Multi-role adaptive interaction method and system based on stomatology knowledge base

Abstract

The embodiment of the application provides a multi-role adaptive interaction method and system based on an oral medicine knowledge base, which comprises the following steps: when a query signal is received, a query role is identified, and query content is acquired; a preset query engine matched with the query role is called, a previously constructed knowledge base is accessed according to the currently called query engine and the query content, and a differential diagnosis set containing multiple diseases with a sorting relationship is acquired; the diseases in the differential diagnosis set are subjected to multi-dimensional differential diagnosis reasoning one by one according to the sorting relationship, and the comprehensive confidence of the reasoning result is correspondingly evaluated; when the comprehensive confidence is not less than a preset threshold, the disease corresponding to the comprehensive confidence is taken as a diagnosis disease, and a query report matched with the query role is output. The application can assist doctors in making professional clinical decisions and provide patients with convenient oral common sense popularization services.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and medical information technology, and in particular to a multi-role adaptive interaction method and system based on an oral medicine knowledge base. Background Technology

[0002] As an important branch of medicine, stomatology possesses a complex, multi-dimensional, and interdisciplinary knowledge system. From basic theory to clinical practice, the knowledge carriers are diverse and dispersed, posing numerous challenges to the efficient delivery of oral healthcare services, as follows:

[0003] The knowledge system of oral medicine encompasses two major sections: basic science and clinical science. Basic science includes descriptions of the microstructure and macroscopic morphology of teeth, periodontal tissues, and maxillofacial bones and muscles in *Oral Anatomy*, and molecular-level analyses of the pathological mechanisms of oral diseases in *Oral Histopathology*. Clinical science includes treatment guidelines for various pulp diseases in *Endodontics*, and selection criteria for different prostheses in *Prosthodontics*. In addition, clinical guidelines published by authoritative international and domestic institutions, such as the American Dental Association (ADA) guidelines for caries prevention and treatment, and the Chinese Stomatological Association's guidelines for periodontal disease treatment, as well as popular science materials for the public, collectively constitute a complete ecosystem of oral medicine knowledge. However, this knowledge is currently fragmented, lacking a unified, structured integration framework—the link between basic theory and clinical application is broken, and treatment recommendations for the same disease in different guidelines lack standardized mapping, resulting in an "island effect" in the knowledge system.

[0004] In clinical settings, physicians often need to collaboratively search multiple sources for decision-making. For example, when faced with a complex case of tooth loss, a physician may need to simultaneously consult the operational guidelines for pulp vitality testing in "Endodontics," the mechanical property parameters of different restorative materials in "Prosthodontics," and the technical requirements for minimally invasive restorations in the latest clinical guidelines. This multi-source retrieval not only consumes a significant amount of clinical time but may also lead to delays or biases in decision-making due to differences in document versions or conflicting statements.

[0005] Patients also face significant challenges in accessing knowledge. Current public oral health information channels exhibit a polarization: professional medical literature is filled with technical terms like "enamel hypoplasia" and "periodontal probing depth," exceeding the comprehension of ordinary patients; while non-authoritative popular science content suffers from simplification and distortion, such as broadly describing "dentin hypersensitivity" as "tooth sensitivity," ignoring its pathological connection to wedge-shaped defects and gingival recession. This information gap leads to cognitive biases among patients regarding oral diseases, such as mistaking "recurrent pericoronitis of wisdom teeth" for "internal heat," delaying treatment; or developing resistance to treatment plans due to a lack of understanding of the principles of "anchorage control in orthodontic treatment."

[0006] Currently, existing oral medicine knowledge bases and related intelligent systems, such as the system described in reference [1], although they have achieved structured storage of knowledge, their retrieval logic is still a static "question-matching answer" mode, which cannot perform dynamic, multi-step reasoning retrieval based on the clinical diagnostic path. For example, when faced with the query "persistent dull pain in tooth 16", such systems can only return a limited number of entries that match "tooth 16" and "dull pain" (such as "pulpitis"), and cannot, like clinicians, further automatically associate and suggest the need to investigate differential diagnoses such as "jaw cyst", "occlusal trauma" or "atypical toothache", lacking in-depth support for the clinical decision-making process. Its essence is a "knowledge retrieval machine" rather than a "decision aid". Although there are studies on dialogue systems for medical scenarios, their design is usually role-agnostic. For example, the application of artificial intelligence in oral diagnosis and treatment mentioned in reference [2] is basically aimed at providing doctors with auxiliary decision-making by using artificial intelligence technology. This makes it impossible for doctors to provide accurate, rigorous, and professional decision support that includes coding and evidence levels, while providing patients with popular, easy-to-understand explanations that eliminate the barriers of professional terminology. The paper "Comparison of the Accuracy of Large Language Models in Question Answering in the Field of Pediatric Oral Preventive Medicine" [3] published by Guan Boyan et al. in 2025 also pointed out that the current general large model technology cannot meet the professional needs of doctors when performing rough retrieval, which further confirms the defects of existing technology in dynamic reasoning and role adaptability.

[0007] The specific references mentioned above are:

[0008] [1] Chen Yifan; Cao Zhanqiang. Construction and application of intelligent triage system for oral specialty [J]. Hospital Management Forum, 2022, 39(6):88-90;

[0009] [2] Tian Erkang; Xiang Qianrong; Zhao Xinran; Peng Jiahan; Shu Rui. Application of artificial intelligence in oral diagnosis and treatment [J]. International Journal of Stomatology, 2021, 48(4):475-484;

[0010] [3] Guan Boyan, Xu Minghe, Zhang Huiqi, et al. Comparison of the accuracy of large language models in question answering in the field of pediatric oral preventive medicine [J]. Prevention and Treatment of Oral Diseases, 2025, 33(4): 313-319. Summary of the Invention

[0011] This application provides a multi-role adaptive interaction method and system based on an oral medicine knowledge base to solve the above-mentioned technical problems.

[0012] The first aspect of this application provides a multi-role adaptive interaction method based on an oral medicine knowledge base, including:

[0013] Upon receiving a query signal, identify the querying role and retrieve the query content;

[0014] Invoke the preset query engine that matches the query role, and access the pre-built knowledge base based on the currently invoked query engine and query content to obtain a set of differential diagnoses containing multiple diseases with sorting relationships;

[0015] Based on the ranking relationship, the disease in the differential diagnosis set is subjected to multi-dimensional differential diagnosis reasoning one by one, and the comprehensive confidence of the reasoning results is evaluated accordingly. When the comprehensive confidence is not less than the preset threshold, the corresponding disease is regarded as the diagnosed disease, and a query report matching the query role is output.

