Knowledge graph and large language model-based intelligent question and answer system for children sleep health

Abstract

The invention discloses a child sleep health intelligent question-answering system based on a knowledge graph and a large language model, and relates to the technical field of artificial intelligence. Nodes of the knowledge graph represent entities of the child sleep information, and edges of the knowledge graph represent association relationships among the entities; utilizing a graph neural network to generate embedded vectors of entities and association relationships in the knowledge graph; converting a question natural language of a user for querying child sleep in the knowledge graph into a query vector for retrieving the knowledge graph; determining an enhanced context corresponding to the question natural language according to the similarity between the query vector and the embedded vector; and inputting the enhanced context and the question natural language into the large language model to obtain an answer result generated by the large language model. According to the method, the multi-source data of child sleep is integrated and structured into the knowledge graph, and then context-aware personalized suggestion generation is realized in combination with the large language model, so that the efficiency of acquiring child sleep information can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to an intelligent question-answering system for children's sleep health based on knowledge graphs and large language models. Background Art

[0002] Healthy sleep for children is crucial for their physical, cognitive, and emotional development. Medical research shows that good sleep quality is closely linked to children's learning ability, attention span, emotional regulation, and immune function. However, the complexity of children's behavioral sleep problems (such as difficulty falling asleep, frequent nighttime awakenings, and early awakenings) and the rapid development of medical knowledge present significant challenges for pediatric healthcare professionals and parents of children with behavioral sleep problems.

[0003] Currently, the knowledge sources that pediatric medical personnel rely on are relatively fragmented, and relevant information is scattered across different resources. Medical personnel need a lot of time to integrate the information; updates lag behind, and traditional publications have a long update cycle and cannot reflect the latest research progress in a timely manner; retrieval is difficult, and in a busy clinical environment, medical personnel find it difficult to quickly retrieve relevant recommendations for specific patient conditions; personalized application is challenging, and there is a lack of clear guidance when applying general guidelines to specific cases. Summary of the Invention

[0004] The main purpose of the embodiments of this application is to propose an intelligent question-and-answer system for children's sleep health based on knowledge graphs and large language models to improve the efficiency of obtaining children's sleep information.

[0005] To achieve the above objectives, one aspect of the present application proposes an intelligent question-answering system for children's sleep health based on a knowledge graph and a large language model. The system includes:

[0006] A knowledge graph construction module is used to construct a knowledge graph based on the children's sleep dataset; wherein each node in the knowledge graph represents an entity of children's sleep information, and the edge represents the relationship between each of the entities;

[0007] A vector embedding module, configured to generate embedding vectors of the entities and the relationships in the knowledge graph using a graph neural network;

[0008] a question conversion module, configured to convert a natural language question of a user querying the knowledge graph regarding children's sleep into a vector for retrieving the knowledge graph as a query vector;

[0009] A context enhancement module, configured to determine an enhanced context corresponding to the natural language of the question based on the similarity between the query vector and the embedding vector;

[0010] The answer module is used to input the enhanced context and the question natural language into the large language model to obtain the answer result generated by the large language model.

[0011] In some embodiments, the knowledge graph construction module includes:

[0012] a data acquisition unit, configured to acquire children's sleep data from multiple data sources;

[0013] a preprocessing unit, configured to perform text preprocessing on the child sleep data;

[0014] An entity recognition unit, configured to recognize and obtain each entity from the children's sleep data after text preprocessing;

[0015] A relationship extraction unit, configured to extract the association relationship between the entities;

[0016] A graph construction unit, configured to construct the knowledge graph using each entity as a node and the corresponding relationship as an edge;

[0017] A formatting unit is used to define attributes for each of the entities and the association relationships in the knowledge graph; wherein the attributes include knowledge source, evidence level, recommendation strength and release date.

[0018] In some embodiments, the vector embedding module includes:

[0019] A graph structure encoding unit, configured to determine the topological relationship between each of the nodes using a graph convolutional network;

[0020] A relationship processing unit, configured to process different types of association relationships using a relationship perception graph attention network;

[0021] A semantic fusion unit, configured to convert the description texts of each of the entities and the association relationships into an initial embedding vector to fuse the attribute information of each of the nodes;

[0022] A context-aware embedding generation unit is used to update each of the nodes and the edges through a multi-layer graph neural network to capture multi-hop relationships, so that the embedding of each node includes a wide range of contexts in the knowledge graph, thereby generating the embedding vector of fixed dimension.

[0023] In some embodiments, the question conversion module includes:

[0024] A question conversion unit is used to extract the target concept and intention in the question natural language, and then convert the question natural language into a vector for retrieving the knowledge graph as the query vector based on the target concept and the intention.

[0025] In some embodiments, the context enhancement module includes:

[0026] a multi-path knowledge retrieval unit, configured to retrieve the related entities based on the similarity between the query vector and the embedding vector, thereby determining a multi-hop path in the knowledge graph and potential related knowledge to obtain structured knowledge; and sorting the structured knowledge according to entity relevance and evidence level;

[0027] The context enhancement unit is configured to convert the retrieved structured knowledge into natural language, and extract the natural language corresponding to the most relevant structured knowledge as the enhanced context.

[0028] In some embodiments, the answer module includes:

[0029] An answering unit, configured to input the enhanced context and the question natural language into a Tongyi Qianwen large language model, and obtain an answer result generated by the Tongyi Qianwen large language model;

[0030] An answer adjustment unit is used to adjust the answer result according to the identity of the user.

