Ear-nose-throat nursing scheme recommendation method based on knowledge graph

Abstract

The invention discloses an ear-nose-throat nursing scheme recommendation method based on a knowledge graph, and relates to the technical field of ear-nose-throat nursing scheme recommendation, and the method comprises the following steps: 1, based on medical research results, clinical cases, treatment methods and nursing key points in the ear-nose-throat field, collecting a data source, and establishing an initial ear-nose-throat knowledge graph based on the collected data source. According to the ear-nose-throat nursing scheme recommendation method based on the knowledge graph, a monitoring mechanism is set, the situation that a nursing scheme is inaccurate due to old knowledge is avoided, expert verification is carried out on the inclusion of new knowledge through expert verification and scientific research trend monitoring, accuracy and reasonability are ensured, newly appearing research hotspots are included in the knowledge graph, and the knowledge graph recommendation efficiency is improved. Therefore, the nursing scheme recommendation is more frontier and more scientific, the newly added knowledge is ensured not to conflict with the existing knowledge graph, and the stability and reliability of the knowledge graph are ensured, so that a solid foundation is provided for generating an accurate nursing scheme.

Description

Technical Field

[0001] The present invention relates to the technical field of ENT nursing program recommendation, and specifically to an ENT nursing program recommendation method based on a knowledge graph. Background Art

[0002] ENT diseases are relatively common in daily life, and effective care is crucial to the patient's recovery. However, the current knowledge graph may be difficult to keep up with the rapid development of the medical field after it is built. New research results, treatment methods and nursing points are constantly emerging, and the existing knowledge graph may not be able to quickly reflect these changes. How to establish an effective monitoring mechanism to timely discover new research, new cases and new expert opinions in the field of ENT, when new knowledge emerges, how to automatically or semi-automatically integrate it into the existing knowledge graph, and how to ensure the accuracy and consistency of the updated knowledge graph, this is a technical problem that needs to be solved urgently. Summary of the invention

[0003] In order to solve the above technical problems, the present invention is implemented by the following technical solutions: a method for recommending ear, nose and throat care plans based on knowledge graph, comprising the following steps: Step 1: Based on the medical research results, clinical cases, treatment methods and nursing points in the field of ENT, data sources are collected, and an initial ENT knowledge graph is established based on the collected data sources; the construction method of the ENT knowledge graph includes entity recognition, relationship extraction and graph representation, wherein entities include disease names, symptoms, treatment methods, nursing measures, drugs and expert names, and relationships describe the association between diseases and symptoms, treatment plans and applicable conditions, and nursing measures and effects; data sources include relevant literature, research reports, medical guidelines and expert opinions on ENT; Step 2: Set up a monitoring mechanism to regularly collect the latest research results, clinical cases, new treatment and care plans in the field of ENT, update the initial ENT knowledge graph, and obtain an updated graph; the monitoring mechanism uses machine learning algorithms to analyze text data from medical journals, scientific research papers, and clinical trial reports to identify new entities and new relationships, and combines the expert review system to verify the accuracy and effectiveness of new knowledge; Step 3: Based on the received patient condition information, combined with the relationship between diseases and nursing measures in the updated atlas, and combined with the patient's individualized condition data, such as symptom severity, medical history, and medication use, a nursing plan is generated through the set nursing plan generation mechanism.

[0004] Preferably, the initial ENT knowledge graph construction includes the following process: Entity recognition: Through medical text processing technology, ENT diseases, symptoms, treatment methods, nursing measures, drugs and expert names are extracted from ENT related literature and research reports, and the initial entity database is constructed using the entity recognition model; Relationship extraction: The deep learning model is used to extract the relationship between diseases and symptoms, treatment methods and applicable conditions, and nursing measures and effects from the entity database. After verification and correction by ENT medical experts, the relationship is clarified to ensure the scientific nature of the knowledge graph. Graph representation: Based on multi-level relationship description, the entity database and association relationships are structured and stored through the graph database to obtain the initial ENT knowledge graph.

[0005] Preferably, the process of building the initial entity database is: Retrieve relevant papers and research reports in the field of ENT from medical academic databases, collect ENT disease treatment guidelines issued by major medical institutions, and organize the opinions and clinical experience summaries published by ENT experts; use data collection tools to collect selected data sources into a unified storage location, and preliminarily classify and organize the collected data, and classify and store literature and research reports according to different types of ENT diseases; store medical guidelines and expert opinions in different folders to obtain text data; Preprocess the collected text data, including noise removal, text formatting and word segmentation. By removing noise, special characters and punctuation errors in the text are removed. By formatting the text, texts of different formats are unified into a standard format. By word segmentation, the text is split into individual words for entity recognition. Specifically, the natural language processing toolkit can be used to segment the text, and the sentence "Symptoms of chronic sinusitis include nasal congestion and runny nose" can be segmented into the words "chronic sinusitis", "symptoms", "includes", "nasal congestion" and "runny nose". The preprocessing of the text data is completed, and the preprocessed text data is used as training data for the model; In the field of otolaryngology, an entity recognition model is constructed based on long short-term memory network, bidirectional long short-term memory network and conditional random field to learn the features in the text and identify the entities; the entity recognition model is used to identify and extract entities from the preprocessed text data; for each document, research report, medical guide and expert opinion, the text is analyzed sentence by sentence to identify the otolaryngological diseases, symptoms, treatment methods, nursing measures, drugs and expert names; the identified and extracted entities are stored in the database to build an initial entity database. During the construction of the initial entity database, a relational database or a non-relational database is selected according to the characteristics of the entity and subsequent usage requirements, and a record is created for each entity in the database, including the name, type and source of the entity, that is, which document or research report it was extracted from.

