Artificial intelligence decision method and system for medical health service platform

Abstract

This invention provides an artificial intelligence decision-making method and system for a healthcare service platform. First, the medical knowledge graph module of the AI ​​decision-making system is activated and a preset set of domain parameter configurations is loaded. Next, the system receives real-time health consultation content and associated historical treatment sequence text pushed by the user from the healthcare service platform. Then, the received content undergoes medical semantic parsing and intent recognition processing to generate a user health problem representation vector and clinical decision-making need classification labels. Based on the medical knowledge graph module and the domain parameter configuration set, multi-round associative reasoning is performed to generate health risk assessment results and clinical intervention strategy suggestions. Finally, based on the health risk assessment results, a platform service response instruction is generated and sent to the business processing module to trigger the health service response process, thereby improving the service quality and efficiency of the healthcare service platform.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically, to an artificial intelligence decision-making method and system for a medical and health service platform. Background Technology

[0002] In the field of healthcare services, with the rapid development of information technology, healthcare service platforms have gradually become important channels for people to obtain health consultations and medical services. However, existing healthcare service platforms face many challenges in handling user health inquiries. On the one hand, traditional platforms mainly rely on human customer service or simple keyword matching to respond to user inquiries. These methods are not only inefficient but also difficult to fully and accurately understand users' complex health needs and symptom descriptions. For example, users may have multiple symptoms simultaneously, and there may be complex relationships between these symptoms, which human customer service or simple matching cannot accurately grasp. On the other hand, when formulating clinical decision-making recommendations, there is a lack of comprehensive analysis and utilization of users' historical medical information. Users' past diagnostic records and treatment intervention records are crucial for the assessment and decision-making of current health problems, but existing platforms often view each consultation in isolation, failing to effectively combine historical information with the current problem, resulting in decision-making recommendations that lack scientific rigor and specificity. In addition, existing platforms lack systematic medical knowledge support when processing health information, making it difficult to conduct in-depth reasoning and analysis, and thus unable to provide users with accurate and reliable health risk assessments and clinical intervention strategy recommendations. Summary of the Invention

[0003] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide an artificial intelligence decision-making method for a healthcare service platform, the method comprising:

[0004] The medical knowledge graph module of the artificial intelligence decision-making system is activated and a preset set of domain parameter configurations is loaded. The set of domain parameter configurations includes medical entity association weights and clinical decision threshold parameters.

[0005] The system receives real-time health consultation content and associated historical treatment sequence text pushed by the user from the medical and health service platform. The health consultation content includes symptom description text and health request text, and the historical treatment sequence text includes past diagnosis records and treatment intervention records arranged in chronological order.

[0006] The received health consultation content and the historical medical treatment sequence text are subjected to medical semantic parsing and intent recognition processing to generate user health problem representation vectors and clinical decision-making need classification tags.

[0007] Based on the entity relationships in the medical knowledge graph module and the domain parameter configuration set, the user health problem representation vector and the clinical decision-making need classification label are subjected to multi-round association reasoning processing to generate health risk assessment results and clinical intervention strategy suggestions.

[0008] Based on the health risk assessment results and the clinical intervention strategy recommendations, a platform service response instruction is generated and sent to the business processing module of the medical and health service platform to trigger the corresponding health service response process.

[0009] In another aspect, embodiments of the present invention also provide an artificial intelligence decision-making system for a medical and health service platform, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0010] Based on the above, this embodiment of the invention, by activating the medical knowledge graph module of the artificial intelligence decision-making system and loading a preset set of domain parameter configurations, receives real-time input of health consultation content and associated historical treatment sequence text from the user. It comprehensively integrates the user's current and historical health information, performs medical semantic parsing and intent recognition processing on the health consultation content and historical treatment sequence text, and can accurately generate user health problem representation vectors and clinical decision-making need classification labels. It deeply understands the user's health status and needs, and performs multi-round association reasoning based on entity relationships and the set of domain parameter configurations in the medical knowledge graph module. It fully utilizes the inherent connections of medical knowledge to generate accurate and reliable health risk assessment results and clinical intervention strategy suggestions. Finally, based on these results, it generates platform service response instructions and triggers corresponding health service response processes, significantly improving the service quality and efficiency of the medical and health service platform and providing users with more personalized and precise health services. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the execution flow of an artificial intelligence decision-making method for a medical and health service platform provided in an embodiment of the present invention.

[0012] Figure 2 This is a schematic diagram of exemplary hardware and software components of an artificial intelligence decision-making system for a healthcare service platform provided in an embodiment of the present invention. Detailed Implementation

[0013] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1This is a flowchart illustrating an artificial intelligence decision-making method for a healthcare service platform according to an embodiment of the present invention. The following is a detailed description of the artificial intelligence decision-making method for a healthcare service platform.

[0014] Step S110: Activate the medical knowledge graph module of the artificial intelligence decision-making system and load the preset domain parameter configuration set, which includes medical entity association weights and clinical decision threshold parameters.

[0015] This embodiment uses a scenario where a user consults a healthcare service platform about a persistent cough as an example to describe the implementation process of the entire method in detail. First, the medical knowledge graph module of the artificial intelligence decision-making system is activated, which stores a massive amount of verified medical knowledge. The activation operation is completed by triggering a preset program instruction. After the instruction is triggered, the basic service processes of the medical knowledge graph module begin to run, including data index loading and initialization of the relational retrieval interface.

[0016] Upon startup of the medical knowledge graph module, a pre-defined set of domain parameter configurations is loaded. These parameters, based on long-term clinical practice, are used to standardize subsequent reasoning and decision-making logic. The medical entity association weights in the domain parameter configuration set measure the strength of the connection between different medical entities, such as the strength of the association between a certain cough symptom and a specific respiratory disease. Clinical decision threshold parameters define the initiation boundaries of different clinical interventions, such as under what circumstances it is necessary to recommend further imaging examinations to the user.

[0017] Step S111: Activate the medical knowledge graph module of the artificial intelligence decision-making system. The medical knowledge graph module includes a set of medical entities, a set of entity attributes, and a set of relationships between entities. The set of medical entities includes disease entities, symptom entities, drug entities, and examination item entities. The set of entity attributes includes the entity's name attribute, classification attribute, and feature description attribute. The set of relationships between entities includes causal relationships, membership relationships, and association relationships.

[0018] Once the medical knowledge graph module is activated, its internal sets of medical entities, entity attributes, and relationships are made available for use. The medical entity set includes disease entities covering various respiratory diseases such as pneumonia, bronchitis, and tuberculosis; symptom entities including cough, sputum production, and fever; drug entities including antibiotics and cough suppressants used to treat respiratory diseases; and examination items such as chest X-rays and complete blood counts.

[0019] In the entity attribute set, the name attribute of each entity is a standard name in the medical field, such as "bacterial pneumonia"; the classification attribute is used to classify entities, such as classifying "bacterial pneumonia" as "infectious pneumonia"; the feature description attribute details the characteristics of the entity, such as the feature description attribute of "bacterial pneumonia" including rapid onset and often accompanied by high fever.

[0020] The set of relationships between entities includes causal relationships such as "bacterial infection leads to bacterial pneumonia"; hierarchical relationships such as "pneumococcal pneumonia belongs to bacterial pneumonia"; and associative relationships such as "cough is associated with pneumonia". These entities and relationships together constitute a complete medical knowledge network.

[0021] Step S112: Retrieve a preset domain parameter configuration set from the parameter storage unit of the artificial intelligence decision-making system. The domain parameter configuration set includes medical entity association weights and clinical decision threshold parameters. The medical entity association weights are used to represent the strength of the association between different medical entities, and the clinical decision threshold parameters are used to determine the initiation conditions of clinical intervention measures.

[0022] When retrieving a preset set of domain parameter configurations from the parameter storage unit of the artificial intelligence decision-making system, it is done through a parameter retrieval interface. The parameter storage unit adopts a distributed storage architecture, storing the domain parameter configuration sets in partitions according to different medical specialties. In this instance, the retrieved set is the domain parameter configuration set related to respiratory diseases.

[0023] The association weights of medical entities exist in the form of a multi-dimensional table. The rows and columns of the table correspond to different medical entities, and the contents of the table represent the strength of the association between the corresponding two entities. For example, the association weight between the entity of cough symptoms and the entity of bacterial pneumonia disease is determined based on the frequency and specificity of the symptom in clinical cases.

[0024] Clinical decision threshold parameters include risk assessment thresholds and intervention initiation thresholds. Risk assessment thresholds are used to classify different levels of health risk, while intervention initiation thresholds specify under what circumstances a specific intervention should be initiated, such as recommending further investigation when the duration and severity of a certain respiratory symptom reach a certain threshold.

[0025] Step S113: Perform parameter integrity verification on the retrieved domain parameter configuration set to check whether there are missing or outlier values ​​in the medical entity association weight and clinical decision threshold parameters.

[0026] The retrieved domain parameter configuration set undergoes parameter integrity verification using preset verification rules. The verification rules first check for any missing cells in the medical entity association weights, specifically whether any association weights between two medical entities are missing.

[0027] Simultaneously, check for any outliers that defy common medical sense, such as extremely high correlation weights between two clearly unrelated entities. For clinical decision threshold parameters, check for missing thresholds, such as whether the initiation threshold for a certain intervention is empty, and for logical contradictions between thresholds, such as a lower intervention initiation threshold for a higher risk level than for a lower risk level.

[0028] If missing or abnormal values ​​are found, the parameter repair mechanism can be triggered to retrieve the corresponding correct parameters from the backup parameter storage unit for supplementation or replacement.

