Intelligent inquiry method, system and device based on knowledge graph and storage medium

Abstract

The application discloses an intelligent inquiry method and system based on a knowledge graph, equipment and a storage medium, and belongs to the technical field of intelligent medical treatment. The method can be applied to an intelligent inquiry scene and comprises the following steps: acquiring a disease description text; performing structural processing on the disease description text, performing semantic retrieval on the structural description data, and performing graph matching processing; fusing and enhancing the retrieved knowledge set and the graph matching data, and performing inquiry based on the fused and enhanced data. In the application, the knowledge graph is utilized, candidate knowledge sets are extracted from the structural description data and the knowledge graph, and the candidate knowledge sets and the knowledge graph are fused and enhanced to ensure that reasoning is based on real medical knowledge, the illusion rate of generated content is reduced, and the reasoning accuracy of existing medical intelligent reasoning is improved.

Description

Technical Field

[0001] This application relates to the fields of smart healthcare technology and intelligent consultation technology, and in particular to an intelligent consultation method, system, device and storage medium based on knowledge graph. Background Technology

[0002] With the rise of large-scale medical models and intelligent consultation technologies, disease-assisted consultation functions can be supported. Generating preliminary consultation results or suggestions based on input descriptions of the patient's condition has been widely studied in clinical decision support. However, most systems rely solely on the semantic generation capabilities of large language models (LLMs), which are prone to generating seemingly reasonable but factually incorrect content (i.e., the "illusion" problem). Factual reasoning is imprecise, and the model's reasoning chain is broken, failing to achieve truly reliable and interpretable intelligent consultation assistance. Therefore, existing technologies suffer from the technical problem of low reasoning accuracy in current medical intelligent reasoning.

[0003] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0004] The main purpose of this application is to provide an intelligent medical consultation method based on knowledge graphs, which aims to solve the technical problem of low reasoning accuracy in existing medical intelligent reasoning.

[0005] Firstly, a knowledge graph-based intelligent consultation method is provided, including:

[0006] Obtain the text describing the user's condition;

[0007] The text describing the illness is processed into a structured format to obtain structured description data.

[0008] Semantic retrieval is performed on the structured description data to obtain a candidate knowledge set;

[0009] Based on the structured description data, graph matching is performed on the preset knowledge graph to obtain graph matching data;

[0010] The candidate knowledge set and the graph matching data are fused to obtain fused enhanced data, wherein the fused enhanced data has both semantic information and structural logic information.

[0011] Based on the fused and enhanced data, the system performs intelligent diagnosis on the user and outputs the diagnosis results.

[0012] In one possible implementation of this application, the step of structuring the disease description text to obtain structured description data includes:

[0013] The text description of the illness is cleaned to obtain a cleaned text description of the illness. The text cleaning process includes adjusting typos and punctuation marks and removing irrelevant characters.

[0014] The cleaned description of the illness was segmented into basic words.

[0015] The basic words are converted into word vectors and the word vectors are input into a preset medical-specific named entity recognition model;

[0016] Based on the medical-specific named entity recognition model, the word vectors are first processed word by word to obtain word tags. Then, based on the preset medical knowledge base and the context information contained in the word tags, the selected word tags with medical significance are extracted as medical elements.

[0017] The medical elements are standardized and coded according to preset categories to obtain structured description data.

[0018] In one possible implementation of this application, the step of performing knowledge fusion on the candidate knowledge set and the graph matching data to obtain fused enhanced data includes:

[0019] Obtain the weights corresponding to the candidate knowledge set and the graph matching data, respectively.

[0020] The complementary information between the candidate knowledge set and the atlas matching data is determined based on multiple dimensions, wherein the candidate knowledge set contains specific clinical manifestations and the atlas matching data contains structured general medical knowledge.

[0021] Determine the knowledge gap information of the candidate knowledge set and the graph matching data, wherein the knowledge gap information includes entity-level gap information, relation-level gap information and attribute-level gap information;

[0022] Based on their respective weights, the candidate knowledge set and the graph matching data are fused using complementary information and knowledge gap information filling to obtain fused enhanced data. In one possible embodiment of this application, the medical elements include at least any one of symptoms, signs, test results, and diseases; the graph matching data is causal chain data matching symptoms and diseases; and the step of performing intelligent consultation on the user based on the fused enhanced data and outputting consultation results includes:

[0023] Based on the fused and enhanced data, a causal chain is determined by the combination of symptom elements, sign elements, test result elements, and disease elements.

[0024] Determine the confidence level for each of the aforementioned causal chains;

[0025] A fact confidence matrix is ​​established based on the confidence levels corresponding to each of the multiple causal chains.

[0026] The system outputs a consultation result to the user based on the description of the illness text. The consultation result is a list of suspected diseases corresponding to multiple causal chains, and each suspected disease in the disease list is equipped with confidence information.

[0027] In one possible implementation of this application, the respective weights include a first weight corresponding to the candidate knowledge set and a second weight corresponding to the graph matching data. Before obtaining the weights corresponding to the candidate knowledge set and the graph matching data, the following steps are included:

[0028] Based on the candidate knowledge set, the text confidence score is determined, wherein the text confidence score is related to quality weight, semantic similarity, and content consistency.

[0029] Based on the map matching data, the map confidence level is determined, wherein the map confidence level is related to path consistency and map coverage;

[0030] Based on the text confidence and the graph confidence, the first weight corresponding to the candidate knowledge set is determined;

[0031] The second weight is determined based on the first weight.

[0032] In one possible implementation of this application, after performing intelligent diagnosis on the user based on the fused enhanced data and outputting the diagnosis results, the process includes:

[0033] The disease names and drug dosages involved in the consultation results are verified to obtain the verification confidence level;

[0034] If the verification confidence level is lower than a preset threshold, the process returns to the step of performing semantic retrieval on the structured description data to obtain a candidate knowledge set, until the verification confidence level is not lower than the preset threshold.

