High blood pressure population knowledge recommendation method and system based on large language model multi-agent

Abstract

The application belongs to the technical field of medical health, and particularly relates to a hypertension group knowledge recommendation method and system based on a large language model multi-agent, which can provide personalized and accurate knowledge recommendation services for hypertension users by constructing a hypertension knowledge question and answer pair library, finely extracting hypertension knowledge question and answer pair information, medical short video information and user personal health portraits, performing video content quality evaluation and matching degree evaluation, and adjusting in combination with user platform interaction data, meet the needs of hypertension patients for health knowledge, improve the accuracy and effectiveness of knowledge recommendation, and help patients with self-health management.

Description

Technical Field

[0001] This invention belongs to the field of medical and health technology, specifically relating to a knowledge recommendation method and system for a hypertension population based on a large language model multi-agent system. Background Technology

[0002] In the context of current global economic integration and technological innovation, big data technology has deeply integrated into all aspects of human life, especially in the healthcare field, which is at a critical juncture of transformation, gradually entering a new era driven by big data. Firstly, medical institutions are using big data analytics systems to deeply mine clinical databases, thereby developing more precise and efficient treatment plans for patients. Secondly, the widespread deployment of electronic medical record systems has digitized patients' medical records, and intelligent matching technology links individual cases with medical teams, ensuring timely and targeted medical services. Furthermore, the combination of big data and cloud computing technologies enables intelligent cloud-based management and optimization of medical resources. Finally, through in-depth analysis of individual differences in data, medical professionals identify specific patient needs and provide personalized medical services, realizing a new "people-centered" healthcare model.

[0003] In the field of chronic disease management, such as hypertension, the public's demand for health knowledge continues to grow. However, the acquisition and application of hypertension health knowledge currently face many challenges. On the one hand, the internet is flooded with a large amount of hypertension-related information of varying quality, making it difficult for users to judge the professionalism and credibility of this information, and they are easily influenced by unprofessional or misleading content. On the other hand, medical and health knowledge exists in various forms such as text, audio, and video, with unstructured or semi-structured data dominating, making it difficult to manage and utilize systematically. For example, the text of doctor-patient Q&A contains rich implicit knowledge, while short medical videos attract users with their intuitive and vivid characteristics, but this content is difficult for machines to effectively identify and extract. In addition, existing knowledge recommendation systems generally lack personalization and are unable to provide accurate and appropriate knowledge content based on the individual differences in the health status of hypertension patients (including blood pressure levels, comorbidities, lifestyle habits, disease stages, etc.). For example, there are significant differences in the required knowledge focus and intervention recommendations for hypertension patients at different risk levels, and traditional generalized recommendation methods cannot meet this refined need. Summary of the Invention

[0004] This invention provides a knowledge recommendation method and system for hypertension patients based on a large language model multi-agent approach. By constructing a hypertension knowledge question-and-answer database, refining the extraction of hypertension knowledge question-and-answer information, medical short video information, and user personal health profiles, and conducting video content quality and matching degree assessments, combined with adjustments based on user platform interaction data, it can provide personalized and accurate knowledge recommendation services for hypertension patients, meet their health knowledge needs, improve the accuracy and effectiveness of knowledge recommendations, and help patients manage their own health.

[0005] A knowledge recommendation method for hypertension groups based on a large language model and multi-agent system includes:

[0006] Based on hypertension knowledge, a hypertension knowledge question-and-answer pair database and a hypertension risk level determination system were constructed. Based on the hypertension knowledge question-and-answer pair database, the knowledge question-and-answer pair texts were classified and structured knowledge question-and-answer pair information was extracted.

[0007] Collect and download short medical videos about hypertension, extract video content, publisher type, video publication time and interaction metadata, and establish video quality assessment standards based on video content, publisher type and interaction metadata to generate video quality scores;

[0008] Based on users' personal health information and user platform interaction data, a user's personal health profile is constructed, and combined with the hypertension risk level assessment system, the current user's hypertension risk level is determined.

[0009] The system performs multi-dimensional matching between users' personal health profiles and structured knowledge question-and-answer pairs, as well as video content, to calculate a matching score. The multi-dimensional matching includes semantic similarity matching and hypertension risk level alignment matching.

[0010] Based on video quality score, matching score, video release time, and user platform interaction data, a personalized video recommendation list is generated, and video recommendations are optimized based on user feedback.

[0011] By constructing a hypertension knowledge question-and-answer database, refining the extraction of hypertension knowledge question-and-answer information, medical short video information, and user personal health profiles, and conducting video content quality and matching degree assessments, combined with adjustments based on user platform interaction data, we can provide personalized and accurate knowledge recommendation services for hypertension users, meet the health knowledge needs of hypertension patients, improve the accuracy and effectiveness of knowledge recommendations, and help patients manage their own health.

[0012] Furthermore, based on hypertension knowledge, a hypertension knowledge question-and-answer pair database and a hypertension risk level determination system are constructed. According to the hypertension knowledge question-and-answer pair database, the knowledge question-and-answer pair texts are classified and structured knowledge question-and-answer pair information is extracted, including:

[0013] Based on hypertension knowledge, hypertension knowledge question-answer pairs are defined, knowledge entity types and relationship types between entities are extracted, and a hypertension knowledge question-answer pair database is constructed; the entity types include: medication precautions, lifestyle interventions, symptoms, lifestyle habits, and efficacy.

[0014] Based on the patient's blood pressure level, comorbidities, and lifestyle habits, a comprehensive risk assessment is conducted to classify hypertension risk levels and construct a hypertension risk level determination system.

[0015] Based on a hypertension knowledge question-and-answer pair database, the knowledge question-and-answer pair texts are classified, and structured knowledge question-and-answer pair information is extracted based on the classification results. The knowledge question-and-answer pair text classifications include popular science question-and-answer pairs and clinical consultation and health management question-and-answer pairs. The clinical consultation and health management question-and-answer pairs include complications and risks, health management, mental health, disease assessment, and treatment and medication.