[0029] Step S114: Perform format conversion processing on the domain parameter configuration set that has passed the integrity check, converting the parameter data into a format that the medical knowledge graph module can recognize.

[0030] The domain parameter configuration set that has passed the integrity check is converted into a format because the parameter storage unit stores parameters in a general structured data format, while the medical knowledge graph module requires a specific graph-compatible format.

[0031] During the conversion process, the entity names in the medical entity association weights are mapped to the corresponding unique identifiers in the medical knowledge graph module, ensuring accurate entity mapping. Simultaneously, the numerical form of the parameters is converted into a numerical encoding format recognizable by the medical knowledge graph module; for example, the textual descriptions of association weights are converted into corresponding numerical range codes.

[0032] The format conversion of clinical decision threshold parameters is similar, converting the threshold description information into structured data that can be parsed by the medical knowledge graph module, so that these parameters can be accurately invoked in the subsequent reasoning process.

[0033] Step S115: Load the converted domain parameter configuration set into the parameter configuration unit of the medical knowledge graph module to complete the initial configuration of the medical knowledge graph module.

[0034] The transformed set of domain parameter configurations is loaded into the parameter configuration unit of the medical knowledge graph module via the parameter loading interface. The parameter configuration unit is a component in the medical knowledge graph module specifically used for storing and managing parameters. During the loading process, parameters are categorized and stored according to their type and purpose. Medical entity association weights are stored in the association weight sub-unit, and clinical decision threshold parameters are stored in the decision threshold sub-unit.

[0035] Once loading is complete, the parameter configuration unit will generate a loading completion signal to notify the medical knowledge graph module that the initialization configuration is complete. At this point, the medical knowledge graph module is ready to make inference decisions based on these parameters.

[0036] Step S116: Activate the self-verification mechanism of the medical knowledge graph module to perform consistency verification between the loaded domain parameter configuration set and the entity relationships in the medical knowledge graph, so that the parameter configuration matches the knowledge graph structure.

[0037] The self-verification mechanism of the medical knowledge graph module is activated. The self-verification mechanism compares the entity relationships in the medical knowledge graph with the medical entity association weights in the loaded domain parameter configuration set.

[0038] For example, if there is a relationship between "cough and pneumonia" in the medical knowledge graph, but the association weight between the cough symptom entity and the pneumonia disease entity in the domain parameter configuration set is zero or extremely low, the self-verification mechanism will identify this inconsistency.

[0039] For clinical decision threshold parameters, the self-verification mechanism checks whether they match the characteristics of entities in the medical knowledge graph, such as whether the severity characteristics of a disease entity and the corresponding intervention initiation threshold are reasonable. If inconsistencies exist, the self-verification mechanism generates a verification report and adjusts the domain parameter configuration set according to preset correction rules until the parameter configuration matches the knowledge graph structure.

[0040] Step S120: Receive health consultation content and associated historical treatment sequence text pushed by the user in real time from the medical and health service platform. The health consultation content includes symptom description text and health request text, and the historical treatment sequence text includes previous diagnosis records and treatment intervention records arranged in chronological order.

[0041] After the medical knowledge graph module completes its initial configuration, it receives information pushed by the healthcare service platform. This information includes real-time health consultation content entered by the user and associated historical medical records.

[0042] In the health consultation content entered by users in real time, the symptom description text may be "I have had a cough every day for the past three weeks, more frequently in the morning and evening, and occasionally accompanied by white phlegm"; the health request text may be "I want to know what is causing this cough, what medicine I need to take, and whether I need to go to the hospital for examination".

[0043] The associated historical medical records are arranged chronologically, for example, "Six months ago, I developed a cough due to a cold and was diagnosed with acute bronchitis. My symptoms improved after taking cough medicine and antibiotics; three years ago, I had a physical examination, and a chest X-ray showed no abnormalities." This information is pushed to the artificial intelligence decision-making system through an encrypted transmission channel to ensure the security of information transmission.

[0044] Step S130: Perform medical semantic parsing and intent recognition processing on the received health consultation content and the historical diagnosis and treatment sequence text to generate a user health problem representation vector and a clinical decision-making requirement classification label.

[0045] After receiving the user's health consultation content and historical diagnosis and treatment sequence text, perform medical semantic parsing and intent recognition processing on them. This process aims to convert unstructured text information into structured feature data for subsequent reasoning and analysis.

[0046] Medical semantic parsing includes identifying, standardizing, and extracting features from medical terms in the text. Intent recognition is to determine the category of the user's health诉求. Through this series of processes, a vector that can comprehensively represent the user's health problem and a clear clinical decision-making requirement classification label are finally generated.

[0047] Step S131: Perform text preprocessing on the symptom description text and health诉求 text in the health consultation content, perform word segmentation on the preprocessed symptom description text, and split the continuous text into an independent sequence of words.

[0048] Perform text preprocessing on the symptom description text and health诉求 text in the health consultation content. The preprocessing includes removing special symbols in the text, such as punctuation marks, emojis, etc.; converting uppercase letters in the text to lowercase letters; removing meaningless function words, such as "de", "le", etc.

[0049] For example, after preprocessing the symptom description text "In the past three weeks, there has been a cough symptom every day, which is more frequent in the morning and evening, and occasionally accompanied by white sputum", we get "In the past three weeks, there has been a cough symptom every day, which is more frequent in the morning and evening, and occasionally accompanied by white sputum".

[0050] Subsequently, perform word segmentation on the preprocessed symptom description text, using a word segmentation algorithm based on a medical dictionary to split the continuous text into an independent sequence of words. After word segmentation of the above preprocessed text, we get the sequence of words "In the past three weeks, every day, there has been, cough, symptom, morning, evening, more, frequent, occasionally, accompanied by, white, sputum".

[0051] Step S132: Perform medical term recognition on the segmented sequence of words, mark the medical terms, and the medical terms include symptom terms, disease terms, and examination terms.

[0052] Perform medical term recognition on the segmented sequence of words, which is achieved through a medical term recognition model. During the recognition process, the model analyzes each word to determine whether it is a medical term and determine its category.

[0053] In the sequence of words “every day for the past three weeks, cough, symptoms, morning, evening, relatively, frequent, occasionally, accompanied by, white, sputum”, “cough” and “sputum” were identified as symptom terms.

[0054] The marked word sequence will indicate the category of each medical term, such as "cough (symptom term), sputum (symptom term)", while non-medical terms will not be marked.

[0055] Step S1321: Input the segmented word sequence into the medical terminology recognition model, which is constructed based on a bidirectional long short-term memory network and a conditional random field algorithm.

[0056] The segmented word sequence is input into the medical terminology recognition model, whose input layer receives each word in the sequence. The model consists of a bidirectional long short-term memory (LSTM) network layer and a conditional random field (CRF) algorithm layer. The LSM layer captures contextual information of the words, while the CRF layer is used to annotate the sequence.

[0057] For example, the word "cough" in a word sequence can be input into the model along with words such as "have" and "symptoms" before and after it, so that the model can determine whether it is a medical term based on the context.

[0058] Step S1322: Convert the word sequence into a word vector sequence through the embedding layer of the medical terminology recognition model, with each word corresponding to a word vector of a fixed dimension.

[0059] The embedding layer of the medical terminology recognition model processes the input word sequence, converting each word into a word vector. The embedding layer assigns a fixed-dimensional vector to each word by searching a pre-defined word vector dictionary; each dimension of this vector reflects the semantic features of the word.

[0060] For example, the word "cough" is converted into a multi-dimensional word vector with the same number of dimensions as the model's preset dimensions. The values ​​in the vector are determined based on the semantic distribution of "cough" in a large-scale medical text corpus.

[0061] After the entire word sequence is processed by the embedding layer, it forms a word vector sequence, with each word vector corresponding to a word in the word sequence.

[0062] Step S1323: Extract contextual features from the word vector sequence using the bidirectional long short-term memory network layer of the medical terminology recognition model to obtain the contextual semantic features of each word.

[0063] The word vector sequence is input into a bidirectional long short-term memory (LSTM) network layer, which consists of a forward LSM network and a backward LSM network. The forward LSM network processes the word sequence from the beginning to the end, capturing the preceding context information; the backward LSM network processes the word sequence from the end to the beginning, capturing the following context information.

[0064] For example, for the word vector corresponding to "cough" in the word vector sequence, the feedforward network will process it in conjunction with the word vector corresponding to "all" before it, and the feedback network will process it in conjunction with the word vector corresponding to "symptoms" after it.

[0065] After being processed by a bidirectional long short-term memory network layer, each word vector is assigned corresponding contextual semantic features, which integrate the semantic information of the word in the entire sequence.

[0066] Step S1324: Sequence labeling of contextual semantic features is performed using the conditional random field algorithm layer of the medical terminology recognition model, and each word is assigned a corresponding term type label, which includes symptom term labels, disease term labels, examination term labels and non-medical term labels.

[0067] The contextual semantic features output by the bidirectional long short-term memory network layer are input into the conditional random field algorithm layer. The conditional random field algorithm layer analyzes the contextual semantic features of each word according to the preset labeling rules and feature functions to determine its corresponding term type label.

[0068] For example, a vector with contextual semantic features of "cough" will be labeled as a symptom term by the conditional random field algorithm layer; while a contextual semantic feature vector corresponding to "in the last three weeks" can be labeled as a non-medical term.

[0069] After the annotation is completed, each word is assigned a corresponding term type label, forming an annotated sequence.

[0070] Step S1325: Post-process the sequence labeling results, merge adjacent words with the same term type label to form complete medical term phrases, match the identified medical term phrases with the medical entity set in the medical knowledge graph module, add corresponding entity identifiers to the successfully matched medical term phrases, and generate a medical term sequence containing entity identifiers.

