Medical information retrieval method based on intelligent medical science popularization platform

Abstract

The invention relates to the technical field of medical information retrieval, and discloses a medical information retrieval method based on an intelligent medical science popularization platform. The method comprises the following steps: acquiring a user retrieval request, and dividing the user retrieval request into a medical concept entity and an association description entity; based on the medical concept entities, extracting core concept nodes and adjacent extension nodes in a preset medical knowledge graph, and generating an initial retrieval range; based on the association description entity, performing semantic filtering on the initial retrieval range, screening out semantic matching nodes, and constructing a target retrieval set; analyzing the medical information in the target retrieval set to generate a medical information optimization evaluation value; and performing priority ranking on the target retrieval set according to the evaluation value, and outputting a ranked medical information retrieval result. According to the method, the matching degree of the retrieval result and the user demand is improved, the user can conveniently and quickly obtain high-quality medical information, the retrieval experience is improved, and the method is suitable for an information retrieval scene of an intelligent medical science popularization platform.

Description

Technical Field

[0001] This invention relates to the field of medical information retrieval technology, specifically a medical information retrieval method based on an intelligent medical science popularization platform. Background Technology

[0002] With increasing health awareness, more and more users are obtaining health-related information through medical science popularization platforms, making efficient and accurate medical information retrieval crucial to meeting user needs. Currently, the information retrieval methods commonly used by medical science popularization platforms largely rely on keyword matching or simple semantic analysis, which have significant limitations when processing user search requests.

[0003] When users submit search requests, they often struggle to use precise medical terminology, resulting in search queries that frequently mix everyday expressions with medical concepts. Existing search methods fail to effectively distinguish between medical concept entities and related descriptive entities in search requests. This leads to either an overly broad search scope, including a large amount of information unrelated to the core medical concept, or an overly narrow scope, omitting relevant extended information. For example, when a user searches for "relief methods for frequent headaches," existing methods may only extract the single medical concept of "headache," ignoring related descriptions such as "frequent" and "relief methods." Consequently, they might return a large amount of information about headache causes and diagnoses, rather than the relief measures the user actually needs, increasing the burden of information filtering for the user.

[0004] While some platforms have incorporated medical knowledge graphs to aid retrieval, most only extract nodes from the knowledge graph based on single medical concept entities, without specifically filtering the extracted nodes. Medical knowledge graphs contain massive amounts of nodes, covering multiple dimensions such as diseases, symptoms, treatment plans, and drugs. Simply extracting core concept nodes and adjacent nodes results in an initial search scope containing a large amount of information that does not match the user's actual needs. For example, for the search request "dietary recommendations for diabetic patients," existing methods, when extracting nodes based on "diabetes," may also include nodes related to the pathogenesis of diabetes and the treatment of complications. These nodes have low relevance to "dietary recommendations," resulting in a low proportion of effective information in the subsequent search results.

[0005] Current search methods lack a scientific evaluation and ranking mechanism for search results. After obtaining the results, they are often sorted by publication time or number of clicks, without considering the professionalism, accuracy, and relevance to user needs. Medical information is unique, with significant differences in quality between different sources, making it difficult for non-professional users to distinguish between true and false information. Sorting based solely on non-professional indicators may result in lower-quality or even erroneous information appearing at the top, potentially leading users to believe such information and causing adverse health effects. Furthermore, when faced with a large number of disordered search results, users need to spend considerable time browsing through them one by one, failing to quickly obtain key information, severely impacting search efficiency and user experience. Summary of the Invention

[0006] The purpose of this invention is to provide a medical information retrieval method based on an intelligent medical science popularization platform, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, this invention provides a medical information retrieval method based on an intelligent medical science popularization platform, the method comprising: Collect user search requests and divide the user search requests into medical concept entities and associated descriptive entities; Based on the medical concept entities, core concept nodes and adjacent extended nodes are extracted from the preset medical knowledge graph to generate an initial search range. Based on the associated description entities, semantic filtering is performed on the initial search range to select semantically matching nodes and construct the target search set; The medical information within the target retrieval set is analyzed to generate optimized evaluation values ​​for the medical information. The target retrieval set is prioritized and sorted according to the medical information optimization evaluation value, and the sorted medical information retrieval results are output.

