Medical science popularization interaction method and system based on double-disease vertical large model

Abstract

The invention provides a medical science popularization interaction method and system based on a double-disease vertical large model. The method comprises the following steps: acquiring a user portrait of a patient; and performing path extraction in a preset double-disease cross knowledge graph according to the user portrait of the patient to obtain a corresponding knowledge path set. The knowledge path set comprises at least one knowledge path, and the knowledge path comprises a plurality of knowledge nodes. Searching corresponding modal content fragments in a content resource library according to the knowledge nodes, and performing weight calculation on each modal content fragment according to the cross risk factor, a preset cognitive load factor and a preset modal recommendation coefficient to obtain a modal weight; and selecting the modal content fragments corresponding to the two maximum modal weights for each knowledge node, combining the modal content fragments into a content unit, splicing all the content units to generate a multi-modal content set, and displaying according to the multi-modal content set. By constructing a patient user portrait, a personalized content recommendation strategy is dynamically generated, and accurate pushing of medical science popularization content is realized.

Description

Technical Field

[0001] This invention belongs to the field of medical science popularization, and in particular relates to a medical science popularization interactive method and system based on a dual-disease vertical large model. Background Technology

[0002] In current specialized hospitals and primary healthcare settings, liver diseases and infectious diseases, as two high-incidence and serious illnesses, are increasingly attracting clinical attention due to their interaction and cross-risk. Disease groups represented by chronic liver diseases such as hepatitis B, cirrhosis, and liver cancer often exhibit unique characteristics in terms of immune function and drug metabolism. When these patients are co-infected with infectious diseases such as COVID-19, dengue fever, and HIV / AIDS, there is a risk of conflicting treatment plans, leading to more complex disease progression and a significantly increased probability of sudden severe illness. Existing medical science popularization systems are mostly designed for single diseases, lacking the capacity for collaborative education for patients with dual diseases. This makes it difficult for patients to accurately understand the interaction mechanisms, medication contraindications, and key points of prevention between the two diseases. The general push strategies used by existing science popularization platforms do not accurately match content with patients' actual conditions, geographical prevalence risks, laboratory test indicators, or health behavior data, easily resulting in irrelevant, missing, or even misleading content. Therefore, how to achieve accurate delivery of medical science popularization content has become an urgent technical problem to be solved. Summary of the Invention

[0003] The purpose of this invention is to design a medical science popularization interactive method and system based on a dual-disease vertical large model, which can achieve accurate delivery of medical science popularization content.

[0004] To achieve the above objectives, the first aspect of this invention provides a medical science popularization interaction method based on a dual-disease vertical large model, the method comprising: Obtain patient user profiles; Based on the patient user profile, path extraction is performed in a preset dual-disease cross-knowledge graph to obtain a corresponding knowledge path set; wherein, the knowledge path set includes at least one knowledge path, and the knowledge path includes multiple knowledge nodes; Based on the knowledge node, the corresponding modal content fragment is searched in the preset content resource library. The modal weight is obtained by calculating the weight of each modal content fragment according to the preset cross-risk factor, preset cognitive load factor and preset modal recommendation coefficient. For each knowledge node, select the two modal content fragments corresponding to the largest modal weights and combine them into a content unit. Concatenate all the content units to generate a multimodal content set and display it according to the multimodal content set.

[0005] Furthermore, after selecting the two modal content fragments corresponding to the largest modal weights for each knowledge node and combining them into a content unit, the method further includes: Generate a corresponding learning task based on each of the content units described; The priority of each learning task is calculated based on the cross-risk factor, the preset fitness score, and the preset specific adjustment factor to obtain the task priority. The learning tasks are sorted from high to low according to their priority to obtain a learning task flow, which is then pushed to the patient user.

[0006] Further, the step of calculating the priority of each learning task based on the cross-risk factor, the preset fitness score, and the preset specific adjustment factor to obtain the task priority includes: The first data is obtained by multiplying the sum of the cross-risk factor, the fitness score, and the specific adjustment factor with a preset value. Add the first data corresponding to all learning tasks together to obtain the second data; The task priority is obtained by calculating the ratio between the first data and the second data.

[0007] Furthermore, after pushing the learning task stream to the patient user, the method further includes: Acquire the patient user's interaction behavior data on the learning task; wherein, the interaction behavior data includes answer accuracy rate, task completion time and error type data; The display method adjustment factor of the learning task is calculated based on the answer accuracy rate, the task completion time, and the preset display method adjustment coefficient. The display method of the learning task is adjusted according to the display method adjustment factor, and feedback content is added to the learning task according to the error type data.

[0008] Furthermore, after adjusting the display method of the learning task according to the display method adjustment factor and adding feedback content to the learning task according to the error type data, the method further includes: Based on the error type data, obtain the extended task corresponding to each learning task; The priority of the extended task is calculated based on the correct answer rate and the task completion time, thus obtaining the priority of the extended task. The extended tasks are sorted from high to low priority to obtain an extended task flow, and the extended task flow is pushed to the patient user.