[0031] In some embodiments, the system further comprises:

[0032] A verification module, configured to receive expert review information on the answer result and update the knowledge graph according to the review information;

[0033] A feedback module is used to receive the user's rating and feedback information on the answer result, and update the knowledge graph according to the rating and the feedback information.

[0034] To achieve the above objectives, another aspect of the present invention proposes an intelligent question-answering method for children's sleep health based on a knowledge graph and a large language model, the method comprising the following steps:

[0035] Constructing a knowledge graph based on a children's sleep dataset; wherein each node in the knowledge graph represents an entity of children's sleep information, and an edge represents an association relationship between each of the entities;

[0036] Generate embedding vectors of the entities and the relationships in the knowledge graph using a graph neural network;

[0037] Converting the natural language of the user's question about children's sleep in the knowledge graph into a vector for retrieving the knowledge graph as a query vector;

[0038] Determining, based on the similarity between the query vector and the embedding vector, an enhanced context corresponding to the natural language of the question;

[0039] The enhanced context and the question natural language are input into a large language model to obtain an answer result generated by the large language model.

[0040] To achieve the above-mentioned purpose, another aspect of an embodiment of the present application provides an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the above-mentioned method when executing the computer program.

[0041] To achieve the above-mentioned purpose, another aspect of an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program implements the above-mentioned method when executed by a processor.

[0042] The embodiments of the present application include at least the following beneficial effects:

[0043] This application can construct a knowledge graph based on a children's sleep dataset; wherein each node in the knowledge graph represents an entity of children's sleep information, and an edge represents an association relationship between each entity; a graph neural network is used to generate embedding vectors of entities and association relationships in the knowledge graph; the natural language of the user's query on the knowledge graph for children's sleep is converted into a vector for retrieving the knowledge graph as a query vector; based on the similarity between the query vector and the embedding vector, an enhanced context corresponding to the natural language of the question is determined; the enhanced context and the natural language of the question are input into a large language model to obtain an answer result generated by the large language model. This application can integrate multi-source data on children's sleep and structure it into a knowledge graph, and then combine it with a large language model to realize context-aware, evidence-based personalized recommendation generation, which can improve the efficiency of obtaining children's sleep information. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0045] Figure 1 A schematic diagram of the structure of the intelligent question-answering system for children's sleep health based on a knowledge graph and a large language model provided in an embodiment of the present application;

[0046] Figure 2 A flowchart of an intelligent question-answering method for children's sleep health based on a knowledge graph and a large language model provided in an embodiment of the present application;

[0047] Figure 3This is an example structural diagram of the intelligent question-and-answer system for children's sleep health based on a knowledge graph and a large language model provided in an embodiment of the present application;

[0048] Figure 4 An example question-and-answer flow chart provided for an embodiment of the present application;

[0049] Figure 5 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of the present application. They are merely examples of devices and methods consistent with some aspects of the embodiments of the present application as detailed in the appended claims.

[0051] It will be understood that the terms "first", "second", etc. used in this application may be used herein to describe various concepts, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the words "if" and "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0052] The terms "at least one", "plurality", "each", "any", etc. used in this application include "at least one", "two" or more, "plurality" or "each", "any" or "any one", "each" or "any one" in the context of the present invention, and "at least one" or "at least one" includes one, two or more, "plurality" or "any one" includes two or more, "each" or "each one" in the context of the present invention, and "any" or "any one

[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application pertains. The terms used herein are for the purpose of describing the embodiments of this application only and are not intended to limit this application.

[0054] Before describing the embodiments of the present application in detail, some of the related technologies involved in the embodiments of the present application are first described as follows:

[0055] At present, there are mainly the following technical solutions to the above problems:

[0056] Related technology 1: Clinical decision support system.

[0057] Evidence-based clinical decision systems such as UpToDate and DynaMed provide systematic medical information queries. These systems typically use keyword search mechanisms and organize medical knowledge in a hierarchical structure.

[0058] Disadvantages of related technology 1: Search results are often large amounts of text, requiring medical staff to spend time screening; lack of contextual understanding capabilities, unable to understand complex or ambiguous natural language queries; professional content targeting children's behavioral sleep problems is relatively limited.

[0059] Related technology 2: Traditional natural language processing system.

[0060] Natural language processing systems based on rules and statistical methods are used for medical text analysis and simple question-answering functions.

[0061] Disadvantages of Related Technology 2: Lack of deep semantic understanding capabilities; limited understanding of medical terminology; inability to handle complex context-related issues; rigid system response and difficulty adapting to different users' expressions.

[0062] Related technology 3: Independent large language model application.

[0063] In recent years, large language models such as GPT and Claude have begun to be applied in the medical field to provide natural language-based query services.

[0064] Disadvantages of Related Technology 3: Lack of structured integration of professional medical knowledge; May produce hallucinations, i.e., generating seemingly reasonable but actually inaccurate information; Difficulty tracing the source of information, which reduces the credibility of medical advice; Inability to obtain the latest medical research results and guidelines in a timely manner.

[0065] Related technology 4: Simple knowledge graph system.

[0066] Knowledge graphs in the medical field are used to organize medical concepts and their relationships.