[0035] Secondly, a knowledge graph-based intelligent consultation system is provided. This system includes an acquisition module, a structured processing module, a semantic retrieval module, a graph matching module, a knowledge fusion module, and an intelligent consultation module.

[0036] The acquisition module is used to acquire the text describing the user's condition.

[0037] The structured processing module is used to perform structured processing on the disease description text to obtain structured description data;

[0038] The semantic retrieval module is used to perform semantic retrieval on the structured description data to obtain a candidate knowledge set;

[0039] The graph matching module is used to perform graph matching on the preset knowledge graph based on the structured description data to obtain graph matching data;

[0040] The knowledge fusion module is used to perform knowledge fusion on the candidate knowledge set and the graph matching data to obtain fused enhanced data, wherein the fused enhanced data has both semantic information and structural logic information.

[0041] The intelligent consultation module is used to conduct intelligent consultations on the user based on the fused and enhanced data, and output the consultation results.

[0042] In one possible implementation of this application, the intelligent consultation module includes a consultation unit, a rehabilitation suggestion unit, and an assessment unit.

[0043] The consultation unit is used to perform intelligent consultation on the user based on the fused enhanced data and output the consultation results;

[0044] The rehabilitation suggestion unit is used to generate feasible rehabilitation suggestions based on the consultation results and a preset knowledge base.

[0045] The evaluation unit is used to verify and evaluate the output of the consultation unit and the output of the rehabilitation suggestion unit, wherein the consultation results output by the consultation unit and the feasible rehabilitation suggestions output by the rehabilitation suggestion unit are cross-validated and consistency-validated.

[0046] Thirdly, a knowledge graph-based intelligent consultation device is provided. The knowledge graph-based intelligent consultation device is an entity node device. The knowledge graph-based intelligent consultation device includes: a memory, a processor, and a knowledge graph-based intelligent consultation program stored in the memory and executable on the processor. The processor executes the knowledge graph-based intelligent consultation program to implement the steps of the knowledge graph-based intelligent consultation method.

[0047] Fourthly, a storage medium is provided, on which a program for implementing a knowledge graph-based intelligent consultation method is stored, wherein when the knowledge graph-based intelligent consultation program is executed by a processor, the steps of the knowledge graph-based intelligent consultation method described above are implemented.

[0048] This application provides a knowledge graph-based intelligent medical consultation method, system, device, and storage medium. Compared to existing technologies where the accuracy of medical intelligent reasoning is low, the method addresses this issue by: acquiring a user's medical condition description text; performing structured processing on the description text to obtain structured description data; performing semantic retrieval on the structured description data to obtain a candidate knowledge set; performing graph matching on a preset knowledge graph based on the structured description data to obtain graph matching data; fusing the candidate knowledge set and the graph matching data to obtain fused enhanced data, wherein the fused enhanced data contains both semantic and structural logical information; and performing intelligent medical consultation on the user based on the fused enhanced data, outputting the consultation results. In this application, by extracting candidate knowledge sets and knowledge graphs from structured description data and fusing them to enhance knowledge, the method ensures that reasoning is based on real medical knowledge, reduces the illusion rate of generated content, and improves the accuracy of existing medical intelligent reasoning. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a flowchart illustrating an intelligent consultation method based on knowledge graphs in one embodiment of this application;

[0051] Figure 2 This is a schematic flowchart of a specific implementation of step S20 in one embodiment of this application;

[0052] Figure 3 yes Figure 1 A schematic diagram of a specific implementation method for step S50;

[0053] Figure 4 yes Figure 1 A flowchart illustrating a specific implementation method prior to step S501;

[0054] Figure 5 yes Figure 1 A schematic diagram of a specific implementation method for step S60;

[0055] Figure 6 yes Figure 9 A flowchart illustrating a specific implementation method following step S60;

[0056] Figure 7 This is a schematic diagram of the structure of an intelligent consultation system based on knowledge graphs in one embodiment of this application;

[0057] Figure 8 This is a schematic diagram of the structure of a computer device according to one embodiment of this application;

[0058] Figure 9 This is another structural schematic diagram of a computer device in one embodiment of this application. Detailed Implementation

[0059] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0060] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0061] It should also be understood that the terminology used in this application specification is for the purpose of describing particular embodiments only and is not intended to limit the application. As used in this application specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0062] It should also be further understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0063] This application provides a knowledge graph-based intelligent consultation method. In the first embodiment of this knowledge graph-based intelligent consultation method, the method is applied to a knowledge graph-based intelligent consultation system, which can be a medical platform. The above-mentioned knowledge graph-based intelligent consultation intelligent question-answering method can be applied to intelligent consultation scenarios or remote consultation scenarios.

[0064] This knowledge graph-based intelligent medical consultation system is a medical intelligent reasoning system based on Retrieval-Augmented Generation (RAG). Before answering questions and generating text, the RAG-based medical intelligent reasoning system retrieves relevant, up-to-date, and accurate information from external knowledge bases. This information is then provided as context to a large language model, enabling the model to generate more accurate, credible, and factual answers. When a user asks a question, the system uses methods such as vector similarity search from specified external knowledge sources to find the most relevant document fragments or information. This retrieved information is combined with the user's original question to form an "enhanced" suggestion. This complete suggestion, containing the original question and relevant background information, is then sent to the large language model, allowing it to generate the final answer.

[0065] Reference Figure 1 The method is applied to a knowledge graph-based intelligent consultation system, and the knowledge graph-based intelligent consultation method includes steps S10-S50:

[0066] In the field of smart healthcare technology, current medical large-scale models and intelligent consultation systems have been widely studied in clinical decision support, and many corresponding products can generate preliminary consultation results or suggestions based on the input description of the patient's condition. However, most systems rely solely on the semantic generation capabilities of large language models (LLMs). LLMs are a core driving force of current artificial intelligence, especially the generative AI wave. However, these models are prone to generating seemingly reasonable but factually incorrect content (i.e., the "illusion" problem), resulting in inaccurate factual reasoning and broken reasoning chains, making it impossible to achieve truly reliable and explainable intelligent consultation assistance. Therefore, existing technologies suffer from the technical problem of low accuracy in current medical intelligent reasoning.