[0016] In some preferred embodiments, the invocation of a preset query engine matched to the query role, based on the currently invoked query engine and the query content, accesses a pre-built knowledge base to obtain a differential diagnosis set containing multiple diseases with ranking relationships, wherein:

[0017] The pre-built knowledge base includes a corpus of translated professional and popular terms, as well as a knowledge graph with evidence level annotations.

[0018] In some preferred embodiments, the knowledge graph database construction process specifically includes:

[0019] Data related to oral medicine knowledge is processed to obtain clean data;

[0020] Knowledge extraction and evidence level labeling are performed on the clean dataset to generate structured knowledge triples in the form of "entity-relationship-attribute" with evidence level labels, specifically:

[0021] Named entity recognition technology is used to extract entities such as disease names, symptoms, and treatment methods from a clean dataset, and semantic associations between entities are determined through relation extraction technology to obtain structured knowledge triples.

[0022] According to the pre-established evidence level labeling rules, the structured knowledge triples are labeled with evidence levels to obtain structured knowledge triples in the form of "entity-relationship-attribute" with evidence level labels;

[0023] The structured knowledge triples are stored in the Neo4j graph database, with nodes representing entities, edges representing relationships, and attributes containing evidence level information, ultimately constructing a structured and queryable oral medicine knowledge graph library with evidence level annotations.

[0024] In some preferred embodiments, the invocation of a preset query engine matched with the query role accesses a pre-built knowledge base based on the currently invoked query engine and the query content to obtain a differential diagnosis set containing multiple diseases with sorting relationships. The preset query engine matched with the query role includes a doctor channel query engine matched with the doctor role.

[0025] Based on the doctor's query engine and the query content, access the pre-built knowledge base to obtain a differential diagnosis set containing multiple diseases with ranking relationships, specifically:

[0026] The query content is located to the corresponding position in the knowledge graph database through a preset encoding mapping matrix, and all symptoms related to the query content are indexed and extracted to obtain a symptom set;

[0027] A recursive segmentation algorithm based on information entropy is adopted to calculate the information entropy of each symptom in the symptom set, select the symptom with the largest information gain as the decision node, and recursively construct a decision tree to sort each symptom according to the disease probability, thereby obtaining a differential diagnosis set containing multiple diseases with sorting relationships.

[0028] In some preferred embodiments, the invocation of a preset query engine matched with the query role accesses a pre-built knowledge base based on the currently invoked query engine and the query content to obtain a differential diagnosis set containing multiple diseases with ranking relationships. The preset query engine matched with the query role includes a patient channel query engine matched with the patient role.

[0029] Based on the patient channel query engine and query content, access the pre-built knowledge base to obtain a differential diagnosis set containing multiple diseases with ranking relationships, specifically:

[0030] By using a pre-defined bidirectional LSTM semantic mapping model and corpus, the translation probability of the query content is obtained. The translated technical term sequence is then sorted according to the translation probability to obtain an initial sequence set.

[0031] The optimal translation sequence is obtained by calculating the initial sequence set using the beam search algorithm;

[0032] Based on the optimal translation sequence and the preset encoding mapping matrix, the corresponding position in the knowledge graph is located, and all symptoms related to the optimal translation sequence are indexed and extracted. According to the sequence relationship of the optimal translation sequence, the diseases corresponding to each symptom are sorted to obtain a differential diagnosis set containing the disease ranking.

[0033] In some preferred embodiments, the process involves performing multi-dimensional differential diagnostic reasoning on each disease in the differential diagnosis set according to the ranking relationship, and evaluating the comprehensive confidence level of the reasoning results. When the comprehensive confidence level is not less than a preset threshold, the corresponding disease is treated as a diagnosed disease, and a query report matching the query role is output, wherein:

[0034] Multidimensional differential diagnostic reasoning, specifically:

[0035] Calculate the match between the disease and the query content, and retain diseases with a match greater than a preset threshold;

[0036] For the retained diseases, cross-validation is performed by calling the associated knowledge of different levels of evidence in the knowledge graph. If the evidence supports each other, the current disease is used as the reasoning result.

[0037] In some preferred embodiments, the step of performing multi-dimensional differential diagnostic reasoning on each disease in the differential diagnosis set according to the ranking relationship and evaluating the comprehensive confidence of the reasoning results, and when the comprehensive confidence is not less than a preset threshold, then the corresponding disease is regarded as a diagnosed disease and a query report matching the query role is output, further includes:

[0038] When the overall confidence level is less than the preset threshold, real-time evidence supplementation and reasoning iteration loop are performed until the overall confidence level is not less than the preset threshold or no more evidence can be obtained. When no more evidence can be obtained, a prompt message is output to guide the user to supplement the evidence according to the prompt.

[0039] In some preferred embodiments, the step of performing multi-dimensional differential diagnosis reasoning on each disease in the differential diagnosis set according to the sorting relationship and evaluating the comprehensive confidence of the reasoning results accordingly, when the comprehensive confidence is not less than a preset threshold, the corresponding disease is regarded as a diagnosed disease and a query report matching the query role is output. The query report includes a multi-step reasoning path graph, which is used to record the input, reasoning logic, output and corresponding evidence level of each step of reasoning.

[0040] In some preferred embodiments, it further includes:

[0041] Monitor clinical guidelines in real time and supplement a pre-built knowledge base with real-time evidence to ensure the timeliness of knowledge.

[0042] As can be seen from the above, the beneficial effects of the embodiments of this application compared with the prior art include at least the following:

[0043] This application embodiment accesses a pre-built knowledge base through a query engine to obtain a differential diagnosis set containing multiple diseases with ranking relationships, and performs multi-dimensional differential diagnosis reasoning on each disease in the differential diagnosis set according to the ranking relationship, that is, simulates the differential diagnosis process of a clinician. Therefore, the accuracy of the query results can be guaranteed.