[0067] Disadvantages of related technology 4: limited ability to interact with natural language; lack of integration of specialized knowledge on children's sleep problems; unfriendly query interface, usually requiring professional technical background; difficulty in handling ambiguous or non-standard expressions.

[0068] Technical problems with related technologies:

[0069] Based on the analysis of the above background technologies, the core technical problems that need to be solved in this application include: how to systematically integrate and structure the scattered knowledge of children's behavioral sleep problems; how to build a system that can understand natural language, accurately retrieve relevant knowledge and generate personalized suggestions; how to ensure that the suggestions provided by the system are based on the latest medical evidence and can be adaptively adjusted to specific patient conditions; how to make the system easy for pediatric medical personnel and parents to understand and use, and lower the threshold for acquiring professional knowledge; how to ensure that the information generated by the system is traceable, verifiable and reliable.

[0070] Reference Figure 1 , the embodiment of the present application provides a children's sleep health intelligent question-answering system based on a knowledge graph and a large language model, the system comprising:

[0071] A knowledge graph construction module is used to construct a knowledge graph based on the children's sleep dataset; wherein each node in the knowledge graph represents an entity of children's sleep information, and the edge represents the relationship between each of the entities;

[0072] A vector embedding module, configured to generate embedding vectors of the entities and the relationships in the knowledge graph using a graph neural network;

[0073] a question conversion module, configured to convert a natural language question of a user querying the knowledge graph regarding children's sleep into a vector for retrieving the knowledge graph as a query vector;

[0074] A context enhancement module, configured to determine an enhanced context corresponding to the natural language of the question based on the similarity between the query vector and the embedding vector;

[0075] The answer module is used to input the enhanced context and the question natural language into the large language model to obtain the answer result generated by the large language model.

[0076] Optionally, the knowledge graph construction module includes:

[0077] a data acquisition unit, configured to acquire children's sleep data from multiple data sources;

[0078] a preprocessing unit, configured to perform text preprocessing on the child sleep data;

[0079] An entity recognition unit, configured to recognize and obtain each entity from the children's sleep data after text preprocessing;

[0080] A relationship extraction unit, configured to extract the association relationship between the entities;

[0081] A graph construction unit, configured to construct the knowledge graph using each entity as a node and the corresponding relationship as an edge;

[0082] A formatting unit is used to define attributes for each of the entities and the association relationships in the knowledge graph; wherein the attributes include knowledge source, evidence level, recommendation strength and release date.

[0083] Optionally, the vector embedding module includes:

[0084] A graph structure encoding unit, configured to determine the topological relationship between each of the nodes using a graph convolutional network;

[0085] A relationship processing unit, configured to process different types of association relationships using a relationship perception graph attention network;

[0086] A semantic fusion unit, configured to convert the description texts of each of the entities and the association relationships into an initial embedding vector to fuse the attribute information of each of the nodes;

[0087] A context-aware embedding generation unit is used to update each of the nodes and the edges through a multi-layer graph neural network to capture multi-hop relationships, so that the embedding of each node includes a wide range of contexts in the knowledge graph, thereby generating the embedding vector of fixed dimension.

[0088] Optionally, the question conversion module includes:

[0089] A question conversion unit is used to extract the target concept and intention in the question natural language, and then convert the question natural language into a vector for retrieving the knowledge graph as the query vector based on the target concept and the intention.

[0090] Optionally, the context enhancement module includes:

[0091] a multi-path knowledge retrieval unit, configured to retrieve the related entities based on the similarity between the query vector and the embedding vector, thereby determining a multi-hop path in the knowledge graph and potential related knowledge to obtain structured knowledge; and sorting the structured knowledge according to entity relevance and evidence level;

[0092] The context enhancement unit is configured to convert the retrieved structured knowledge into natural language, and extract the natural language corresponding to the most relevant structured knowledge as the enhanced context.

[0093] Optionally, the answer module includes:

[0094] An answering unit, configured to input the enhanced context and the question natural language into a Tongyi Qianwen large language model, and obtain an answer result generated by the Tongyi Qianwen large language model;

[0095] An answer adjustment unit is used to adjust the answer result according to the identity of the user.

[0096] Optionally, the system further comprises:

[0097] A verification module, configured to receive expert review information on the answer result and update the knowledge graph according to the review information;

[0098] A feedback module is used to receive the user's rating and feedback information on the answer result, and update the knowledge graph according to the rating and the feedback information.

[0099] Reference Figure 2 This embodiment also provides a method for intelligent question-answering of children's sleep health based on a knowledge graph and a large language model. The method may include but is not limited to S1 to S5, as follows:

[0100] S1: Construct a knowledge graph based on the children's sleep dataset; wherein each node in the knowledge graph represents an entity of children's sleep information, and the edge represents the relationship between each entity;

[0101] S2: Generate embedding vectors of the entities and the relationships in the knowledge graph using a graph neural network;

[0102] S3: Converting the natural language of the user's question about children's sleep in the knowledge graph into a vector for retrieving the knowledge graph as a query vector;

[0103] S4: Determine, based on the similarity between the query vector and the embedding vector, an enhanced context corresponding to the natural language of the question;

[0104] S5: Input the enhanced context and the question natural language into a large language model to obtain an answer result generated by the large language model.

[0105] Next, the solution of the embodiment of the present application will be introduced and explained in detail with reference to specific application examples.