[0071] The sequence labeling results are post-processed to check for adjacent words with the same term type label. For example, if both "white" and "sputum" are labeled as symptom terms and are adjacent to each other, they are merged into the complete medical term phrase "white sputum".

[0072] Subsequently, these medical terminology phrases are matched with the medical entity set in the medical knowledge graph module. For example, "cough" corresponds to "cough symptom entity" in the medical entity set, and "white sputum" corresponds to "white sputum symptom entity".

[0073] After a successful match, a corresponding entity identifier is added to each medical term phrase, such as "cough (entity identifier: S001)" and "white sputum (entity identifier: S002)", and finally a medical term sequence containing entity identifiers is generated.

[0074] Step S1326: Statistically count the number and location of the different types of medical terms identified, and generate a medical term distribution report.

[0075] The statistical analysis included the number of symptom terms, disease terms, and examination terms identified. For example, two symptom terms were identified: "cough" and "white sputum," while no disease or examination terms were found.

[0076] Simultaneously, the position of each medical term in the original text is recorded; for example, "cough" appears as the 4th and 9th word in the symptom description text. Based on this statistical information, a medical term distribution report is generated, clearly listing the quantity and location information of each type of medical term.

[0077] Step S133: Standardize the tagged medical terms and map non-standard terms to standard medical entity names in the medical knowledge graph module.

[0078] After identifying medical terms, they are standardized. Some users may use non-standard terms to describe symptoms, such as describing "coughing up phlegm" as "having phlegm." These non-standard terms need to be mapped to standard medical entity names in the medical knowledge graph module.

[0079] By comparing the thesaurus and terminology mapping rules in the medical knowledge graph module, "having phlegm" was mapped to the standard medical entity name "coughing up phlegm." This standardization process ensured the consistency of terminology, avoiding the impact of inconsistent terminology on subsequent feature extraction and reasoning analysis.

[0080] Step S134: Based on the standardized medical terminology, extract symptom features from the symptom description text, including symptom location features, symptom nature features, and symptom duration features.

[0081] Based on standardized medical terminology, symptom features are extracted from symptom description text. Symptom location features refer to the body part where the symptom occurs, such as "respiratory tract" for cough; symptom nature features include the form and severity of the symptom, such as "frequent" and "occasionally accompanied by white sputum" describing the nature of the cough; symptom duration features are the length of time the symptom lasts, such as "for the past three weeks" indicating that the cough has lasted for about three weeks.

[0082] Each symptom feature is extracted as multi-dimensional feature data. For example, the symptom duration feature includes not only the duration but also the frequency of attacks, such as "every day" or "frequent in the morning and evening," which together constitute the multi-dimensional data of the symptom duration feature.

[0083] Step S135: Perform intent classification processing on the health request text in the health consultation content to identify the user's health request intent category, which includes diagnostic consultation, treatment advice, prevention and health care, and examination guidance.

[0084] The intent classification process for health-related text requests is performed using a text classification model. The text request, "I want to know what's causing this cough, what medication I need, and whether I need to go to the hospital for examination," is input into the text classification model. The model analyzes the text and identifies key information.

[0085] "Want to know the reason" corresponds to diagnostic consultation, "What medicine should I take?" corresponds to treatment advice, and "Do I need to do an examination?" corresponds to examination guidance. Therefore, the user's health request intent categories include diagnostic consultation, treatment advice, and examination guidance. Each category has a corresponding identifier, which is used to generate clinical decision-making need classification tags later.

[0086] Step S136: Extract time-series features from the historical diagnosis and treatment sequence text, analyze the changing trends of previous diagnosis records and treatment intervention records over time, and obtain historical diagnosis and treatment time-series features.

[0087] Temporal features were extracted from historical medical records, and past diagnostic and treatment intervention records were sorted out in chronological order. For example, the diagnosis of acute bronchitis and the corresponding treatment intervention six months ago are temporally related to the current cough symptoms.

[0088] Analyze the effects of treatment interventions, such as symptom relief after taking cough medicine and antibiotics six months ago, and record information such as the time period and degree of symptom relief. Also, pay attention to the time interval between two medical records, such as the time span between a physical examination three years ago and the diagnosis of acute bronchitis six months ago, and the time span between acute bronchitis six months ago and the current cough symptoms.

[0089] This time information is linked to diagnostic results and treatment measures to form multi-dimensional historical medical time-series features. For example, a feature matrix can be constructed that includes time nodes, diagnostic entities, treatment entities, and symptom change trends. Each time node corresponds to a specific medical event, and symptom change trends are reflected by the differences in symptom descriptions at different time points. These features collectively reflect the evolution of a user's health status over time.

[0090] Step S137: The extracted symptom features, health demand intent categories, and historical treatment time sequence features are fused to generate a user health problem representation vector.

[0091] The extracted symptom features, health concern intention categories, and historical treatment time-series features are fused. First, the dimensions of each feature are adjusted to ensure they have a matching number of dimensions in the feature space. The multi-dimensional data of symptom features, the label data of health concern intention categories (converted into multi-dimensional vectors through one-hot encoding), and the multi-dimensional matrix of historical treatment time-series features are adjusted to the same dimensionality level.

[0092] Subsequently, a feature concatenation method was used for fusion, sequentially piecing together the multi-dimensional data of the three features to form a longer multi-dimensional vector, namely the user health problem representation vector. This user health problem representation vector contains detailed information on symptoms, the user's needs, and the evolution of historical health conditions, comprehensively representing the user's current health problems. For example, the first half of the vector is the multi-dimensional data of symptom features, the middle part is the encoding vector of health need intent category, and the second half is the multi-dimensional data of historical treatment time sequence features.

[0093] Step S138: Determine the clinical decision-making needs classification labels based on the user's health problem representation vector and health demand intent category.

[0094] Based on the user's health problem representation vector and health demand intent category, clinical decision-making need classification labels are determined. These labels represent a further refinement and standardization of user health demands, determined by considering the specific characteristics of the user's health problems.

[0095] For example, a user's health-related intent categories include diagnostic consultation, treatment advice, and examination guidance. Combined with the user's health problem representation vector showing a prolonged cough (three weeks) accompanied by specific sputum characteristics, the clinical decision-making need classification label might be determined as "prioritizing etiological diagnosis, and providing targeted treatment recommendations and examination plans based on the diagnostic results." These labels use standardized terminology, facilitating the medical knowledge graph module's accurate understanding of the user's decision-making needs during subsequent reasoning.

[0096] Step S140: Based on the entity association relationships in the medical knowledge graph module and the domain parameter configuration set, perform multi-round association reasoning processing on the user health problem representation vector and the clinical decision-making need classification label to generate health risk assessment results and clinical intervention strategy suggestions.

[0097] Based on the entity relationships and domain parameter configuration sets in the medical knowledge graph module, multi-round association reasoning is performed on the user's health problem representation vector and clinical decision-making need classification labels. This multi-round reasoning aims to progressively and deeply analyze the user's health problems, from initial symptom matching to disease diagnosis, and then to risk assessment and intervention recommendations, continuously optimizing the reasoning results.

[0098] Each round of reasoning builds upon the results of previous reasoning, combining entity relationships and domain parameters in the medical knowledge graph to conduct more detailed analysis and judgment, ultimately generating accurate health risk assessment results and clinical intervention strategy recommendations.

[0099] Step S141: Determine the initial reasoning direction and reasoning depth parameters for multi-round associative reasoning based on the clinical decision-making needs classification labels.

[0100] The initial reasoning direction and depth parameters for multi-round associative reasoning are determined based on the classification labels of clinical decision-making needs. The initial reasoning direction is determined by the core demands of clinical decision-making needs. For example, if the label emphasizes "prioritizing etiological diagnosis," then the initial reasoning direction is to infer from the symptom entity to the disease entity.

[0101] The inference depth parameter is determined based on the complexity of the problem and the level of detail required for clinical decision-making. For example, for a cough lasting three weeks with specific characteristics, the inference depth parameter is set to a level sufficient to cover both directly and indirectly related diseases, ensuring that no important potential causes are overlooked. The inference depth parameter determines the number of layers to traverse in the medical knowledge graph, such as starting with the symptom entity, traversing to the directly related disease entities, and then traversing to other entities associated with these disease entities (such as complication entities, etiology entities, etc.).

[0102] Step S142: Based on the symptom features in the user's health problem representation vector, perform entity matching in the medical knowledge graph module to find the corresponding symptom entity nodes.

[0103] Based on symptom features in the user's health problem representation vector, entity matching is performed in the medical knowledge graph module. Key symptom terms, such as "cough" and "white sputum," are extracted from the symptom features, and these terms are compared with the symptom entity set in the medical knowledge graph module.

[0104] The terminology matching algorithm identifies symptom entity nodes that perfectly match or are highly similar to these symptom terms, such as "cough symptom entity" and "white sputum symptom entity." During the matching process, the name attribute and feature description attribute in the entity attribute set are referenced to ensure accuracy and avoid mismatches caused by synonyms or near-synonyms. For example, if "sputum" appears in the symptom features, it will be matched as the "cough sputum symptom entity."

[0105] Step S143: Starting from the matched symptom entity node, generate an initial set of associated entity paths based on the entity relationship set in the medical knowledge graph module and the medical entity association weight in the domain parameter configuration set. The initial set of associated entity paths includes disease entity nodes and examination item entity nodes that are directly connected from the symptom entity node.

[0106] Starting with the matched symptom entity nodes, an initial set of associated entity paths is generated based on the entity relationship set in the medical knowledge graph module and the medical entity association weights in the domain parameter configuration set. Starting with the "cough symptom entity" and the "white sputum symptom entity," entity nodes in the medical knowledge graph that have direct associations with them are searched.