[0067] Step S10: Obtain the text describing the user's condition;

[0068] Retrieve the user's medical condition description text, which is generally unstructured. This description text can specifically describe medical data, such as medical records or consultation logs.

[0069] Example: Description of the user's condition: "The patient has had a frequent cough and low-grade fever for the past week, and a CT scan showed shadows in the lungs."

[0070] Step S20: Perform structured processing on the disease description text to obtain structured description data;

[0071] Structured processing refers to the process of extracting, organizing, and transforming unstructured or semi-structured information according to a clear logical framework, fixed attribute categories, or data models, so that it becomes standard format data that is machine-readable, queryable, and analyzable.

[0072] The RAG-based medical intelligent reasoning system performs structured processing on the medical condition description text, parsing, segmenting, vectorizing, and indexing to obtain a vector database or index file. This is done to facilitate machine retrieval and matching, providing a computable knowledge base.

[0073] In intelligent medical consultation scenarios, the fact structuring module of a knowledge graph-based intelligent medical consultation system processes the medical condition description text into structured descriptive data. This module is responsible for transforming the input unstructured medical condition description into a computable structured fact table. Through structured descriptive data, structured fact extraction, and inference chain recording, the entire medical consultation decision-making process becomes traceable and verifiable, enhancing the interpretability of reasoning.

[0074] Reference Figure 2 Step S20 includes steps S201 to S205:

[0075] Step S201: Perform text cleaning processing on the disease description text to obtain the cleaned disease description text. The text cleaning processing includes adjusting the typos and punctuation marks and removing irrelevant characters.

[0076] Step S202: Perform word segmentation on the cleaned description of the illness to obtain basic words;

[0077] Step S203: Convert the basic words into word vectors and input the word vectors into a preset medical-specific named entity recognition model;

[0078] Step S204: Based on the medical-specific named entity recognition model, the word vector is first processed word by word to obtain word tags. Then, based on the preset medical knowledge base and the context information contained in the word tags, the selected word tags with medical significance are extracted as medical elements.

[0079] The core steps of the fact structuring module include, first, medical entity recognition and standardization. Using a medical-specific named entity recognition model, medical-related elements are extracted from the natural language of the medical condition description text.

[0080] As an example, medical-related elements include symptoms, signs, lab results, and diagnosis.

[0081] Step S205: Standardize and encode the medical elements according to preset categories to obtain structured description data.

[0082] The input text undergoes initial cleaning and standardization. This includes correcting obvious typos, standardizing punctuation, and removing irrelevant characters to ensure proper text formatting and create favorable conditions for model recognition. The text is then broken down into smaller basic units such as words or subwords, and these units are converted into word vectors that the model can understand. This process preserves the positional information of each word in the original text. The preprocessed numerical sequence is fed into a medical-specific named entity recognition model. Each word is analyzed for its encoding and contextual information, and a label is predicted and assigned to each word in the sequence. These labels typically follow these criteria: B-XXX: Indicates the beginning of a specific type of entity (Begin), such as B-symptom, B-disease. I-XXX: Indicates the interior of a specific type of entity (Inside), i.e., the subsequent part of the entity. O: Indicates that the word does not belong to any entity of interest (Outside). Based on a pre-set medical knowledge base and trained patterns, the model determines which word sequences together constitute a meaningful medical entity and identifies its specific category: symptoms, signs, test results, and medical history. Consecutive words with B- and I- labels are merged into a complete entity. For example, symptoms B-"cough" and I-"sneezes" are merged into the complete symptom "cough". At the same time, any special characters or spaces that may have been mistakenly introduced are removed. Finally, all identified entities are categorized according to preset classes, and a structured list of medical elements, such as structured data (e.g., JSON), is output for further processing and analysis by the computer system.

[0083] The core steps of the fact structuring module also include standardizing the extraction results of medically relevant elements. This standardization is achieved through a unified medical terminology standard.

[0084] As an example, after extracting medical elements from the input medical condition description text, standardizing and encoding them, and outputting structured facts:

[0085] {

[0086] "Symptom": ["cough", "low-grade fever"],

[0087] "ExamFinding": ["Lung shadow"],

[0088] "Duration": "7 days"

[0089] }

[0090] Step S30: Perform semantic retrieval on the structured description data to obtain a candidate knowledge set;

[0091] Traditional RAG (Research Aggregator) relies solely on single-layer vector retrieval. This application proposes a "hierarchical semantic-knowledge graph hybrid RAG mechanism" to improve the accuracy and robustness of medical fact-based reasoning. The multi-layer RAG retrieval and knowledge graph fusion of the hierarchical semantic-knowledge graph hybrid RAG mechanism significantly reduces the illusion rate of generated content, ensuring that reasoning is based on real medical knowledge and achieving high knowledge accuracy.

[0092] In the field of smart healthcare technology, enhanced knowledge representations are generated through the knowledge fusion module in a knowledge graph-based intelligent consultation system.

[0093] The knowledge fusion module is structured into three sub-layers: semantic vector retrieval layer, structured graph matching layer, and knowledge fusion layer.

[0094] In the semantic vector retrieval layer, medical knowledge texts that are closest to the semantics of structured facts, such as literature abstracts and clinical guidelines, are retrieved through medical semantic embedding models (such as BioBERT and MedCPT) to generate candidate knowledge sets.

[0095] Step S40: Perform graph matching on the preset knowledge graph based on the structured description data to obtain graph matching data;

[0096] In the structured graph matching layer, based on the medical knowledge graph, the facts of the patient's condition are matched with standard disease nodes.