[0106] This embodiment proposes an automated query answering system based on the fusion of knowledge graph retrieval enhancement generation (GraphRAG) and large language model (LLM), specifically designed to improve the ability of pediatric medical providers to identify and manage children's behavioral sleep problems. The system integrates a structured knowledge graph of children's behavioral sleep problems with the "Thousand Questions" large language model to achieve context-aware, evidence-based personalized recommendation generation. The system architecture of this embodiment is as follows: Figure 3 As shown, an example question-answering flow chart based on the system of this embodiment is as follows Figure 4 shown.

[0107] The system's core solutions include: a children's sleep health knowledge graph that systematically integrates information on children's behavioral sleep problems and best practices from textbooks, peer-reviewed literature, evidence-based clinical decision systems, and clinical guidelines; graph neural network embedding that generates embedding vectors for each node and edge in the knowledge graph, capturing the semantic and structural context within the graph; a retrieval enhancement generation engine that retrieves relevant knowledge graph information based on user queries and enhances the input context of the large language model; a natural language understanding and generation module based on "Tongyi Qianwen" that processes natural language queries and combines retrieved structured knowledge to generate personalized and accurate answers; a verification and feedback mechanism that ensures that the recommendations generated by the system comply with medical ethics and clinical practice standards.

[0108] Next, the solution of this embodiment is described in detail.

[0109] 1. Construction of children's sleep health knowledge map.

[0110] The knowledge graph construction process includes the following steps:

[0111] The first step is knowledge source collection and preprocessing. Textbooks, peer-reviewed literature, clinical decision systems and clinical guidelines related to children's behavioral sleep problems are systematically collected, and the text is preprocessed, including word segmentation, part-of-speech tagging, named entity recognition, etc.

[0112] Secondly, entity and relationship extraction is performed, using natural language processing technology to identify key entities, such as the types of behavioral sleep problems, symptoms, intervention measures, applicable age, etc., and extract complex relationships between entities, such as "cause", "improve", "taboo", etc., to construct a directed labeled graph, in which nodes represent entities and edges represent relationships.

[0113] Then, knowledge fusion and verification are carried out to integrate knowledge from different sources, resolve conflicts and redundancies, verify the accuracy and completeness of the knowledge graph through medical experts, establish a traceability mechanism for knowledge entries, and retain the original information source.

[0114] Finally, the knowledge graph is formatted to store the knowledge in a graph database format (such as Neo4j), defining attributes for each entity and relationship, such as evidence level, recommendation strength, publication date, etc.

[0115] 2. Graph neural network embedding generation.

[0116] In order to effectively integrate the knowledge graph with the large language model, this embodiment uses Graph Neural Networks (GNN) to generate embedding vectors for entities and relationships in the knowledge graph:

[0117] In terms of graph structure encoding, Graph Convolutional Networks (GCN) are used to capture the topological relationships between nodes, and Relational Graph Attention Networks (RGANs) are used to process different types of relationships.

[0118] In terms of semantic information fusion, the text descriptions of entities and relationships are converted into initial embedding vectors, and the attribute information of nodes, such as entity type, evidence level, etc., are fused, and the representation of nodes and edges is updated through the message passing mechanism.

[0119] In terms of context-aware embedding generation, the representations of nodes and edges are iteratively updated through a multi-layer graph neural network to capture multi-hop relationships, so that the embedding of each node contains its extensive context in the graph, and a fixed-dimensional embedding vector is generated for subsequent similarity calculation and retrieval.

[0120] 3. Retrieval enhancement generation.

[0121] The Retrieval-Augmented Generation (RAG) module is a bridge connecting the knowledge graph and the large language model:

[0122] Query understanding and conversion includes analyzing user natural language queries, extracting key concepts and intents, and converting queries into vector representations suitable for knowledge graph retrieval.

[0123] Multi-path knowledge retrieval includes retrieving related entities based on embedding vector similarity, exploring multi-hop paths in the graph structure, discovering potentially relevant knowledge, and sorting retrieval results according to entity relevance and evidence level.

[0124] Context enhancement involves converting the retrieved structured knowledge into natural language descriptions, selectively extracting the most relevant knowledge points to avoid information overload, and constructing enhanced context input that includes query-related expertise.

[0125] 4. Large language model integration and response generation.

[0126] This example uses the "Tongyi Qianwen" large language model as the core of natural language understanding and generation:

[0127] The enhanced input processing stage receives the enhanced context and original user query from the RAG module and analyzes the specific needs and context of the query.

[0128] The expertise-aware generation process generates professional and accurate answers based on enhanced context, ensuring that the answers are directly targeted at user questions, avoiding irrelevant content, and adjusting the professionalism and comprehensibility of the answers based on the user role (medical professional or parent).

[0129] Evidence support and source tracking features provide evidence support for key recommendations, preserve knowledge sources, enable users to access original literature, and clearly distinguish between evidence-based recommendations and general recommendations.

[0130] The uncertainty handling mechanism clearly expresses the level of certainty of knowledge (such as strong recommendation, limited evidence support, etc.) and clearly states the limitations when facing questions that are beyond the scope of knowledge.

[0131] 5. System verification and feedback mechanism.

[0132] To ensure the reliability and practicality of the system, this embodiment includes the following verification and feedback mechanisms:

[0133] Expert Review Medical experts regularly review a sample of system-generated responses to assess their accuracy, completeness, and applicability.