[0107] Based on the set of relationships between entities, these directly related entities may include disease entities (such as bronchitis, pneumonia, etc.) and examination item entities (such as chest X-ray examination, sputum analysis, etc.). Simultaneously, by combining the association weights of medical entities, entities with association weights reaching a certain level are selected to form initial associated entity paths. Each path consists of a starting symptom entity, an association type, and associated entities, such as "cough symptom entity - association - bronchitis disease entity" and "white sputum symptom entity - association - sputum analysis examination item entity," etc. These paths collectively constitute the initial set of associated entity paths.

[0108] Step S144: Calculate the path weight for each associated entity path in the initial associated entity path set. The path weight is the product of the association weights between the entities in the path.

[0109] Calculate the path weight for each associated entity path in the initial set of associated entity paths. Each path consists of one or more associations between entities. For a path containing multiple associations (such as symptom entity-disease entity-complication entity), the path weight is the product of the association weights between the entities.

[0110] For example, if a path is "cough symptom entity - association weight W1 - bronchitis disease entity - association weight W2 - pneumonia complication entity", then the weight of this path is the product of W1 and W2. For paths containing only one association (such as a symptom entity directly associated with a disease entity), the path weight is that single association weight. Using the above method, the comprehensive weight of each path is calculated for subsequent path selection.

[0111] Step S145: Filter out the associated entity paths whose path weights are greater than the preset path weight threshold to form a candidate inference path set.

[0112] Paths to associated entities with a path weight greater than a preset path weight threshold are selected to form a set of candidate inference paths. The preset path weight threshold is a parameter in the domain parameter configuration set, set according to the rigor requirements of clinical decision-making.

[0113] The calculated weight of each path is compared with a threshold, and paths with weights exceeding the threshold are retained. For example, if the preset path weight threshold is a certain value, paths with calculated weights greater than that value are retained, while other paths are excluded. Through this filtering process, the paths in the candidate inference path set are all closely related to the user's symptoms, reducing the workload of subsequent inferences while ensuring the accuracy of the inference.

[0114] Step S146: Based on the disease entity nodes in the candidate reasoning path set, retrieve the corresponding disease feature set from the medical knowledge graph module. The disease feature set includes typical symptom features, high-risk factor features, and complication features.

[0115] Based on the disease entity nodes in the candidate reasoning path set, the corresponding disease feature set is retrieved from the medical knowledge graph module. For each disease entity node in the candidate reasoning path set, such as bronchitis or pneumonia, the entity attribute set of the medical knowledge graph module is accessed.

[0116] The typical symptom features, high-risk factor features, and complication features of the disease entity are extracted from the entity attribute set. These features are stored in the form of structured text descriptions. For example, typical symptom features of the bronchitis disease entity include cough and sputum; high-risk factor features include smoking history and air pollution exposure; and complication features include pneumonia and chronic obstructive pulmonary disease. These features are then organized into a disease feature set, with each disease entity node corresponding to an independent disease feature set.

[0117] Step S1461: Traverse the candidate reasoning path set, extract all disease entity nodes contained therein, remove duplicate disease entity nodes, and form a candidate disease entity set.

[0118] Traverse the set of candidate reasoning paths, examine the entity nodes contained in each path one by one, and identify the disease entity nodes. For example, in the paths "cough symptom entity - association - bronchitis disease entity" and "white sputum symptom entity - association - pneumonia disease entity", extract "bronchitis disease entity" and "pneumonia disease entity".

[0119] Subsequently, duplicate disease entity nodes are removed; if the same disease entity node appears in multiple paths, only one instance is retained. This process forms a candidate disease entity set, which contains all possible disease entities associated with the user's symptoms through high-weighted paths.

[0120] Step S1462: For each disease entity node in the candidate disease entity set, retrieve the corresponding disease feature description information from the entity attribute set of the medical knowledge graph module.

[0121] For each disease entity node in the candidate disease entity set, the corresponding disease feature description information is retrieved from the entity attribute set of the medical knowledge graph module. Using the entity's unique identifier, the storage location of each disease entity node in the entity attribute set is located, and the feature description attribute content at that location is extracted.

[0122] This descriptive information provides a detailed description of the disease's characteristics, including its clinical manifestations, causes, and potential complications. For example, retrieving the descriptive information for the "bronchitis disease entity" might include phrases such as "This disease often manifests as cough and sputum production, which may be accompanied by wheezing. It is mostly caused by viral or bacterial infections, and long-term recurrent attacks may lead to chronic bronchitis."

[0123] Step S1463: Analyze the retrieved disease feature description information and extract typical symptom features, which include descriptions of common clinical manifestations and signs of the disease.

[0124] The retrieved disease feature description information is analyzed to extract typical symptom features. Natural language processing technology is then used to perform semantic analysis on the feature description information, identifying the portions describing common clinical manifestations and signs of the disease.

[0125] For example, from the feature description information of the "bronchitis disease entity," typical symptom features such as "cough, sputum, and wheezing" are extracted. These features are common clinical manifestations of the disease. For each disease entity node, the extracted typical symptom features are presented in a multi-dimensional form, including information such as the name of the symptom, frequency of occurrence, and severity level, forming a set of typical symptom features for the disease.

[0126] Step S1464: Extract high-risk factor features from the disease feature description information, wherein the high-risk factor features include risk factors and triggering conditions related to the occurrence of the disease.

[0127] High-risk factor features are extracted from disease description information. Similarly, semantic analysis is used to identify risk factors and triggering conditions associated with disease occurrence. For example, from the feature description information of the "bronchitis disease entity," high-risk factor features such as "viral infection, bacterial infection, smoking, and air pollution" are extracted.

[0128] These characteristics include environmental factors, lifestyle factors, and infection factors. Each high-risk factor has corresponding descriptive information. For example, the description of "smoking" may include "long-term smoking increases the risk of disease." Together, they constitute a set of high-risk factor characteristics for the disease.

[0129] Step S1465: Extract complication features from the disease feature description information, wherein the complication features include accompanying diseases and complication symptoms that may occur during the development of the disease.

[0130] Extracting complication features from disease feature descriptions. Analyzing the content in the feature descriptions regarding possible accompanying diseases and complications during disease development. For example, extracting complication features such as "chronic bronchitis, pneumonia, and pulmonary heart disease" from the feature descriptions of the "bronchitis disease entity".

[0131] These complication features indicate other health problems that may arise if the disease is not treated or controlled in a timely manner. Each complication feature also contains corresponding related information, such as the probability of occurrence and the association mechanism with the primary disease, forming a set of complication features of the disease.

[0132] Step S1466: Standardize the extracted typical symptom features, high-risk factor features, and complication features, and unify the feature description format and terminology.

[0133] The extracted typical symptom features, high-risk factor features, and complication features were standardized. Referring to the standard terminology system in the medical knowledge graph module, non-standard terms in the feature descriptions were converted into standard terms, thus unifying the feature description format.

[0134] For example, if "cough caused by an itchy throat" is among the typical symptoms of a disease, it will be standardized as "throat discomfort accompanied by cough"; "frequent smoking" will be standardized as "long-term smoking history". Standardization ensures the comparability of features between different disease entities, facilitating subsequent similarity calculations with user health problem representation vectors.

[0135] Step S1467: Assign corresponding feature weights to each disease feature type, with typical symptom feature weights being higher than high-risk factor feature weights, and high-risk factor feature weights being higher than complication feature weights.

[0136] Each disease feature type is assigned a corresponding feature weight. Based on the importance of different features in disease diagnosis, typical symptom features play a major role in disease identification and are therefore assigned a higher weight; high-risk factor features are next, with a lower weight than typical symptom features; complication features are relatively less important in initial diagnosis and have the lowest weight.

[0137] These feature weights are part of a domain parameter configuration set, determined through clinical practice data. For example, the weights for typical symptom features are set to a high range, the weights for high-risk factor features are set to a medium range, and the weights for complication features are set to a low range. The sum of the three weights is a fixed value (e.g., 1) to ensure the appropriateness of the weights in subsequent calculations.

[0138] Step S1468: Combine the standardized typical symptom features, high-risk factor features, and complication features, along with their corresponding feature weights, to form a complete set of disease features.

[0139] The standardized typical symptom features, high-risk factor features, and complication features, along with their corresponding feature weights, are combined to form a complete disease feature set. Each disease feature set contains three subsets (typical symptom feature subset, high-risk factor feature subset, and complication feature subset), and each feature in the subset has a corresponding weight value.

[0140] For example, in the disease feature set of the "bronchitis disease entity", the typical symptom feature subset includes "cough (weight Wt1), sputum (weight Wt2)", etc., the high-risk factor feature subset includes "smoking (weight Wg1), infection (weight Wg2)", etc., and the complication feature subset includes "pneumonia (weight Wb1)", etc., where Wt1 and Wt2 are greater than Wg1 and Wg2, and Wg1 and Wg2 are greater than Wb1.

[0141] Step S147: Calculate the similarity between the retrieved disease feature set and the user's health problem representation vector to generate a disease matching score.

[0142] The retrieved disease feature set is compared with the user's health problem representation vector to calculate the similarity and generate a disease matching score. By comparing the similarity between disease features and user symptom features, historical medical records, etc., the degree of association between the disease and the user's current health problem is determined.

[0143] Similarity calculation is performed from multiple dimensions, including the degree of matching of typical symptoms and the degree of conformity of high-risk factors. Then, a comprehensive score is calculated by combining feature weights to obtain the disease matching score.

[0144] Step S1471: Convert the typical symptom features, high-risk factor features, and complication features in the disease feature set into feature vector form to form a disease feature vector.