[0097] The pre-built knowledge graph is an existing structured medical knowledge base, which can be used for self-built professional knowledge graphs. The pre-built knowledge graph consists of nodes and edges. Graph matching involves "entity alignment" and relation mapping of structured descriptive data. Structured entities extracted from the text are linked to standardized nodes in the knowledge graph, and the implicit contextual relationships between entities are mapped to the semantic relationships defined in the knowledge graph.

[0098] In some embodiments, a bidirectional matching algorithm is employed. This algorithm calculates the shortest path starting from the fact node and traces backward from the disease node to the symptom node, ultimately calculating a structural consistency score. The bidirectional constraint of the bidirectional matching algorithm avoids chaotic one-to-many matching, forcing the algorithm to find a more globally consistent solution. It considers not only matching from the symptom node to the disease node but also simultaneously matching from the disease node to the symptom node, ensuring consistency and uniqueness of the matching through bidirectional verification. The bidirectional matching algorithm can find a stable set of matchings, ensuring that, from both perspectives, the paired entities are each other's optimal choice, thereby reducing ambiguity and misalignment.

[0099] The pre-built knowledge graph is a medical knowledge base, constructed in the early stages of AI-assisted consultation systems like the "Intelligent Medical Assistant." These systems are designed to improve consultation efficiency. The constructed medical knowledge base forms the foundation for retrieval. The preliminary preparation stage of the Intelligent Medical Assistant consultation system includes integrating authoritative medical textbooks (such as *Internal Medicine* and *Diagnostics*), clinical guidelines (such as various guidelines published by the Chinese Medical Association), drug instructions, massive amounts of de-identified and annotated electronic medical records (EHRs), and medical literature databases. Unstructured text, such as textbook passages, is transformed into structured, machine-readable knowledge. For example, a large "disease-symptom-examination-treatment" graph is constructed, where each disease is associated with a series of typical symptoms, accompanying symptoms, rare symptoms, and their probabilities of occurrence. Key laboratory test indicators and imaging features are also linked. Consultation criteria are defined, such as requiring symptom A + sign B, or test value C greater than X. Differential consultation information is included, identifying which diseases are easily confused with and how to differentiate them.

[0100] Step S50: Perform knowledge fusion on the candidate knowledge set and the graph matching data to obtain fused enhanced data, wherein the fused enhanced data has both semantic information and structural logic information;

[0101] Knowledge fusion aims to integrate knowledge from different sources to generate a more complete, accurate, and consistent knowledge representation than any single source. The core of knowledge fusion involving the candidate knowledge set and the graph matching data lies in effectively integrating the retrieved candidate knowledge set with the graph matching data. The retrieved candidate knowledge set is typically text fragments, while the graph matching data is usually structured and standardized knowledge.

[0102] Enhanced knowledge representations are generated through a weighted fusion mechanism at the knowledge fusion layer. The candidate knowledge set and the graph matching data are then fused to obtain fused enhanced data, which contains both semantic and structural logical information.

[0103] Reference Figure 3 Step S50 includes steps S501-S504:

[0104] Step S501: Obtain the weights corresponding to the candidate knowledge set and the graph matching data respectively;

[0105] In some embodiments, when performing knowledge fusion on the candidate knowledge set and the graph matching data, both the candidate knowledge set and the graph matching data are assigned their respective weights.

[0106] Obtain the weights corresponding to the candidate knowledge set and the graph matching data. The respective weights include a first weight corresponding to the candidate knowledge set and a second weight corresponding to the graph matching data.

[0107] Before step S501, refer to Figure 4 This includes steps A1 through A4:

[0108] Step A1: Based on the candidate knowledge set, determine the text confidence score, wherein the text confidence score is related to quality weight, semantic similarity, and content consistency.

[0109] The first weight is adaptively determined based on the factual confidence matrix. The first weight of the candidate knowledge set is related to the text confidence and graph confidence. Text confidence is related to quality weight, semantic similarity, and content consistency. The quality weight represents the influence factor of the candidate knowledge set's document quality. When the textual evidence is more reliable, the first weight approaches 1, indicating a greater reliance on the candidate knowledge set. When the graph structure is more credible, the first weight approaches 0, indicating a greater reliance on graph matching data. The quality weight represents the authority and timeliness of the source. For example, different journal levels and publication times result in different document quality and therefore different quality weights. Semantic similarity represents the semantic similarity between the text content in the candidate knowledge set and the input structured disease description text facts. Content consistency indicates whether a symptom-disease relationship is supported. Based on the quality weight, semantic similarity, and content consistency of the candidate knowledge set, the text confidence is determined.

[0110] Step A2: Based on the map matching data, determine the map confidence level, wherein the map confidence level is related to path consistency and map coverage;

[0111] The confidence level of a disease map is related to path consistency and map coverage. Path consistency refers to the existence of a symptom → disease path. Map coverage refers to the proportion of nodes that are hit by a symptom.

[0112] The confidence level of the map is determined based on path consistency and map coverage.

[0113] Step A3: Based on the text confidence and the graph confidence, determine the first weight corresponding to the candidate knowledge set;

[0114] Automatically balance text confidence and graph confidence using maximum normalization:

[0115] Here, represents the first weight, is the text confidence score, and is the graph confidence score. Therefore, when the text evidence is more reliable, the first weight approaches 1, indicating a greater reliance on the candidate knowledge set. When the graph structure is more reliable, the first weight approaches 0, indicating a greater reliance on the graph matching data.

[0116] Step A4: Determine the second weight based on the first weight.

[0117] Based on the text confidence score and the graph confidence score, a first weight is determined for each candidate knowledge set. This first weight represents the weight of the text knowledge in the final fusion. Based on this first weight, a second weight is determined. The second weight is the graph knowledge weight.

[0118] Step S502: Determine the complementary information of the candidate knowledge set and the atlas matching data based on multiple dimensions, wherein the candidate knowledge set contains specific clinical manifestations and the atlas matching data contains structured general medical knowledge.

