to perform a comprehension test on the real-time user until a fuzzy variant question that matches the user's cognitive ability is obtained, and progressive polling is initiated.

[0083] Furthermore, the progressive polling module 30 is used to perform the following operation steps: Backtrack the first clinical diagnosis-oriented variant operation rule of the first fuzzy variant problem; use the first clinical diagnosis-oriented variant operation rule to perform semantic reorganization of the first polling benchmark problem to generate the first fuzzy polling problem.

[0084] Furthermore, the membership mapping module 40 is used to perform the following operation steps: Extract the first symptom entity from the first fuzzy variant problem; extract the first attribute modification, the first spatiotemporal feature, and the first confirmation intensity from the first polling response, wherein the first symptom entity, the first attribute modification, the first spatiotemporal feature, and the first confirmation intensity constitute the first symptom feature quadruple; preconstruct the symptom pathology pattern matrix; calculate the weighted cosine similarity between the first symptom feature quadruple and the symptom pathology pattern matrix, and output the first initial membership vector.

[0085] Furthermore, the confidence correction module 50 is used to perform the following operation steps: Based on the real-time user's physiological characteristics and cognitive baseline, a response delay baseline interval sequence is matched; the delay deviation percentage sequence of the response delay baseline interval sequence and the response delay sequence is calculated; after correcting the initial membership vector sequence for medical logic conflicts, the delay deviation percentage sequence is used for confidence correction, and the confidence membership vector sequence is output.

[0086] Through the foregoing detailed description of the AI-driven intelligent guidance method for gynecological medical resources, those skilled in the art can clearly understand the AI-driven intelligent guidance system for gynecological medical resources in this embodiment. Since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and relevant parts can be referred to the method section.

[0087] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An AI-driven intelligent guidance method for gynecological medical resources, characterized in that, The method includes: After receiving the initial symptom description from real-time user input, multiple symptom keywords are extracted and matched against a gynecological symptom association map. The related questions are semantically reorganized to output multiple fuzzy variant questions; Using the multiple fuzzy variant questions as the starting point for polling interaction, progressive polling is performed on the real-time user to obtain a polling question-answer sequence; Perform symptom pattern membership mapping on the question-and-answer pair sequence to obtain an initial membership vector sequence; Based on the response delay sequence of the question-and-answer pair sequence, the confidence correction of the initial membership vector sequence is performed, and the confidence membership vector sequence is output. The question-and-answer sequence is subjected to question semantic clustering to obtain K groups of symptom semantic question clusters with K core symptom patterns; Based on the K groups of symptom semantic question cluster mappings, the confidence membership vector sequence is aggregated to obtain K confidence membership vector sequences; The K core symptom patterns and K confidence membership vector sequences are used to match real-time triage strategies in the hierarchical triage decision rule base; The real-time triage strategy is sent to the triage terminal as a reference for the dynamic scheduling of gynecological medical resources.

2. The AI-driven intelligent guidance method for gynecological medical resources as described in claim 1, characterized in that, After receiving an initial symptom description from a real-time user, the method extracts multiple symptom keywords and matches them against a gynecological symptom association graph. The initial symptom descriptions after text cleaning were traversed using a gynecological stop word list, and non-symptom semantic filtering was performed to obtain several non-precise descriptive keywords. Perform medical semantic standardization mapping on the multiple imprecise descriptive keywords, and output the multiple symptom keywords; Using the multiple symptom keywords as source nodes, a minimum correlation connected subgraph search is performed on the gynecological symptom association graph to obtain the symptom association connected subgraph. The association problem is matched based on the pathological nodes of the symptom-related connected subgraph.

3. The AI-driven intelligent guidance method for gynecological medical resources as described in claim 1, characterized in that, The method involves semantically reorganizing the associated questions to output multiple fuzzy variant questions, and includes: The aforementioned related issues are broken down into multiple medical semantic elements; Based on the multiple attribute dimensions of the aforementioned multiple medical semantic elements, synonym matching and expansion are performed in the multi-level synonym database of gynecological terms to obtain multiple sets of expanded semantic elements; After combining and enumerating the multiple sets of extended semantic elements, a clinical consultation template is used to perform grammatical compliance reorganization to construct multiple intermediate semantic expressions; Perform clinical diagnosis-oriented variant operations based on the sentence type of the plurality of intermediate semantic expressions to generate the plurality of fuzzy variant questions, wherein the fuzzy variant questions are used for intent equivalence verification using the associated questions.

4. The AI-driven intelligent guidance method for gynecological medical resources as described in claim 3, characterized in that, Based on multiple attribute dimensions of the aforementioned multiple medical semantic elements, synonym matching and expansion are performed in a multi-level synonym database for gynecological terminology to obtain multiple sets of expanded semantic elements. The method includes: The first medical semantic element is classified and labeled in four dimensions to generate the first attribute dimension, wherein the first attribute dimension covers one or more of the anatomical dimension, degree dimension, time dimension and feature dimension. The corresponding sub-library of the gynecological terminology multi-level synonym library is activated according to the first attribute dimension, and the first medical semantic element is used to perform three-level term matching expansion to obtain the first set of candidate semantic elements. The gynecological terminology multi-level synonym library includes an anatomical location mapping library, a symptom intensity grading library, a symptom temporal pattern library, and a symptom feature description library. Cross-dimensional conflict detection is performed progressively on the first group of candidate semantic elements to generate the first group of extended semantic elements.