[0044] Meanwhile, this application embodiment sets up a query engine that matches the query role and performs multi-dimensional identification and diagnostic reasoning to output query results that match the query role. Obviously, this setting can automatically adjust the knowledge presentation method, language professionalism and reasoning depth according to the different query roles (such as doctors or patients), that is, it can achieve query output that is adaptive to the needs or professional level of the query role.

[0045] A second aspect of this application provides a multi-role adaptive interaction system based on an oral medicine knowledge base, comprising:

[0046] The query role recognition module is used to identify the query role and obtain the query content when a query signal is received.

[0047] The matching query module is used to call a preset query engine that matches the query role. Based on the currently called query engine and query content, it accesses a pre-built knowledge base to obtain a set of differential diagnoses containing multiple diseases with sorting relationships.

[0048] The query optimization module is used to perform multi-dimensional differential diagnosis reasoning on each disease in the differential diagnosis set according to the sorting relationship and evaluate the comprehensive confidence of the reasoning results. When the comprehensive confidence is not less than a preset threshold, the corresponding disease is regarded as the diagnosed disease and a query report matching the query role is output.

[0049] A third aspect of this application provides a terminal 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 implement the steps of the method described in the first aspect.

[0050] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described in the first aspect.

[0051] The fifth aspect of this application provides a computer program product that, when run on a terminal, causes the terminal to perform the steps of the method described in the first aspect.

[0052] It should be understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a flowchart illustrating a multi-role adaptive interaction method based on a dental knowledge base provided in one embodiment of this application.

[0055] Figure 2 This is a schematic diagram of the overall workflow of a multi-role adaptive interaction method based on an oral medicine knowledge base provided in another embodiment of this application;

[0056] Figure 3 This is a schematic diagram illustrating the core data source of the knowledge base provided in one embodiment of this application;

[0057] Figure 4 This is a schematic diagram of the structure of a multi-role adaptive interaction system based on an oral medicine knowledge base provided in an embodiment of this application;

[0058] Figure 5 This is a schematic diagram of the terminal structure provided in the embodiments of this application. Detailed Implementation

[0059] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

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

[0061] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this 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 be understood that the term "and / or" as used in this application specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0063] It should be understood that the sequence number of each step in this embodiment 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 this application embodiment.

[0064] To illustrate the technical solution described in this application, specific embodiments are provided below.

[0065] like Figure 1 As shown, a multi-role adaptive interaction method based on an oral medicine knowledge base is proposed, which includes the following steps:

[0066] Step 1: When a query signal is received, identify the querying role and obtain the query content;

[0067] Step 2: Invoke the preset query engine that matches the query role. Based on the currently invoked query engine and query content, access the pre-built knowledge base to obtain a set of differential diagnoses containing multiple diseases with sorting relationships.

[0068] Step 3: Based on the ranking relationship, perform multi-dimensional differential diagnosis reasoning for each disease in the differential diagnosis set and evaluate the comprehensive confidence of the reasoning results. When the comprehensive confidence is not less than the preset threshold, the corresponding disease is regarded as the diagnosed disease, and a query report matching the query role is output.

[0069] As can be seen, the embodiments of this application access a pre-built knowledge base through a query engine to obtain a differential diagnosis set containing multiple diseases with ranking relationships, and perform multi-dimensional differential diagnosis reasoning on each disease in the differential diagnosis set according to the ranking relationship, that is, simulate the differential diagnosis process of a clinician. Therefore, the accuracy of the query results can be guaranteed.

[0070] Meanwhile, this embodiment of the application sets up a query engine that matches the query role and performs multi-dimensional differential diagnostic reasoning to output query results that match the query role. Clearly, this setting can automatically adjust the knowledge presentation method, language professionalism, and reasoning depth according to different query roles (such as doctor or patient), that is, it can achieve query output that adapts to the needs or professional level of the query role. For example:

[0071] When the query role is a doctor, it adapts to their professional thinking and decision-making path, and receives professional query content by simulating the clinical reasoning process (such as generating differential diagnosis through decision trees) and providing evidence-based levels to assist them in making rigorous clinical decisions, rather than simply outputting search results.

[0072] When the query role is a patient: Adapting to their cognitive level and information needs, the received query content in colloquial terms is converted into professional terminology, while the output content transforms professional medical concepts and complex treatment plans into everyday language and easy-to-understand popular science knowledge, eliminating terminology barriers.

[0073] In some embodiments, step 1 involves identifying the query role and obtaining the query content when a query signal is received, wherein the query role includes doctors and patients;

[0074] For example, doctors or patients can be provided with unified identity authentication to identify doctors and patients and grant corresponding operating permissions and different processing mechanisms.

[0075] When using, such as Figure 2 As shown, doctors or patients can enter their queries through the search interface. Doctors' queries use professional terminology, while patients' queries use colloquial expressions. After entering their queries, patients can trigger a query signal by clicking the query button on the search interface. At this point, the current query role can be identified based on the identity authentication information, and the query content can be obtained.

[0076] In some embodiments, step 2 involves invoking a preset query engine that matches the query role, accessing a pre-built knowledge base based on the currently invoked query engine and the query content, and obtaining a set of differential diagnoses containing multiple diseases with ranking relationships.

[0077] In this embodiment, the pre-built knowledge base includes a corpus of translated professional and popular terms, as well as a knowledge graph library with evidence level annotations.

[0078] Example corpora of translated technical and colloquial terms are shown in the table below:

[0079] Glaze demineralization White spots on the surface of teeth Public Oral Health Guide Periodontal flap surgery Minimally invasive treatment of the gums Chinese Stomatological Association Popular Science Standards

[0080] For example, the knowledge graph library construction process may include the following steps:

[0081] Step 201: Collect data related to oral medicine knowledge and generate the original dataset of the oral medicine knowledge system, such as... Figure 3 As shown, Figure 3 In order to collect oral health-related textbooks, Figure 3It displays a non-exclusive list of representative authoritative works that serve as the core data source of the knowledge base, aiming to intuitively demonstrate the breadth, depth, and authority of the knowledge sources;

[0082] Specifically, one can collect various oral medicine professional books, authoritative medical journal articles, clinical guidelines, expert experience and other materials, including extracting key knowledge points from classic textbooks such as "Oral and Maxillofacial Surgery", "Prosthodontics" and "Endodontics".

[0083] In practice, a professional medical literature collection team can be formed to collect professional books on oral medicine, covering all branches of the discipline; and in cooperation with dental hospitals, real case data (such as 5,000 cases) and expert experience summaries from clinical practice can be obtained.