[0016] By defining hypertension knowledge question-and-answer pairs, a knowledge hypertension knowledge question-and-answer pair database is constructed to provide basic data for subsequent knowledge processing and recommendation. Furthermore, by classifying the text through knowledge question-and-answer pairs, the knowledge is structured, which facilitates subsequent knowledge matching and recommendation and improves the efficiency of knowledge utilization.

[0017] Furthermore, based on the patient's blood pressure level, comorbidities, and lifestyle habits, a comprehensive risk assessment is conducted to classify hypertension risk levels and construct a hypertension risk level determination system, including:

[0018] Obtain the patient's blood pressure level score; the blood pressure level score is assigned a value based on the blood pressure value.

[0019] Obtain a score for the patient's comorbidities; the score for comorbidities is determined by the degree of prognostic impact of high-risk disease factors.

[0020] Obtain the patient's lifestyle habit score; the patient's lifestyle habit score is determined through undesirable behaviors and their degree of harm.

[0021] The patient's comprehensive risk score is calculated by combining the patient's blood pressure level score, the patient's comorbidity score, and the patient's lifestyle habit score.

[0022] The patient's hypertension risk level is determined based on a weighted calculation of the patient's comprehensive risk score and the hypertension risk level classification criteria. The hypertension risk level includes low risk, intermediate risk, intermediate-high risk, high risk, high-very high risk, and very high risk.

[0023] By considering multiple factors to assess patient risk and combining them with hypertension risk grading standards, the patient's hypertension risk level is determined, accurately reflecting the severity of the patient's hypertension and health risk, and providing a basis for personalized knowledge recommendations.

[0024] Furthermore, the process of classifying knowledge question-and-answer pairs based on a hypertension knowledge question-and-answer pair database, and extracting structured knowledge question-and-answer pair information based on the classification results, includes:

[0025] Based on popular science Q&A text information, extract information from specific groups and brief questions, brief replies and suggestions to generate structured knowledge Q&A information;

[0026] Based on clinical consultation and health management question-and-answer text information, a large language model is used to extract patient question-and-answer information and generate the first stage of extracted information.

[0027] Based on clinical consultation and health management Q&A text information, the one-prompt method was used to systematically extract patient Q&A information, generate second-stage extracted information, and combine it with the hypertension risk level assessment system to determine the patient's hypertension risk level;

[0028] Based on a multi-model collaboration and cross-validation mechanism, the consistency of the information extracted in the first stage and the information extracted in the second stage is compared, and the optimal structured knowledge question-answer pair information is integrated and output.

[0029] Different processing methods are used to process the text information of clinical consultation and health management questions and answers, and consistency comparison is used to extract knowledge information more comprehensively and accurately, so as to provide data support for subsequent matching.

[0030] Furthermore, the process involves collecting and downloading short medical videos related to hypertension, extracting video content, publisher type, video publication time, and interaction metadata, and establishing video quality assessment standards based on the video content, publisher type, and interaction metadata to generate a video quality score, including:

[0031] Based on preset prompt word templates, the system automatically identifies, downloads, and parses short medical videos about hypertension, extracting video content, publisher type, video publication time, and interactive metadata, and converting the video content into structured video text information; the interactive metadata includes the number of likes, comments, and shares.

[0032] Based on prompt word templates, health behavior suggestions are extracted from structured video text information to construct an intermediate semantic layer for knowledge recommendation;

[0033] Calculate video influence score based on video interaction metadata;

[0034] Determine the video's professionalism score based on the type of video publisher;

[0035] Based on the video content, and combined with the DISCERN scale for assessing the quality of medical information, a video content score is calculated.

[0036] The video quality score is calculated by weighting the video influence score, video professionalism score, and video content score.

[0037] By extracting information from medical short videos and combining it with information from different dimensions to assess video quality, it is easier to select high-quality videos, thereby ensuring that the recommended video content is reliable and professional, and improving the quality of knowledge acquired by users.

[0038] Furthermore, the video content score is calculated based on the video content and in conjunction with the DISCERN scale for medical information quality assessment, including:

[0039] Multiple medical experts were selected to conduct multi-dimensional evaluations of the video content of each medical short video, and the DISCERN scale for medical information quality assessment was used to calculate the DISCERN score of each medical expert for the current medical short video. The multi-dimensional evaluation included clarity assessment, reliability assessment, objectivity assessment, adequacy assessment, and expressiveness assessment.

[0040] Fleiss's Kappa coefficient is introduced to conduct a cross-dimensional consistency test on the rating category selection of medical short videos by multiple medical experts, determine the average consistency of the current medical short videos, and generate consistency quantification results;

[0041] The consistency quantification results are compared with a preset threshold, and the video content score of the current medical short video is determined based on the comparison results. When the consistency quantification results reach the preset threshold, the average DISCERN score of each medical expert for the current medical short video is used as the video content score of the current medical short video. When the consistency quantification results do not reach the preset threshold, the DISCERN score of the review experts for the current medical short video is used as the video content score of the current medical short video.

[0042] By introducing medical experts to conduct multi-dimensional evaluations of video content, the quality of video content can be evaluated professionally and comprehensively. Based on the cross-dimensional consistency test results, the reliability and accuracy of video content scoring can be ensured, thereby improving video quality evaluation standards.

[0043] Furthermore, the process of matching user personal health profiles with structured knowledge question-and-answer pairs of information and video content across multiple dimensions to calculate a matching score includes:

[0044] Integrate structured knowledge question-and-answer pairs and video content into knowledge entries;

[0045] A large language model is used to calculate the cosine similarity between a user's personal health profile and knowledge items, and knowledge items with cosine similarity higher than the semantic similarity threshold are selected and retained as semantic similarity matching results.

[0046] A large language model is used to analyze the user's personal health profile, identify the user's intent, and match the knowledge items corresponding to the hypertension risk level based on the user intent, and use them as the risk level alignment matching result;

[0047] The matching score is calculated by combining the semantic similarity matching results and the danger level alignment matching results.

[0048] By combining users' personal health profiles with knowledge items, the system measures the degree of match between knowledge items and users, thereby improving the accuracy of recommended videos.