[0134] User feedback collection allows users to rate answers and provide feedback, collecting common points of confusion for specific types of questions.

[0135] System self-assessment can automatically assess possible uncertainties in responses and identify responses that require further verification by experts.

[0136] The knowledge base is dynamically updated to update the knowledge graph according to new research results and clinical guidelines, and optimize retrieval and generation strategies based on user feedback.

[0137] In summary, this embodiment includes the following technical solutions:

[0138] (1) Knowledge graph and large language model fusion architecture: Innovatively integrate the structured knowledge graph in the field of children's sleep health with the "Thousand Questions" large language model to form a new type of medical intelligent question-answering system.

[0139] (2) Application of GraphRAG technology: Graph Retrieval Enhanced Generation (GraphRAG) technology was applied for the first time in the field of pediatric medicine, enhancing the understanding and response capabilities of large language models through graph structure retrieval.

[0140] (3) Graph neural network embedding: Graph neural networks are used to generate semantically rich embedding vectors for entities and relationships in the knowledge graph, capturing the complex semantic and structural context within the graph.

[0141] (4) Multi-path knowledge retrieval: Exploring multi-hop paths in the knowledge graph based on user queries, discovering potentially relevant medical knowledge, and providing a comprehensive perspective on the problem.

[0142] (5) Context-aware answer generation: Automatically adjust the professionalism and operability of the answer according to the user role (medical professional or parent) to achieve personalized knowledge transfer.

[0143] More specifically:

[0144] (1) Knowledge graph construction method: A knowledge graph construction method specifically for children's behavioral sleep problems, including the specific processes of entity and relationship extraction, knowledge fusion and verification.

[0145] (2) Graph embedding and large language model integration method: a technical solution that integrates the structured knowledge embedding generated by graph neural networks with the input context of a large language model.

[0146] (3) Specific implementation of retrieval enhancement generation: semantic analysis based on user queries, multi-path knowledge retrieval, retrieval result sorting and screening, and implementation methods of context enhancement.

[0147] (4) User role-aware answer generation: A method that automatically adjusts the professionalism of answers, terminology usage, and actionability of suggestions based on whether the user is a medical professional or a parent.

[0148] (5) Evidence level labeling and traceability mechanism: Automatically label the evidence level and recommendation strength for medical recommendations generated by the system and retain the knowledge source information.

[0149] (6) Knowledge graph dynamic update mechanism: a method to continuously update the content and structure of the knowledge graph based on new research results, clinical guidelines and user feedback.

[0150] This example combines knowledge graph retrieval enhancement and generation technology with a large language model to provide an innovative solution for the identification and management of behavioral sleep problems in children, with the following significant advantages and benefits:

[0151] 1. Compared with the existing technology, this embodiment achieves a systematic integration of knowledge related to behavioral sleep problems in children. The comprehensiveness of the knowledge in this embodiment is reflected in the integration of professional knowledge from multiple sources such as textbooks, peer-reviewed literature, clinical decision-making systems, and clinical guidelines to form a comprehensive knowledge network. The scattered information is organized into interconnected entities and relationship networks through the knowledge graph, making the logical relationship between knowledge clear and visual, and realizing structured representation. The system captures the complex semantic associations between knowledge through graph neural networks, going beyond simple keyword matching and realizing deep knowledge representation. Each knowledge entry retains its source information, ensuring that medical advice can be traced back to reliable professional literature, and providing a complete evidence tracing mechanism.

[0152] 2. Breakthrough application of retrieval-enhanced generation.

[0153] This embodiment uses graph retrieval enhancement generation technology to achieve multiple technological breakthroughs. The system has context-aware retrieval capabilities, and intelligently retrieves relevant knowledge based on the actual context of the user's query, surpassing traditional keyword matching. Through multi-hop exploration of the graph structure, potentially relevant medical knowledge is discovered, providing a comprehensive perspective on the problem and realizing multi-path exploration. The system realizes precise knowledge fusion, selecting only the most relevant knowledge points to enhance the large language model input, avoiding interference from irrelevant information. In addition, the system can also identify blank areas in the knowledge graph, clearly indicating the uncertainty of information, and effectively identifying knowledge gaps.

[0154] 3. Improve the efficiency and accuracy of medical staff's decision-making.

[0155] Compared with existing technical solutions, this embodiment provides medical personnel with more efficient and accurate decision-making support. The system can shorten the decision-making time, and medical staff can quickly obtain professional knowledge for specific patient conditions. Preliminary tests show that the average query response time does not exceed 3 seconds. The system provides professional advice in a natural language interactive manner, which reduces the difficulty of acquiring sleep medicine expertise and effectively lowers the threshold for acquiring professional knowledge. Based on a unified knowledge graph and standardized processing procedures, the consistency of diagnosis and treatment among different medical staff is improved. The suggestions generated by the system clearly mark the level of evidence and the strength of recommendation, helping medical staff make evidence-based medical decisions and providing evidence-supported suggestions.

[0156] 4. Personalized suggestions and flexible responses.

[0157] This embodiment addresses the shortcomings of traditional methods in terms of personalization. The system can adjust recommendations based on factors such as the patient's age, developmental stage, and comorbidities, fully considering the patient's characteristics. For ineffective methods that have been tried, the system can provide alternative intervention plans, achieving adaptation of intervention strategies. The system adjusts the professionalism and operability of answers based on whether the user is a professional medical staff or a parent, and has the ability to perceive the user's role. In addition, the system supports queries and answers in multiple languages, meeting the needs of different groups of people and providing multilingual support.