[0006] Preferably, the process of constructing an entity recognition model based on a long short-term memory network, a bidirectional long short-term memory network and a conditional random field is as follows: sequentially constructing a long short-term memory network layer, a bidirectional long short-term memory network layer and a conditional random field layer; First, construct the long short-term memory network layer: input the preprocessed text data into the long short-term memory network layer, convert each word into a word vector, and update the internal cell state and hidden state according to the current input state and the state at the previous moment; The flow of information and the update of memory are controlled through the input gate, forget gate and output gate; new information is added to the cell state based on the input gate, old information is forgotten based on the forget gate, and information is output to the next layer based on the output gate; Then, a bidirectional long short-term memory network layer is constructed: based on the capture of text context information, a bidirectional long short-term memory network layer is added after the long short-term memory network layer; the bidirectional long short-term memory network layer is composed of two forward and backward long short-term memory networks, which process sequence data from the beginning and end of the text respectively. Finally, the outputs of the two directions are spliced ​​to obtain a comprehensive representation of each word, thereby outputting the feature representation of each word; in this way, the contextual information of the word can be considered at the same time, improving the model's ability to extract text features; Finally, the conditional random field layer is constructed: the conditional random field layer is added on top of the output of the bidirectional long short-term memory network layer; the conditional random field layer receives the feature representation of each word output by the bidirectional long short-term memory network layer, and then calculates the probability of each possible labeling sequence based on the transition probability and emission probability learned from the training data; finally, the labeling sequence with the highest probability is selected as the output of the model; the conditional random field layer determines the labeling of each word based on the input text features and the learned probability model, thereby identifying the entities in the text, including disease names and symptoms; thus, an entity recognition model is constructed, and the labeling sequence output by the conditional random field layer is used as the output of the model; Conditional random fields are a probabilistic model for sequence labeling that can take into account the dependencies between labels of the entire sequence and output the optimal labeling sequence.

[0007] Preferably, the entity recognition model is trained: first, a log-likelihood loss function is selected, the logarithmic difference between the probability of the label sequence predicted by the model and the probability of the true label sequence is calculated, and the parameters of the model are optimized by minimizing the logarithmic difference; then, an adaptive moment estimation algorithm is selected, and the learning rate of the adaptive moment estimation algorithm is set to 0.001, and the momentum parameters are set to 0.9 and 0.999, so as to control the convergence speed and stability of the optimization process; the log-likelihood loss function is minimized by the adaptive moment estimation algorithm, and the parameters of the model are updated; finally, the pre-processed text data is input into the model as training data, and the output of the model is calculated by forward propagation, the error between the label sequence and the true label sequence is calculated according to the log-likelihood loss function, and then the gradient is calculated by the back propagation algorithm to update the parameters of the model; this process is repeated until the log-likelihood loss function converges or reaches a predetermined number of training times; the entity recognition model after the parameters of the trained model are updated is obtained, and during the training, the training data can be divided into multiple batches for batch training to improve the training efficiency; Evaluate and optimize the entity recognition model after parameter update: For entity recognition tasks, use evaluation indicators to evaluate the performance of the model; the evaluation indicators include accuracy, recall and F1 value; accuracy refers to the ratio of the number of entities correctly recognized by the model to the total number of recognized entities; recall refers to the ratio of the number of entities correctly recognized by the model to the number of actual entities; F1 value is the harmonic mean of accuracy and recall, which comprehensively considers the accuracy and completeness of the model; In order to improve the generalization ability of the model, the cross-validation method is used; the training data is divided into multiple subsets, and one subset is selected each time as the validation set, and the remaining subsets are used as the training set for model training and evaluation; this process is repeated until each subset is used as a validation set once; finally, the average value of the evaluation indicators on each validation set is calculated as the final performance indicator of the model; Grid search, random search, and Bayesian optimization are used to adjust the hidden layer size of the long short-term memory network, the hidden layer size of the bidirectional long short-term memory network, and the regularization parameter of the conditional random field to optimize the performance of the model, observe the performance of the model on the validation set, select the hyperparameter combination with the best performance, and obtain the hyperparameter combination.

[0008] Preferably, the process of learning features in text and identifying entities is as follows: The trained and optimized model is used to automatically learn features in the text; the long short-term memory network layer and the bidirectional long short-term memory network layer are used to extract the contextual information and semantic features of words, and the conditional random field layer is used to consider the dependencies between labels, so as to learn more comprehensive and accurate text features; the learned features are then applied to entity recognition tasks in the field of ENT; when new text data is input, the model automatically identifies the entities therein, including disease names, symptoms, treatments, and nursing measures.

[0009] Preferably, the process of obtaining the initial ENT knowledge graph is as follows: Based on the goal of building a knowledge graph, the entity types involved are clearly defined, including disease names, symptoms, treatment methods, nursing measures, drugs, and expert names. At the same time, the relationships between entities are determined: the relationship between diseases and symptoms, the relationship between treatment plans and applicable conditions, and the relationship between nursing measures and effects. Design the hierarchical structure of the knowledge graph based on entity types and association relationships; divide the knowledge graph into different levels: disease level, symptom level, treatment level, and nursing level; each level contains corresponding entity types and is interconnected through association relationships; Different graph databases are evaluated based on performance factors to obtain evaluation results, where performance factors include data storage capacity, query performance, scalability, and ease of use; based on the evaluation results, a graph database that matches the ENT knowledge graph is selected: if the amount of data in the knowledge graph is large, the storage capacity and scalability of the database are considered; if complex queries and analysis are required, the query performance and functions of the database need to be considered; based on the continuous growth of the amount of data in the ENT knowledge graph and the need for efficient queries and analysis, Neo4j is selected as the graph database because of its good scalability and powerful query language; using the API provided by the graph database, the entities in the entity database are converted into nodes in the graph database, and corresponding edges are created based on the relationships between the entities; Set corresponding attributes for each node and edge to store more information; set attributes for disease nodes: disease name, cause of disease, common symptoms; set attributes for symptom nodes: symptom description, severity; set attributes for edges: relationship type, association strength, convert entity and relationship data into nodes and edges in the graph database in the above way, and use Cypher query language to write scripts to batch import entity and relationship data into the database for structured storage to obtain the initial ENT knowledge graph; ensure the accuracy and completeness of the data, and the correct establishment of the relationship between nodes and edges, then use Neo4j's query language Cypher to query the symptoms and treatment methods of specific diseases, and check whether the results are consistent with expectations; at the same time, use Neo4j's visualization tool to view the structure of the knowledge graph and the relationship between nodes; perform data verification on the knowledge graph imported into the graph database, check whether the number, type, and attributes of nodes and edges are correct, and whether the relationship is as expected, to ensure the accuracy and completeness of the data; if the knowledge graph has a large amount of data or the query performance is not ideal, improve the query performance by creating an index; if the query often involves a specific type of node or relationship, create an index for the node or relationship to quickly locate and retrieve data.