[0008] Preferably, the step of analyzing the medical information within the target retrieval set and generating a medical information optimization evaluation value includes: Multi-dimensional feature extraction is performed on the medical information within the target retrieval set to obtain the authoritative feature value, timeliness feature value, and audience suitability feature value of the medical information; The authoritativeness, timeliness, and audience suitability of the medical information are weighted and calculated to generate a comprehensive evaluation value for the medical information. The comprehensive evaluation value of the medical information is compared with a preset evaluation threshold range. If the comprehensive evaluation value of the medical information is lower than the lower limit of the evaluation threshold range, a content credibility warning signal is generated. Based on the content credibility warning signal, the cross-modal association path density value and node information update frequency value of the semantic matching node are obtained; The cross-modal associated path density value and the node information update frequency value are coupled and calculated to generate a dynamic adjustment factor; Based on the aforementioned dynamic adjustment factor, the comprehensive evaluation value of medical information is corrected to obtain an optimized evaluation value of medical information.

[0009] Preferably, the specific method for generating the initial search range is as follows: Identify the main node position of the medical concept entity in the medical knowledge graph, and extract the first-level related nodes directly connected to the main node and their attribute labels; Traverse the interaction paths of the first-level associated nodes to obtain all second-level associated nodes and their relationship types; The main node, first-level associated nodes, and second-level associated nodes are merged to form the initial search range.

[0010] Preferably, the specific method for constructing the target retrieval set is as follows: Parse the syntactic structure of the associated descriptive entities and extract the keyword weight distribution and logical connection relationships; Calculate the semantic similarity value between each node within the initial retrieval range and the keyword weight distribution; Nodes with semantic similarity values ​​greater than a preset similarity threshold are selected and marked as semantically matched nodes; All semantically matching nodes and their associated edges are aggregated to form the target retrieval set.

[0011] The preferred method for multi-dimensional feature extraction is as follows: Extract the literature citation level and source institution level of the semantic matching node to generate the authority feature value of the medical information; Extract the latest revision timestamp and historical update cycle of the semantic matching node to generate the timeliness feature value; Extract the user's historical feedback data and reading difficulty coefficient of the semantic matching node to generate the audience suitability feature value.

[0012] Preferably, the specific method for generating the comprehensive evaluation value of medical information is as follows: The first weighting coefficient of the authoritative feature value of the medical information, the second weighting coefficient of the timeliness feature value, and the third weighting coefficient of the audience suitability feature value are set. The comprehensive evaluation value of the medical information is obtained by multiplying the authoritative feature value of the medical information by the first weighting coefficient, the timeliness feature value by the second weighting coefficient, and the audience suitability feature value by the third weighting coefficient, and then summing them up.

[0013] Preferably, the specific method for obtaining the cross-modal association path density value and node information update frequency value is as follows: The number of image and text data paths, video data paths, and clinical case data paths associated with the semantic matching nodes is counted, and the cross-modal association path density value is calculated. Monitor the number of version changes of the semantic matching node within a preset time window, and generate the node information update frequency value.

[0014] Preferably, the specific method for generating the dynamic adjustment factor is as follows: The modal coverage coefficient is obtained by calculating the ratio between the cross-modal associated path density value and the path density benchmark value. The node information update frequency value and the update frequency benchmark value are logarithmically normalized to obtain the timeliness enhancement coefficient; The modal coverage coefficient is multiplied by the time-effect enhancement coefficient to generate the dynamic adjustment factor.

[0015] Preferably, the specific method for correcting the comprehensive evaluation value of medical information to obtain the optimized evaluation value of medical information is as follows: The comprehensive evaluation value of the medical information is added to the dynamic adjustment factor to obtain the optimized evaluation value of the medical information.

[0016] Preferably, the specific method for prioritizing the target retrieval set based on the medical information optimization evaluation value is as follows: The nodes in the target retrieval set are sorted from high to low according to the medical information optimization evaluation value; When the medical information optimization evaluation values ​​are the same, a secondary sort is performed based on the cross-modal association path density value.

[0017] Compared with the prior art, the beneficial effects of the present invention are: In the search scope determination stage, this method first divides user search requests into medical concept entities and associated descriptive entities. This division accurately captures the core and supplementary information of user search needs, avoiding search bias caused by existing methods' inability to distinguish between the two types of entities. Based on medical concept entities, core concept nodes and adjacent extended nodes are extracted from a pre-set medical knowledge graph. This ensures that the search scope revolves around the core medical concepts while also covering potentially useful information related to the core concepts through adjacent extended nodes, avoiding the omission of key content due to an overly narrow search scope. For example, when a user searches for "nursing measures for children's colds," the method can accurately extract the medical concept entity "children's colds" and the associated descriptive entity "nursing measures." Based on "children's colds," core nodes and adjacent extended nodes such as "nursing care for cold symptoms" and "diet during a cold" are extracted, laying the foundation for subsequent accurate searches.