[0009] Furthermore, obtaining the patient user profile includes: Obtain the patient's disease tags, disease stage, geographic risk tags, and cognitive level; The patient user profile is obtained by combining the disease tag, the disease stage, the geographical risk tag, and the cognitive level.

[0010] Further, the step of calculating the modal weight by weighting each modal content segment based on a preset cross-risk factor, a preset cognitive load factor, and a preset modal recommendation coefficient includes: The third data is obtained by multiplying the cross-risk factor, the cognitive load factor, and the modal recommendation coefficient. The third data corresponding to all modal content fragments is added together to obtain the fourth data; The modal weights are obtained by calculating the ratio between the third data and the fourth data.

[0011] In a second aspect, the present invention provides a medical science popularization interactive system based on a dual-disease vertical large model, the system comprising: The acquisition unit is used to acquire patient user profiles. The extraction unit is used to extract paths from a preset dual-disease cross-knowledge graph based on the patient user profile to obtain a corresponding set of knowledge paths; wherein, the set of knowledge paths includes at least one knowledge path, and the knowledge path includes multiple knowledge nodes; The calculation unit is used to search for the corresponding modal content fragment in the preset content resource library according to the knowledge node, and to calculate the weight of each modal content fragment according to the preset cross-risk factor, the preset cognitive load factor and the preset modal recommendation coefficient to obtain the modal weight. The display unit is used to select the two modal content fragments corresponding to the largest modal weights for each knowledge node and combine them into a content unit, splice all the content units to generate a multimodal content set, and display the multimodal content set.

[0012] In a third aspect of the invention, an electronic device is provided, the electronic device including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the method described in the first aspect above.

[0013] In a fourth aspect of the invention, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect.

[0014] The beneficial technical effects of the present invention are at least as follows: To address the aforementioned issues, this invention provides a medical science popularization interaction method and system based on a dual-disease vertical large-scale model. Its core lies in constructing knowledge association logic to achieve a structured expression of treatment conflicts, medication contraindications, and concurrent risks. To address the problem that existing push mechanisms cannot incorporate individual circumstances, this invention introduces a multi-source health data fusion method, comprehensively considering the patient's disease type, disease stage, geographical prevalence risk, and cognitive level to dynamically generate personalized content recommendation strategies. To overcome the comprehension barriers caused by monotonous content formats, this invention designs a multimodal content generation mechanism, integrating text, images, animations, interactive tasks, and other methods to improve knowledge accessibility and memorability. Addressing the current situation of low user participation and difficulty in assessing educational effectiveness, this invention constructs a task-based interactive learning module and a behavior response tracking mechanism, capable of dynamically monitoring users' comprehension levels and changes in health behaviors, and continuously optimizing the content push logic based on the monitoring results. Through these multi-layered system integrations, this invention can achieve truly highly adaptable, highly participatory, and highly responsive intelligent medical science popularization education for patients with dual diseases, effectively filling the technological gaps in existing systems regarding insufficient adaptability and lack of closed-loop management capabilities. Attached Figure Description

[0015] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0016] Figure 1 This is a flowchart of a medical science popularization interaction method based on a dual-disease vertical large model provided in the embodiments of this application.

[0017] Figure 2 yes Figure 1 The flowchart of step S101.

[0018] Figure 3 This is a flowchart of a medical science popularization interaction method based on a dual-disease vertical large model provided in another embodiment of this application.

[0019] Figure 4 This is a flowchart of a medical science popularization interaction method based on a dual-disease vertical large model provided in the third embodiment of this application.

[0020] Figure 5 This is a flowchart of a medical science popularization interaction method based on a dual-disease vertical large model provided in the fourth embodiment of this application.

[0021] Figure 6 This is a schematic diagram of the structure of the medical science popularization interactive system based on a dual-disease vertical large model provided in the embodiments of this application. Detailed Implementation

[0022] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0023] Please refer to Figure 1 , Figure 1 This is a flowchart of a medical science popularization interaction method based on a dual-disease vertical large model provided in an embodiment of this application. Figure 1 The method may include, but is not limited to, steps S101 to S104.

[0024] Step S101: Obtain patient user profile; Step S102: Based on the patient user profile, extract the path in the preset dual-disease cross-knowledge graph to obtain the corresponding knowledge path set; wherein, the knowledge path set includes at least one knowledge path, and the knowledge path includes multiple knowledge nodes. Step S103: Based on the knowledge node, search for the corresponding modal content fragment in the preset content resource library, and calculate the weight of each modal content fragment according to the preset cross-risk factor, preset cognitive load factor and preset modal recommendation coefficient to obtain the modal weight. Step S104: For each knowledge node, select the modal content fragments corresponding to the two largest modal weights and combine them into a content unit. Then, splice all the content units together to generate a multimodal content set and display it based on the multimodal content set.