[0010] Preferably, the monitoring mechanism during updating the knowledge graph is as follows: Real-time data collection: Use crawler technology to regularly collect the latest data sources from medical journals, academic conference reports, and online medical forums, and use natural language processing technology to parse these data, extract potential new entities and new relationships, and merge new entities and new relationships into new knowledge; Expert verification: Before new knowledge is incorporated into the knowledge graph, it is verified by ENT medical experts, and the accuracy and rationality of the automatically extracted new entities and new relationships are reviewed to obtain review results to ensure that the newly introduced knowledge is in line with the latest nursing research results and practical experience in the medical field; Automatic update of knowledge graph: Through the knowledge update process within the system, based on the coherence and completeness of the knowledge graph, after the review results are correct, and on the premise that there is no conflict between the association between the new knowledge and the existing knowledge graph, the new knowledge verified by otolaryngology experts is integrated into the existing initial otolaryngology knowledge graph, the existing data is updated and corrected, the knowledge graph is improved, and the updated graph is obtained.

[0011] Preferably, during the updating of the knowledge graph, scientific research trends and hot topics in the field of otolaryngology are also monitored, and analysis is performed based on the scientific research trends and hot topics to extract emerging research hotspots. At the same time, early warning signals are generated based on the emergence of research hotspots, and the research hotspots are merged into new knowledge based on the early warning signals to participate in the generation process of the updated graph, which can provide early warning of potential important research results and quickly merge emerging knowledge into the updated graph.

[0012] Preferably, during the process of obtaining the nursing plan, a nursing plan generation mechanism is set, and the nursing plan generation mechanism generates the nursing plan through the following steps: First, for patient condition information input, the user enters the patient's basic information, including age, gender, medical history, and current symptoms, and then extracts disease entities that match the patient's symptoms from the knowledge graph; Secondly, for the selection of nursing measures, the system extracts a list of nursing measures applicable to the patient's condition from the knowledge graph based on the disease entity and the associated nursing measure relationship; Then, multi-dimensional optimization is performed, and the best nursing combination plan is calculated through optimization algorithms based on the applicable conditions of nursing measures, possible side effects, and individual differences of patients; Finally, the plan is output, combining the patient's needs and the latest nursing research results to generate the final nursing plan to ensure the scientificity, effectiveness and safety of the nursing plan.

[0013] The present invention provides an ENT nursing plan recommendation method based on knowledge graph, which has the following beneficial effects: This ENT nursing plan recommendation method based on knowledge graph sets up a monitoring mechanism to regularly collect the latest research results, clinical cases, the latest treatment and nursing plans in the field of ENT, and update the initial knowledge graph; this can ensure that the recommended nursing plans are always based on the latest medical knowledge and research progress, and avoid inaccurate nursing plans due to outdated knowledge; through expert verification and scientific research trend monitoring, new knowledge is verified by ENT medical experts before being incorporated into the knowledge graph, ensuring accuracy and rationality; at the same time, scientific research trends and hot topics are monitored, and newly emerging research hotspots are promptly incorporated into the knowledge graph, making nursing plan recommendations more cutting-edge and scientific; it is also updated based on the coherence and integrity of the knowledge graph: the automatic update process of the knowledge graph is based on its coherence and integrity, ensuring that the newly added knowledge does not conflict with the existing knowledge graph, ensuring the stability and reliability of the knowledge graph, and thus providing a solid foundation for generating accurate nursing plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a flowchart of a method for recommending ENT care plans based on a knowledge graph in the present invention. DETAILED DESCRIPTION

[0015] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The embodiments of the present invention are provided for the purpose of illustration and description, and are not intended to be exhaustive or to limit the present invention to the disclosed forms. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the present invention, and to enable those of ordinary skill in the art to understand the present invention and thereby design various embodiments with various modifications suitable for specific uses.

[0016] like Figure 1 As shown, the present invention provides a technical solution: a method for recommending an ear, nose and throat care plan based on a knowledge graph, comprising the following steps: Step 1: Based on the medical research results, clinical cases, treatment methods and nursing points in the field of ENT, data sources are collected, and the initial ENT knowledge graph is established based on the collected data sources; the construction method of the ENT knowledge graph includes entity recognition, relationship extraction and graph representation, among which entities include disease names, symptoms, treatment methods, nursing measures, drugs and expert names, and relationships describe the relationship between diseases and symptoms, treatment plans and applicable conditions, and nursing measures and effects; data sources include relevant literature, research reports, medical guidelines and expert opinions on ENT; Step 2: Set up a monitoring mechanism to regularly collect the latest research results, clinical cases, new treatment and care plans in the field of ENT, update the initial ENT knowledge graph, and obtain an updated graph; the monitoring mechanism uses machine learning algorithms to analyze text data from medical journals, scientific research papers, and clinical trial reports to identify new entities and new relationships, and combines the expert review system to verify the accuracy and effectiveness of new knowledge; Step 3: Based on the received patient condition information, combined with the relationship between diseases and nursing measures in the updated atlas, and combined with the patient's individualized condition data, such as symptom severity, medical history, and medication use, a nursing plan is generated through the set nursing plan generation mechanism.