5. The AI-driven intelligent guidance method for gynecological medical resources as described in claim 3, characterized in that, Using the multiple fuzzy variant questions as the starting point for polling interaction, progressive polling is performed on the real-time user to obtain a polling question-answer sequence. The method includes: Randomly and without replacement, a first fuzzy variant problem is drawn from the plurality of fuzzy variant problems; Receive the first polling response from the real-time user to the first fuzzy variant question; By comparing the first fuzzy variant question with the first polling response, the first core element coverage is generated; If the coverage of the first core element is higher than the preset medical credibility threshold, then the first polling baseline question is generated based on the core elements not covered by the first polling response for the first fuzzy variant question, wherein the first polling response and the first fuzzy variant question constitute the first polling response pair; The first fuzzy polling problem is obtained by semantically reorganizing the first polling benchmark problem; Based on the real-time user's second round of query response, a second round of query baseline question is generated for the core elements not covered by the first fuzzy round of query question, wherein the second round of query response and the first fuzzy round of query question constitute a second round of query-response pair; Similarly, perform progressive polling on the real-time users until the core elements that are not covered are an empty set, and output the polling response sequence.

6. The AI-driven intelligent guidance method for gynecological medical resources as described in claim 5, characterized in that, The method further includes: If the coverage of the first core element is lower than the preset medical credibility threshold, then a second fuzzy variant problem is randomly drawn from the plurality of fuzzy variant problems without replacement; Receive the first updated response from the real-time user to the second fuzzy variant question; By comparing the second fuzzy variant problem and the first update response, a second core element coverage calculation is generated. By analogy, questions are iteratively extracted from the multiple fuzzy variant questions to conduct real-time user comprehension tests until a fuzzy variant question that matches the user's cognitive ability is obtained, and then progressive polling is initiated.

7. The AI-driven intelligent guidance method for gynecological medical resources as described in claim 5, characterized in that, The method involves semantically reorganizing the first polling baseline problem to obtain the first fuzzy polling problem, the method comprising: Retrospectively examine the first clinical diagnosis-oriented variant operation rules for the first fuzzy variant problem; The semantics of the first polling baseline problem are reorganized using the first clinical diagnosis-oriented variant operation rule to generate the first fuzzy polling problem.

8. The AI-driven intelligent guidance method for gynecological medical resources as described in claim 5, characterized in that, The method involves performing symptom pattern membership mapping on the question-and-answer pair sequence to obtain an initial membership vector sequence, the method comprising: Extract the first symptom entity from the first fuzzy variant problem; Extract the first attribute modifier, the first spatiotemporal feature, and the first confirmation intensity from the first polling response, wherein the first symptom entity, the first attribute modifier, the first spatiotemporal feature, and the first confirmation intensity constitute the first symptom feature quadruple; Pre-constructed symptom-pathology pattern matrix; Calculate the weighted cosine similarity between the first symptom feature quadruple and the symptom pathology pattern matrix, and output the first initial membership vector.

9. The AI-driven intelligent guidance method for gynecological medical resources as described in claim 1, characterized in that, The method includes: performing confidence correction on the initial membership vector sequence based on the response delay sequence of the question-and-answer pair sequence, and outputting a confidence membership vector sequence. Based on the physiological characteristics and cognitive baseline of the real-time user, a response delay baseline interval sequence is matched; Calculate the response delay reference interval sequence and the response delay sequence delay deviation percentage sequence; After correcting for medical logic conflicts in the initial membership vector sequence, confidence correction is performed using the time delay deviation percentage sequence, and the confidence membership vector sequence is output.

10. An AI-driven intelligent guidance system for gynecological medical resources, characterized in that: For implementing the AI-driven intelligent guidance method for gynecological medical resources according to any one of claims 1-9, the system comprises: The related question matching module is used to receive the initial symptom description input by the user in real time, extract multiple symptom keywords and match related questions in the gynecological symptom association map; The semantic restructuring module is used to perform semantic restructuring on the associated questions and output multiple fuzzy variant questions; The progressive polling module is used to perform progressive polling on the real-time user, starting from the multiple fuzzy variant questions, to obtain a polling response sequence. The membership mapping module is used to perform symptom pattern membership mapping on the question-answer pair sequence to obtain an initial membership degree vector sequence. The confidence correction module is used to perform confidence correction on the initial membership vector sequence based on the response delay sequence of the polling response pair sequence, and output a confidence membership vector sequence. The clustering and grouping module is used to perform question semantic clustering and grouping on the question-answer pair sequence to obtain K groups of symptom semantic question clusters with K core symptom patterns; The mapping and aggregation module is used to map and aggregate the confidence membership vector sequence according to the K groups of symptom semantic question clusters to obtain K confidence membership vector sequences; The strategy matching module is used to match real-time triage strategies in the hierarchical triage decision rule base using the K core symptom patterns and K confidence membership vector sequences. The strategy sending module is used to send the real-time triage strategy to the triage terminal as a reference for the dynamic scheduling of gynecological medical resources.

Images