[0025] Please see Figure 2 In some embodiments, step S101 may include, but is not limited to, steps S201 to S202: Step S201: Obtain the patient's disease label, disease stage, geographic risk label, and cognitive level; Step S202: Based on disease tags, disease stages, geographical risk tags, and cognitive levels, a patient user profile is obtained by combining these elements.

[0026] In step S201 of some embodiments, the input data comes from the Hospital Information System (HIS), Electronic Medical Record System (EMR), Hospital Health Education Platform, and In-Hospital Geographic Epidemiological Monitoring Platform. All of these are existing standard data, and no new data is collected. The input data includes the patient's disease label, disease stage, geographic risk label, and cognitive level, wherein: (Disease Tags): The ICD code is automatically extracted from the primary diagnosis field of inpatient or outpatient care. The system matches it with a standard disease database and converts the code into tags, such as "chronic hepatitis B", "decompensated cirrhosis", "dengue fever", "AIDS", etc., which serve as the basis for dual-disease identification.

[0027] (Disease Stage): Extract stage markers from fields such as outpatient summary and inpatient enrollment forms filled out by doctors, such as "mid-stage of antiviral treatment for hepatitis B", "3 months after liver cancer surgery", and "HIV viral load increase stage". The fields are structured fields (not free text) and can be extracted through template matching.

[0028] (Geographical Risk Label): The system determines whether the patient is currently in a "dengue fever high-incidence area" or "HIV high-monitoring area" by fuzzy matching of the patient's address field with the administrative region, and by connecting to the municipal CDC API or the hospital's local epidemiological reporting system. The matching result is a Boolean variable or a multi-level risk level.

[0029] (Cognitive Level): After logging into the health education system for the first time, patients need to complete a set of 10 questions and answers set by the hospital's knowledge base. The questions cover basic protection, medication logic, and common sense about disease transmission. The test scores are automatically classified by the system as "Beginner (0-4 points)", "Intermediate (5-7 points)" and "Advanced (8-10 points)".

[0030] In step S202 of some embodiments, after collecting the above four types of variables, they are spliced ​​together to form a patient user profile with a unified structure: ; in, This represents a patient user profile. Indicates a disease label. Indicates the stage of the disease. Indicates geographical risk label, Indicates cognitive level.

[0031] Through the aforementioned steps S201 to S202, the aim is to construct a structured patient user profile that can be used for knowledge reasoning for patients in scenarios involving the cross-management of liver diseases and infectious diseases. .

[0032] In step S102 of some embodiments, after the patient user profile is generated, the system calls the hospital knowledge graph service module. This module is based on a pre-constructed dual-disease cross-knowledge graph. Perform path matching. Dual-disease cross-knowledge graph. This is a knowledge graph developed collaboratively by the hospital's medical expert team and AI engineering team. Graph nodes include symptoms, indicators, concurrency mechanisms, drug contraindications, and cross-risk events. Edges represent medical knowledge relationships such as causality, concurrency, and constraints, and edge weights are the cross-risk strength (0–1) assigned by expert scores or case statistics. A path extraction function is used to extract patient user profiles from the cross-disease knowledge graph, finding corresponding knowledge paths and obtaining a set of knowledge paths. The path extraction function is shown below: ; in, Profiling patients The corresponding set of knowledge paths is represented as an ordered list of knowledge node sequences, with each path having a length of 1 to 5 knowledge nodes. The path extraction function is executed on the graph. Based on After pruning the graph structure of each field, edge weights and the maximum path are selected by breadth-first search.

[0033] The path extraction logic employs a graph structure pruning + BFS search algorithm, prioritizing the output of node sequences with high cross-risk paths. All paths must contain at least one liver disease-related node, one infectious disease-related node, and one cross-risk node. In one example, a user... The following criteria must be met: the disease is "chronic hepatitis B," the stage is "mid-stage antiviral therapy," the resident is in a dengue-endemic area, and the cognitive ability is "basic." The system-retrieved atlas contains the following knowledge paths: Knowledge Path 1: "Elevated ALT" → "Body temperature above 38℃" → "Early symptoms of dengue fever" → "Risk of overlapping diagnoses" → "Misuse of acetaminophen" Knowledge Path 2: "Mid-stage antiviral treatment" → "Liver function fluctuation period" → "Increased risk of infectious complications" → "High incidence of dengue fever" → "3 times higher probability of severe illness".

[0034] In step S103 of some embodiments, the aim is to base the analysis on the patient user profile. With knowledge path Select and combine suitable multimodal science popularization content, including text, images, videos, and animations, from a pre-set content resource library to construct a multimodal content collection with consistent structure and personalized expression. Compared to traditional disease education systems, this step not only focuses on content matching but also on the consistency between content structure and knowledge path. Furthermore, it introduces innovative modal control mechanisms and cross-risk content priority factors to address the unique knowledge structures in cross-disease scenarios (liver disease and infectious diseases)—such as medication contraindications, overlapping symptoms, and misjudgments of transmission mechanisms. This ensures that the generated science popularization content possesses medical structural alignment, interactive guidance, and patient cognitive adaptability.