[0017] The initial ENT knowledge graph construction includes the following processes: Entity recognition: Through medical text processing technology, ENT diseases, symptoms, treatment methods, nursing measures, drugs and expert names are extracted from ENT related literature and research reports, and the initial entity database is constructed using the entity recognition model; Relationship extraction: The deep learning model is used to extract the relationship between diseases and symptoms, treatment methods and applicable conditions, and nursing measures and effects from the entity database. After verification and correction by ENT medical experts, the relationship is clarified to ensure the scientific nature of the knowledge graph. Graph representation: Based on multi-level relationship description, the entity database and association relationships are structured and stored through the graph database to obtain the initial ENT knowledge graph. It should be further explained that in the specific implementation process, the process of constructing the initial ENT knowledge graph is clarified, including entity recognition, relationship extraction and graph representation, which provides clear steps and methods for building a comprehensive and accurate knowledge graph. The association relationships verified by medical experts improve the reliability and authority of the knowledge graph, and provide a solid foundation for subsequent nursing plan recommendations.

[0018] The process of building the initial entity database: Retrieve relevant papers and research reports in the field of ENT from medical academic databases, collect ENT disease treatment guidelines issued by major medical institutions, and organize the opinions and clinical experience summaries published by ENT experts; use data collection tools to collect selected data sources into a unified storage location, and preliminarily classify and organize the collected data, and classify and store literature and research reports according to different types of ENT diseases; store medical guidelines and expert opinions in different folders to obtain text data; Preprocess the collected text data, including noise removal, text formatting and word segmentation. By removing noise, special characters and punctuation errors in the text are removed. By formatting the text, texts of different formats are unified into a standard format. By word segmentation, the text is split into individual words for entity recognition. Specifically, the natural language processing toolkit can be used to segment the text, and the sentence "Symptoms of chronic sinusitis include nasal congestion and runny nose" can be segmented into the words "chronic sinusitis", "symptoms", "includes", "nasal congestion" and "runny nose". The preprocessing of the text data is completed, and the preprocessed text data is used as training data for the model; In the field of ENT, an entity recognition model is constructed based on long short-term memory network, bidirectional long short-term memory network and conditional random field to learn the features in the text and identify the entities; the entity recognition model is used to identify and extract the preprocessed text data; for each document, research report, medical guide and expert opinion, the text is analyzed sentence by sentence to identify the ENT diseases, symptoms, treatment methods, nursing measures, drugs and expert names; it should be further explained that in the specific implementation process, in a research report on allergic rhinitis, the disease name entity "allergic rhinitis", the symptom entities of "sneezing", "runny nose" and "nasal congestion", and the drug entity of "antihistamine" are identified, and the identified and extracted entities are stored in the database to build the initial entity database. During the construction of the initial entity database, the relational entity is selected according to the characteristics of the entity and subsequent usage requirements. A database or a non-relational database is used to build a record, and in the database, a record is created for each entity, including the name and type of the entity, such as disease, symptoms, treatment method, and source, that is, from which literature or research report the entity is extracted; it should be further explained that in the specific implementation process, there is a record in the database as "Entity Name: Allergic Rhinitis; Type: Disease; Source: A Medical Journal Article"; through the above process, an initial ENT field entity database can be constructed, which provides a basis for the subsequent construction of a knowledge graph; it should be further explained that in the specific implementation process, data is collected from multiple reliable data sources and classified and sorted, which ensures the comprehensiveness and accuracy of the data, provides a guarantee for building a high-quality entity database, and uses advanced deep learning technology to build an entity recognition model, which can automatically extract entities from text data and improve the efficiency and accuracy of entity recognition.

[0019] The process of constructing an entity recognition model based on long short-term memory network, bidirectional long short-term memory network and conditional random field is as follows: Construct the long short-term memory network layer, the bidirectional long short-term memory network layer and the conditional random field layer in sequence; First, build the long short-term memory network layer: Input the preprocessed text data into the long short-term memory network layer, convert each word into a word vector, and update the internal cell state and hidden state according to the current input state and the state at the previous moment; The flow of information and the update of memory are controlled through the input gate, forget gate and output gate; new information is added to the cell state based on the input gate, old information is forgotten based on the forget gate, and information is output to the next layer based on the output gate; Then construct a bidirectional long short-term memory network layer: Based on the capture of text context information, a bidirectional long short-term memory network layer is added after the long short-term memory network layer; the bidirectional long short-term memory network layer is composed of two forward and backward long short-term memory networks, which process sequence data from the beginning and end of the text respectively, so as to output the feature representation of each word; in this way, the context information of the words can be considered at the same time, and the model's ability to extract text features can be improved; it should be further explained that in the specific implementation process, for a sentence "the patient has tinnitus symptoms", the forward long short-term memory network processes each word in turn starting from "patient", and the backward long short-term memory network processes each word in reverse starting from "symptoms"; finally, the outputs of the two directions are spliced ​​to obtain a comprehensive representation of each word; Finally, construct the conditional random field layer: A conditional random field layer is added on top of the output of the bidirectional long short-term memory network layer; the conditional random field layer receives the feature representation of each word output by the bidirectional long short-term memory network layer, and then calculates the probability of each possible labeling sequence based on the transition probability and emission probability learned from the training data; finally, the labeling sequence with the highest probability is selected as the output of the model; the conditional random field layer determines the labeling of each word based on the input text features and the learned probability model, thereby identifying entities in the text, including disease names and symptoms; thereby, an entity recognition model is constructed, and the labeling sequence output by the conditional random field layer is used as the output of the model; Conditional random field is a probabilistic model for sequence labeling, which can consider the dependencies between labels of the entire sequence, so as to output the optimal labeling sequence; it should be further explained that, in the specific implementation process, for the entity recognition task, the labeling sequence can be: "B-Disease I-Disease OOO B-Symptom I-Symptom"; where "B" represents the beginning of the entity, "I" represents the inside of the entity, and "O" represents the non-entity; it should be further explained that, in the specific implementation process, the long short-term memory network layer, the bidirectional long short-term memory network layer and the conditional random field layer are constructed in sequence, which can fully learn the features in the text and improve the performance of entity recognition. The functions of each layer are clear and they cooperate with each other, which can effectively capture the contextual information of the text and the dependencies between labels, and provide strong technical support for the accurate recognition of entities.