[0018] In the semantic filtering stage, the initial search scope is filtered based on associated descriptive entities. This effectively eliminates nodes that do not match the user's actual needs, further narrowing the search scope and improving search accuracy. Compared to existing methods that rely solely on keyword filtering or non-targeted filtering, this semantic filtering stage fully integrates the associated descriptive information in the user's search request, making the filtered target search set more closely match the user's needs. For example, when a user searches for "suitable exercises for patients with hypertension," the associated descriptive entity "suitable exercises" can filter irrelevant nodes such as "causes of hypertension" and "medical treatment for hypertension" in the initial search scope, retaining only relevant nodes such as "types of exercise for patients with hypertension" and "intensity of exercise for patients with hypertension," reducing the interference of useless information on the user.

[0019] In the information evaluation and ranking stage, by analyzing the medical information within the target search set and generating optimized evaluation values, multiple factors such as the professionalism, accuracy, completeness, and relevance to user needs of the information can be comprehensively considered, rather than relying on a single indicator. Prioritizing search results based on these evaluation values ​​ensures that higher-quality information that better meets user needs appears at the top, facilitating quick access to key information and improving search efficiency and user experience. For non-professional users, there's no need to sift through vast amounts of information; focusing on the top-ranked information provides reliable medical knowledge, reducing the possibility of health risks from obtaining incorrect or low-quality information.

[0020] This method fully leverages the structured advantages of a pre-defined medical knowledge graph and combines it with semantic analysis technology to achieve intelligent processing throughout the entire process, from query request parsing to result output, thereby enhancing the intelligence level of information retrieval on medical science popularization platforms. Compared to traditional retrieval methods, this method is better suited to users' diverse and personalized retrieval needs. Whether users submit retrieval requests using everyday expressions or professional terminology, it can provide high-quality retrieval results through precise entity segmentation, filtering, and evaluation, further enhancing the practical value of medical science popularization platforms and helping more users conveniently and accurately obtain the medical information they need. Attached Figure Description

[0021] Figure 1 This is a schematic diagram illustrating the working principle of the medical information retrieval method based on an intelligent medical science popularization platform described in this invention. Figure 2 Flowchart of the method for optimizing evaluation values ​​to generate medical information; Figure 3 Flowchart illustrating the specific steps for generating the initial search range; Figure 4 A flowchart illustrating the specific methods for multi-dimensional feature extraction. Detailed Implementation

[0022] 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.

[0023] Please see Figure 1 This invention provides a medical information retrieval method based on an intelligent medical science popularization platform, the method comprising: The system collects user search requests submitted through an input interface and uses natural language processing (NLP) to parse and categorize these requests into medical concept entities and associated descriptive entities. Medical concept entities refer to core medical terms in the search, such as disease names or drug names; associated descriptive entities include additional modifiers, symptom descriptions, or contextual information. Based on these medical concept entities, node matching is performed in a pre-defined medical knowledge graph to extract core concept nodes corresponding to the medical concept entities. Furthermore, adjacent extended nodes directly connected to these core concept nodes are extracted to generate an initial search scope. The pre-defined medical knowledge graph stores medical concept nodes and their relationships in a graph structure. Based on the associated descriptive entities, a semantic similarity calculation model is used to semantically filter the nodes within the initial search scope, selecting nodes that semantically match the associated descriptive entities to construct a target search set. The medical information within the target search set is analyzed, and through multi-dimensional feature extraction and calculation, a medical information optimization evaluation value is generated. Based on the medical information optimization evaluation value, the nodes in the target search set are prioritized, and the sorted medical information search results are finally returned through an output interface.