[0049] Furthermore, the multi-dimensional matching of user personal health profiles with structured knowledge question-and-answer pairs and video content also includes knowledge graph path association matching. When the hypertension knowledge question-and-answer pair database is constructed in the form of a graph, the association paths between user personal health profiles and structured knowledge question-and-answer pairs are mined based on graph embedding technology, and the shortest path or the association path with the highest weight in the association paths is identified to obtain potential association knowledge, which is then used as the knowledge graph path association matching result.

[0050] Furthermore, the process of generating a personalized video recommendation list based on video quality score, matching score, video release time, and user platform interaction data, and optimizing video recommendations based on user feedback behavior, includes:

[0051] Determine the timeliness of a video based on its release date;

[0052] Based on user platform interaction data, collaborative filtering is used to learn implicit user feedback and determine user preferences;

[0053] By combining video quality score, matching score, video timeliness, and user preferences, a video recommendation score is calculated, and the videos are sorted to generate a personalized video recommendation list with the reasons for the recommendation.

[0054] Collect user feedback on recommended videos and optimize those videos.

[0055] By considering multiple factors to calculate video recommendation scores, a personalized video recommendation list that meets user needs and preferences can be generated. Furthermore, video recommendations can be optimized based on user feedback, thereby improving the accuracy of video recommendations and ultimately enhancing the user experience.

[0056] A system for recommending knowledge about a hypertension population based on a large language model multi-agent approach includes:

[0057] The knowledge acquisition module is used to build a hypertension knowledge question-and-answer pair database and a hypertension risk level determination system based on hypertension knowledge, and to classify and extract structured knowledge question-and-answer pair information based on the hypertension knowledge question-and-answer pair database.

[0058] The video acquisition and quality assessment module is used to collect and download short medical videos on hypertension, extract video content, publisher type, video publication time and interactive metadata, and establish video quality assessment standards based on video content, publisher type and interactive metadata to generate video quality scores.

[0059] The user information acquisition module is used to construct a user's personal health profile based on the user's personal health information and user platform interaction data, and to determine the current user's hypertension risk level in conjunction with the hypertension risk level assessment system.

[0060] The matching and calculation module is used to perform multi-dimensional matching between the user's personal health profile and structured knowledge question and answer information and video content, and calculate the matching degree score; the multi-dimensional matching includes semantic similarity matching and hypertension risk level alignment matching;

[0061] The recommendation and optimization module is used to generate a personalized video recommendation list based on video quality score, matching score, video release time, and user platform interaction data, and to optimize video recommendations based on user feedback.

[0062] The beneficial effects of this invention are as follows:

[0063] This invention constructs a hypertension knowledge question-and-answer database, refines the extraction of hypertension knowledge question-and-answer information, medical short video information, and user personal health profiles, conducts video content quality assessment and matching degree assessment, and adjusts it in conjunction with user platform interaction data. This enables it to provide personalized and accurate knowledge recommendation services for hypertension users, meet the health knowledge needs of hypertension patients, improve the accuracy and effectiveness of knowledge recommendations, and help patients manage their own health. Attached Figure Description

[0064] Figure 1 This is a flowchart of the present invention;

[0065] Figure 2 The diagram shown is a framework for knowledge recommendation methods for people with hypertension.

[0066] Figure 3 A schematic diagram illustrating the treatment inquiry in a hypertension knowledge Q&A session.

[0067] Figure 4 This is a physical illustration of treatment suggestions based on a Q&A section on hypertension.

[0068] Figure 5 A schematic diagram of the optimal structured knowledge question answering framework for information extraction;

[0069] Figure 6 A flowchart illustrating the quality assessment system for medical short videos;

[0070] Figure 7 A schematic diagram of the framework for a personalized video recommendation list;

[0071] Figure 8 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0072] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using other structures and / or functionalities besides one or more of the aspects set forth herein.

[0074] In addition, specific details are provided in the following description to facilitate a thorough understanding of the examples, and those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0075] Example 1

[0076] Figure 1 This paper presents a knowledge recommendation method for hypertension patients based on a large language model and multi-agent approach. By constructing a hypertension knowledge question-and-answer pair database, it extracts hypertension knowledge question-and-answer pair information, medical short video information, and user personal health profiles in a refined manner. It then conducts video content quality assessment and matching degree assessment, and adjusts the method by combining user platform interaction data. This approach can provide personalized and accurate knowledge recommendation services for hypertension patients, meet their health knowledge needs, improve the accuracy and effectiveness of knowledge recommendations, and help patients manage their own health. Figure 2 The diagram shown illustrates the framework for a knowledge recommendation method for people with hypertension, which includes the following steps:

[0077] S1: Based on hypertension knowledge, construct a hypertension knowledge question-and-answer pair database and a hypertension risk level determination system, and classify and extract structured knowledge question-and-answer pair information according to the hypertension knowledge question-and-answer pair database;

[0078] S21: Based on hypertension knowledge, define hypertension knowledge question-answer pairs, extract knowledge entity types and relationship types between entities, and construct a hypertension knowledge question-answer pair library;

[0079] In this embodiment, the entity types include treatment inquiry entities and treatment suggestion entities.

[0080] Among them, such as Figure 3 As shown, the treatment recommendations include patient gender, patient age, patient special population, patient hypertension type, patient family history, patient symptoms, patient comorbidities, patient blood pressure, patient lifestyle habits, patient medication (tablet name and dosage), patient medication effect, patient medication side effects, patient hypertension risk level, and patient mood.

[0081] Among them, such as Figure 4 As shown, the treatment recommendations include the doctor-recommended medication (pill name and dosage), the doctor's objective description of the disease symptoms, the efficacy of the doctor-recommended medication, the side effects of the doctor-recommended medication, the doctor-recommended examinations, the doctor's precautions for medication use, and lifestyle interventions.