[0020] The entity recognition model is trained as follows: First, the log-likelihood loss function is selected, and the logarithmic difference between the probability of the label sequence predicted by the model and the probability of the true label sequence is calculated, and the parameters of the model are optimized by minimizing the logarithmic difference; then, the adaptive moment estimation algorithm is selected, and the learning rate of the adaptive moment estimation algorithm is set to 0.001, and the momentum parameters are set to 0.9 and 0.999 to control the convergence speed and stability of the optimization process; the log-likelihood loss function is minimized by the adaptive moment estimation algorithm, and the parameters of the model are updated; finally, the preprocessed text data is input into the model as training data, and the output of the model is calculated by forward propagation. The error between the labeled sequence and the true labeled sequence is calculated according to the log-likelihood loss function, and then the gradient is calculated by the back-propagation algorithm to update the parameters of the model; this process is repeated until the log-likelihood loss function converges or the predetermined number of training times is reached; the parameter of the trained model is obtained, and the entity recognition model after the parameters are updated. During the training, the training data can be divided into multiple batches for batch training to improve the training efficiency; it should be further explained that in the specific implementation process, a batch of text data is randomly selected from the training data each time and input into the model for training; after multiple iterations, the model gradually learns the characteristics in the text and the rules of entity annotation;

[0021] For entity recognition tasks, evaluation indicators are used to evaluate the performance of the model; the evaluation indicators include accuracy, recall and F1 value; among them, accuracy refers to the ratio of the number of entities correctly recognized by the model to the total number of recognized entities; recall refers to the ratio of the number of entities correctly recognized by the model to the number of actual entities; F1 value is the harmonic mean of accuracy and recall, which comprehensively considers the accuracy and completeness of the model; In order to improve the generalization ability of the model, the cross-validation method is used; the training data is divided into multiple subsets, and one subset is selected each time as the validation set, and the remaining subsets are used as the training set for model training and evaluation; this process is repeated until each subset is used as a validation set once; finally, the average value of the evaluation indicators on each validation set is calculated as the final performance indicator of the model; Grid search, random search and Bayesian optimization are used to adjust the hidden layer size of the long short-term memory network, the hidden layer size of the bidirectional long short-term memory network and the regularization parameter of the conditional random field respectively to optimize the performance of the model, observe the performance of the model on the validation set, select the hyperparameter combination with the best performance, and obtain the hyperparameter combination; it should be further explained that in the specific implementation process, different hidden layer sizes and regularization parameters are tried, the performance of the model on the validation set is observed, and the hyperparameter combination with the best performance is selected, the hidden layer size, such as 64, 128, 256; regularization parameters, such as 0.01, 0.1, 1; it should be further explained that in the specific implementation process, selecting appropriate loss functions and optimization algorithms to train the entity recognition model can effectively optimize the parameters of the model and improve the performance of the model. The cross-validation and hyperparameter adjustment methods can be used to evaluate the generalization ability of the model, find the best hyperparameter combination, and further improve the accuracy and stability of the model.

[0022] The process of learning features in text and identifying entities is as follows: the trained and optimized model is used to automatically learn features in text; the long short-term memory network layer and the bidirectional long short-term memory network layer are used to extract the contextual information and semantic features of words, and the conditional random field layer is used to consider the dependencies between labels, so as to learn more comprehensive and accurate text features; it should be further explained that in the specific implementation process, for a sentence containing "the patient has tinnitus symptoms", the model can learn the association between the word "tinnitus" and "symptoms", as well as the role of the words "patient" and "appearance" in the sentence, so as to accurately identify "tinnitus" as a symptom entity; the learned features are then applied It is used for entity recognition tasks in the field of otolaryngology; when new text data is input, the model automatically identifies the entities therein, including disease names, symptoms, treatment methods, and nursing measures; it should be further explained that in the specific implementation process, for a new medical document, the model can quickly identify the various otolaryngological diseases and symptoms mentioned therein, and provide accurate entity information for building a knowledge graph; it should be further explained that in the specific implementation process, the trained and optimized model can automatically learn the features in the text and apply it to entity recognition tasks in the field of otolaryngology, which improves the efficiency and accuracy of entity recognition, provides accurate entity information for building a knowledge graph, and lays the foundation for subsequent nursing plan recommendations.