[0035] Specifically, the content generation system maintains a content resource library. Each knowledge node All of them establish mapping relationships with multiple modal content fragments, including text and images. ,voice ,video Interactive animation Each modal content fragment is accompanied by a modal attribute tag (readability, cognitive load, risk level, etc.).

[0036] Knowledge Path Each knowledge node in The system retrieves content from the resource library. Extract all corresponding modal content fragments and then analyze them based on the cognitive level in the patient user profile. Calculate its modal weights with the risk characteristics of the two diseases. To address the strong overlap and asymmetric cognitive needs associated with certain knowledge points in dual-disease scenarios, a cross-risk factor was introduced into the calculation of modal weights. and cognitive load factor The modal weight is calculated by weighting each modal content segment based on preset cross-risk factors, preset cognitive load factors, and preset modal recommendation coefficients. Specifically, the cross-risk factor, cognitive load factor, and modal recommendation coefficient are multiplied to obtain the third data; the third data corresponding to all modal content segments are added together to obtain the fourth data; and the ratio of the third data and the fourth data is calculated to obtain the modal weight. The formula is shown below: ; in, For knowledge nodes modal content fragments Modal weights; For knowledge nodes The cross-risk factor is derived from the association weights between liver disease and infectious diseases in the graph edge weights, with a value range of [1,3]. The more obvious the cross-traffic, the larger the value (e.g., nodes related to "vaccine conflict"). =3); Modal content fragment Cognitive load factors, labeled by the hospital's health education team, such as in the animation. =0.5, Images and Text =1.0; Cognitive level User modality The modal recommendation coefficients are derived from the hospital knowledge base recommendation table (e.g., animation priority for novice users). Advanced users prioritize images and text. The denominator is the normalization coefficient, ensuring... . Indicated in the same knowledge node Excluding the current modal content fragment Other modal content fragments, This represents other modal content fragments. The corresponding weighting factor, This indicates the cognitive level of the user regarding other modal content fragments. The recommendation rating.

[0037] By using the risk level of the knowledge node itself as a modulatory factor for modal expression intensity, for example, for a knowledge node "contraindications of dengue vaccine in the context of hepatitis B", due to its high cross-risk ( =3), even if the user has a high level of cognitive ability, the system will appropriately reduce the ratio of text and images and prioritize the enhancement of animated demonstrations or case videos to improve the accuracy of understanding and attention.

[0038] In step S104 of some embodiments, in the modal weights Once determined, select each knowledge node. The two largest modal content segments combine to form a content unit. and the content units corresponding to all knowledge nodes according to The sequence of elements is concatenated to generate a multimodal content collection:

[0039] in, For knowledge nodes The two modal content segments with the highest modal weights (such as animation + text and images); It is a collection of multimodal content, an ordered list, where each item is a combination of a knowledge node and a fragment of its modal content; To ensure the content is assembled according to the knowledge path sequence and to align the content structure with the medical knowledge structure.

[0040] Furthermore, the collection of multimodal content will be displayed to enable patients to accurately understand the interaction mechanism between the two diseases, medication contraindications, and key points of protection.

[0041] In actual deployment, the system will preload The modal content fragment indexes of all knowledge nodes involved are used to calculate the score of each modal content fragment before dynamic content combination is performed. For example, a user... For the "post-liver cancer surgery + HIV recovery period + low risk of dengue fever + intermediate level of cognition" knowledge path, which includes knowledge nodes such as "fluctuations in liver function + ARV drug metabolism + precautions for vaccination", the system will output a content chain mainly consisting of text, images and case videos, emphasizing drug interaction mechanisms and specific vaccine response strategies.

[0042] Through this knowledge path-driven and modal weight dynamic adjustment approach, the system outputs... It not only covers the necessary knowledge nodes, but also adapts the content expression to the risk level of medical knowledge and the user's cognitive ability. This is a unity of personalization and knowledge integrity that traditional medical science popularization systems cannot achieve.

[0043] Steps S101 to S104 of this embodiment involve obtaining a patient user profile. Based on the patient user profile, paths are extracted from a preset dual-disease cross-knowledge graph to obtain a corresponding knowledge path set. The knowledge path set includes at least one knowledge path, and each knowledge path includes multiple knowledge nodes. Based on the knowledge node, corresponding modal content fragments are searched in a preset content resource library. Each modal content fragment is weighted according to preset cross-risk factors, preset cognitive load factors, and preset modal recommendation coefficients to obtain modal weights. For each knowledge node, the two modal content fragments with the highest modal weights are selected and combined into a content unit. All content units are then concatenated to generate a multimodal content set, which is then displayed. By constructing a patient user profile, a personalized content recommendation strategy is dynamically generated. To overcome the comprehension barriers caused by a single content format, a multimodal content generation mechanism is designed, integrating various methods such as text, images, animation, and interactive tasks to improve knowledge accessibility and memorability, achieving accurate delivery of medical science popularization content.