[0082] S22: Based on the patient's blood pressure level, comorbidities, and lifestyle habits, conduct a comprehensive risk assessment, classify hypertension risk levels, and construct a hypertension risk level determination system;

[0083] S221: Obtain the patient's blood pressure level score;

[0084] In this embodiment, based on the authoritative definition of hypertension in the "Guidelines for the Prevention and Treatment of Hypertension in China," the degree of abnormality in systolic blood pressure (SBP) and diastolic blood pressure (DBP) is used as a preliminary risk indicator. Blood pressure level scores are assigned as follows: 0 points for normal blood pressure (SBP < 120 mmHg and DBP < 80 mmHg), 1 point for high-normal blood pressure (SBP 120-139 mmHg or DBP 80-89 mmHg), 2 points for stage 1 hypertension (SBP 140-159 mmHg or DBP 90-99 mmHg), 3 points for stage 2 hypertension (SBP 160-179 mmHg or DBP 100-109 mmHg), and 4 points for stage 3 hypertension (SBP ≥ 180 mmHg or DBP ≥ 110 mmHg).

[0085] S222: Obtain a score for the patient's comorbidities;

[0086] In this embodiment, it is determined whether the patient has high-risk disease factors such as diabetes, chronic kidney disease, coronary heart disease, stroke, heart failure, peripheral artery disease, and hyperlipidemia. For each high-risk disease factor present, different weight scores are assigned according to the degree of influence of the high-risk disease factor on the prognosis of hypertension, thereby determining the score of comorbidity.

[0087] S223: Obtain patient's lifestyle habit score;

[0088] In this embodiment, it is determined whether the patient has unhealthy behaviors such as smoking, excessive drinking, high-salt diet, lack of exercise, excessive mental stress, or obesity (BMI≥28 kg / m²). The patient's lifestyle habit score is set based on the degree of harm of the unhealthy behaviors, and the patient's lifestyle habit score increases by 0.5-1 points for each unhealthy behavior.

[0089] S224: Combine the patient's blood pressure level score, the patient's comorbidity score, and the patient's lifestyle habit score to calculate the patient's comprehensive risk score;

[0090] S225: Based on the weighted calculation of the patient's comprehensive risk score, combined with the hypertension risk level classification criteria, the patient's hypertension risk level is determined;

[0091] S2251: When a patient's comprehensive risk score is 0-2, the hypertension risk level is defined as low risk.

[0092] S2252: When a patient's comprehensive risk score is 2.1 to 4, the hypertension risk level is defined as intermediate.

[0093] S2253: When a patient's comprehensive risk score is 4.1 to 6, the hypertension risk level is defined as intermediate to high risk;

[0094] S2254: When a patient's comprehensive risk score is 6.1 to 8, the hypertension risk level is defined as high risk;

[0095] S2255: When a patient's comprehensive risk score is 8.1 to 10, the hypertension risk level is defined as high to very high risk;

[0096] S2256: When a patient's comprehensive risk score is 10 or higher, the hypertension risk level is defined as very high risk.

[0097] S23: Based on the hypertension knowledge question-answer pair database, perform knowledge question-answer pair text classification, and extract structured knowledge question-answer pair information based on the knowledge question-answer pair text classification results;

[0098] In this embodiment, the knowledge-based question and answer text classification includes popular science question and answer, clinical consultation and health management question and answer, and the clinical consultation and health management question and answer includes complications and risks, health management, mental health, disease assessment, and treatment and medication.

[0099] S231: Based on popular science Q&A text information, a large language model is used to extract information on specific groups of people, brief questions, brief replies and suggestions to generate structured knowledge Q&A information;

[0100] Specifically, specific groups and brief questions are extracted from the question text, and brief replies and suggestions are extracted from the answer text. The extracted information is then integrated to form structured knowledge question and answer information.

[0101] S232: Extract the optimal structured knowledge question-and-answer pairs based on clinical consultation and health management question-and-answer text information;

[0102] S2321: Based on clinical consultation and health management question-and-answer text information, a large language model is used to extract patient question-and-answer information and generate the first stage of extracted information;

[0103] Specifically, based on clinical consultation and health management Q&A text information, the system extracts the patient's basic personal information and medication usage from the question text, and analyzes and grades the patient's emotions based on a preset emotion level standard. It also extracts drug treatment plans, non-drug treatment suggestions, and objective disease symptoms from the answer text, and integrates and outputs the information extracted in the first stage.

[0104] S2322: Based on clinical consultation and health management Q&A text information, the one-prompt method is used to systematically extract patient Q&A information, generate second-stage extracted information, and combine it with the hypertension risk level assessment system to determine the patient's hypertension risk level;

[0105] Specifically, based on clinical consultation and health management Q&A text information, information such as the patient's special population, blood pressure, lifestyle habits, and comorbidities are extracted from the patient's basic personal information; medication information (pigment name, dosage and administration) and medication effects are extracted from medication use; and side effects of medication are extracted from the drug treatment plan. The extracted information is then integrated and output as the second stage of information extraction. Finally, the patient's hypertension risk level is determined based on the hypertension risk level assessment system.

[0106] S2323: Based on a multi-model collaboration and cross-validation mechanism, the consistency of the information extracted in the first stage and the information extracted in the second stage is compared, and the optimal structured knowledge question-answer pair information is integrated and output.

[0107] Specifically, based on a multi-model collaboration and cross-validation mechanism, and according to multi-dimensional standards of entity content consistency, medical semantic integrity, and logical rationality, the information extracted in the first stage and the information extracted in the second stage are compared for consistency. That is, based on the same field, if the extracted information is consistent, the consistent extracted information is retained; if the extracted information is inconsistent, it is re-extracted based on the original clinical consultation and health management question-and-answer text information. Based on the multi-dimensional standards of entity content consistency, medical semantic integrity, and logical rationality, the better extraction result is selected as the optimal structured knowledge question-and-answer pair information.

[0108] Figure 5 The diagram shows the optimal structured knowledge question-answering information extraction framework.