[0119] This study analyzes the complementarity of the two types of data in terms of information type, granularity, and dimension. The candidate knowledge set contains specific clinical manifestations of patients, but may be incomplete, noisy, and lack in-depth medical knowledge. The graph matching data contains structured medical knowledge, such as typical symptoms, treatments, and complications of diseases, but lacks patient specificity.

[0120] Step S503: Determine the knowledge gap information of the candidate knowledge set and the graph matching data, wherein the knowledge gap information includes entity-level gap information, relation-level gap information and attribute-level gap information;

[0121] Knowledge gaps are identified by comparing two types of data. Entity-level gaps refer to entities mentioned in the candidate knowledge set but lacking relevant information in the knowledge graph, or entities with related relationships in the graph but not mentioned in the candidate knowledge set. Relationship-level gaps refer to relationships between entities that are not explicitly defined in the candidate knowledge set but exist as typical relationships in the knowledge graph, or relationships mentioned in the candidate knowledge set but not supported by the knowledge graph. Attribute-level gaps refer to entity attribute information, such as severity and duration, described in the candidate knowledge set but lacking corresponding attribute values ​​in the knowledge graph.

[0122] Step S504: Based on their respective weights, the candidate knowledge set and the graph matching data are subjected to the fusion processing of complementary information and the filling processing of knowledge gap information to obtain fused enhanced data.

[0123] The candidate knowledge set and the graph matching data are fused using complementary information: From the graph matching data, other entities related to entities in the candidate knowledge set, such as other typical symptoms and complications of the same disease, are extracted and added to the fused data as supplementary entities. For example, if the candidate knowledge set contains "cough," and the graph shows that typical symptoms of "pneumonia" also include "fever" and "sputum," then "fever" and "sputum" are added as candidate symptoms. Typical relationships existing in the graph are added to the fused data, especially when entities in the candidate knowledge set have strong correlations in the graph. For example, if the candidate knowledge set contains "diabetes" and "polydipsia," and the graph shows that "diabetes" and "polydipsia" have a "typical symptom" relationship, then this relationship is added to the fused data. Attributes of entities in the graph, such as disease definitions, normal value ranges, and treatment principles, are combined with attributes in the candidate knowledge set, such as specific numerical values.

[0124] The candidate knowledge set and the graph matching data are processed to fill knowledge gaps. Using the relational paths in the graph, entities or relationships that are not mentioned in the candidate knowledge set but may exist are inferred. If entity A exists in the candidate knowledge set, and A often appears alongside B and C in the graph, but C exists in the candidate knowledge set but B is missing, then B can be considered for inclusion. Entity combinations in the candidate knowledge set are matched with common patterns in the graph. If a disease pattern is matched, but other typical manifestations of that disease are missing in the candidate knowledge set, then a potential knowledge gap is indicated.

[0125] The candidate knowledge set, graph matching data, and new knowledge obtained through complementation and filling are integrated to form a unified knowledge representation.

[0126] As an example, the sum of the first weight and the second weight is 1.

[0127] Here, represents enhanced knowledge, represents the textual knowledge representation obtained from the semantic vector retrieval layer, and represents the knowledge vector obtained from the structured graph matching layer. Based on the first weight and the second weight, knowledge fusion is performed on the candidate knowledge set and the graph matching data to obtain fused enhanced data. When the textual evidence is more reliable, the first weight approaches 1, indicating a greater reliance on the candidate knowledge set. When the graph structure is more credible, the first weight approaches 0, indicating a greater reliance on the graph matching data.

[0128] Based on the candidate knowledge set and the graph matching data, conflict detection and resolution are first performed, then information completion and enhancement are performed, and finally the data is fused and enhanced.

[0129] Step S60: Perform intelligent diagnosis on the user based on the fused and enhanced data, and output the diagnosis results.

[0130] The intelligent consultation module in the knowledge graph-based intelligent consultation system performs disease reasoning consultation to obtain consultation results. The intelligent consultation module includes a consultation unit, a rehabilitation suggestion unit, and an evaluation unit. The consultation unit is used to perform intelligent consultation on the user based on the fused enhanced data and output consultation results; the rehabilitation suggestion unit is used to generate feasible rehabilitation suggestions based on the consultation results and a preset knowledge base.

[0131] An evaluation unit is used to verify and evaluate the output of the consultation unit and the output of the rehabilitation suggestion unit, wherein the consultation results output by the consultation unit and the feasible rehabilitation suggestions output by the rehabilitation suggestion unit are cross-validated and consistency-validated.

[0132] The medical elements include at least one of the following: symptoms, signs, test results, and disease. The atlas matching data is causal chain data matching symptoms and diseases, referring to... Figure 5 Step S60 includes steps S601 to S604:

[0133] Step S601: Based on the fused and enhanced data, determine the causal chain obtained by combining symptom elements, sign elements, test result elements, and disease elements;

[0134] Based on the fused and enhanced data, a causal chain is established.

[0135] Extract relevant entities such as symptoms, signs, test results, and diseases from the fused and enhanced data, and the relationships between them. Construct paths connecting these entities using existing relationships in a pre-defined knowledge graph.

[0136] In medical terms, a typical medical history chain might be symptoms / signs / test results → indications or associations → disease. However, it can also be the reverse: a disease causes certain symptoms or signs. In some embodiments, a causal chain is established, where symptoms, etc., lead to a disease. This causal chain may involve intermediate nodes; for example, an abnormal test result such as high blood sugar indicates a diabetic pathology, and diabetes may then lead to a series of symptoms such as polydipsia. The chain begins with symptoms, signs, or test results, and ends with the disease.

[0137] A causal chain between symptoms and disease is established using a relation extraction model. This causal chain is derived from a combination of symptom elements, sign elements, test result elements, and disease elements. Different combinations of these elements can yield different causal chains depending on the specific circumstances.