[0044] Please refer to Figure 3 In some embodiments, after step S105, the medical science popularization interaction method based on the dual-disease vertical large model may also include, but is not limited to, steps S301 to S303: Step S301: Generate corresponding learning tasks for each content unit; Step S302: Calculate the priority of each learning task based on the cross-risk factor, the preset fitness score, and the preset specific adjustment factor to obtain the task priority; Step S303: Sort the learning tasks from high to low according to their priority to obtain the learning task flow, and push the learning task flow to the patient user.

[0045] In step S301 of some embodiments, according to Each knowledge node in Generate corresponding learning tasks The tasks can take various forms, such as multiple-choice questions, true / false questions, case studies, and symptom identification questions. The generated learning tasks will depend on the user's cognitive level. Cross-risk factors of each node in the knowledge path The difficulty and presentation format are determined by this. In one example, the generation of learning tasks is based on each knowledge node and its corresponding content unit. First, knowledge nodes are read from the knowledge graph. Medical semantic types (such as "symptom recognition", "drug contraindications", "transmission prevention and control") are identified, and then the corresponding content units for that knowledge node are retrieved from the content resource library. Next, based on the cross-risk factors of knowledge nodes... With user cognitive level The system automatically selects a matching task format from the task template library: for example, high-risk tasks are prioritized for case analysis or multi-step reasoning questions, while low-risk tasks are matched with true / false or multiple-choice questions. During the generation process, the system will... The core content (such as text paragraphs, video clips, or key question-and-answer points) is embedded in the task template, automatically generating the question stem, options, and correct answer annotations, thus forming a complete learning task. This approach ensures that each task is closely centered around a knowledge node, maintaining consistency in medical content while dynamically matching task difficulty with the user's cognitive level.

[0046] In step S302 of some embodiments, in order to better adapt to the cross-risk in dual-disease scenarios, this step introduces a dynamic task priority adjustment mechanism based on knowledge point risk and cognitive adaptability, that is, the task priority of each learning task. It depends not only on the user's cognitive ability It also considers the cross-risk factors of the knowledge nodes involved in the task. Specifically, the task priority is calculated by multiplying the cross-risk factor, fitness score, and specific adjustment factor by a preset value to obtain the first data; summing the first data for all learning tasks to obtain the second data; and then calculating the ratio between the first and second data to obtain the task priority. The formula is as follows: ; in, For learning tasks The task priority indicates the importance of the task in the task flow. The higher the priority, the earlier the task will be presented to the user. Knowledge nodes in the graph The cross-risk factor represents the potential threat that the knowledge node poses to the user's health. The value ranges from [1,3]. A higher value indicates that the knowledge node has a greater impact on the management of dual diseases (e.g., drug contraindications, cross symptoms, etc.). For user cognitive level For learning tasks The fitness score is usually determined by the hospital education team based on the difficulty of the task design and the user's cognitive ability. A higher score indicates better fitness. The value indicates that the task is more suitable for the user; For knowledge node-based Specific adjustment factors are used to introduce additional risk and complexity considerations. For example, nodes prone to cross-infection may be assigned... =0.5, while the common disease node =0. In addition to learning tasks Other learning tasks, express Cross-risk factors corresponding to knowledge nodes Indicates the user's cognitive level fitness score express The specific adjustment factor corresponding to the knowledge node, with the normalization coefficient in the denominator, ensures that the sum of the priorities of all tasks is 1.

[0047] By combining the cross-risk of each node in the knowledge path with the user's cognitive level, and a specific moderating factor This effectively adjusts the task order, prioritizing tasks that are high-risk and require deep understanding for users. This dynamic priority adjustment mechanism is particularly suitable for complex disease management scenarios, improving user comprehension while ensuring that tasks are not overly difficult or complicated for users.

[0048] In step S303 of some embodiments, according to task priority The calculation results are used to generate a learning task flow in descending order of task priority. The learning task flow will be pushed to patient users. The learning task flow will cover all knowledge nodes in the knowledge path and will be sorted according to task priority. The selection of task types will be based on the knowledge nodes. Types of questions (e.g., diagnostic problems, symptom assessment, vaccination recommendations, etc.) and user cognitive level This determines whether to use interactive task types such as true / false questions, case-based reasoning, or video Q&A. Each task will have a certain feedback mechanism, such as dynamically evaluating the correctness of the user's answers and their comprehension, which can automatically adjust the difficulty or presentation of the next task.

[0049] For example, if a user performs well in the "Hepatitis B vaccination" task, the system will move to the next stage and provide more challenging tasks, such as comparisons with COVID-19 vaccination or correlation analysis of drug contraindications. For users with lower cognitive levels, the task difficulty may be reduced to basic symptom recognition tasks, such as "how to determine the manifestations of active hepatitis B" or "early symptoms of dengue fever during the high fever stage."