[0109] S2: Collect and download short medical videos about hypertension, extract video content, publisher type, video publication time and interaction metadata, and establish video quality assessment standards based on video content, publisher type and interaction metadata to generate video quality scores;

[0110] S21: Based on preset prompt word templates, automatically identify, download and parse short medical videos about hypertension, extract video content, publisher type, video publication time and interactive metadata, and convert the video content into structured video text information;

[0111] In this embodiment, the interactive metadata includes the number of likes, comments, and shares.

[0112] S22: Based on prompt word templates, extract health behavior suggestions from structured video text information to construct an intermediate semantic layer for knowledge recommendation;

[0113] Specifically, based on prompt word templates, health behavior suggestions such as exercise advice, psychological adjustment methods, dietary management plans, and common cognitive misconceptions are extracted from structured video text information to construct an intermediate semantic layer for knowledge recommendation.

[0114] S23: Calculate the video influence score based on video interaction metadata;

[0115] Among them, the video interaction metadata based on the number of likes, comments, and shares is normalized, and its calculation expression is as follows:

[0116] ;

[0117] ;

[0118] ;

[0119] In the formula, Indicates the normalized number of... The number of likes for each medical short video; Indicates the first The number of likes for each medical short video; This represents the maximum number of likes for all videos. This represents the minimum number of likes for all videos. Indicates the normalized number of... The number of comments on each medical short video; Indicates the first The number of comments on each medical short video; This represents the maximum number of comments across all videos. This represents the minimum number of comments for all videos. Indicates the normalized number of... The number of reposts for each medical short video; Indicates the first The number of reposts for each medical short video; This represents the maximum number of times all videos have been forwarded. This represents the minimum number of times all videos have been forwarded.

[0120] The expression for calculating the video influence score is as follows:

[0121] ;

[0122] In the formula, Indicates the first The video influence score of each medical short video; Indicates the weight of the number of likes; Indicates the weight of the number of comments; Indicates the weight of the number of forwards;

[0123] S24: Determine the video's professionalism score based on the type of video publisher;

[0124] The video publishers include medical professionals, professional institutions, and individual users. A video professionalism score is assigned to each type of video publisher, with the score ranging from 0 to 1.

[0125] In this embodiment, the video professionalism score for medical workers is set to 1.0; the video professionalism score for professional institutions is set to 0.8; and the video professionalism score for individual users is set to 0.3.

[0126] S25: Calculate the video content score based on the video content and the DISCERN scale for assessing the quality of medical information;

[0127] S251: Select multiple medical experts to conduct multi-dimensional evaluations of the video content of each medical short video, and combine the DISCERN scale for medical information quality assessment to calculate the DISCERN score of each medical expert for the current medical short video.

[0128] In this embodiment, the multi-dimensional evaluation includes clarity assessment, reliability assessment, objectivity assessment, adequacy assessment, and expressiveness assessment. Clarity assessment determines whether the medical short video clearly states its objective; reliability assessment determines whether the medical short video mentions reliable information sources; objectivity assessment determines whether the medical short video contains advertisements, exaggerations, or misleading information; adequacy assessment determines whether the medical short video systematically explains the treatment or management plan for hypertension; and expressiveness assessment determines whether the language of the medical short video is easy to understand and suitable for the general public.

[0129] In this embodiment, five medical experts are selected to evaluate the content of each medical short video based on five dimensions, with each dimension having a scoring standard of 0, 0.5, and 1 point. The scores from the five dimensions are summed to obtain the medical expert's DISCERN score for the current medical short video, and the calculation expression is as follows:

[0130] The formula for calculating the overall DISCERN score of current medical short videos by a single medical expert is as follows:

[0131] ;

[0132] In the formula, Indicates the first The first medical expert in the The DISCERN rating of a medical short video; Indicates the first The first medical expert in the The first medical short video Scoring across multiple dimensions; Indicates the number of evaluation dimensions. , Indicates the total number of evaluation dimensions; Indicates the number of medical experts. , Indicates the total number of medical experts;

[0133] S252: Introduce Fleiss's Kappa coefficient to conduct a cross-dimensional consistency test on the rating category selection of medical short videos by multiple medical experts, determine the average consistency of the current medical short videos, and generate consistency quantification results;

[0134] Among them, Fleiss's Kappa coefficient is used to quantify the inter-group consistency level of multiple medical experts on a discrete scoring task, and the overall average consistency coefficient of the current medical short videos is calculated and used as the consistency quantification result. The expression is as follows:

[0135] ;

[0136] In the formula, Indicates the first The overall consistency coefficient of the medical short videos; Indicates the first The first medical short video in The actual degree of consistency across each dimension is expressed as follows: , Indicates the rating category, , This indicates the total number of rating categories. Indicates the first The first medical short video in The first dimension The number of times each rating category was selected by experts; Indicates the first The first medical short video in The expression for the degree of expected consistency across all dimensions is: ;

[0137] S253: Compare the consistency quantification results with the preset threshold, and determine the video content score of the current medical short video based on the comparison results;

[0138] S2531: When the consistency quantification result reaches the preset threshold, the average DISCERN score of each medical expert for the current medical short video will be used as the video content score of the current medical short video.

[0139] When the consistency quantification result reaches the preset threshold, that is At that time, the calculation expression for the video content score of the current medical short video is:

[0140] ;

[0141] In the formula, Indicates the first The final DISCERN overall score for each medical short video; This indicates the total number of experts who participated in the final scoring; This represents a preset threshold; in this embodiment, ;

[0142] S2532: When the consistency quantification result does not reach the preset threshold, the DISCERN score of the review experts on the current medical short video will be used as the video content score of the current medical short video.

[0143] When the consistency quantification result does not reach the preset threshold, i.e. At that time, the calculation expression for the video content score of the current medical short video is:

[0144] ;

[0145] In the formula, Indicates the review experts For the The DISCERN rating of a medical short video;

[0146] S26: Combine video influence score, video professionalism score, and video content score to calculate the video quality score with weights;

[0147] Based on video influence score, video professionalism score, and video content score, after normalization, the video quality score of the current medical short video is calculated, and its expression is as follows:

[0148] ;

[0149] In the formula, Indicates the first The final video quality score for each medical short video; Indicates the normalized number of... The influence score of a medical short video; Indicates the normalized number of... The weighting of the influence score for each medical short video; This indicates the video's professionalism score; The weighting of the video's professionalism score; Indicates the normalized number of... Video content rating for each medical short video; Indicates the normalized number of... The weighting of video content scores for each medical short video; ;

[0150] Figure 6 The diagram shown is a flowchart of the medical short video quality assessment system.