[0138] Specifically, a directed graph G is extracted from the fused and enhanced data. An allowed set of relations R is defined, where relations point from symptoms, signs, and test results to diseases. For each disease node d, all nodes s of type symptom, sign, and test result are found such that there exists a path from s to d, and the relation type of each edge on the path is in R. For each such node s, the path is recorded and combined to form a causal chain.

[0139] As an example, a causal chain is established: "cough → bronchitis" and "low-grade fever + lung shadow → pneumonia".

[0140] Step S602: Determine the confidence level of each of the causal chains;

[0141] As an example, the confidence level of each of the aforementioned causal chains is determined. "Cough → Bronchitis (confidence level 0.72)", "Low-grade fever + lung shadow → Pneumonia (confidence level 0.88)".

[0142] Step S603: Based on the confidence levels corresponding to each of the multiple causal chains, establish a fact confidence matrix;

[0143] Cross-modal evidence fusion was employed to weight and integrate electronic medical records, test results, and literature search results, and a fact confidence matrix was constructed based on the confidence levels corresponding to each of the multiple causal chains.

[0144] Step S604: Output the consultation results based on the description of the illness text to the user, wherein the consultation results are a list of suspected diseases corresponding to multiple causal chains, and each suspected disease in the disease list is equipped with confidence information.

[0145] The system outputs a consultation result to the user based on the description of the illness text. The consultation result is a list of suspected diseases corresponding to multiple causal chains, and each suspected disease in the disease list is equipped with confidence information.

[0146] In some embodiments, assessing the quality of creative responses is challenging. An innovation generation mechanism is provided. This embodiment introduces a multi-perspective simulation mechanism, generating multiple treatment plans by simulating doctors from different departments, such as pulmonology, infectious diseases, and radiology. The system automatically aggregates and compares the consistency and coverage of these plans. An innovation score is calculated using an information gain-based metric.

[0147] Here, the entropy value of the multi-agent solution set is represented, reflecting knowledge diversity, calculated by the probability distribution entropy of the output solutions from multiple departments. The entropy of the corresponding solution in the benchmark knowledge base is calculated by the prior distribution entropy of each solution in medical guidelines and standardized pathways. The significance of the innovation score is that the solutions proposed by the multi-agent team are more "diverse" than the guidelines, indicating higher innovation. If the consistency with the guidelines is high, the innovation is lower, but the authority is high. This quantifies the ability to "innovate without excessively deviating from clinical norms." For cross-departmental intelligent collaboration, the multi-agent structure supports multi-perspective interaction and solution integration, enhancing innovation and solution diversity.

[0148] This embodiment boasts high accuracy in factual reasoning. Most existing systems rely solely on the semantic generation capabilities of large language models, leading to the generation of seemingly reasonable but factually incorrect content. This embodiment employs structured data, addressing the shortcomings of structured reasoning from real medical knowledge bases. Existing technologies primarily provide natural language text as search results; this embodiment can extract specialized medical entities such as symptoms, signs, medications, and consultations from unstructured medical records, preventing breaks in the reasoning chain. Furthermore, this embodiment features cross-domain knowledge fusion and conflict verification mechanisms, employing multidisciplinary physician assessments for complex cases to avoid generation from a single perspective.

[0149] Therefore, the medical intelligent reasoning system of this application achieves truly reliable and explainable intelligent consultation assistance in three stages: "fact understanding—structured reasoning—multi-perspective innovation". Through a four-layer innovative mechanism of "structured fact extraction + multi-stage RAG retrieval + multi-agent reasoning collaboration + knowledge alignment optimization", the system realizes intelligent reasoning and innovative generation throughout the entire process from disease description to consultation suggestions.

[0150] After step S60, refer to Figure 6 This includes steps S70-S80:

[0151] Step S70: Perform fact verification on the disease names and drug dosages involved in the consultation results to obtain the verification confidence level;

[0152] The knowledge graph-based intelligent medical consultation system also includes a knowledge consistency optimization module. This module is used to reduce potential factual biases in reasoning and generation. It includes a two-step mechanism: reducing factual bias through knowledge alignment optimization and a fact verification feedback loop. Contrastive learning optimizes the consistency between the linguistic representation and the knowledge graph representation, ensuring that the generated content is semantically and logically aligned. After the model generates output, the system automatically calls a medical fact-checker to verify key information such as disease names and drug dosages. If the confidence level is below a threshold, it backtracks to the knowledge retrieval stage for regeneration. This mechanism forms a closed-loop optimization chain from generation → verification → correction → regeneration. The loss function is defined as:

[0153] This is a semantic embedding vector for the generated text. Using a medical-specific semantic model such as BioBERT, the input includes consultation results such as consultation suggestions, and the output is a multi-dimensional semantic vector. This is a structured embedding vector for the knowledge graph. Based on a medical knowledge graph such as symptom-disease-examination-drug structure, a graph neural network is used to treat symptoms, diseases, and examinations as nodes. Based on the edge weights of the knowledge graph, a graph neural network is used to generate node embeddings, and the node embeddings are aggregated to form the final structured embedding vector of the knowledge graph.

[0154] After conducting intelligent medical consultations on users and obtaining the consultation results, the disease names and drug dosages involved in the consultation results are verified to obtain the verification confidence level. The role of the loss function is to make the semantics of the generated text more consistent with the structure of the medical knowledge graph, thereby reducing AI "illusions" and ensuring the reliability of clinical logic.

[0155] Step S80: If the verification confidence is lower than a preset threshold, return to the step of performing semantic retrieval on the structured description data to obtain a candidate knowledge set, until the verification confidence is not lower than the preset threshold.