[0050] Furthermore, to accommodate complex cross-disciplinary knowledge points, a regularization mechanism is employed to prevent excessive bias towards a particular type of knowledge point or task in the learning task flow. This regularization mechanism introduces regularization terms during task generation to ensure a balanced distribution of task content across different knowledge points, preventing tasks from concentrating on only a few high-risk nodes. The regularization term is expressed as follows: ; in, The regularization term for the learning task flow represents the balance of the task flow; the smaller the value, the more balanced the task distribution. This is a regularization hyperparameter used to control the balance of tasks. It is usually adjusted experimentally to ensure that the complexity and difficulty of tasks do not deviate drastically.

[0051] Through regularization, the system can automatically adjust the task flow to avoid a single disease or task type from affecting the user's overall learning process, thereby improving the breadth and depth of learning outcomes.

[0052] Steps S301 to S303 shown in this embodiment are based on the patient user profile. and knowledge path Build personalized interactive learning task flow This process guides users to gradually master and understand the complex medical knowledge in the context of dual-disease crossover scenarios through dynamically generated reasoning tasks, case application scenarios, and multiple-choice questions. In particular, this step introduces a task priority adjustment mechanism based on path risk assessment, hierarchical cognitive task design, and an interactive feedback model to ensure that users receive the most suitable educational experience in each task, fully exploring and addressing the complex medical risks brought about by dual-disease crossover.

[0053] Please refer to Figure 4 In some embodiments, after step S303, the medical science popularization interaction method based on the dual-disease vertical large model may also include, but is not limited to, steps S401 to S403: Step S401: Obtain the patient user's interaction behavior data on the learning task; wherein, the interaction behavior data includes the correct answer rate, task completion time and error type data; Step S402: Calculate the display method adjustment factor for the learning task based on the correct answer rate, task completion time, and preset display method adjustment coefficient; Step S403: Adjust the display method of the learning task according to the display method adjustment factor, and add feedback content to the learning task according to the error type data.

[0054] In step S401 of some embodiments, the display method of each learning task is adjusted using patient user interaction behavior data. First, patient user interaction behavior data in each learning task is recorded, specifically: Answer accuracy : Indicates that the user is on a learning task The percentage of correct answers. If the accuracy rate is low, the task may need a more intuitive or interactive presentation.

[0055] Task completion time The time required for the user to complete the task. If the time is too long, it may indicate that the user does not fully understand the task, and additional supplementary content or simplification of the task is needed.

[0056] Error type data The system analyzes the types of errors users make in tasks. For example, if a user makes multiple errors in a "drug contraindications" task, the system will add visual content about drug contraindications, such as animations or diagrams.

[0057] In steps S402 to S403 of some embodiments, the learning task is calculated by adjusting coefficients based on the correct answer rate, task completion time, and preset display method. The display method adjustment factor is shown in the following formula: ; in, For learning tasks The display method adjustment factor determines whether the display format of the task needs to be changed. The larger the value, the more interactive or visual the display method needs to be. For learning tasks The accuracy rate of answering questions, the percentage of users who answered the task correctly; The task completion time reflects the user's level of mastery over the task; A coefficient is used to adjust the display method and balance the weight between the accuracy of answering questions and the time to complete the task. It is usually set to 0.5 to indicate that equal attention is paid to the accuracy of answering questions and the time to complete the task. This indicates other learning tasks besides the current learning task. express The accuracy rate of answering questions, express The display method adjustment coefficient, express The task completion time; the denominator is the normalization coefficient of all tasks, ensuring that the sum of the display mode adjustment factors of all tasks is 1.

[0058] The presentation of learning tasks is adjusted based on a display method adjustment factor. This means that the presentation of each task will be tailored to the user's interaction performance. Adjustments to the task presentation method include the content format (text, images, animation, video), the frequency of interactive feedback, and the simplification or complexity of the task. The task presentation will dynamically change based on task priority, user performance, and the needs of the task type. If a user performs poorly (e.g., low accuracy or high completion time), the presentation method will lean more towards text, images, animation, or interactive feedback to help the user better understand the task content; conversely, for high-performing users, the system will appropriately reduce interactive feedback and increase content depth.

[0059] In addition, to accommodate complex knowledge points involving two diseases, the system also introduced error type data. If a user frequently makes mistakes on a specific type of knowledge point (such as drug contraindications), the system will add more feedback to the task, such as interactive case explanations and scenario simulations, to help the user master complex knowledge through repeated practice.

[0060] Steps S401 to S403 shown in this embodiment are based on the learning task flow. By combining user interaction data during learning tasks, the system dynamically adjusts the task display method. Specifically, based on user performance during the learning process (such as task completion, answer accuracy, and answer time), the system optimizes the presentation of learning tasks (such as text, video, animation, and interactive feedback) to ensure that each task is displayed in a way that best suits the user's cognitive level and task completion status, thereby improving the user's learning effectiveness and task completion rate.