[0151] S3: Based on the user's personal health information and user platform interaction data, construct the user's personal health profile, and combine it with the hypertension risk level assessment system to determine the current user's hypertension risk level;

[0152] In this embodiment, the user's personal health information is the personal health information actively entered by the user, which includes gender, age, special population, family history, symptoms, comorbidities, blood pressure, lifestyle habits, patient medication (pigment name and dosage), medication effects, and medication side effects; user platform interaction data includes user's historical questions, historical browsing history, user likes, favorites and shares, and user viewing time.

[0153] Based on the user's personal health profile and hypertension risk level assessment system, the system automatically updates the user's current hypertension risk level.

[0154] S4: Match users' personal health profiles with structured knowledge Q&A information and video content from multiple dimensions, and calculate the matching score;

[0155] S41: Integrate structured knowledge question-and-answer pairs and video content into knowledge entries;

[0156] In this embodiment, structured knowledge question-and-answer pairs and video content are combined to form knowledge entries. Specifically, knowledge entries include the questions in the knowledge question-and-answer pairs, summaries of the answers, the main body of the video text, and video tags.

[0157] S42: Use a large language model to calculate the cosine similarity between the user's personal health profile and the knowledge entries, and filter and retain the knowledge entries with a cosine similarity higher than the semantic similarity threshold, and use them as the semantic similarity matching results;

[0158] In this embodiment, the pre-trained large language model BERT is used to calculate the cosine similarity of the embedding vectors between keywords and phrases in a user's personal health profile and knowledge entries. This cosine similarity is then compared with a semantic similarity threshold. Knowledge entries with cosine similarity higher than the semantic similarity threshold are retained and used as the semantic similarity matching results. By selecting knowledge entries with high similarity, the content topics become more relevant.

[0159] S43: Use a large language model to analyze the user's personal health profile, identify the user's intent, and match the knowledge items corresponding to the hypertension risk level based on the user's intent, and use them as the risk level alignment matching result;

[0160] In this embodiment, a large language model is used to analyze the user's historical questions or actively input personal health information, and multi-classification or sequence labeling tasks are used to identify the user's deep health needs. Knowledge-based question answering is then used to classify the text and identify the user's specific intent.

[0161] Based on the user's current hypertension risk level, priority is given to recommending knowledge content corresponding to that risk level. For example, low-risk users are recommended "healthy lifestyles for preventing hypertension," while high-risk users are recommended "precautions for hypertension medication." When boundary situations arise or when there is a desire to improve health literacy, knowledge content related to adjacent hypertension risk levels is appropriately recommended, but the recommendation still prioritizes the user's current hypertension risk level. At the same time, the focus of the recommended knowledge content is adjusted based on different hypertension risk levels. For example, low-risk users are emphasized in health education and lifestyle interventions, medium-risk users are emphasized in early intervention and monitoring, and high- and very high-risk users are emphasized in medication management, prevention and management of complications, and specialist treatment recommendations.

[0162] S44: When the hypertension knowledge question and answer database is constructed in the form of a graph, based on graph embedding technology, the association path between the user's personal health profile and the structured knowledge question and answer information is mined, knowledge graph path association matching is performed, and the shortest path or the association path with the highest weight in the association path is identified to obtain potential association knowledge, which is used as the knowledge graph path association matching result.

[0163] In this embodiment, when the hypertension knowledge question-and-answer database is constructed in graph form, the TransE graph embedding technology is introduced to discover the association paths between entities in the user's personal health profile and entities in the structured knowledge question-and-answer information. The shortest path or the highest-weighted association path is identified, and potential related knowledge is recommended, which is then used as the knowledge graph path association matching result. For example, if the user's personal health profile involves "diabetes," knowledge such as "dietary recommendations for hypertension combined with diabetes" can be obtained through knowledge graph path association matching.

[0164] S45: Calculate the matching score by combining the semantic similarity matching results, the knowledge graph path association matching results, and the danger level alignment matching results.

[0165] The formula for calculating the matching score is as follows:

[0166] ;

[0167] In the formula, Indicates the matching score; This represents the semantic similarity matching result, which is obtained through cosine similarity. Indicates the weight of the semantic similarity matching result; This represents the path association matching result of the knowledge graph, which is obtained through the graph association degree. This indicates the weight of the knowledge graph path association matching results; This indicates the alignment result of the hazard level, i.e., when perfectly aligned. When partially aligned When completely misaligned ; This indicates the weight of the risk level alignment matching result;

[0168] S5: Based on video quality score, matching score, video release time, and user platform interaction data, generate a personalized video recommendation list and optimize video recommendations based on user feedback.

[0169] S51: Determine the timeliness of a video based on its release time;

[0170] In this embodiment, for time-sensitive health information (guideline updates, new drug developments), newer knowledge content is prioritized; for basic science knowledge, the weight of timeliness is appropriately reduced based on a time decay function. The expression for the time decay function is:

[0171] ;

[0172] In the formula, Indicates the timeliness of the video; This indicates the time difference between the video's release time and the current time. This represents the attenuation coefficient.

[0173] S52: Based on user platform interaction data, collaborative filtering is used to learn implicit user feedback and determine user preferences;

[0174] In this embodiment, based on user platform interaction data, the system learns users' preferences for specific types, formats, or themes of content, and uses collaborative filtering to learn implicit user feedback, assigning higher weight to knowledge content that users prefer.

[0175] S53: Combine video quality score, matching score, video timeliness, and user preferences to calculate video recommendation score, sort and generate a personalized video recommendation list, and add the recommendation reasons;

[0176] The expression for the video recommendation score is:

[0177] ;

[0178] In the formula, This indicates the video recommendation score; This indicates the video quality score; Indicates user preferences; Indicates the weight of the matching score; Indicates the weight of the video quality score; Indicates the weight of video timeliness; Indicates the weight of user preferences, and .