[0084] Step 202: Clean the original dataset of the oral medicine knowledge system to obtain a clean dataset;

[0085] For example, for the original dataset of oral medicine knowledge system, the SimHash algorithm is used to calculate the text hash value. The hash value is compared to determine whether the text is duplicated. Duplicate text is deleted. Oral medicine experts are invited to review and correct incorrect professional terminology and inaccurate medical knowledge, and finally a clean dataset is obtained.

[0086] Step 203 involves extracting knowledge and labeling evidence levels on the clean dataset to generate structured knowledge triples in the form of "entity-relationship-attribute" with evidence level labels. This may include the following steps:

[0087] Step 2031: Use named entity recognition technology to extract entities such as disease names, symptoms, and treatment methods from a clean dataset, and use relation extraction technology to determine the semantic relationships between entities to obtain structured knowledge triples.

[0088] Step 2032: According to the pre-established evidence level labeling rules, perform initial evidence level labeling on the structured knowledge triples, where:

[0089] The pre-established rules for classifying evidence levels can be:

[0090] Referring to the GRADE evidence-based medicine grading system and considering the characteristics of oral medicine, a three-level evidence grading standard was developed:

[0091] Grade A (High-quality evidence): From authoritative textbooks;

[0092] Level B (Medium-quality evidence): Knowledge from expert consensus or case-control studies;

[0093] Level C (low-quality evidence): Knowledge derived from case reports or the personal experience of experts;

[0094] Step 2033: Invite evidence-based medicine experts to review and make necessary corrections to the initial evidence level labeling to ensure the accuracy of the evidence level labeling, and finally obtain structured knowledge triples in the form of "entity-relationship-attribute" with evidence level labeling;

[0095] Step 2034: Store the structured knowledge triples in the Neo4j graph database, with nodes representing entities, edges representing relationships, and attributes containing evidence level information, ultimately constructing a structured, queryable oral medicine knowledge graph library with evidence level annotations.

[0096] In this embodiment, a preset query engine matching the query role is invoked. Based on the currently invoked query engine and the query content, a pre-built knowledge base is accessed to obtain a differential diagnosis set containing multiple diseases with sorting relationships. The preset query engine matching the query role includes a doctor channel query engine matching the doctor role and a patient channel query engine matching the patient role.

[0097] For example, accessing a pre-built knowledge base to obtain a differential diagnosis set containing multiple diseases with ranking relationships, based on the doctor channel query engine and query content, may include the following steps:

[0098] Step 211: The query content is located at the corresponding position in the knowledge graph database using a preset encoding mapping matrix, and all symptoms related to the query content are extracted by indexing to obtain a symptom set, for example:

[0099] The doctor enters either an ICD-10 disease code (e.g., "K03.0" represents dental caries) or an FDI tooth position identifier (e.g., "16" represents the upper right first molar).

[0100] The preset encoding mapping matrix can be represented by the following formula:

[0101]

[0102] In the above formula, The preset encoding mapping matrix is ​​C, which is the ICD-10 encoding set, T, which is the FDI tooth position set, and K, which is the knowledge base index set. Through this encoding mapping matrix, the corresponding disease-tooth position related knowledge unit in the knowledge graph library can be directly located.

[0103] Once the corresponding location in the knowledge graph is located, the knowledge graph is retrieved based on index K to extract all symptoms related to the target disease / tooth position (such as symptoms of caries including "white spots on teeth" and "cavities").

[0104] Step 212: Using a recursive segmentation algorithm based on information entropy, calculate the information entropy of each symptom in the symptom set, select the symptom with the largest information gain as the decision node, recursively construct a decision tree, and sort each symptom according to the disease probability to obtain a differential diagnosis set containing multiple diseases with sorting relationships, as a preliminary diagnosis result.

[0105] For example, let the input symptom set be... The information entropy of each symptom can be calculated using the following formula:

[0106]

[0107] In the above formula, Let sᵢ be the disease probability distribution corresponding to symptom sᵢ, and p be the probability of the disease corresponding to the symptom;

[0108] In addition, parameters can be optimized through 5-fold cross-validation (test set accuracy ≥ 85%), and the final output will include a preliminary diagnostic result that includes a disease probability ranking.

[0109] To provide a clearer understanding of the solutions provided in the embodiments of this application, the following exemplary descriptions of the construction process of the encoding mapping matrix and the training process of the clinical decision tree are provided:

[0110] The process of constructing the encoding mapping matrix may include the following:

[0111] Collect ICD-10 oral disease codes (a total of 128 sub-codes) and FDI tooth position representation (a 2-digit coding system), and establish a mapping table;

[0112] Data cleaning was performed using the Python pandas library to remove invalid codes (such as obsolete tooth position codes). Cosine similarity was used to calculate the correlation between codes, and mapping pairs with similarity ≥ 0.9 were retained to form an initial matrix.

[0113] Three experts in oral medical informatics were invited to review the matrix and correct coding mismatches (such as correcting the incorrect association between "K03.0" (dental caries) and "16" (right upper first molar) to the correct tooth position combination), ultimately generating 326 valid mapping relationships.

[0114] The clinical decision tree training process may include the following steps:

[0115] We selected 5,000 dental outpatient cases (covering 30 common diseases), of which 4,000 cases were used as training data and 1,000 cases were used as test data. We extracted symptom-disease label pairs (e.g., "night pain + pain from hot or cold stimuli → pulpitis").

[0116] An information entropy-based decision tree algorithm was implemented using the Scikit-learn library. The maximum depth was set to 10, and the minimum number of leaf node samples was 5. Parameters were optimized using 5-fold cross-validation. During training, the information gain of each symptom was calculated in each iteration (e.g., the information gain for "spontaneous pain" was 0.72, and for "occlusal pain" it was 0.68). Symptoms with high information gain were prioritized as split nodes. The final decision tree achieved an accuracy of 89.3% on the test set.