[0158] 5. System scalability and knowledge updating.

[0159] This embodiment designs an innovative knowledge update mechanism. The system supports the continuous integration of new research results and clinical guidelines, maintains the timeliness of knowledge, and realizes the dynamic update of the knowledge graph. Through the user feedback mechanism, the retrieval and generation strategies are continuously optimized to improve system performance, forming a circular optimization mechanism. The core architecture can be expanded to other pediatric health problem areas, such as developmental behavioral problems, nutritional health, etc., and has good field scalability. The system can also integrate the local clinical pathways and standards of medical institutions to achieve personalized deployment and support the integration of local knowledge in hospitals.

[0160] 6. Clinical verification results.

[0161] In the pilot study, the system demonstrated significant practical value. After using the system, medical staff's confidence in managing sleep problems increased significantly (the score increased from 5.2 / 10 to 8.7 / 10), resulting in improved medical staff satisfaction. Patient consultation time was shortened by an average of 24%, while more comprehensive information was provided, improving diagnosis and treatment efficiency. The standardization of the hospital's sleep problem handling process was significantly improved, reducing differences in diagnosis and treatment between individual doctors and promoting standardized practice. The personalized parent education materials generated by the system improved parent compliance and intervention effectiveness, enhancing the effectiveness of parent education.

[0162] In addition, this application also provides the following alternatives:

[0163] Collaborative consulting system based on multi-agent:

[0164] 1. System Architecture

[0165] By building a cluster of expert agents, each focused on a specific sleep problem area (such as difficulty falling asleep, night terrors, and sleep-disordered breathing), a coordinator agent was designed to analyze user questions, call relevant expert agents, and integrate responses from multiple sources. A conversation history agent was also established to track the user's consultation process and provide consistent recommendations.

[0166] 2. Search Enhancement: Utilize a hybrid search strategy, combining semantic search and structured queries (e.g., SPARQL). Implement causal-specific searches, automatically constructing causal chains for "why" questions. Introduce differentiated search scopes, employing different depths and breadths of knowledge retrieval for both professional and general users.

[0167] Although this alternative solution is innovative compared to the technical solution of this embodiment, it has the following limitations: First, this solution significantly increases the complexity of the system. Building and maintaining multiple professional agents is costly, and each agent requires an independent knowledge base and reasoning mechanism. The coordination and information transmission mechanism between multiple agents is complex, and information gaps or conflicts are prone to occur. When the system generates an error, it becomes more difficult to locate the source of the problem because the problem may involve the interaction process of multiple agents.

[0168] Knowledge consistency is also a significant challenge. Distributing knowledge across multiple agents can easily lead to fragmentation, and improper management of knowledge boundaries between agents can create reasoning blind spots. Medical knowledge updates require coordinated, synchronized updates from multiple agents, otherwise conflicting recommendations may result. Furthermore, knowledge overlap between agents in different fields inevitably increases storage and maintenance costs. Furthermore, multi-agent systems face responsiveness issues. The coordinator requires time to analyze questions, dispatch expert agents, and synthesize responses, increasing system response time. Running multiple specialized agents simultaneously requires additional computing resources, especially in high-concurrency scenarios. For simple questions, initiating multi-agent collaboration can actually reduce system efficiency and waste resources. From a user experience perspective, multi-agent collaboration can increase user wait times. The splicing of content generated by multiple agents can lack natural and fluent expression. Users may need to interact with multiple "experts" simultaneously, increasing cognitive burden. Finally, deployment and implementation present obstacles. The system is highly technically dependent, requiring a mature agent collaboration framework and scheduling mechanism. Complex multi-agent decision-making processes can reduce system interpretability. Compared to the tight integration of a single knowledge graph and a large language model, a multi-agent approach may be an over-engineered solution to complex problems.

[0169] In comparison, the retrieval enhancement generation model architecture based on the unified knowledge graph in the technical solution of this embodiment is simpler, more consistent and easier to maintain, especially in the medical professional field, where the accuracy and consistency of knowledge are particularly important.

[0170] In addition, this application also provides more specific implementation plans, which are as follows:

[0171] Example 1: Sleep health consultation scenario for pediatric medical staff.

[0172] This example demonstrates how the system can assist pediatric medical staff in answering inquiries about sleep problems in 3-year-old children:

[0173] The system deployment environment includes the hardware environment (deployed on a cloud server, configured with an 8-core CPU, 32GB of memory, and 500GB of SSD storage), the software environment (Linux operating system, Neo4j graph database, and PyTorch deep learning framework), and the network environment (providing a Web API interface through the HTTPS protocol to support hospital intranet access).

[0174] The knowledge graph construction collects 15 pediatric sleep medicine textbooks, 200+ peer-reviewed articles, and 5 clinical guideline systems, extracts 1,500+ behavioral sleep problem-related entities and 3,000+ relationships, and constructs a knowledge graph, which includes nodes and their relationships such as sleep problem classification, symptom manifestations, assessment methods, and intervention strategies.

[0175] User query example: "A 3-year-old boy takes more than an hour to fall asleep each night and must have his parents by his bedside, otherwise he will cry and make a fuss. Even after falling asleep, he will wake up 2-3 times during the night and need his parents to comfort him each time. His parents have tried storytelling and playing white noise, but the results are limited. What suggestions can be made to improve this situation?"