[0145] Typical symptom features, high-risk factor features, and complication features in the disease feature set are converted into feature vectors. Each feature (such as "cough" in typical symptoms and "smoking" in high-risk factors) is converted into a fixed-dimensional vector using word embedding technology. Then, all feature vectors under the same feature type are averaged or concatenated to form a comprehensive vector for that feature type.

[0146] For example, all feature vectors in the typical symptom feature subset are concatenated into a typical symptom feature vector, feature vectors in the high-risk factor feature subset are concatenated into a high-risk factor feature vector, and feature vectors in the complication feature subset are concatenated into a complication feature vector. Finally, these three vectors are concatenated in sequence to form a disease feature vector, which contains all the feature information of the disease.

[0147] Step S1472: Adjust the dimensions of the user's health problem representation vector so that it has the same number of dimensions as the disease feature vector.

[0148] The dimensions of the user health problem representation vector are adjusted to have the same number of dimensions as the disease feature vector. This is done through feature mapping or dimensional expansion / compression methods.

[0149] For example, if the dimension of the user health problem representation vector is lower than that of the disease feature vector, interpolation is used to expand its dimension, and the added dimension value is filled according to the numerical trend of adjacent dimensions; if the dimension is higher than that of the disease feature vector, methods such as principal component analysis are used to compress the dimension, retain the main feature information, and ensure that the adjusted user health problem representation vector and the disease feature vector are completely consistent in dimension, so as to perform similarity calculation.

[0150] Step S1473: Calculate the cosine similarity between the disease feature vector and the user health problem representation vector to obtain the basic similarity score.

[0151] The cosine similarity between the disease feature vector and the user health problem representation vector is calculated to obtain a basic similarity score. Cosine similarity measures the degree of similarity between two vectors by the angle between them in the feature space; the smaller the angle, the higher the similarity.

[0152] During the calculation, the corresponding dimension values ​​of the two vectors are multiplied, summed, and then divided by the product of the magnitudes of the two vectors. For example, each dimension value of the disease feature vector is multiplied by the corresponding dimension value of the user health problem representation vector, and all the products are summed to obtain a total. At the same time, the magnitude of each vector (the square root of the sum of the squares of the dimension values) is calculated. Finally, the sum is divided by the product of the two magnitudes, and the result is the cosine similarity, which is the basic similarity score.

[0153] Step S1474: Based on the weights of various features in the disease feature set, the basic similarity scores are weighted, with the matching degree of typical symptom features having the highest weight, followed by the matching degree of high-risk factor features, and the matching degree of complication features having the lowest weight.

[0154] Based on the weights of various features in the disease feature set, the basic similarity scores are weighted. First, the similarity between the typical symptom feature vector and the corresponding symptom part of the user's health problem representation vector, the similarity between the high-risk factor feature vector and the corresponding historical factor part of the user's health problem representation vector, and the similarity between the complication feature vector and the corresponding potential symptom part of the user's health problem representation vector are calculated.

[0155] Then, the similarity scores of these three dimensions are multiplied by their respective feature weights (weights for typical symptoms, high-risk factors, and complications), and the three products are summed to obtain a weighted overall similarity score. Since typical symptoms have the highest weight, they have the greatest impact on the overall score, followed by high-risk factors, and finally complications.

[0156] Step S1475: Calculate the weighted similarity score as the preliminary disease matching score, normalize the preliminary disease matching score, and sort the disease entity nodes in the candidate disease entity set according to the normalized disease matching score.

[0157] After calculating the weighted similarity score, this total similarity score serves as the preliminary disease matching score. The preliminary disease matching score is a multi-dimensional numerical combination that encompasses the matching status of typical symptom features, high-risk factor features, and complication features, as well as the overall matching result.

[0158] Normalizing the initial disease matching scores aims to unify the scores from different dimensions into the same numerical range for cross-sectional comparison. During normalization, based on the maximum and minimum scores for each dimension, a linear transformation is used to map the score of each dimension to a specific numerical range. For example, scores matching typical symptom characteristics, high-risk factor characteristics, and complication characteristics, which originally had different ranges, are all transformed to the same numerical range, making the scores comparable across dimensions.

[0159] After normalization, the disease entity nodes in the candidate disease entity set are ranked according to the normalized disease matching score. The ranking is based on the comprehensive normalized score; the higher the score, the higher the ranking of the disease entity node. Simultaneously, the specific details of the scores across various dimensions are also considered during the ranking process. If the comprehensive scores of two disease entity nodes are similar, the matching scores of typical symptom features are further compared, and so on, ensuring that the ranking results accurately reflect the degree of matching between disease entities and user health issues.

[0160] Step S1476: Extract a preset number of disease entity nodes that are ranked at the top, and after selecting them as key disease entities, generate a disease matching report that includes disease entity identifiers, matching scores, and matching details of each feature.

[0161] After sorting the disease entity nodes in the candidate disease entity set, a predetermined number of disease entity nodes are extracted from the top of the sorted list and identified as key disease entities of concern. These key disease entities are disease types that are highly correlated with the user's current health problems and require further in-depth analysis in subsequent reasoning processes.

[0162] When generating a disease matching report, the report first lists the unique identifier of each key disease entity. This unique identifier corresponds one-to-one with the disease entity node in the medical knowledge graph module, ensuring that the corresponding disease entity information can be accurately traced.

[0163] Secondly, the report will record in detail the matching score for each key disease entity, including the comprehensive normalized score as well as the normalized scores for typical symptom characteristics, high-risk factor characteristics, and complication characteristics.

[0164] Finally, the feature matching details section will specify the matching details between the user's health problem representation vector and the disease feature set in each feature dimension, such as which typical symptoms match, which high-risk factors are associated, and which complication features show corresponding signs. Through the above disease matching report, the matching status between each disease entity and the user's health problem can be clearly presented.

[0165] Step S148: Based on the disease matching score and clinical decision-making needs classification label, determine the core disease entity node for the first round of reasoning. Using the core disease entity node as the center, conduct the second round of association reasoning to find the treatment intervention entity node and prognostic assessment entity node associated with the core disease entity node.

[0166] Based on the normalized disease matching score and clinical decision-making need classification labels in the disease matching report, the core disease entity nodes for the first round of inference are determined. Among the disease entities of key concern, the disease entity node with the highest comprehensive matching score and the best fit with the diagnostic consultation need category in the clinical decision-making need classification labels is selected as the core disease entity node. For example, if the clinical decision-making need is mainly diagnostic consultation, and a certain disease entity has an extremely high matching degree in typical symptom characteristics and the highest comprehensive score, then it is determined as the core disease entity node.

[0167] Centered on the core disease entity node, a second round of association reasoning is conducted. Using the entity relationship set in the medical knowledge graph module, treatment intervention entity nodes and prognostic assessment entity nodes that are associated with the core disease entity node are retrieved. Treatment intervention entity nodes include entities such as various treatment methods, medication regimens, and treatment cycles for the core disease; prognostic assessment entity nodes cover entities such as the disease's possible progression, recovery probability, and potential sequelae.

[0168] During the search process, based on the medical entity association weights in the domain parameter configuration set, entity nodes with higher association strength are prioritized for screening, ensuring that the retrieved treatment intervention entity nodes and prognostic assessment entity nodes are most closely related to the core disease entity nodes. Simultaneously, relevant entity nodes are retrieved in a targeted manner, taking into account treatment suggestion and examination guidance needs in the clinical decision-making needs classification tags.

[0169] Step S149: Based on the clinical decision threshold parameters in the domain parameter configuration set and the results of the second round of correlation inference, generate preliminary health risk assessment results and clinical intervention strategy recommendations.

[0170] Based on the clinical decision threshold parameters in the domain parameter configuration set and the treatment intervention entity nodes and prognostic assessment entity nodes obtained from the second round of correlation inference, preliminary health risk assessment results and clinical intervention strategy recommendations are generated. The clinical decision threshold parameters provide the criteria for assessing health risks and determining intervention measures, while the results of the second round of correlation inference provide the specific analytical objects.

[0171] When generating preliminary health risk assessment results, information from the prognostic assessment entity node is referenced, and risk grading standards in the clinical decision threshold parameters are combined to assess the health risks the user may face. The assessment includes multiple dimensions such as disease severity, progression rate, and potential complication risks. Each dimension is judged based on corresponding threshold parameters, ultimately forming a multi-dimensional preliminary health risk assessment result.

[0172] For clinical intervention strategy recommendations, based on the information in the treatment intervention entity node and the intervention initiation conditions in the clinical decision threshold parameters, suitable intervention measures are selected for the current situation. These intervention measures cover multiple aspects such as drug therapy, non-drug therapy, and lifestyle modifications. The selection of each intervention measure must meet the corresponding threshold conditions, such as initiating a certain type of drug therapy when the symptom severity reaches a certain level. At the same time, combined with the treatment recommendation and examination guidance needs in the clinical decision needs classification tags, the intervention measures are prioritized to form preliminary clinical intervention strategy recommendations.

[0173] Step S1491: Analyze the clinical decision threshold parameters in the domain parameter configuration set and extract the health risk classification threshold and intervention initiation threshold.

[0174] The clinical decision threshold parameters in the domain parameter configuration set are analyzed to extract health risk grading thresholds and intervention initiation thresholds. Health risk grading thresholds are a set of multi-dimensional judgment criteria used to classify different levels of health risk, each level corresponding to a different stage of disease development and severity. For example, multiple threshold intervals are set based on dimensions such as symptom duration, symptom severity, and the presence of signs of complications, with each interval corresponding to a health risk level.