[0023] The process of obtaining the initial ENT knowledge graph is as follows: Based on the goal of building a knowledge graph, the entity types involved include disease name, symptoms, treatment methods, nursing measures, drugs and expert names. At the same time, the relationship between entities is determined: the association between disease and symptoms, the relationship between treatment plans and applicable conditions, and the relationship between nursing measures and effects. It should be further explained that in the specific implementation process, "allergic rhinitis" is a disease name entity, "sneezing", "runny nose" and "nasal congestion" are related symptom entities, "antihistamine treatment" is a treatment method entity, and "avoiding contact with allergens" is a nursing measure entity. The relationship between them can be expressed as "allergic rhinitis" disease will have the symptom of "sneezing", and "antihistamine treatment" is applicable to "allergic rhinitis"; According to entity types and association relationships, the hierarchical structure of the knowledge graph is designed; the knowledge graph is divided into different levels: disease layer, symptom layer, treatment layer, and nursing layer; each level contains corresponding entity types and is interconnected through association relationships; it should be further explained that in the specific implementation process, the disease layer contains various ENT disease name entities, and the symptom layer contains various symptom entities. The entities in the disease layer are connected to the entities in the symptom layer through the "has symptoms" relationship; similarly, the entities in the treatment layer and the nursing layer are connected to the entities in the disease layer through the "treatment method" and "nursing measures" relationships respectively; Different graph databases are evaluated based on performance factors to obtain evaluation results, where performance factors include data storage capacity, query performance, scalability and ease of use; based on the evaluation results, a graph database matching the ENT knowledge graph is selected: if the amount of data in the knowledge graph is large, the storage capacity and scalability of the database are considered; if complex queries and analysis are required, the query performance and functions of the database need to be considered; based on the growing amount of data in the ENT knowledge graph and the need for efficient queries and analysis, Neo4j is selected as the graph database because it has good scalability and powerful query language; using the API provided by the graph database, the entities in the entity database are converted into nodes in the graph database, and corresponding edges are created based on the relationship between the entities; it should be further explained that in the specific implementation process, for the "allergic rhinitis" disease entity, a node can be created in the graph database and its type can be marked as "disease"; for the "sneezing" symptom entity, a node is also created and its type is marked as "symptom"; then, an edge is created from the "allergic rhinitis" node to the "sneezing" node, and its relationship is marked as "symptomatic"; Set corresponding attributes for each node and edge to store more information; set attributes for disease nodes: disease name, cause of disease, common symptoms; set attributes for symptom nodes: symptom description, severity; set attributes for edges: relationship type, association strength. It should be further explained that in the specific implementation process, for the "allergic rhinitis" node, the attribute "name" can be set to "allergic rhinitis", "cause of disease" can be set to "allergen stimulation", and "common symptoms" can be set to "sneezing, runny nose, nasal congestion"; for the "symptomatic" edge, the attribute "association strength" can be set to "strong", indicating that allergic rhinitis is closely associated with these symptoms. Convert the entity and relationship data into nodes and edges in the graph database in the above manner, and use the Cypher query language to write scripts to batch import the entity and relationship data into the database for structured storage to obtain the initial ENT knowledge graph; ensure the accuracy and completeness of the data, and the correct establishment of the relationship between nodes and edges, and then use Neo4j's query language Cypher to query the symptoms and treatment methods of specific diseases. method to check whether the results are consistent with expectations; at the same time, use Neo4j's visualization tool to view the structure of the knowledge graph and the relationship between nodes; perform data verification on the knowledge graph imported into the graph database to check whether the number, type, and attributes of nodes and edges are correct, and whether the relationship is as expected, to ensure the accuracy and completeness of the data; if the knowledge graph has a large amount of data or the query performance is not ideal, create an index to improve the query performance; if the query often involves a specific type of node or relationship, create an index for the node or relationship to quickly locate and retrieve data; through the above process, an initial ENT knowledge graph can be obtained, and it is continuously optimized and improved to provide strong support for medical research, clinical diagnosis and treatment in the field of ENT; it needs to be further explained that in the specific implementation process, a reasonable design of the knowledge graph hierarchical structure can clearly display the relationship between entities, facilitate user query and understanding, and by selecting a suitable graph database and optimizing it, the storage capacity and query performance of the knowledge graph can be improved, providing technical support for the rapid generation of nursing plans.

[0024] The monitoring mechanism during updating the knowledge graph is as follows: Real-time data collection: Use crawler technology to regularly collect the latest data sources from medical journals, academic conference reports, and online medical forums, and use natural language processing technology to parse these data, extract potential new entities and new relationships, and merge new entities and new relationships into new knowledge; Expert verification: Before new knowledge is incorporated into the knowledge graph, it is verified by ENT medical experts, and the accuracy and rationality of the automatically extracted new entities and new relationships are reviewed to obtain review results to ensure that the newly introduced knowledge is in line with the latest nursing research results and practical experience in the medical field; Automatic update of knowledge graph: through the knowledge update process within the system, based on the coherence and completeness of the knowledge graph, based on the correctness of the review results, and based on the premise that there is no conflict in the relationship between the new knowledge and the existing knowledge graph, the new knowledge verified by otolaryngology medical experts is integrated into the existing initial otolaryngology knowledge graph, the existing data is updated and corrected, the knowledge graph is improved, and the updated graph is obtained; it should be further explained that in the specific implementation process, when new research results are published or clinical cases appear, new entities of diseases, symptoms, and treatment methods can be added to the knowledge graph, and corresponding relationships can be established; at the same time, the attributes of existing entities are updated to reflect the latest medical knowledge and research progress; it should be further explained that in the specific implementation process, the monitoring mechanism can collect the latest data sources in real time, and automatically update the knowledge graph after expert verification, ensuring that the knowledge graph is always kept up to date and accurate, improving the timeliness and reliability of nursing plan recommendations, and providing better nursing services for patients.

[0025] During the updating of the knowledge graph, the scientific research trends and hot topics in the field of otolaryngology are also monitored, and analysis is conducted based on the scientific research trends and hot topics to extract emerging research hotspots. At the same time, early warning signals are generated based on the emergence of research hotspots, and the research hotspots are merged into new knowledge based on the early warning signals to participate in the generation process of the updated graph, which can provide early warning of potential important research results and quickly merge emerging knowledge into the updated graph. It needs to be further explained that in the specific implementation process, by monitoring scientific research trends and hot topics, new research hotspots can be discovered in a timely manner and incorporated into the knowledge graph updating process, making the nursing plan recommendations more cutting-edge and scientific, and adapting to the development and changes in medical research.

[0026] During the process of obtaining the nursing plan, a nursing plan generation mechanism is set, and the nursing plan generation mechanism generates the nursing plan through the following steps: First, for patient condition information input, the user enters the patient's basic information, including age, gender, medical history, and current symptoms, and then extracts disease entities that match the patient's symptoms from the knowledge graph; Secondly, for the selection of nursing measures, the system extracts a list of nursing measures applicable to the patient's condition from the knowledge graph based on the disease entity and the associated nursing measure relationship; Then, multi-dimensional optimization is performed, and the best nursing combination plan is calculated through optimization algorithms based on the applicable conditions of nursing measures, possible side effects, and individual differences of patients; Finally, the plan is output, combining the patient's needs and the latest nursing research results to generate the final nursing plan to ensure the scientificity, effectiveness and safety of the nursing plan.