[0117] In this embodiment, accessing a pre-built knowledge base based on the patient channel query engine and query content to obtain a differential diagnosis set containing multiple diseases with ranking relationships may include the following steps:

[0118] Step 221: Using a pre-defined bidirectional LSTM semantic mapping model and corpus, obtain the translation probability of the query content, and sort the translated terminology sequence according to the translation probability to obtain an initial sequence set;

[0119] For example, let the sequence of technical terms be... If the sequence is simply expressed as N={n1,n2,...,nᵦ}, then the translation probability can be calculated as follows:

[0120]

[0121] Step 222: Calculate the initial sequence set using the beam search algorithm to obtain the optimal translation sequence, for example, to achieve accurate conversion between terms such as "dental caries" and "tooth decay", "enamel demineralization" and "white spots on teeth".

[0122] Step 223: Locate the corresponding position in the knowledge graph database based on the optimal translation sequence and the preset encoding mapping matrix, and extract all symptoms related to the optimal translation sequence. Sort the diseases corresponding to each symptom according to the sequence relationship of the optimal translation sequence to obtain a differential diagnosis set containing the disease ranking, which serves as preliminary popular science content.

[0123] It should be noted that, in this embodiment of the application, the presence of treatment risks in the translated sequence can also be detected through context awareness. If treatment risks exist, the translation of "contraindications → unsuitable conditions" is automatically triggered.

[0124] To provide a clearer understanding of the solutions provided in the embodiments of this application, the training process of the bidirectional LSTM semantic mapping model is described below as follows:

[0125] A corpus containing 30,000 professional-popular terminology pairs (derived from doctor-patient dialogue records and popular science literature) was constructed and divided into a training set and a validation set in an 8:2 ratio.

[0126] A bidirectional LSTM model was built using the TensorFlow framework, with the hidden layer dimension set to 256, the embedding dimension to 128, and the dropout rate to 0.3.

[0127] The Adam optimizer was used during training, with an initial learning rate of 0.001, which decayed by 10% every 5 epochs, for a total of 50 epochs.

[0128] The optimal translation sequence was solved by beam search algorithm (beam size=5), and the BLEU value reached 0.82 on the validation set, achieving accurate conversion of terms such as "subgingival scaling → deep cleaning".

[0129] Clearly, the embodiments of this application, by constructing an oral knowledge base and a dual-channel query engine, can achieve fast and accurate oral knowledge retrieval. This can not only assist doctors in clinical decision-making but also provide patients with convenient oral common sense popularization services. In other words, it realizes a dual-role service of precise clinical decision-making by doctors and popular science popularization for patients, solving the problems of knowledge fragmentation, inefficient retrieval, and poor role adaptability. It has the advantages of high efficiency, scalability, and evidence-based support.

[0130] In some embodiments, step 3 involves performing multi-dimensional differential diagnostic reasoning on each disease in the differential diagnosis set according to the ranking relationship, and evaluating the comprehensive confidence level of the reasoning results. When the comprehensive confidence level is not less than a preset threshold, the corresponding disease is taken as the diagnosed disease, and a query report matching the query role is output, wherein:

[0131] Multidimensional differential diagnostic reasoning may include the following steps:

[0132] Step 301: Calculate the matching degree between the disease and the query content, and retain diseases with a matching degree greater than a preset threshold. Specifically, this can be done as follows:

[0133] Based on key symptom weights and a preset formula, the match degree between the disease and the query content is calculated, and diseases with a match degree greater than a preset threshold are retained.

[0134] Key symptom weights can be pre-defined based on the degree to which evidence explicitly describes the strength of the association between the symptom and the disease, as well as the level of evidence. For example:

[0135] The initial weights for key symptoms are based on explicit descriptions of the "symptom-disease association strength" in Level A evidence (authoritative textbooks) and Level B evidence (clinical guidelines). If a clear definition can be found and the semantic similarity is above 90%, the weight is set to 0.8; for example...

[0136] "The typical symptoms of pulpitis are spontaneous pain at night and persistent pain from hot and cold stimuli." Therefore, the weight of spontaneous pain at night on the symptoms of pulpitis is set at 0.8 (typical symptom), and the weight of pain from hot and cold stimuli on the symptoms of pulpitis is set at 0.7 (auxiliary symptom).

[0137] Furthermore, the weights of key symptoms can be pre-stored in the form of a "symptom-disease weight dictionary", for example: {"spontaneous nocturnal pain": {"pulpitis": 0.8, "dental caries": 0.1}}, which is reviewed and developed by dentists;

[0138] The preset formula can be expressed by the following formula:

[0139]

[0140] In the above formula, As a candidate disease, is the symptom weight, Sim(·) is the symptom semantic similarity, and diseases with a score greater than 0.7 are retained first.

[0141] Step 302: For the retained diseases, cross-validate the related knowledge of different evidence levels in the knowledge graph base. If the evidence supports each other, the current disease is used as the reasoning result. If there is a conflict between the evidence, supplementary evidence retrieval is performed.

[0142] For example, regarding "pulpitis," first verify the Level A evidence ("Clinical Classification and Corresponding Symptoms of Pulpitis" in "Endodontics"), then match it with Level B evidence ("The Correlation Between Nocturnal Pain and Irreversible Pulpitis" in the 2024 "Guidelines for the Diagnosis and Treatment of Pulp Diseases"). If the two types of evidence consistently support "Nocturnal pain in tooth 16 → Irreversible pulpitis," the confidence of the inference is increased. If there is a conflict (such as a Level B piece of evidence mentioning that "Occlusal trauma may also cause nocturnal pain"), then automatically supplement the search with "Key Points for Differentiating Occlusal Trauma from Pulpitis" (such as "Occlusal trauma does not cause pain from hot or cold stimuli") to further exclude interfering diseases.

[0143] This embodiment may also include:

[0144] Step 303: Adapt and adjust the inference results to fit the clinical scenario, for example:

[0145] By combining real-time operational variables (such as patient age and past medical history, and prioritizing the exclusion of "senile pulpitis" for pediatric patients) and constraints on diagnostic and treatment resources (such as prioritizing the recommendation of "conventional root canal treatment options that can be carried out" in the inference process if the primary hospital does not have a root canal microscope), the inference direction is dynamically adjusted to ensure that the inference results are consistent with the actual clinical scenario.

[0146] In practical applications, context-aware interfaces can be set up to receive additional information from the user (doctor / patient) (such as patient age and available devices) and incorporate it as a filtering condition into the graph query and reasoning process.