[0175] Intervention initiation thresholds are threshold conditions set for various treatment interventions, and different interventions have different initiation thresholds. For example, the initiation threshold for a certain type of drug treatment may include the severity of symptoms reaching a certain level or the patient having no history of allergy to the drug; the initiation threshold for a certain type of examination may include the duration of symptoms exceeding a certain period or the presence of specific high-risk factors.

[0176] During the extraction process, it is necessary to ensure the completeness and accuracy of the health risk classification thresholds and intervention initiation thresholds, with each threshold clearly corresponding to a specific judgment dimension and applicable scope.

[0177] Step S1492: Based on the prognostic assessment entity nodes in the second round of association reasoning results, determine the risk factors and prognostic indicators related to the core disease entity nodes. Based on the risk factors and prognostic indicators, and combined with the symptom features in the user's health problem representation vector and the relevant records in the historical diagnosis and treatment sequence text, calculate the health risk assessment value.

[0178] Based on the prognostic assessment entity nodes in the second round of association inference results, risk factors and prognostic indicators related to the core disease entity nodes are identified. Risk factors include various factors that may aggravate the disease, such as other underlying diseases of the user and unhealthy lifestyle habits; prognostic indicators are used to assess the disease progression trend and treatment effectiveness, such as the speed of symptom improvement and changes in relevant physiological indicators.

[0179] Based on these risk factors and prognostic indicators, and combined with symptom characteristics in the user's health problem representation vector, such as symptom severity, duration, and progression, as well as relevant records in historical medical records, such as past treatment effects and the presence of similar medical histories, a health risk assessment value is calculated. During the calculation process, each risk factor and prognostic indicator is assigned a corresponding weight according to its importance, and then a comprehensive score is obtained by combining this with the user's specific circumstances, resulting in a multi-dimensional health risk assessment value. This health risk assessment value covers multiple aspects of risk information and can comprehensively reflect the user's current health risk status.

[0180] Step S1493: Compare the health risk assessment value with the health risk grading threshold to determine the corresponding health risk level. Based on the health risk level and the core disease entity node, retrieve the corresponding set of clinical intervention measures from the medical knowledge graph module, including drug treatment measures, non-drug treatment measures and lifestyle adjustment recommendations.

[0181] The calculated health risk assessment value is compared with a health risk grading threshold, and the corresponding health risk level is determined based on the comparison result. If multiple dimensions of the health risk assessment value fall within the range corresponding to a certain health risk grading threshold, then the health risk level corresponding to that range is determined as the current health risk level.

[0182] Based on the determined health risk level and core disease entity nodes, a corresponding set of clinical intervention measures is retrieved from the medical knowledge graph module. Different combinations of health risk levels and core disease entity nodes correspond to different sets of clinical intervention measures. These sets include pharmacological treatments, such as applicable drug types, dosages, and frequency of administration; non-pharmacological treatments, such as physical therapy methods and rehabilitation training programs; and lifestyle modification recommendations, such as dietary recommendations, exercise recommendations, and sleep schedule adjustments. These intervention measures are specifically designed for the core disease entities and the current health risk level, possessing strong relevance and applicability.

[0183] Step S1494: For each intervention in the retrieved set of clinical interventions, assess whether its applicability conditions are met based on the intervention activation threshold in the clinical decision threshold parameter.

[0184] For each intervention in the retrieved set of clinical interventions, its applicability is assessed based on the intervention activation threshold in the clinical decision threshold parameters to determine if its conditions are met. The applicability conditions for each intervention include multiple judgment dimensions, each with a corresponding threshold standard.

[0185] For example, the applicable conditions for a certain drug treatment might include the user's symptom severity reaching a certain threshold, the absence of a history of drug allergy, and liver and kidney function indicators within a certain normal range. During the evaluation process, each dimension of the applicable conditions for each intervention is checked one by one to determine whether the user's current condition meets the corresponding threshold standard. If all dimensions of the conditions are met, the applicable conditions for the intervention are valid; if any dimension of the conditions is not met, the applicable conditions for the intervention are invalid.

[0186] Through the above assessment, intervention measures that meet the applicable conditions are selected to ensure that the subsequent recommended intervention measures are safe and effective.

[0187] Step S1495: Select clinical intervention measures that meet the applicable conditions to form a set of candidate intervention measures. Prioritize the intervention measures in the set of candidate intervention measures based on the health risk level, the effectiveness of the intervention measures, and the difficulty of implementation.

[0188] Clinical interventions that meet the applicable conditions are selected to form a candidate intervention set. The interventions in the candidate intervention set are all suitable for the user in the current situation, but they differ in implementation order and importance, therefore, they need to be prioritized.

[0189] Prioritization is based on three dimensions: health risk level, effectiveness of intervention measures, and difficulty of implementation. The higher the health risk level, the more priority should be given to intervention measures that can quickly reduce the risk; the effectiveness of intervention measures refers to the degree to which the measures improve the condition and alleviate symptoms, and measures with more significant effects have higher priority; the difficulty of implementation considers factors such as the ease and cost for users to implement the measures, and measures with lower difficulty of implementation have higher priority.

[0190] During the ranking process, each of the three dimensions is assigned a corresponding weight, and the priority score of each intervention measure is calculated. Then, the intervention measures in the candidate intervention measure set are ranked according to their scores, with intervention measures with higher scores ranked first and recommended to users first.

[0191] Step S1496: Combining the health risk level and the ranked set of candidate intervention measures, generate preliminary health risk assessment results and clinical intervention strategy recommendations, add corresponding decision-making basis explanations to the preliminary health risk assessment results and clinical intervention strategy recommendations, and explain the origin of the assessment results and recommended measures.

[0192] By combining the health risk level and the ranked set of candidate interventions, preliminary health risk assessment results and clinical intervention strategy recommendations are generated. The preliminary health risk assessment results describe in detail the user's current health risk level, main risk factors, and possible development trends. Each conclusion is based on a comparison between the previously calculated health risk assessment value and the health risk grading threshold.

[0193] The recommended clinical intervention strategies are listed in order of priority from the set of candidate interventions, including the specific content, implementation method, and expected effects of each intervention. At the same time, corresponding decision-making justifications are added to the preliminary health risk assessment results and clinical intervention strategy recommendations.

[0194] The explanation of the decision-making basis, when elaborating on the origin of the health risk assessment results, can include details of which prognostic assessment entity nodes' information was referenced, the calculation process of the health risk assessment value, and its comparison with health risk grading thresholds. When explaining the origin of the clinical intervention strategy recommendations, it can include the assessment process of the applicable conditions for each intervention measure, the basis for prioritization, etc. Through these explanations of the decision-making basis, users and relevant personnel on the healthcare service platform can understand the rationality and scientific validity of the assessment results and recommended measures.

[0195] Step S1410: Combine the treatment intervention records in the historical diagnosis and treatment sequence text to verify the rationality of the preliminary health risk assessment results and clinical intervention strategy recommendations. Adjust the inference parameters according to the verification results, perform a third round of correlation inference optimization, and finally generate accurate health risk assessment results and clinical intervention strategy recommendations.

[0196] By combining treatment intervention records from historical medical records, the rationality of preliminary health risk assessment results and clinical intervention strategy recommendations is verified. Treatment intervention records from historical medical records include information such as the user's previous treatment methods, medications used, and treatment effects. This information reflects the user's response to different treatment measures and historical changes in their physical condition.

[0197] Reasonableness verification is mainly carried out from the following aspects: checking whether the preliminary health risk assessment results are consistent with the user's historical health status and treatment response; judging whether the recommended clinical intervention measures conflict with the user's previous treatment intervention records, such as whether drugs for which the user has had allergic reactions are recommended; and assessing the continuity and effectiveness of the intervention measures, such as whether the currently recommended intervention measures are consistent with previously effective treatment measures and whether they can avoid repeating ineffective treatments.

[0198] Based on the results of the rationality check, the inference parameters are adjusted, such as adjusting the association weights of medical entities and the weights of each dimension in the calculation of health risk assessment values. Then, a third round of association inference optimization is performed based on the adjusted parameters, re-searching relevant entity nodes, recalculating health risk assessment values, and re-screening and sorting intervention measures. Through the above optimization process, potential irrationalities in the preliminary results are eliminated, ultimately generating accurate health risk assessment results and clinical intervention strategy recommendations.

[0199] Step S150: Generate a platform service response instruction based on the health risk assessment results and the clinical intervention strategy recommendations, and send the platform service response instruction to the business processing module of the medical and health service platform to trigger the corresponding health service response process.

[0200] After obtaining accurate health risk assessment results and clinical intervention strategy recommendations, it is necessary to generate platform service response instructions based on this information and send them to the business processing module of the healthcare service platform to trigger the corresponding health service response process. Platform service response instructions serve as a bridge connecting the artificial intelligence decision-making system and the business processing module of the healthcare service platform, translating decision results into concrete service actions.

[0201] Step S151: Analyze the health risk assessment results and extract the health risk level, main risk factors, and prognostic assessment conclusions.

[0202] Analyze the health risk assessment results to extract key information, including the health risk level, major risk factors, and prognostic assessment conclusion. The health risk level is an overall judgment of the degree of risk to the user's current health status and is an important basis for subsequent service responses; major risk factors are the key factors that lead to this health risk level, clarifying the aspects that need to be focused on and intervened in; the prognostic assessment conclusion predicts the possible development trend of the disease and the prospect of recovery.

[0203] During the extraction process, the accuracy and completeness of this information are ensured, with each information point corresponding to specific content in the health risk assessment results to avoid omissions or misinterpretations. After extraction, this information is organized into a structured data format to facilitate the subsequent construction of platform service response instructions.