[0179] Videos are sorted from highest to lowest based on their recommendation scores to facilitate recommendations to users.

[0180] In this embodiment, the sorted knowledge items are recommended in the form of a structured list to directly present the summary of the structured knowledge question and answer pair, the title and brief description of the medical short video, and to mark key information such as video quality score and video professionalism; at the same time, a recommendation reason is generated for each recommended knowledge content to explain the suitability of the knowledge item with the current user.

[0181] Furthermore, in practical applications, various content formats such as text, images, and video links can be flexibly adopted according to user preferences to enhance the user experience.

[0182] Figure 7 A schematic diagram of the framework for a personalized video recommendation list.

[0183] S54: Collect user feedback on recommended videos and optimize recommended videos.

[0184] Based on user feedback such as click-through rate, dwell time, collection, sharing, and user-initiated comments on recommended videos, we continuously optimize recommended videos to improve the accuracy of recommended knowledge content and user satisfaction.

[0185] Example 2

[0186] Based on the same design concept, such as Figure 8 As shown, this embodiment provides a knowledge recommendation system for hypertension groups based on a large language model multi-agent system, including a knowledge acquisition module, a video acquisition and quality assessment module, a user information acquisition module, a matching and calculation module, and a recommendation and optimization module.

[0187] Specifically, the knowledge acquisition module is used to build a hypertension knowledge question-and-answer pair database and a hypertension risk level determination system based on hypertension knowledge, and to classify and extract structured knowledge question-and-answer pair information based on the hypertension knowledge question-and-answer pair database.

[0188] Specifically, the video acquisition and quality assessment module is used to collect and download short medical videos on hypertension, extract video content, publisher type, video publication time and interactive metadata, and establish video quality assessment standards based on video content, publisher type and interactive metadata to generate video quality scores.

[0189] Specifically, the user information acquisition module is used to construct a user's personal health profile based on the user's personal health information and user platform interaction data, and to determine the current user's hypertension risk level in conjunction with the hypertension risk level assessment system.

[0190] Specifically, the matching and calculation module is used to perform multi-dimensional matching between the user's personal health profile and structured knowledge question and answer information and video content, and calculate the matching degree score; the multi-dimensional matching includes semantic similarity matching and hypertension risk level alignment matching;

[0191] Specifically, the recommendation and optimization module is used to generate a personalized video recommendation list based on video quality score, matching score, video release time, and user platform interaction data, and to optimize video recommendations based on user feedback behavior.

[0192] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0193] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A knowledge recommendation method for a hypertension population based on a large language model multi-agent system, characterized in that, include: Based on hypertension knowledge, a hypertension knowledge question-and-answer pair database and a hypertension risk level assessment system were constructed. Furthermore, based on the hypertension knowledge question-and-answer pair database, the text of the knowledge question-and-answer pairs was classified and structured information was extracted, including: Based on hypertension knowledge, hypertension knowledge question-answer pairs are defined, knowledge entity types and relationship types between entities are extracted, and a hypertension knowledge question-answer pair database is constructed; entity types include treatment inquiry entities and treatment suggestion entities; The treatment inquiry includes patient gender, patient age, patient special population, patient hypertension type, patient family history, patient symptoms, patient comorbidities, patient blood pressure, patient lifestyle habits, patient medication, patient medication effects, patient medication side effects, patient hypertension risk level, and patient mood. Treatment recommendations include doctor-recommended medications, objective descriptions of disease symptoms by the doctor, efficacy of doctor-recommended medications, side effects of doctor-recommended pills, recommended examinations by the doctor, precautions for medication use pointed out by the doctor, and lifestyle interventions; Based on the patient's blood pressure level, comorbidities, and lifestyle habits, a comprehensive risk assessment is conducted to classify hypertension risk levels and construct a hypertension risk level determination system, including: Obtain the patient's blood pressure level score; the blood pressure level score includes 0 points for normal blood pressure, 1 point for high normal blood pressure, 2 points for grade 1 hypertension, 3 points for grade 2 hypertension, and 4 points for grade 3 hypertension; Obtain the patient's comorbidity score; determine whether the patient has high-risk disease factors such as diabetes, chronic kidney disease, coronary heart disease, stroke, heart failure, peripheral artery disease, and hyperlipidemia. For each high-risk disease factor present, assign different weight scores according to the degree of influence of the high-risk disease factor on the prognosis of hypertension, and then determine the comorbidity score. Obtain the patient's lifestyle habit score; determine whether the patient has unhealthy behaviors such as smoking, excessive drinking, high-salt diet, lack of exercise, excessive mental stress, or obesity, and set the patient's lifestyle habit score based on the degree of harm of unhealthy behaviors. For each unhealthy behavior, the patient's lifestyle habit score increases by 0.5-1 points. The patient's comprehensive risk score is calculated by combining the patient's blood pressure level score, the patient's comorbidity score, and the patient's lifestyle habit score. Based on the weighted calculation of the patient's comprehensive risk score and combined with the hypertension risk level classification criteria, the patient's hypertension risk level is determined. When a patient's comprehensive risk score is 0-2, the hypertension risk level is defined as low risk. When a patient's comprehensive risk score is 2.1 to 4, the hypertension risk level is defined as intermediate. When a patient's comprehensive risk score is 4.1 to 6, the hypertension risk level is defined as intermediate to high risk. When a patient's comprehensive risk score is 6.1 to 8, the hypertension risk level is defined as high risk. When a patient's comprehensive risk score is 8.1 to 10, the risk level of hypertension is defined as high to very high risk. When a patient's comprehensive risk score is 10 or higher, the hypertension risk level is defined as very high risk. Based on a hypertension knowledge question-and-answer pair database, text classification of knowledge question-and-answer pairs is performed, and structured information extraction of knowledge question-and-answer pairs is carried out based on the text classification results, including: Based on popular science Q&A text information, a large language model is used to extract information on specific groups of people, brief questions, brief replies and suggestions to generate structured knowledge Q&A information; Based on clinical consultation and health management question-and-answer text information, the optimal structured knowledge question-and-answer pairs are extracted, including: Based on clinical consultation and health management question-and-answer text information, a large language model is used to extract patient question-and-answer information and generate the first stage of extracted information. Based on clinical consultation and health management Q&A text information, the one-prompt method was used to systematically extract patient Q&A information, generate second-stage extracted information, and combine it with the hypertension risk level assessment system to determine the patient's hypertension risk level; Based on a multi-model collaboration and cross-validation mechanism, the consistency of the information extracted in the first stage and the information extracted in the second stage is compared, and the optimal structured knowledge question-answer pair information is integrated and output. The knowledge-based Q&A text classification includes popular science Q&A and clinical consultation and health management Q&A. The clinical consultation and health management Q&A category includes complications and risks, health management, mental health, disease assessment, and treatment and medication. Collect and download short medical videos about hypertension, extract video content, publisher type, video publication time and interaction metadata, and establish video quality assessment standards based on video content, publisher type and interaction metadata to generate video quality scores; Based on users' personal health information and user platform interaction data, a user's personal health profile is constructed, and combined with the hypertension risk level assessment system, the current user's hypertension risk level is determined. The system performs multi-dimensional matching of user personal health profiles with structured knowledge question-and-answer pairs, information, and video content, and calculates a matching score, including: Integrate structured knowledge question-and-answer pairs and video content into knowledge entries; A large language model is used to calculate the cosine similarity between a user's personal health profile and knowledge items, and knowledge items with cosine similarity higher than the semantic similarity threshold are selected and retained as semantic similarity matching results. A large language model is used to analyze the user's personal health profile, identify the user's intent, and match the knowledge items corresponding to the hypertension risk level based on the user intent, and use them as the risk level alignment matching result; When the hypertension knowledge question and answer database is constructed in the form of a graph, the association path between the user's personal health profile and the structured knowledge question and answer information is mined based on graph embedding technology. The shortest path or the association path with the highest weight is identified in the association path to obtain potential related knowledge, which is used as the knowledge graph path association matching result. The matching score is calculated by combining semantic similarity matching results, knowledge graph path association matching results, and danger level alignment matching results. Based on video quality score, matching score, video release time, and user platform interaction data, a personalized video recommendation list is generated, and video recommendations are optimized based on user feedback, including: Determine the timeliness of a video based on its release date; Based on user platform interaction data, collaborative filtering is used to learn implicit user feedback and determine user preferences; By combining video quality score, matching score, video timeliness, and user preferences, a video recommendation score is calculated, and the videos are sorted to generate a personalized video recommendation list with the reasons for the recommendation. Collect user feedback on recommended videos and optimize those videos.