[0061] Please refer to Figure 5 In some embodiments, after step S403, the medical science popularization interaction method based on the dual-disease vertical large model may also include, but is not limited to, steps S501 to S503: Step S501: Obtain the extended task corresponding to each learning task based on the error type data; Step S502: Calculate the priority of the learning task on the preset extended task based on the answer accuracy and task completion time to obtain the priority of the extended task. Step S503: Sort the extended tasks from high to low according to their priority to obtain the extended task flow, and push the extended task flow to the patient user.

[0062] In step S501 of some embodiments, the system can determine which knowledge points the user has weaknesses based on error type data, and generate corresponding extension tasks for learning tasks that the user has not fully mastered. These extension tasks allow the user to enter an efficient and personalized follow-up learning phase after completing the current learning task, enabling them to continue mastering more related knowledge points. In one example, the extension tasks are dynamically generated based on the user's actual learning performance. After the user completes the learning task, the system analyzes their error type data to determine which knowledge points or concepts are weak. Subsequently, it locates nodes with high relevance to these knowledge points in the knowledge graph, such as hierarchical concepts, causal relationships, or nodes related to concurrent risks, and retrieves task templates matching these nodes from the task template library. If the user has a high error rate or takes a long time to complete a certain type of task (such as drug contraindication judgment questions or transmission mechanism analysis questions), the system automatically generates extension tasks from related knowledge tasks of the corresponding category. The content is mostly supplementary or reinforcing training of the original task's knowledge points, such as changing to diagrammatic explanations, case analyses, or situational question-and-answer sessions, to help the user make up for cognitive gaps and consolidate understanding.

[0063] In some embodiments, steps S502 to S503, the calculation of the extended task priority is based on the user's performance. That is, the priority of the extended task is calculated based on the correct answer rate and task completion time, resulting in the extended task priority. As shown in the following formula: ; in, To extend the mission The priority of the extended task; the larger the value, the more likely the extended task should be recommended first. For users in learning tasks The accuracy rate reflects the user's mastery of the task. Complete learning tasks for users The longer the time required, the more likely the user is to encounter difficulties with the task. and For adjustment coefficients, Reflecting the importance of accuracy, Weights reflecting time. They are usually set to be equal, for example... = =0.5, to balance the difficulty of the task and the completion status; In addition to learning tasks Other learning tasks, express The accuracy rate of answering questions, express Task completion time; the denominator is a normalization coefficient to ensure the priority of all deferred tasks. The total is 1.

[0064] Through this calculation, the system can assign appropriate priorities to each extended task based on the user's performance in the learning task. If the user's mastery of a certain knowledge point is poor (e.g., low accuracy or high completion time), the system will prioritize the extended tasks related to that knowledge point at the beginning of the learning path so that the user can focus on overcoming these weaknesses.

[0065] Next, the extended tasks are sorted from highest to lowest priority to obtain an extended task flow, which is then recommended to the patient user. These extended tasks are dynamically adjusted according to the user's learning progress to ensure that the recommended path is continuously optimized and closely aligned with the user's cognitive needs. The extended task flow generation function is expressed as follows: ; in, The extended task flow is a task recommendation path generated based on the priority of extended tasks and user performance, sorted by priority; The function generates the most suitable learning path based on the priority and error type data of the extended task flow. To extend task priority, used to determine the recommended order; Error type data is used to identify the types of errors users make in learning tasks, thereby recommending corresponding supplementary learning content.

[0066] Steps S501 to S503 of this embodiment generate a personalized extended task flow based on interactive behavior data. The purpose of the extended task flow is to allow users to enter an efficient and personalized follow-up learning stage after completing the current learning task, so as to continue to master more related knowledge points. The extended task flow will automatically recommend the next learning content based on the user's weaknesses, learning progress, and task performance, ensuring the continuity and effectiveness of learning.

[0067] Please see Figure 6 This application also provides a medical science popularization interactive system based on a dual-disease vertical large model, which can realize the above-mentioned medical science popularization interactive method based on a dual-disease vertical large model. The system includes: Acquisition unit 601 is used to acquire patient user profiles; Extraction unit 602 is used to extract paths from a preset dual-disease cross-knowledge graph based on the patient user profile to obtain a corresponding knowledge path set; wherein, the knowledge path set includes at least one knowledge path, and the knowledge path includes multiple knowledge nodes. The calculation unit 603 is used to find the corresponding modal content fragment in the preset content resource library according to the knowledge node, and to calculate the weight of each modal content fragment according to the preset cross-risk factor, the preset cognitive load factor and the preset modal recommendation coefficient to obtain the modal weight. Display unit 604 is used to select the modal content fragments corresponding to the two largest modal weights for each knowledge node and combine them into a content unit. All content units are spliced ​​together to generate a multimodal content set, which is then displayed.