[0147] In this embodiment, step 3 involves performing multi-dimensional differential diagnostic reasoning on each disease in the differential diagnosis set according to the ranking relationship, and evaluating the comprehensive confidence level of the reasoning results. When the comprehensive confidence level is not less than a preset threshold (e.g., 0.8), the corresponding disease is taken as the diagnosed disease, and a query report matching the query role is output. This step may also include:

[0148] When the overall confidence level is less than the preset threshold, real-time evidence supplementation and reasoning iteration loop will be performed until the overall confidence level is not less than the preset threshold or no more evidence can be obtained. When no more evidence can be obtained, a prompt message will be output to guide the user to supplement the evidence according to the prompt.

[0149] The real-time evidence is not stored in the current local knowledge base. To distinguish the contribution of the local knowledge base (pre-built knowledge base) from the real-time knowledge source, the fusion weight of the two types of knowledge is calculated in the following way:

[0150]

[0151] In the above formula, α represents the domain expert experience coefficient (0 < α < 1, generally taken as 0.7). , These are the time-sensitivity decay factors for local knowledge and real-time knowledge, respectively, satisfying λ. l +λᵣ=1, used to dynamically adjust the time relevance weight of knowledge;

[0152] Secondly, the semantic similarity between knowledge and query is calculated. The matching degree between the two types of knowledge sources and the user query (Q) is calculated using the cosine similarity function Sim(·), as follows:

[0153] Convert the local knowledge fragment (Kl), real-time knowledge fragment (Kr), and user query (Q) into vectors respectively;

[0154] Calculate the cosine angle between vectors, with a value ranging from 0 to 1. The closer the value is to 1, the stronger the semantic association between the knowledge and the query (e.g., Sim(Kl,Q)=0.9 indicates a high degree of match between local knowledge and the query).

[0155] Finally, based on the combination of fusion weights and semantic similarity, a quantified comprehensive confidence score is obtained, as shown in the following formula:

[0156]

[0157] Furthermore, for ease of understanding, the iterative loop logic provided in the embodiments of this application is explained below:

[0158] Identifying missing evidence: Analyzing weak links in the current reasoning chain, such as "the candidate disease is dental caries, but there is a lack of symptom information on 'cavity depth'", or "the evidence only contains content from A-level textbooks and is not supported by the latest guidelines";

[0159] Multi-source evidence retrieval: For missing items, dynamically search the local knowledge base (Kl) and real-time knowledge sources (Kr, such as clinical guideline databases) to obtain supplementary evidence;

[0160] The reasoning process is iteratively optimized by incorporating new evidence into the reasoning flow and re-executing multi-dimensional differential diagnostic reasoning. This cycle can be repeated multiple times until the confidence level is reached or no more evidence can be obtained (at which point the user is prompted to supplement information, such as "it is recommended to take an X-ray to determine the extent of the caries"), guiding the doctor to improve the input information and restarting the reasoning process.

[0161] In some embodiments, step 3 involves performing multi-dimensional differential diagnostic reasoning on each disease in the differential diagnosis set according to the ranking relationship, and evaluating the comprehensive confidence level of the reasoning results. When the comprehensive confidence level is not less than a preset threshold, the corresponding disease is taken as the diagnosed disease, and a query report matching the query role is output. The query report includes a multi-step reasoning path graph, which records the input (symptoms / evidence), reasoning logic (filtering / verification rules), output (candidate disease / exclusion result), and corresponding evidence level for each step of reasoning. For example:

[0162] Initial input: Nocturnal pain in tooth 16, pain from hot or cold stimuli → Candidate set: dental caries (Grade A), pulpitis (Grade A), occlusal trauma (Grade B);

[0163] Filtering: "Nighttime spontaneous pain" weight 0.8 → Exclude occlusal trauma (no such symptom) → Candidate set: dental caries, pulpitis;

[0164] Verification: Level A evidence "Pulpitis is always accompanied by nighttime pain, while dental caries usually does not cause nighttime pain" → exclude dental caries → candidate set: pulpitis;

[0165] Supplementary evidence: A search of the 2025 guidelines confirms "nocturnal pain + pain from hot or cold stimuli → irreversible pulpitis" → confidence level 0.92;

[0166] The final output is "Irreversible pulpitis in tooth 16 (Level of evidence: Level A textbook + 2025 Level B guidelines, confidence level 0.92)", along with a recommended treatment plan (such as "root canal treatment").

[0167] Clearly, the above settings correspond to the clinical "diagnosis record stage," enabling the solidification of a clear and unambiguous reasoning path and the tracing of evidence throughout the entire process.

[0168] In some embodiments, the multi-role adaptive interaction method based on a dental knowledge base may further include:

[0169] Step 4: Monitor clinical guidelines in real time and add real-time evidence to the knowledge base to ensure the timeliness of knowledge.

[0170] Based on the same inventive concept, embodiments of this application also provide a multi-role adaptive interaction system based on an oral medicine knowledge base, such as... Figure 4 As shown, the system includes:

[0171] The query role recognition module 1 is used to identify the query role and obtain the query content when a query signal is received;

[0172] Matching query module 2 is used to call a preset query engine that matches the query role. Based on the currently called query engine and query content, it accesses a pre-built knowledge base to obtain a set of differential diagnoses containing multiple diseases with sorting relationships.

[0173] The query optimization module 3 is used to perform multi-dimensional differential diagnosis reasoning on each disease in the differential diagnosis set according to the sorting relationship and evaluate the comprehensive confidence of the reasoning results. When the comprehensive confidence is not less than a preset threshold, the corresponding disease is regarded as the diagnosed disease and a query report matching the query role is output.

[0174] As can be seen, the embodiments of this application access a pre-built knowledge base through a query engine to obtain a differential diagnosis set containing multiple diseases with ranking relationships, and perform multi-dimensional differential diagnosis reasoning on each disease in the differential diagnosis set according to the ranking relationship, that is, simulate the differential diagnosis process of a clinician. Therefore, the accuracy of the query results can be guaranteed.

[0175] Meanwhile, this application embodiment sets up a query engine that matches the query role and performs multi-dimensional identification and diagnostic reasoning to output query results that match the query role. Obviously, this setting can automatically adjust the knowledge presentation method, language professionalism and reasoning depth according to the different query roles (such as doctors or patients), that is, it can achieve query output that is adaptive to the needs or professional level of the query role.