2. The knowledge recommendation method for hypertension groups based on a large language model multi-agent system according to claim 1, characterized in that, The process involves collecting and downloading short medical videos related to hypertension, extracting video content, publisher type, video posting time, and interaction metadata, and establishing video quality assessment standards based on the video content, publisher type, and interaction metadata to generate a video quality score, including: Based on preset prompt word templates, the system automatically identifies, downloads, and parses short medical videos about hypertension, extracting video content, publisher type, video publication time, and interactive metadata, and converting the video content into structured video text information; the interactive metadata includes the number of likes, comments, and shares. Based on prompt word templates, health behavior suggestions are extracted from structured video text information to construct an intermediate semantic layer for knowledge recommendation; Calculate video influence score based on video interaction metadata; Determine the video's professionalism score based on the type of video publisher; Based on the video content, and combined with the DISCERN scale for assessing the quality of medical information, a video content score is calculated. The video quality score is calculated by weighting the video influence score, video professionalism score, and video content score.

3. The knowledge recommendation method for a hypertension group based on a large language model multi-agent according to claim 2, characterized in that, The video content score is calculated based on the video content and the DISCERN scale for medical information quality assessment, including: Multiple medical experts were selected to conduct multi-dimensional evaluations of the video content of each medical short video, and the DISCERN scale for medical information quality assessment was used to calculate the DISCERN score of each medical expert for the current medical short video. The multi-dimensional evaluation included clarity assessment, reliability assessment, objectivity assessment, adequacy assessment, and expressiveness assessment. Fleiss's Kappa coefficient is introduced to conduct a cross-dimensional consistency test on the rating category selection of medical short videos by multiple medical experts, determine the average consistency of the current medical short videos, and generate consistency quantification results; The consistency quantification results are compared with a preset threshold, and the video content score of the current medical short video is determined based on the comparison results. When the consistency quantification results reach the preset threshold, the average DISCERN score of each medical expert for the current medical short video is used as the video content score of the current medical short video. When the consistency quantification results do not reach the preset threshold, the DISCERN score of the review experts for the current medical short video is used as the video content score of the current medical short video.

4. A system for implementing the knowledge recommendation method for hypertension groups based on a large language model multi-agent as described in claim 1, characterized in that, include: The knowledge acquisition module is used to build a hypertension knowledge question-and-answer pair database and a hypertension risk level determination system based on hypertension knowledge, and to classify and extract structured knowledge question-and-answer pair information based on the hypertension knowledge question-and-answer pair database. The video acquisition and quality assessment module is used to collect and download short medical videos on hypertension, extract video content, publisher type, video publication time and interactive metadata, and establish video quality assessment standards based on video content, publisher type and interactive metadata to generate video quality scores. The user information acquisition module is used to construct a user's personal health profile based on the user's personal health information and user platform interaction data, and to determine the current user's hypertension risk level in conjunction with the hypertension risk level assessment system. The matching and calculation module is used to perform multi-dimensional matching between the user's personal health profile and structured knowledge question and answer information and video content, and calculate the matching degree score; the multi-dimensional matching includes semantic similarity matching and hypertension risk level alignment matching; The recommendation and optimization module is used to generate a personalized video recommendation list based on video quality score, matching score, video release time, and user platform interaction data, and to optimize video recommendations based on user feedback.

Images