[0204] Step S152: Analyze the clinical intervention strategy recommendations and extract the recommended intervention types, implementation steps, and precautions.

[0205] The analysis of clinical intervention strategy recommendations extracts the types of interventions, implementation steps, and precautions. Intervention types include pharmacological therapy, non-pharmacological therapy, and lifestyle modifications, clarifying the specific direction of the service response. The implementation steps detail the operational procedures and sequence of each intervention to ensure correct implementation by users. Precautions include potential problems and contraindications during implementation, ensuring safe implementation of the interventions.

[0206] Similarly, the accuracy and completeness of the information must be ensured during the extraction process. This information should be organized into a structured data format and matched with the information extracted from the health risk assessment results, together serving as the core content for building platform service response instructions.

[0207] Step S153: Construct a basic framework for platform service response instructions according to the instruction format requirements of the medical and health service platform. The basic framework includes an instruction header, an instruction body, and an instruction tail. The instruction header contains an instruction identifier and a generation time. The instruction body contains specific service content. The instruction tail contains verification information.

[0208] Based on the instruction format requirements of the healthcare service platform, a basic framework for platform service response instructions is constructed. Instruction format requirements are specifications established by the healthcare service platform to ensure that instructions can be correctly parsed and processed, including the instruction structure, data type, and encoding method.

[0209] The basic framework of a platform service response command includes a command header, command body, and command tail. The command header contains a command identifier and a generation time. The command identifier is a unique identifier used to distinguish different commands; the generation time records when the command was created, facilitating traceability and management. The command body is the core of the command, containing the specific service content, namely various information extracted from health risk assessment results and clinical intervention strategy recommendations. The command tail contains verification information used to verify whether the command has been tampered with or corrupted during transmission, ensuring the integrity and security of the command.

[0210] Build a basic framework according to these requirements to prepare for filling in the specific content later.

[0211] Step S154: Fill the extracted health risk level, main risk factors and prognostic assessment conclusions into the health assessment area of ​​the instruction body.

[0212] The extracted health risk level, main risk factors, and prognostic assessment conclusions are then populated into the health assessment area of ​​the instruction body. The health assessment area is a section of the instruction body specifically designed to display health risk-related information, which is arranged and presented according to a preset format.

[0213] For example, first list the health risk levels, then list the main risk factors in sequence, with each risk factor accompanied by a brief description, and finally present the prognostic assessment conclusion. During the data entry process, ensure that the information format conforms to the requirements of the instruction body, and that the text descriptions are clear, accurate, and easy for the business processing modules of the healthcare service platform and users to understand.

[0214] Step S155: Fill the intervention suggestion area of ​​the instruction body with the extracted intervention type, implementation steps and precautions.

[0215] Fill the extracted intervention type, implementation steps, and precautions into the intervention suggestion area of ​​the instruction body. The intervention suggestion area is the part of the instruction body used to display information related to the intervention measures, and it should also be filled in according to the preset format.

[0216] For each intervention, first describe its type, then list the implementation steps in detail, and finally note any precautions. Multiple interventions should be arranged according to priority, ensuring that important interventions are presented first. During the data entry process, ensure the information is logical and coherent to make the intervention recommendations actionable.

[0217] Step S156: Add user identification information to the platform service response command, and determine the priority of the platform service response command according to the preset command priority rules and the health risk level.

[0218] Add user identification information to the platform service response command. This user identification information is a unique identifier for the user associated with the health consultation content, used by the healthcare service platform to accurately identify the user corresponding to the command. During the addition process, the identification information is extracted from the received user information and filled in according to the required position and format of the command, ensuring that the user identification information is accurately embedded into the platform service response command.

[0219] The platform determines the priority of service response instructions based on preset instruction priority rules and the health risk level. These preset priority rules are formulated based on factors such as health risk level and the urgency of the user's inquiry. The higher the health risk level, the higher the corresponding instruction priority; under the same health risk level, the more obvious the urgency in the user's inquiry, the higher the priority will be.

[0220] For example, when the health risk level is high, the platform service response instruction is prioritized to ensure that the instruction can be processed by the healthcare service platform first and quickly trigger the corresponding health service response process. After determining the priority, the priority information is added to the designated field in the instruction header, which, together with the instruction identifier and generation time, constitutes the complete instruction header.

[0221] Step S157: Perform format verification on the generated platform service response command, encrypt the platform service response command that passes the verification, and generate summary information of the platform service response command. The summary information contains a brief description of the key content of the command.

[0222] The generated platform service response commands are format-validated according to the command format standards stipulated by the healthcare service platform. This includes checking the completeness of the command header, body, and footer, and verifying that the information in each field conforms to the format requirements. For example, the length and character type of the command identifier are correct, the information in each area of ​​the command body is filled according to the preset format, and the presence of validation information in the footer.

[0223] If the format validation fails, return to the corresponding step for modification and adjustment until the instruction format fully meets the requirements. Platform service response instructions that pass validation are encrypted using an encryption algorithm to prevent unauthorized acquisition or tampering during transmission, ensuring information security.

[0224] After encryption, a summary of the platform service response instruction is generated. This summary is obtained by extracting and compressing the encrypted instruction content and includes a brief description of key information such as the health risk level and the type of primary intervention. Generating this summary helps the healthcare service platform quickly understand the core content of the instruction, improving processing efficiency.

[0225] Step S158: Send the platform service response instruction to the business processing module of the medical and health service platform to trigger the corresponding health service response process.

[0226] After generating, verifying, encrypting, and generating a summary of the platform service response command, the command is sent to the business processing module of the healthcare service platform. The sending process utilizes a dedicated communication interface to ensure the stability and timeliness of data transmission.

[0227] After receiving a service response instruction from the platform, the business processing module of the healthcare service platform decrypts and parses the instruction, triggering the corresponding health service response process based on its content. For example, if the instruction contains high-priority intervention suggestions, the business processing module will prioritize arranging for medical staff to communicate with the user and provide further diagnosis and treatment guidance; if the instruction suggests a certain examination, the business processing module will schedule the corresponding examination for the user and provide the appointment information back to the user.

[0228] Figure 2The illustration shows exemplary hardware and software components of an artificial intelligence decision-making system 100 applied to a healthcare service platform, which can implement the ideas of this application, according to some embodiments of this application. For example, processor 120 can be used in the artificial intelligence decision-making system 100 applied to a healthcare service platform and to perform the functions in this application.

[0229] For example, an AI decision-making system 100 for a healthcare service platform may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the AI ​​decision-making system 100 for a healthcare service platform may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The AI ​​decision-making system 100 for a healthcare service platform also includes an I / O interface 150 between the computer and other input / output devices.

[0230] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the artificial intelligence decision-making method applied to the medical and health service platform as described above is implemented.

[0231] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. An artificial intelligence decision-making method for a healthcare service platform, characterized in that, The method includes: The medical knowledge graph module of the artificial intelligence decision-making system is activated and a preset set of domain parameter configurations is loaded. The set of domain parameter configurations includes medical entity association weights and clinical decision threshold parameters. The system receives real-time health consultation content and associated historical treatment sequence text pushed by the user from the medical and health service platform. The health consultation content includes symptom description text and health request text, and the historical treatment sequence text includes past diagnosis records and treatment intervention records arranged in chronological order. The received health consultation content and the historical medical treatment sequence text are subjected to medical semantic parsing and intent recognition processing to generate user health problem representation vectors and clinical decision-making need classification tags. Based on the entity relationships in the medical knowledge graph module and the domain parameter configuration set, the user health problem representation vector and the clinical decision-making need classification label are subjected to multi-round association reasoning processing to generate health risk assessment results and clinical intervention strategy suggestions. Based on the health risk assessment results and the clinical intervention strategy recommendations, a platform service response instruction is generated and sent to the business processing module of the medical and health service platform to trigger the corresponding health service response process.

2. The artificial intelligence decision-making method for a medical and health service platform according to claim 1, characterized in that, The medical knowledge graph module that initiates the artificial intelligence decision-making system and loads a preset set of domain parameter configurations includes: The medical knowledge graph module of the artificial intelligence decision-making system is activated. The medical knowledge graph module includes a set of medical entities, a set of entity attributes, and a set of relationships between entities. The set of medical entities includes disease entities, symptom entities, drug entities, and examination item entities. The set of entity attributes includes the entity's name attribute, classification attribute, and feature description attribute. The set of relationships between entities includes causal relationships, membership relationships, and association relationships. A preset set of domain parameter configurations is retrieved from the parameter storage unit of the artificial intelligence decision-making system. The set of domain parameter configurations includes medical entity association weights and clinical decision threshold parameters. The medical entity association weights are used to represent the strength of the association between different medical entities, and the clinical decision threshold parameters are used to determine the initiation conditions of clinical intervention measures. The retrieved domain parameter configuration set is subjected to parameter integrity verification processing to check whether there are missing or abnormal values ​​in the medical entity association weight and clinical decision threshold parameters; The domain parameter configuration set that has passed integrity verification is subjected to format conversion processing to convert the parameter data into a format that can be recognized by the medical knowledge graph module; The converted domain parameter configuration set is loaded into the parameter configuration unit of the medical knowledge graph module to complete the initial configuration of the medical knowledge graph module; The self-verification mechanism of the medical knowledge graph module is activated to verify the consistency between the loaded domain parameter configuration set and the entity relationships in the medical knowledge graph, so that the parameter configuration matches the knowledge graph structure.