[0176] The system processing flow includes query understanding (identifying key entities, attempted methods, and query intent), knowledge graph retrieval (retrieval of relevant nodes and intervention strategies), context enhancement (integration of relevant knowledge points), and answer generation (generation of structured answers).

[0177] The system output example includes a professional assessment, evidence-based intervention recommendations, and specific implementation steps. The system can identify this as a typical combination of sleep-related disorders and sleep-onset association disorders, recommend a progressive independent sleep training method suitable for a 3-year-old, and provide a specific implementation plan, including progressive separation techniques and positive reinforcement strategies. The system also labels the strength of evidence for each recommendation based on the level of evidence in the knowledge graph and provides a replicable intervention process and parent education points tailored to the needs of medical professionals.

[0178] Example 2: Personalized suggestions for parents in usage scenarios.

[0179] This example demonstrates how the system can provide easy-to-understand personalized advice to parents of a 6-year-old child with sleep problems:

[0180] Example user query: "My 6-year-old daughter has been resisting going to bed for the past month, delaying it by an hour or two every night. She says she's worried there's a monster in her closet, and we've checked it numerous times. She behaves normally in school during the day, but becomes extremely anxious before bed. How can I help her?"

[0181] The system processing flow includes user role identification (adjusting the professionalism and expression of the answer), knowledge graph retrieval and analysis (retrieval of relevant nodes and analysis) and answer generation strategy (providing step-by-step suggestions using calm and supportive language).

[0182] The system's sample output specifically addresses parental concerns, explaining the problem and solution in layman's terms. Key content includes a user-friendly explanation of the common and developmental causes of bedtime fears in six-year-olds, practical coping strategies like "Monster Spray" and "Brave Diary," guidance on establishing a stable bedtime routine and environmental adjustments, parental techniques for managing emotions and responding positively, and clear guidance on seeking professional help.

[0183] The system is able to assess the nature and severity of the problem based on subtle clues in the parents' statements (such as "started in the last month" and "normal performance at school during the day"), and provide suggestions that are consistent with professional guidelines but expressed in a way that parents can understand.

[0184] Example 3: Knowledge management and training scenarios in medical institutions.

[0185] This example demonstrates how the system supports knowledge management and medical staff training in medical institutions:

[0186] Application scenarios include deployment in a pediatric residency training program at a children's hospital, system integration into the hospital's electronic medical record system as a clinical decision support tool, and providing a case-based learning model as a training resource.

[0187] The system configuration includes knowledge graph expansion (incorporating the hospital's local clinical pathways and standards), user authority classification (providing information of varying depths based on the roles of medical staff), and integration of patient electronic medical record data interfaces (supporting contextualized recommendations).

[0188] A typical usage process is for a resident to inquire about the assessment methods for treating patients with complex sleep problems: "An 8-year-old patient with a history of attention deficit hyperactivity disorder currently complains of difficulty falling asleep and having frequent dreams at night. Conventional sleep hygiene education has limited effect. What additional assessments and possible intervention directions are needed?"

[0189] Systematic processing includes identifying key factors, retrieving relevant knowledge pathways, integrating the hospital's existing assessment processes and referral criteria, and providing references to evidence-based intervention strategies and professional guidelines.

[0190] The answers include a structured assessment process (including recommended screening tools and assessment scales), synergistic considerations of medication and behavioral interventions, clear criteria for referral to a psychiatrist or sleep specialist, special considerations for sleep interventions suitable for patients with ADHD, and the strength of relevant medical evidence and the level of recommendation.

[0191] In terms of training support functions, the system can generate simulated cases and corresponding processing paths, support case-based learning modules, generate decision trees and processing flowcharts, provide knowledge self-test questions and case analysis exercises, and adjust the content depth and difficulty according to the user's knowledge level and learning progress.

[0192] This example demonstrates the application value of the system in clinical teaching and knowledge management, emphasizing its potential as a learning tool to help medical institutions standardize the treatment process for children's sleep problems.

[0193] The present application also provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method of the present application. The electronic device can be any smart terminal, such as a tablet computer or an in-vehicle computer.

[0194] It can be understood that the contents of the above method embodiments are all applicable to the embodiments of the present device, the functions specifically implemented by the embodiments of the present device are the same as those of the method of the present application, and the beneficial effects achieved are also the same as those achieved by the method of the present application.

[0195] See also Figure 5 , Figure 5 The hardware structure of an electronic device according to another embodiment is shown. The electronic device includes:

[0196] The processor 501 may be implemented as a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is configured to execute relevant programs to implement the technical solutions provided in the embodiments of the present application.

[0197] The memory 502 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 502 can store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 502 and is called by the processor 501 to execute the methods of the embodiments of this application.

[0198] Input / output interface 503, used to implement information input and output;

[0199] Communication interface 504, used to implement communication interaction between this device and other devices, which can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WiFi, Bluetooth, etc.);

[0200] Bus 505 , which transmits information between various components of the device (e.g., processor 501 , memory 502 , input / output interface 503 , and communication interface 504 );

[0201] The processor 501 , the memory 502 , the input / output interface 503 and the communication interface 504 are connected to each other in communication within the device via a bus 505 .