[0027] It should be further explained that, during the specific implementation process, the nursing plan generation mechanism takes into account the individual differences and multi-dimensional factors of the patients, and can generate the best nursing combination plan, thereby improving the personalization and effectiveness of the nursing plan. It combines patient needs and the latest research results to provide patients with more considerate and professional nursing services.

[0028] Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field and related fields without creative work should fall within the scope of protection of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention are implemented according to the conventional means in the field unless otherwise specified and limited.

Claims

1. A method for recommending ENT care plans based on knowledge graph, characterized in that: The following steps are involved: Step 1: Based on the medical research results, clinical cases, treatment methods and nursing points in the field of ENT, data sources are collected, and an initial ENT knowledge graph is established based on the collected data sources; the construction method of the ENT knowledge graph includes entity recognition, relationship extraction and graph representation, wherein entities include disease names, symptoms, treatment methods, nursing measures, drugs and expert names, and relationships describe the association between diseases and symptoms, treatment plans and applicable conditions, and nursing measures and effects; data sources include relevant literature, research reports, medical guidelines and expert opinions on ENT; Step 2: Set up a monitoring mechanism to regularly collect the latest research results, clinical cases, new treatment and care plans in the field of ENT, update the initial ENT knowledge graph, and obtain an updated graph; the monitoring mechanism uses machine learning algorithms to analyze text data from medical journals, scientific research papers, and clinical trial reports to identify new entities and new relationships, and combines the expert review system to verify the accuracy and effectiveness of new knowledge; Step 3: Based on the received patient condition information, combined with the association between diseases and nursing measures in the updated atlas, combined with the patient's individualized condition data, a nursing plan is generated through the set nursing plan generation mechanism.

2. The method for recommending ENT nursing solutions based on knowledge graph according to claim 1, characterized in that: The initial ENT knowledge graph construction includes the following processes: Entity recognition: Through medical text processing technology, ENT diseases, symptoms, treatment methods, nursing measures, drugs and expert names are extracted from ENT related literature and research reports, and the initial entity database is constructed using the entity recognition model; Relationship extraction: The deep learning model is used to extract the relationship between diseases and symptoms, treatment methods and applicable conditions, and nursing measures and effects from the entity database, and the relationship is clarified through verification and correction by ENT medical experts; Graph representation: Based on multi-level relationship description, the entity database and association relationships are structured and stored through the graph database to obtain the initial ENT knowledge graph.

3. The method for recommending ENT nursing solutions based on knowledge graph according to claim 2, characterized in that: The process of building the initial entity database: Retrieve relevant papers and research reports in the field of ENT from medical academic databases, collect ENT disease treatment guidelines issued by major medical institutions, and organize the opinions and clinical experience summaries published by ENT experts; use data collection tools to collect selected data sources into a unified storage location, and preliminarily classify and organize the collected data, and classify and store the literature and research reports according to different types of ENT diseases; The medical guidelines and expert opinions are stored in different folders to obtain text data; Preprocess the collected text data, including noise removal, text formatting and word segmentation. By removing noise, special characters and punctuation errors in the text are removed. By formatting the text, texts in different formats are unified into a standard format. By word segmentation, the text is split into individual words to complete the preprocessing of the text data. The preprocessed text data is used as training data for the model. In the field of otolaryngology, an entity recognition model is constructed based on long short-term memory network, bidirectional long short-term memory network and conditional random field to learn the features in the text and identify the entities; the entity recognition model is used to perform entity recognition and extraction on the preprocessed text data; for each document, research report, medical guide and expert opinion, the text is analyzed sentence by sentence to identify the otolaryngology diseases, symptoms, treatment methods, nursing measures, drugs and expert names, and the identified and extracted entities are stored in the database to build an initial entity database. During the construction of the initial entity database, a relational database or a non-relational database is selected according to the characteristics of the entity and subsequent usage requirements, and a record is created in the database for each entity, including the name, type and source of the entity.

4. The method for recommending an ear, nose and throat care plan based on a knowledge graph according to claim 3, characterized in that: The process of constructing an entity recognition model based on long short-term memory network, bidirectional long short-term memory network and conditional random field is as follows: Construct the long short-term memory network layer, the bidirectional long short-term memory network layer and the conditional random field layer in sequence; First, build the long short-term memory network layer: Input the preprocessed text data into the long short-term memory network layer, convert each word into a word vector, and update the internal cell state and hidden state according to the current input state and the state at the previous moment; The flow of information and the update of memory are controlled through the input gate, forget gate and output gate; new information is added to the cell state based on the input gate, old information is forgotten based on the forget gate, and information is output to the next layer based on the output gate; Then construct a bidirectional long short-term memory network layer: Based on the capture of text context information, a bidirectional long short-term memory network layer is added after the long short-term memory network layer; the bidirectional long short-term memory network layer is composed of two long short-term memory networks, the forward and backward ones, which process the sequence data from the beginning and the end of the text respectively, so as to output the feature representation of each word; Finally, construct the conditional random field layer: A conditional random field layer is added on top of the output of the bidirectional long short-term memory network layer; the conditional random field layer receives the feature representation of each word output by the bidirectional long short-term memory network layer, and then calculates the probability of each possible labeling sequence based on the transition probability and emission probability learned from the training data; finally, the labeling sequence with the highest probability is selected as the output of the model; thereby, an entity recognition model is constructed, and the labeling sequence output by the conditional random field layer is used as the output of the model.