[0068] The specific implementation of this medical science popularization interactive system based on a dual-disease vertical large model is basically the same as the specific implementation of the medical science popularization interactive method based on the dual-disease vertical large model described above, and will not be repeated here.

[0069] The above are preferred embodiments of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A medical popular science interaction method based on a double-disease vertical large model, characterized in that, The method comprises: acquiring a patient user portrait; extracting a path in a preset double-disease cross knowledge graph according to the patient user portrait to obtain a corresponding knowledge path set; wherein the knowledge path set comprises at least one knowledge path, and the knowledge path comprises a plurality of knowledge nodes; finding corresponding modal content segments in a preset content resource library according to the knowledge nodes, and calculating the weight of each modal content segment according to a preset cross risk factor, a preset cognitive load factor and a preset modal recommendation coefficient to obtain a modal weight; combining modal content segments corresponding to the two largest modal weights of each knowledge node to form a content unit, splicing all the content units to generate a multi-modal content set, and displaying the multi-modal content set. 2.The medical popular science interaction method based on a double-disease vertical large model according to claim 1, wherein, After the combination of the modal content segments corresponding to the two largest modal weights of each knowledge node to form a content unit, the method further comprises: generating a corresponding learning task for each content unit; calculating the priority of each learning task according to the cross risk factor, a preset fitness score and a preset specific adjustment factor to obtain a task priority; sorting the learning tasks from high to low according to the task priority to obtain a learning task flow, and pushing the learning task flow to the patient user. 3.The medical popular science interaction method based on the double-disease vertical large model according to claim 2, characterized in that, The calculation of the priority of each learning task according to the cross risk factor, a preset fitness score and a preset specific adjustment factor to obtain a task priority comprises: multiplying the sum of the cross risk factor, the fitness score and the specific adjustment factor by a preset value to obtain first data; adding the first data corresponding to all learning tasks to obtain second data; calculating the ratio of the first data and the second data to obtain the task priority. 4.The medical popular science interaction method based on the double-disease vertical large model according to claim 2, characterized in that, After the learning task flow is pushed to the patient user, the method further comprises: obtaining interaction behavior data of the patient user on the learning task; wherein the interaction behavior data comprises a correct answer rate, a task completion time and an error type data; calculating a display mode adjustment factor of the learning task according to the correct answer rate, the task completion time and a preset display mode adjustment coefficient; adjusting the display mode of the learning task according to the display mode adjustment factor, and adding feedback content to the learning task according to the error type data.

5. The medical popular science interaction method based on the double-disease vertical large model according to claim 4, characterized in that, After the display mode adjustment factor is used to adjust the display mode of the learning task and feedback content is added to the learning task according to the error type data, the method further comprises: obtaining an extension task corresponding to each learning task according to the error type data; calculating the priority of the extension task according to the correct answer rate and the task completion time to obtain an extension task priority; sorting the extension tasks from high to low according to the extension task priority to obtain an extension task flow, and pushing the extension task flow to the patient user. 6.The medical popular science interaction method based on a double-disease vertical large model according to claim 1, wherein, The acquisition of the patient user portrait comprises: Obtaining a disease label, a disease stage, a geographical risk label and a cognitive level of a patient; According to the disease label, the disease stage, the geographical risk label and the cognitive level, splicing is performed to obtain the patient user portrait. 7.The medical popularization interaction method based on a double-disease vertical large model according to claim 1, characterized in that, The weight calculation of each modal content segment according to the preset cross-risk factor, the preset cognitive load factor and the preset modal recommendation coefficient includes: The cross-risk factor, the cognitive load factor and the modal recommendation coefficient are multiplied to obtain third data; All modal content segments corresponding to the third data are added to obtain fourth data; The third data and the fourth data are ratio calculated to obtain the modal weight.

8. A medical popularization interaction system based on a double-disease vertical large model, characterized in that, The system comprises: An acquisition unit for acquiring a patient user portrait; An extraction unit for extracting a corresponding knowledge path set from a preset double-disease cross-knowledge graph according to the patient user portrait; wherein the knowledge path set comprises at least one knowledge path, and the knowledge path comprises a plurality of knowledge nodes; A calculation unit for finding corresponding modal content segments in a preset content resource library according to the knowledge nodes, and calculating the weight of each modal content segment according to a preset cross-risk factor, a preset cognitive load factor and a preset modal recommendation coefficient to obtain a modal weight; A display unit for selecting two largest modal content segment combinations corresponding to the modal weight for each knowledge node to form a content unit, splicing all the content units to generate a multi-modal content set, and displaying the multi-modal content set.

9. An electronic device, comprising: The electronic device comprises a memory and a processor, the memory stores a computer program, and the processor executes the computer program to realize the medical popular science interaction method based on the double-disease vertical large model according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to realize the medical popular science interaction method based on the double-disease vertical large model according to any one of claims 1 to 7.

Images