[0176] The multi-role adaptive interaction system based on a dental medicine knowledge base provided in this application adopts a layered architecture design concept, clearly defining the hierarchical relationship and data interaction interface of the core modules. The bottom layer is the data layer, responsible for data interaction with the knowledge base (Neo4j graph database), defining standardized APIs for data query and update. The middle layer is the business logic layer, containing core modules such as a dual-channel query engine and dynamic reasoning mechanism. Module decoupling is achieved through a microservice architecture, and cross-module communication is carried out using the gRPC protocol. The upper layer is the application layer, responsible for the development of user interaction interfaces for doctors and patients, supporting adaptation to web and mobile terminals. The architecture design uses UML modeling tools to draw module dependency diagrams and data flow diagrams to ensure that the responsibilities of each module are clear and the interaction logic is rigorous.

[0177] The integration and testing of the multi-role adaptive interaction system based on a dental medicine knowledge base provided in this application embodiment are briefly described below:

[0178] Module integration: The system uses containerization technology to deploy each core module (including knowledge base, inference engine, front-end and back-end applications), and uses service orchestration tools to achieve high concurrency support and inter-module collaboration. The system provides a unified identity authentication mechanism to distinguish between doctor and patient roles and grant corresponding operation permissions.

[0179] System testing: The system performance is verified through clinical trials. On the doctor's side, the focus is on evaluating query response time and the accuracy of diagnostic suggestions, while on the patient's side, the focus is on evaluating terminology comprehension and ease of operation. Based on the test feedback, the system is iteratively optimized to ensure its reliability and user experience in practical applications.

[0180] According to test experiments, the multi-role adaptive interaction system based on the oral medicine knowledge base provided in this application has at least the following beneficial effects:

[0181] Improve knowledge acquisition efficiency: Doctors and patients can quickly obtain the information they need from a large amount of oral health knowledge, saving time and improving work and medical efficiency;

[0182] Supporting clinical decision-making: Providing doctors with accurate knowledge support and diagnostic suggestions helps improve diagnostic accuracy and treatment outcomes;

[0183] Popularizing oral health knowledge: Facilitating patients' access to authoritative oral health information, enhancing their awareness of oral health, and promoting the prevention of oral diseases.

[0184] Scalability: The knowledge base and intelligent agent have good scalability, and can continuously update and improve the knowledge content to adapt to the development and changes in the field of oral medicine.

[0185] This application also provides a terminal, such as... Figure 5As shown, the terminal of this embodiment includes: at least one processor 50 ( Figure 5 (Only one is shown in the diagram), memory 51, and computer program 52 stored in said memory 51 and executable on said at least one processor 50, wherein said processor 50 executes said computer program 52 to implement the steps in any of the above method embodiments.

[0186] The terminal 5 can be a computing device such as a desktop computer, laptop, handheld computer, or cloud server. The terminal 5 may include, but is not limited to, a processor 50 and a memory 51. Those skilled in the art will understand that... Figure 5 This is merely an example of terminal 5 and does not constitute a limitation on terminal 5. It may include more or fewer components than shown, or combine certain components, or different components. For example, the terminal may also include input / output devices, network access devices, buses, etc.

[0187] The processor 50 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0188] The memory 51 can be an internal storage unit of the terminal 5, such as a hard disk or memory of the terminal 5. The memory 51 can also be an external storage device of the terminal 5, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the terminal 5. Furthermore, the memory 51 can include both internal storage units and external storage devices of the terminal 5. The memory 51 is used to store the computer program and other programs and data required by the terminal. The memory 51 can also be used to temporarily store data that has been output or will be output.

[0189] In specific implementations, the terminals described in the embodiments of this application include, but are not limited to, other portable devices such as mobile phones, laptop computers, or tablet computers with touch-sensitive surfaces (e.g., touchscreen displays and / or touchpads). It should also be understood that in some embodiments, the device is not a portable communication device, but a desktop computer with touch-sensitive surfaces (e.g., touchscreen displays and / or touchpads).

[0190] The terminal supports a variety of applications, such as one or more of the following: drawing applications, presentation applications, word processing applications, website creation applications, disc burning applications, spreadsheet applications, game applications, telephone applications, video conferencing applications, email applications, instant messaging applications, exercise support applications, photo management applications, digital camera applications, digital camcorder applications, web browsing applications, digital music player applications, and / or digital video player applications.

[0191] Various applications that can run on a terminal can use at least one common physical user interface device, such as a touch-sensitive surface. One or more functions of the touch-sensitive surface and the corresponding information displayed on the terminal can be adjusted and / or changed between and / or within applications. In this way, the terminal's common physical architecture (e.g., the touch-sensitive surface) can support various applications with user interfaces that are intuitive and transparent to the user.

[0192] 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 merely 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. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0193] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0194] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0195] In the embodiments provided in this application, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling or direct coupling or communication connection may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0196] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0197] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0198] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0199] The methods described in this application can be implemented in whole or in part by a computer program product. When the computer program product is run on a terminal, the terminal executes the steps in the various method embodiments described above.

[0200] 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 multi-role adaptive interaction method based on a dental medicine knowledge base, characterized in that, include: When a query signal is received, the query role is identified based on unified identity authentication, and the query content is obtained. The query roles include doctors and patients. The system invokes a preset query engine that matches the query role. Based on the currently invoked query engine and the query content, it accesses a pre-built knowledge base to obtain a differential diagnosis set containing multiple diseases with ranking relationships. The pre-built knowledge base includes a corpus of translated professional and popular terms, as well as a knowledge graph library with evidence level annotations. Among them, the preset query engines that match the query role include the doctor channel query engine that matches the doctor role and the patient channel query engine that matches the patient role. Based on the doctor's query engine and the query content, access the pre-built knowledge base to obtain a differential diagnosis set containing multiple diseases with ranking relationships, specifically: The query content is located to the corresponding position in the knowledge graph database through a preset encoding mapping matrix, and all symptoms related to the query content are indexed and extracted to obtain a symptom set; A recursive segmentation algorithm based on information entropy is adopted to calculate the information entropy of each symptom in the symptom set, select the symptom with the largest information gain as the decision node, and recursively construct a decision tree to sort each symptom according to the disease probability, thereby obtaining a differential diagnosis set containing multiple diseases with sorting relationships. Based on the patient channel query engine and query content, access the pre-built knowledge base to obtain a differential diagnosis set containing multiple diseases with ranking relationships, specifically: By using a pre-defined bidirectional LSTM semantic mapping model and a corpus of professional and colloquial terminology translations, the translation probability of the query content is obtained. Based on the translation probability, the translated professional terminology sequence is sorted to obtain an initial sequence set. The optimal translation sequence is obtained by calculating the initial sequence set using the beam search algorithm; Based on the optimal translation sequence and the preset encoding mapping matrix, the corresponding position in the knowledge graph is located, and all symptoms related to the optimal translation sequence are indexed and extracted. According to the sequence relationship of the optimal translation sequence, the diseases corresponding to each symptom are sorted to obtain a differential diagnosis set containing the disease ranking. Based on the ranking relationship, the disease in the differential diagnosis set is subjected to multi-dimensional differential diagnosis reasoning one by one, and the comprehensive confidence of the reasoning results is evaluated accordingly. When the comprehensive confidence is not less than the preset threshold, the corresponding disease is regarded as the diagnosed disease, and a query report matching the query role is output.