3. The artificial intelligence decision-making method for a medical and health service platform according to claim 1, characterized in that, The process of performing medical semantic analysis and intent recognition on the received health consultation content and the historical medical treatment sequence text to generate user health problem representation vectors and clinical decision-making need classification tags includes: The symptom description text and health request text in the health consultation content are preprocessed, and the preprocessed symptom description text is segmented into word sequences to divide the continuous text into independent word sequences. Medical terminology is identified and marked from the segmented word sequence. The medical terms include symptom terms, disease terms, and examination terms. Standardize the tagged medical terms and map non-standard terms to standard medical entity names in the medical knowledge graph module; Based on standardized medical terminology, symptom features are extracted from symptom description text, including symptom location features, symptom nature features, and symptom duration features. The health request text in the health consultation content is processed for intent classification to identify the user's health request intent category, which includes diagnostic consultation, treatment advice, prevention and health care, and examination guidance. Temporal features are extracted from the historical diagnosis and treatment sequence text, and the changing trends of past diagnosis records and treatment intervention records over time are analyzed to obtain historical diagnosis and treatment temporal features. The extracted symptom features, health concern intention categories, and historical treatment time sequence features are fused to generate a user health problem representation vector; Based on the user's health problem representation vector and health demand intent category, determine the clinical decision-making need classification label.

4. The artificial intelligence decision-making method for a medical and health service platform according to claim 3, characterized in that, The process of identifying and marking medical terms in the segmented word sequence includes: The segmented word sequence is input into the medical terminology recognition model, which is constructed based on a bidirectional long short-term memory network and a conditional random field algorithm. The embedding layer of the medical terminology recognition model converts word sequences into word vector sequences, with each word corresponding to a word vector of a fixed dimension. The contextual features of each word are obtained by extracting contextual semantic features from the word vector sequence through the bidirectional long short-term memory network layer of the medical terminology recognition model. The contextual semantic features are sequentially labeled using the conditional random field layer of the medical terminology recognition model, and each word is assigned a corresponding term type label, which includes symptom term labels, disease term labels, examination term labels and non-medical term labels. The sequence labeling results are post-processed to merge adjacent words with the same term type label to form complete medical term phrases. The identified medical term phrases are then matched with the medical entity set in the medical knowledge graph module. Corresponding entity identifiers are added to the successfully matched medical term phrases to generate medical term sequences containing entity identifiers. The number and location of different types of medical terms are statistically identified, and a medical term distribution report is generated.

5. The artificial intelligence decision-making method for a medical and health service platform according to claim 1, characterized in that, The process involves multi-round association reasoning processing of the user's health problem representation vector and the clinical decision-making need classification labels based on the entity relationships in the medical knowledge graph module and the domain parameter configuration set, generating health risk assessment results and clinical intervention strategy suggestions, including: The initial reasoning direction and reasoning depth parameters for multi-round associative reasoning are determined based on the clinical decision-making needs classification labels. Based on the symptom features in the user's health problem representation vector, entity matching is performed in the medical knowledge graph module to find the corresponding symptom entity nodes. Starting from the matched symptom entity node, an initial set of associated entity paths is generated based on the set of inter-entity relationships in the medical knowledge graph module and the medical entity association weights in the domain parameter configuration set. The initial set of associated entity paths includes disease entity nodes and examination item entity nodes that are directly associated with the symptom entity node. For each associated entity path in the initial set of associated entity paths, calculate the path weight. The path weight is the product of the association weights between the entities in the path. Filter out associated entity paths whose path weight is greater than a preset path weight threshold to form a candidate inference path set; Based on the disease entity nodes in the candidate reasoning path set, the corresponding disease feature set is retrieved from the medical knowledge graph module. The disease feature set includes typical symptom features, high-risk factor features, and complication features. The similarity between the retrieved disease feature set and the user's health problem representation vector is calculated to generate a disease matching score. Based on disease matching scores and clinical decision-making needs classification labels, the core disease entity nodes for the first round of reasoning are identified. Then, the second round of association reasoning is carried out with the core disease entity nodes as the center to find treatment intervention entity nodes and prognostic assessment entity nodes that are associated with the core disease entity nodes. Based on the clinical decision threshold parameters in the domain parameter configuration set and the results of the second round of correlation inference, preliminary health risk assessment results and clinical intervention strategy recommendations are generated. By combining treatment intervention records in historical medical records, the preliminary health risk assessment results and clinical intervention strategy recommendations are validated for rationality. Based on the validation results, the inference parameters are adjusted, and a third round of correlation inference optimization is performed to ultimately generate accurate health risk assessment results and clinical intervention strategy recommendations.

6. The artificial intelligence decision-making method for a healthcare service platform according to claim 5, characterized in that, The step of retrieving the corresponding disease feature set from the medical knowledge graph module based on the disease entity nodes in the candidate reasoning path set includes: Traverse the candidate reasoning path set, extract all disease entity nodes contained therein, remove duplicate disease entity nodes, and form a candidate disease entity set; For each disease entity node in the candidate disease entity set, retrieve the corresponding disease feature description information from the entity attribute set of the medical knowledge graph module; The retrieved disease feature description information is analyzed, and typical symptom features are extracted. The typical symptom features include descriptions of common clinical manifestations and signs of the disease. High-risk factor features are extracted from disease characteristic description information, and the high-risk factor features include risk factors and triggering conditions related to the occurrence of the disease; Extract complication features from disease feature description information, wherein the complication features include accompanying diseases and complication symptoms that may occur during the disease development process; The extracted typical symptom features, high-risk factor features, and complication features are standardized to unify the feature description format and terminology. Each disease feature type is assigned a corresponding feature weight, with typical symptom features having a higher weight than high-risk factor features, and high-risk factor features having a higher weight than complication features. The standardized typical symptom features, high-risk factor features, and complication features, along with their corresponding feature weights, are combined to form a complete set of disease features.

7. The artificial intelligence decision-making method for a medical and health service platform according to claim 5, characterized in that, The step of calculating the similarity between the retrieved disease feature set and the user's health problem representation vector to generate a disease matching score includes: The typical symptom features, high-risk factor features, and complication features in the disease feature set are converted into feature vectors to form disease feature vectors. Adjust the dimensions of the user's health problem representation vector to make it have the same number of dimensions as the disease feature vector; Calculate the cosine similarity between the disease feature vector and the user health problem representation vector to obtain the basic similarity score; Based on the weights of various features in the disease feature set, the basic similarity scores are weighted, with typical symptom features having the highest matching weight, high-risk factor features having the second highest matching weight, and complication features having the lowest matching weight. Calculate the weighted similarity score as the preliminary disease matching score, normalize the preliminary disease matching score, and sort the disease entity nodes in the candidate disease entity set according to the normalized disease matching score. Extract a preset number of disease entity nodes from the top of the sorted list, and then use them as the key disease entities to generate a disease matching report that includes disease entity identifiers, matching scores, and matching details of each feature.

8. The artificial intelligence decision-making method for a medical and health service platform according to claim 5, characterized in that, The process involves generating preliminary health risk assessment results and clinical intervention strategy recommendations based on the clinical decision threshold parameters in the domain parameter configuration set and the results of the second round of correlation inference, including: The clinical decision threshold parameters in the domain parameter configuration set are analyzed to extract health risk classification thresholds and intervention initiation thresholds. Based on the prognostic assessment entity nodes in the second round of association inference results, risk factors and prognostic indicators related to the core disease entity nodes are determined. Based on the risk factors and prognostic indicators, combined with the symptom features in the user's health problem representation vector and the relevant records in the historical diagnosis and treatment sequence text, the health risk assessment value is calculated. The health risk assessment value is compared with the health risk classification threshold to determine the corresponding health risk level. Based on the health risk level and the core disease entity node, the corresponding set of clinical intervention measures is retrieved from the medical knowledge graph module, including drug treatment measures, non-drug treatment measures and lifestyle adjustment suggestions. For each intervention in the retrieved set of clinical interventions, assess whether its applicability conditions are met based on the intervention activation threshold in the clinical decision threshold parameters; Clinical interventions that meet the applicable conditions are selected to form a set of candidate interventions. The interventions in the set of candidate interventions are prioritized based on the health risk level, the effectiveness of the intervention, and the difficulty of implementation. By combining the health risk level and the ranked set of candidate interventions, preliminary health risk assessment results and clinical intervention strategy recommendations are generated. Corresponding decision-making explanations are added to the preliminary health risk assessment results and clinical intervention strategy recommendations to explain the origin of the assessment results and recommended measures.

9. The artificial intelligence decision-making method for a medical and health service platform according to claim 1, characterized in that, The step of generating platform service response instructions based on the health risk assessment results and the clinical intervention strategy recommendations includes: Analyze the health risk assessment results to extract the health risk level, main risk factors, and prognostic assessment conclusions; Analyze clinical intervention strategy recommendations, and extract the recommended intervention types, implementation steps, and precautions; Based on the instruction format requirements of the healthcare service platform, a basic framework for platform service response instructions is constructed. The basic framework includes an instruction header, an instruction body, and an instruction tail. The instruction header contains an instruction identifier and a generation time. The instruction body contains specific service content. The instruction tail contains verification information. Fill the extracted health risk level, main risk factors and prognostic assessment conclusions into the health assessment area of ​​the instruction body; Fill the extracted intervention type, implementation steps, and precautions into the intervention suggestion area of ​​the instruction body; Add user identification information to platform service response commands, and determine the priority of platform service response commands based on preset command priority rules and health risk levels; The generated platform service response command is format-validated, and the platform service response command that passes the validation is encrypted to generate a summary information of the platform service response command. The summary information contains a brief description of the key content of the command.

10. An artificial intelligence decision-making system for a healthcare service platform, characterized in that, The device includes a processor and a memory, the memory being connected to the processor. The memory is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the memory to implement the artificial intelligence decision-making method applied to a medical and health service platform as described in any one of claims 1-9.

Images