[0156] If the verification confidence level is lower than a preset threshold, the system returns to the step of performing semantic retrieval on the structured description data to obtain a candidate knowledge set, until the verification confidence level is not lower than the preset threshold. A fact verification loop is introduced, enabling the system to have self-correction and knowledge update capabilities. Compared to traditional RAG systems, this system improves the accuracy of diagnosis and the usability of innovative suggestions in typical tasks such as disease prediction or treatment plan generation, significantly enhancing the "knowledge-practice consistency" and clinical usability of large-scale models in the medical field.

[0157] This application provides a knowledge graph-based intelligent medical consultation method, system, device, and storage medium. Compared to existing technologies where the accuracy of medical intelligent reasoning is low, the method addresses this issue by: acquiring a user's medical condition description text; performing structured processing on the description text to obtain structured description data; performing semantic retrieval on the structured description data to obtain a candidate knowledge set; performing graph matching on a preset knowledge graph based on the structured description data to obtain graph matching data; fusing the candidate knowledge set and the graph matching data to obtain fused enhanced data, wherein the fused enhanced data contains both semantic and structural logical information; and performing intelligent medical consultation on the user based on the fused enhanced data, outputting the consultation results. In this application, by extracting candidate knowledge sets and knowledge graphs from structured description data and fusing them to enhance knowledge, the method ensures that reasoning is based on real medical knowledge, reduces the illusion rate of generated content, and improves the accuracy of existing medical intelligent reasoning.

[0158] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0159] In one embodiment, a knowledge graph-based intelligent consultation system is provided, which corresponds one-to-one with the knowledge graph-based intelligent consultation methods described in the above embodiments. For example... Figure 7 As shown, the knowledge graph-based intelligent consultation system includes an acquisition module 101, a structured processing module 102, a semantic retrieval module 103, a graph matching module 104, a knowledge fusion module 105, and an intelligent consultation module 106. Detailed descriptions of each functional module are as follows:

[0160] Module 101 is used to acquire the text describing the user's condition.

[0161] The structured processing module 102 is used to perform structured processing on the disease description text to obtain structured description data;

[0162] Semantic retrieval module 103 is used to perform semantic retrieval on the structured description data to obtain a candidate knowledge set;

[0163] Graph matching module 104 is used to perform graph matching on a preset knowledge graph based on the structured description data to obtain graph matching data;

[0164] The knowledge fusion module 105 is used to perform knowledge fusion on the candidate knowledge set and the graph matching data to obtain fused enhanced data, wherein the fused enhanced data has both semantic information and structural logic information.

[0165] The intelligent consultation module 106 is used to conduct intelligent consultations on the user based on the fused enhanced data and output the consultation results.

[0166] In one possible implementation of this application, the intelligent consultation module 103 includes a consultation unit, a rehabilitation suggestion unit, and an evaluation unit. The consultation unit is used to perform intelligent consultation on the user based on the fused enhanced data and output consultation results. The rehabilitation suggestion unit is used to generate feasible rehabilitation suggestions based on the consultation results and a preset knowledge base. The evaluation unit is used to verify and evaluate the output of the consultation unit and the output of the rehabilitation suggestion unit, wherein the consultation results output by the consultation unit and the feasible rehabilitation suggestions output by the rehabilitation suggestion unit are cross-validated and consistency-validated.

[0167] This application provides an intelligent medical consultation system based on a knowledge graph. Compared to existing technologies where medical intelligent reasoning suffers from low accuracy, this system acquires a text describing a user's condition; performs structured processing on the text to obtain structured description data; performs semantic retrieval on the structured description data to obtain a candidate knowledge set; performs graph matching on a preset knowledge graph based on the structured description data to obtain graph matching data; fuses the candidate knowledge set and the graph matching data to obtain fused enhanced data, wherein the fused enhanced data contains both semantic and structural logical information; and performs intelligent medical consultation on the user based on the fused enhanced data, outputting the consultation results. In this application, by extracting candidate knowledge sets and knowledge graphs from structured description data and fusing them to enhance knowledge, the system ensures that reasoning is based on real medical knowledge, reduces the illusion rate of generated content, and improves the accuracy of existing medical intelligent reasoning.

[0168] Specific limitations regarding knowledge graph-based intelligent consultation systems can be found in the above section on limitations of knowledge graph-based intelligent consultation methods, and will not be repeated here. The modules in the aforementioned knowledge graph-based intelligent consultation system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.

[0169] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 8 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a knowledge graph-based intelligent medical consultation method on the server side.

[0170] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 9As shown, the computer device includes a processor, memory, network 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 network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the client-side functions or steps of a knowledge graph-based intelligent medical consultation method.

[0171] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps:

[0172] Obtain the text describing the user's condition;

[0173] The text describing the illness is processed into a structured format to obtain structured description data.

[0174] Semantic retrieval is performed on the structured description data to obtain a candidate knowledge set;

[0175] Based on the structured description data, graph matching is performed on the preset knowledge graph to obtain graph matching data;

[0176] The candidate knowledge set and the graph matching data are fused to obtain fused enhanced data, wherein the fused enhanced data has both semantic information and structural logic information.

[0177] Based on the fused and enhanced data, the system performs intelligent diagnosis on the user and outputs the diagnosis results.

[0178] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0179] Obtain the text describing the user's condition;

[0180] The text describing the illness is processed into a structured format to obtain structured description data.

[0181] Semantic retrieval is performed on the structured description data to obtain a candidate knowledge set;

[0182] Based on the structured description data, graph matching is performed on the preset knowledge graph to obtain graph matching data;

[0183] The candidate knowledge set and the graph matching data are fused to obtain fused enhanced data, wherein the fused enhanced data has both semantic information and structural logic information.

[0184] Based on the fused and enhanced data, the system performs intelligent diagnosis on the user and outputs the diagnosis results.