5. The method for recommending ENT care plans based on knowledge graph according to claim 4, characterized in that: Train the entity recognition model: First, the log-likelihood loss function is selected, the logarithmic difference between the probability of the annotation sequence predicted by the model and the probability of the true annotation sequence is calculated, and the parameters of the model are optimized by minimizing the logarithmic difference; then, the adaptive moment estimation algorithm is selected, and the learning rate of the adaptive moment estimation algorithm is set to 0.001, and the momentum parameters are set to 0.9 and 0.999; the log-likelihood loss function is minimized by the adaptive moment estimation algorithm, and the parameters of the model are updated; finally, the preprocessed text data is input into the model as training data, and the output of the model is calculated by forward propagation, and the error between the annotation sequence and the true annotation sequence is calculated according to the log-likelihood loss function, and then the gradient is calculated by the back-propagation algorithm to update the parameters of the model; this process is repeated until the log-likelihood loss function converges; the entity recognition model after the parameters of the trained model are updated is obtained; Evaluate and optimize the entity recognition model after parameter update: For entity recognition tasks, evaluation indicators are used to evaluate the performance of the model; the evaluation indicators include accuracy, recall and F1 value; among them, accuracy refers to the ratio of the number of entities correctly recognized by the model to the total number of recognized entities; recall refers to the ratio of the number of entities correctly recognized by the model to the number of actual entities; F1 value is the harmonic mean of accuracy and recall; Use the cross-validation method; divide the training data into multiple subsets, select one subset as the validation set each time, and use the remaining subsets as the training set for model training and evaluation; repeat this process until each subset is used as a validation set once; finally, calculate the average of the evaluation indicators on each validation set as the final performance indicator of the model; Grid search, random search and Bayesian optimization are used to adjust the hidden layer size of the long short-term memory network, the hidden layer size of the bidirectional long short-term memory network and the regularization parameter of the conditional random field respectively, and the performance of the model on the validation set is observed to obtain the hyperparameter combination.

6. The method for recommending an ear, nose and throat care plan based on a knowledge graph according to claim 3, characterized in that: The process of learning features in text and identifying entities is as follows: The trained and optimized model is used to automatically learn features in the text; the long short-term memory network layer and the bidirectional long short-term memory network layer are used to extract the contextual information and semantic features of words, and the conditional random field layer is used to consider the dependencies between labels. The learned features are then applied to entity recognition tasks in the field of ENT; when new text data is input, the model automatically identifies the entities therein, including disease names, symptoms, treatments, and nursing measures.

7. The method for recommending ENT care plans based on knowledge graph according to claim 6, characterized in that: The process of obtaining the initial ENT knowledge graph is as follows: Based on the goal of building a knowledge graph, the entity types involved are clearly defined, including disease names, symptoms, treatment methods, nursing measures, drugs, and expert names. At the same time, the relationships between entities are determined: the relationship between diseases and symptoms, the relationship between treatment plans and applicable conditions, and the relationship between nursing measures and effects. Design the hierarchical structure of the knowledge graph based on entity types and association relationships; divide the knowledge graph into different levels: disease level, symptom level, treatment level, and nursing level; each level contains corresponding entity types and is interconnected through association relationships; Different graph databases are evaluated based on performance factors to obtain evaluation results, where performance factors include data storage capacity, query performance, scalability, and ease of use; based on the evaluation results, a graph database that matches the ENT knowledge graph is selected; based on the growing amount of data in the ENT knowledge graph and the need for efficient query and analysis, Neo4j is selected as the graph database, and the API provided by the graph database is used to convert entities in the entity database into nodes in the graph database, and create corresponding edges based on the relationships between entities; Set corresponding attributes for each node and edge; set attributes for disease nodes: disease name, cause of disease, common symptoms; set attributes for symptom nodes: symptom description, severity; set attributes for edges: relationship type, association strength, convert entity and relationship data into nodes and edges in the graph database in the above way, and use Cypher query language to write scripts to batch import entity and relationship data into the database for structured storage to obtain the initial ENT knowledge graph; then use Neo4j's query language Cypher to query the symptoms and treatments of specific diseases to check whether the results are consistent with expectations; at the same time, use Neo4j's visualization tools to view the structure of the knowledge graph and the relationship between nodes; if the knowledge graph has a large amount of data or the query performance is not ideal, create an index; if the query often involves a specific type of node or relationship, create an index for the node or relationship.

8. The method for recommending ENT care plans based on knowledge graph according to claim 7, characterized in that: The monitoring mechanism during updating the knowledge graph is as follows: Real-time data collection: Use crawler technology to regularly collect the latest data sources from medical journals, academic conference reports, and online medical forums, and use natural language processing technology to parse these data, extract potential new entities and new relationships, and merge new entities and new relationships into new knowledge; Expert verification: Before new knowledge is incorporated into the knowledge graph, it is verified by ENT experts, who review the accuracy and rationality of the automatically extracted new entities and new relationships to obtain the review results; Automatic update of knowledge graph: Through the knowledge update process within the system, based on the coherence and completeness of the knowledge graph, after the review results are correct, and on the premise that there is no conflict between the association between the new knowledge and the existing knowledge graph, the new knowledge verified by otolaryngology experts is integrated into the existing initial otolaryngology knowledge graph, the existing data is updated and corrected, the knowledge graph is improved, and the updated graph is obtained.

9. The method for recommending ENT nursing solutions based on knowledge graph according to claim 8, characterized in that: During the updating of the knowledge graph, we also monitor the scientific research trends and hot topics in the field of otolaryngology, analyze them, extract emerging research hotspots, generate early warning signals based on the emergence of research hotspots, merge the research hotspots into new knowledge based on the early warning signals, and participate in the generation process of updating the graph.

10. The method for recommending ENT nursing plans based on knowledge graph according to claim 9, characterized in that: During the process of obtaining the nursing plan, a nursing plan generation mechanism is set, and the nursing plan generation mechanism generates the nursing plan through the following steps: First, for patient condition information input, the user enters the patient's basic information, including age, gender, medical history, and current symptoms, and then extracts disease entities that match the patient's symptoms from the knowledge graph; Secondly, for the selection of nursing measures, the system extracts a list of nursing measures applicable to the patient's condition from the knowledge graph based on the disease entity and the associated nursing measure relationship; Then, multi-dimensional optimization is performed, and the best nursing combination plan is calculated through optimization algorithms based on the applicable conditions of nursing measures, possible side effects, and individual differences of patients; Finally, the plan is output and the final nursing plan is generated by combining the patient's needs and the latest nursing research results.

Images