2. The method according to claim 1, characterized in that, The knowledge graph database construction process is as follows: Data related to oral medicine knowledge is processed to obtain clean data; Knowledge extraction and evidence level labeling are performed on the clean dataset to generate structured knowledge triples in the form of "entity-relationship-attribute" with evidence level labels, specifically: Named entity recognition technology is used to extract entities such as disease names, symptoms, and treatment methods from a clean dataset, and semantic associations between entities are determined through relation extraction technology to obtain structured knowledge triples. According to the pre-established evidence level labeling rules, the structured knowledge triples are labeled with evidence levels to obtain structured knowledge triples in the form of "entity-relationship-attribute" with evidence level labels; The structured knowledge triples are stored in the Neo4j graph database, with nodes representing entities, edges representing relationships, and attributes containing evidence level information, ultimately constructing a structured and queryable oral medicine knowledge graph library with evidence level annotations.

3. The method according to claim 1, characterized in that, The process involves performing multi-dimensional differential diagnostic reasoning on each disease in the differential diagnosis set according to the ranking relationship, and evaluating the comprehensive confidence level of the reasoning results. When the comprehensive confidence level is not less than a preset threshold, the corresponding disease is taken as the diagnosed disease, and a query report matching the query role is output, wherein: Multidimensional differential diagnostic reasoning, specifically: Calculate the match between the disease and the query content, and retain diseases with a match greater than a preset threshold; For the retained diseases, cross-validation is performed by calling the associated knowledge of different levels of evidence in the knowledge graph. If the evidence supports each other, the current disease is used as the reasoning result.

4. The method according to claim 1, characterized in that, The process of performing multi-dimensional differential diagnostic reasoning on each disease in the differential diagnosis set according to the ranking relationship and evaluating the comprehensive confidence of the reasoning results, wherein when the comprehensive confidence is not less than a preset threshold, the corresponding disease is regarded as a diagnosed disease and a query report matching the query role is output, further includes: When the overall confidence level is less than the preset threshold, real-time evidence supplementation and reasoning iteration loop are performed until the overall confidence level is not less than the preset threshold or no more evidence can be obtained. When no more evidence can be obtained, a prompt message is output to guide the user to supplement the evidence according to the prompt.

5. The method according to claim 1, characterized in that, The process involves performing multi-dimensional differential diagnosis reasoning on each disease in the differential diagnosis set according to the ranking relationship, and evaluating the comprehensive confidence level of the reasoning results. When the comprehensive confidence level is not less than a preset threshold, the corresponding disease is taken as the diagnosed disease, and a query report matching the query role is output. The query report includes a multi-step reasoning path graph, which records the input, reasoning logic, output, and corresponding evidence level of each step of reasoning.

6. The method according to claim 1, characterized in that, Also includes: Monitor clinical guidelines in real time and supplement a pre-built knowledge base with real-time evidence to ensure the timeliness of knowledge.

7. A multi-role adaptive interactive system based on a dental medicine knowledge base, characterized in that, include: The query role recognition module is used to identify the query role based on unified identity authentication when a query signal is received, and to obtain the query content. The query roles include doctors and patients. The matching query module is used to call a preset query engine that matches the query role. Based on the currently called query engine and query content, it accesses a pre-built knowledge base to obtain a differential diagnosis set containing multiple diseases with ranking relationships. The pre-built knowledge base includes a corpus of translated professional and popular terms, as well as a knowledge graph library with evidence level annotations. Among them, the preset query engines that match the query role include the doctor channel query engine that matches the doctor role and the patient channel query engine that matches the patient role. Based on the doctor's query engine and the query content, access the pre-built knowledge base to obtain a differential diagnosis set containing multiple diseases with ranking relationships, specifically: The query content is located to the corresponding position in the knowledge graph database through a preset encoding mapping matrix, and all symptoms related to the query content are indexed and extracted to obtain a symptom set; A recursive segmentation algorithm based on information entropy is adopted to calculate the information entropy of each symptom in the symptom set, select the symptom with the largest information gain as the decision node, and recursively construct a decision tree to sort each symptom according to the disease probability, thereby obtaining a differential diagnosis set containing multiple diseases with sorting relationships. Based on the patient channel query engine and query content, access the pre-built knowledge base to obtain a differential diagnosis set containing multiple diseases with ranking relationships, specifically: By using a pre-defined bidirectional LSTM semantic mapping model and a corpus of professional and colloquial terminology translations, the translation probability of the query content is obtained. Based on the translation probability, the translated professional terminology sequence is sorted to obtain an initial sequence set. The optimal translation sequence is obtained by calculating the initial sequence set using the beam search algorithm; Based on the optimal translation sequence and the preset encoding mapping matrix, the corresponding position in the knowledge graph is located, and all symptoms related to the optimal translation sequence are indexed and extracted. According to the sequence relationship of the optimal translation sequence, the diseases corresponding to each symptom are sorted to obtain a differential diagnosis set containing the disease ranking. The query optimization module is used to perform multi-dimensional differential diagnosis reasoning on each disease in the differential diagnosis set according to the sorting relationship and evaluate the comprehensive confidence of the reasoning results. When the comprehensive confidence is not less than a preset threshold, the corresponding disease is regarded as the diagnosed disease and a query report matching the query role is output.

Images