[0185] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0186] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other storage media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0187] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0188] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A knowledge graph-based intelligent consultation method, characterized in that, The knowledge graph-based intelligent consultation method includes: Obtain the text describing the user's condition; The text describing the illness is processed into a structured format to obtain structured description data. Semantic retrieval is performed on the structured description data to obtain a candidate knowledge set; Based on the structured description data, graph matching is performed on the preset knowledge graph to obtain graph matching data; The candidate knowledge set and the graph matching data are fused to obtain fused enhanced data, wherein the fused enhanced data has both semantic information and structural logic information. Based on the fused and enhanced data, the system performs intelligent diagnosis on the user and outputs the diagnosis results.

2. The intelligent consultation method based on knowledge graphs according to claim 1, characterized in that, The step of structuring the text describing the illness to obtain structured description data includes: The text description of the illness is cleaned to obtain a cleaned text description of the illness. The text cleaning process includes adjusting typos and punctuation marks and removing irrelevant characters. The cleaned description of the illness was segmented into basic words. The basic words are converted into word vectors and the word vectors are input into a preset medical-specific named entity recognition model; Based on the medical-specific named entity recognition model, the word vectors are first processed word by word to obtain word tags. Then, based on the preset medical knowledge base and the context information contained in the word tags, the selected word tags with medical significance are extracted as medical elements. The medical elements are standardized and coded according to preset categories to obtain structured description data.

3. The intelligent consultation method based on knowledge graphs according to claim 2, characterized in that, The medical elements include at least one of symptoms, signs, test results, and diseases. The atlas matching data is causal chain data matching symptoms and diseases. The intelligent consultation of the user based on the fused and enhanced data, and the output of consultation results, include: Based on the fused and enhanced data, a causal chain is determined by the combination of symptom elements, sign elements, test result elements, and disease elements. Determine the confidence level for each of the aforementioned causal chains; A fact confidence matrix is ​​established based on the confidence levels corresponding to each of the multiple causal chains. The system outputs a consultation result to the user based on the description of the illness text. The consultation result is a list of suspected diseases corresponding to multiple causal chains, and each suspected disease in the list is equipped with confidence information.

4. The intelligent consultation method based on knowledge graphs according to claim 1, characterized in that, The knowledge fusion of the candidate knowledge set and the graph matching data to obtain fused enhanced data includes: Obtain the weights corresponding to the candidate knowledge set and the graph matching data, respectively; The complementary information between the candidate knowledge set and the atlas matching data is determined based on multiple dimensions, wherein the candidate knowledge set contains specific clinical manifestations and the atlas matching data contains structured general medical knowledge. Determine the knowledge gap information of the candidate knowledge set and the graph matching data, wherein the knowledge gap information includes entity-level gap information, relation-level gap information and attribute-level gap information; Based on their respective weights, the candidate knowledge set and the graph matching data are fused together using complementary information and filled with knowledge gap information to obtain fused enhanced data.

5. The intelligent consultation method based on knowledge graphs according to claim 4, characterized in that, The respective weights include a first weight corresponding to the candidate knowledge set and a second weight corresponding to the graph matching data. Before obtaining the respective weights of the candidate knowledge set and the graph matching data, the process includes: Based on the candidate knowledge set, the text confidence score is determined, wherein the text confidence score is related to quality weight, semantic similarity, and content consistency. Based on the map matching data, the map confidence level is determined, wherein the map confidence level is related to path consistency and map coverage; Based on the text confidence and the graph confidence, the first weight corresponding to the candidate knowledge set is determined; The second weight is determined based on the first weight.

6. The intelligent consultation method based on knowledge graphs according to claim 1, characterized in that, After performing intelligent diagnosis on the user based on the fused enhanced data and outputting the diagnosis results, the process includes: The disease names and drug dosages involved in the consultation results are verified to obtain the verification confidence level; If the verification confidence level is lower than a preset threshold, the process returns to the step of performing semantic retrieval on the structured description data to obtain a candidate knowledge set, until the verification confidence level is not lower than the preset threshold.

7. A knowledge graph-based intelligent consultation system, characterized in that, The knowledge graph-based intelligent consultation system includes an acquisition module, a structured processing module, a semantic retrieval module, a graph matching module, a knowledge fusion module, and an intelligent consultation module. The acquisition module is used to acquire the text describing the user's condition. The structured processing module is used to perform structured processing on the disease description text to obtain structured description data; The semantic retrieval module is used to perform semantic retrieval on the structured description data to obtain a candidate knowledge set; The graph matching module is used to perform graph matching on the preset knowledge graph based on the structured description data to obtain graph matching data; The knowledge fusion module is used to perform knowledge fusion on the candidate knowledge set and the graph matching data to obtain fused enhanced data, wherein the fused enhanced data has both semantic information and structural logic information. The intelligent consultation module is used to conduct intelligent consultations on the user based on the fused enhanced data and output the consultation results.

8. The knowledge graph-based intelligent consultation system according to claim 7, characterized in that, The intelligent consultation module includes a consultation unit, a rehabilitation suggestion unit, and an assessment unit. The consultation unit is used to perform intelligent consultation on the user based on the fused enhanced data and output the consultation results; The rehabilitation suggestion unit is used to generate feasible rehabilitation suggestions based on the consultation results and a preset knowledge base. The evaluation unit is used to verify and evaluate the output of the consultation unit and the output of the rehabilitation suggestion unit, wherein the consultation results output by the consultation unit and the feasible rehabilitation suggestions output by the rehabilitation suggestion unit are cross-validated and consistency-validated.

9. A knowledge graph-based intelligent medical consultation device, characterized in that, The method includes a memory, a processor, and a knowledge graph-based intelligent consultation program stored in the memory and executable on the processor. The processor executes the knowledge graph-based intelligent consultation program to implement the steps of the knowledge graph-based intelligent consultation method according to any one of claims 1 to 6.

10. A storage medium, characterized in that, The storage medium stores a program for implementing a knowledge graph-based intelligent consultation method, which is executed by a processor to implement the steps of the knowledge graph-based intelligent consultation method as described in any one of claims 1 to 6.

Images