[0202] An embodiment of the present application further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the method of the present application is implemented.

[0203] It can be understood that the contents of the above method embodiments are all applicable to the present storage medium embodiment, the functions specifically implemented by the present storage medium embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0204] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0205] The embodiments described in the embodiments of this application are intended to more clearly illustrate the technical solutions of the embodiments of this application and do not constitute a limitation on the technical solutions provided by the embodiments of this application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0206] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0207] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, i.e., they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment.

[0208] Those skilled in the art will appreciate that all or some of the steps in the methods, systems, and functional modules / units in the devices disclosed above may be implemented as software, firmware, hardware, or appropriate combinations thereof.

[0209] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0210] It should be understood that in this application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0211] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but are not intended to limit the scope of the present invention. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and essence of the present invention should be within the scope of the present invention.

Claims

1. The intelligent question-answering system for children's sleep health based on knowledge graph and large language model is characterized by: The system comprises: A knowledge graph construction module is used to construct a knowledge graph based on the children's sleep dataset; wherein each node in the knowledge graph represents an entity of children's sleep information, and the edge represents the relationship between each of the entities; A vector embedding module, configured to generate embedding vectors of the entities and the relationships in the knowledge graph using a graph neural network; a question conversion module, configured to convert a natural language question of a user querying the knowledge graph regarding children's sleep into a vector for retrieving the knowledge graph as a query vector; A context enhancement module, configured to determine an enhanced context corresponding to the natural language of the question based on the similarity between the query vector and the embedding vector; The answer module is used to input the enhanced context and the question natural language into the large language model to obtain the answer result generated by the large language model.

2. The children's sleep health intelligent question-answering system based on knowledge graph and large language model according to claim 1 is characterized in that: The knowledge graph construction module includes: a data acquisition unit, configured to acquire children's sleep data from multiple data sources; a preprocessing unit, configured to perform text preprocessing on the child sleep data; An entity recognition unit, configured to recognize and obtain each entity from the children's sleep data after text preprocessing; A relationship extraction unit, configured to extract the association relationship between the entities; A graph construction unit, configured to construct the knowledge graph using each entity as a node and the corresponding relationship as an edge; A formatting unit is used to define attributes for each of the entities and the association relationships in the knowledge graph; wherein the attributes include knowledge source, evidence level, recommendation strength and release date.

3. The children's sleep health intelligent question-answering system based on knowledge graph and large language model according to claim 1 is characterized in that: The vector embedding module includes: A graph structure encoding unit, configured to determine the topological relationship between each of the nodes using a graph convolutional network; A relationship processing unit, configured to process different types of association relationships using a relationship perception graph attention network; A semantic fusion unit, configured to convert the description texts of each of the entities and the association relationships into an initial embedding vector to fuse the attribute information of each of the nodes; A context-aware embedding generation unit is used to update each of the nodes and the edges through a multi-layer graph neural network to capture multi-hop relationships, so that the embedding of each node includes a wide range of contexts in the knowledge graph, thereby generating the embedding vector of fixed dimension.

4. The children's sleep health intelligent question-answering system based on knowledge graph and large language model according to claim 1 is characterized in that: The question conversion module includes: A question conversion unit is used to extract the target concept and intention in the question natural language, and then convert the question natural language into a vector for retrieving the knowledge graph as the query vector based on the target concept and the intention.

5. The children's sleep health intelligent question-answering system based on knowledge graph and large language model according to claim 1 is characterized in that: The context enhancement module includes: a multi-path knowledge retrieval unit, configured to retrieve the related entities based on the similarity between the query vector and the embedding vector, thereby determining a multi-hop path in the knowledge graph and potential related knowledge to obtain structured knowledge; and sorting the structured knowledge according to entity relevance and evidence level; The context enhancement unit is configured to convert the retrieved structured knowledge into natural language, and extract the natural language corresponding to the most relevant structured knowledge as the enhanced context.

6. The children's sleep health intelligent question-answering system based on knowledge graph and large language model according to claim 1 is characterized in that: The answer module includes: An answering unit, configured to input the enhanced context and the question natural language into a Tongyi Qianwen large language model, and obtain an answer result generated by the Tongyi Qianwen large language model; An answer adjustment unit is used to adjust the answer result according to the identity of the user.

7. The children's sleep health intelligent question-answering system based on knowledge graph and large language model according to any one of claims 1 to 6, characterized in that: The system further comprises: A verification module, configured to receive expert review information on the answer result and update the knowledge graph according to the review information; A feedback module is used to receive the user's rating and feedback information on the answer result, and update the knowledge graph according to the rating and the feedback information.

8. An intelligent question-answering method for children's sleep health based on knowledge graph and large language model, characterized by: The method comprises the following steps: Constructing a knowledge graph based on a children's sleep dataset; wherein each node in the knowledge graph represents an entity of children's sleep information, and an edge represents an association relationship between each of the entities; Generate embedding vectors of the entities and the relationships in the knowledge graph using a graph neural network; Converting the natural language of the user's question about children's sleep in the knowledge graph into a vector for retrieving the knowledge graph as a query vector; Determining, based on the similarity between the query vector and the embedding vector, an enhanced context corresponding to the natural language of the question; The enhanced context and the question natural language are input into a large language model to obtain an answer result generated by the large language model.

9. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the method according to claim 8 when executing the computer program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to claim 8 is implemented.

Images