[0024] Example 1: See Figure 2 This demonstrates the process of analyzing medical information within a target retrieval set and generating optimized evaluation values ​​for that information. This is clearly illustrated through a specific user search example. Assuming the user's search request is "the latest drug treatment plan for type 2 diabetes," the system first divides this request into the medical concept entity "type 2 diabetes" and the associated descriptive entity "latest drug treatment plan." Following these initial steps, the system has extracted relevant nodes from the medical knowledge graph based on the core concept of "type 2 diabetes" and, using semantic filtering based on "latest drug treatment plan," constructs a target retrieval set containing various drug information, clinical guideline nodes, and research literature nodes.

[0025] The analysis of the target retrieval set began with multi-dimensional feature extraction. The system processed the authority feature value of medical information. For a node in the set related to "metformin," the system extracted its citation level; this node was cited by the American Diabetes Association clinical guidelines, indicating a high citation level. It also extracted the level of the source institution; this information originated from a top-tier medical research institution. Based on the preset authority scoring rules, the system generated a high medical information authority feature value for this node. For another node related to a traditional Chinese medicine formula, its literature was only cited by a local journal, and the source institution was a general research institute; therefore, the generated authority feature value was relatively low. The system extracted the timeliness feature value, which is crucial for the requirement of "the latest drug treatment plan." The system reads the latest revision timestamp of each node. A node about GLP-1 receptor agonists showed its timestamp to be updated within the last three months, while another node about sulfonylureas showed its latest revision timestamp to be two years ago. Simultaneously, the system analyzes the historical update cycles of these nodes. The GLP-1 receptor agonist node has frequent update records in the past year, while the sulfonylurea drug node has a longer update cycle. Based on this, the system generates a higher timeliness feature value for the former and a lower timeliness feature value for the latter. The system extracts audience suitability feature values. The platform calls on users' historical feedback data. For example, most users without professional medical backgrounds gave high ratings and positive feedback to the text and images of the "metformin" node, and its reading difficulty coefficient was assessed to be at a medium level, suitable for the general public. Conversely, a node containing a large amount of molecular formulas and raw clinical trial data, although professional and accurate, has a low click-through rate and poor reading completion rate, indicating an extremely high reading difficulty coefficient. Therefore, the system generates a higher audience suitability feature value for the former and a lower audience suitability feature value for the latter.

[0026] After feature extraction, the system performs weighted calculations to generate a comprehensive evaluation value for medical information. In the system's preset weight configuration, given the special nature of medical information, the authoritative feature is assigned the highest first weight coefficient, the timeliness feature is assigned the second weight coefficient, and the audience suitability feature is assigned the third weight coefficient. Each node's three feature values ​​are multiplied by their corresponding weight coefficients, and the three products are then summed to obtain a preliminary comprehensive evaluation value for medical information. In this example, the GLP-1 receptor agonist node, due to its high authority, high timeliness, and moderate audience suitability, obtained a high comprehensive evaluation value. Conversely, the node with strong technical content but poor readability, despite potentially high authority and timeliness, suffers from extremely low audience suitability, resulting in a significantly lower comprehensive evaluation value after weighted calculation.

[0027] The system automatically compares this comprehensive evaluation value with a preset evaluation threshold range. For example, if a node concerning a rare herbal remedy has an ambiguous information source, a low authority value, and a low timeliness value due to its age, the calculated comprehensive evaluation value of the medical information will fall below the preset credibility threshold range. The system will then generate a content credibility warning signal for this node.

[0028] Based on this warning signal, the system triggers further data acquisition and analysis processes. The system will count the number of cross-modal data paths associated with the semantic matching node, that is, check whether the node is associated with multiple forms of content such as graphic interpretations, popular science videos, or clinical cases. If the node only has a simple text description, its cross-modal association path density value will be very low. At the same time, the system monitors the information update frequency of the node within the most recent 24-month time window and finds that its version change count is zero, thus generating an extremely low node information update frequency value.

[0029] The system performs coupled calculations to generate a dynamic adjustment factor. It compares the calculated cross-modal path density value with the average path density of all nodes within the system (the path density baseline) to obtain a modal coverage coefficient, which in this example is much less than 1. Simultaneously, it logarithmically normalizes the node information update frequency value with the median update frequency of that type of node (the update frequency baseline) to obtain a timeliness enhancement coefficient, which in this example is also extremely low. Multiplying these two coefficients generates a dynamic adjustment factor less than 1, reflecting the node's deficiencies in information dimension and timeliness.

[0030] The system uses this dynamic adjustment factor to correct the initially low overall evaluation value of medical information. The correction method involves adding the overall evaluation value to the dynamic adjustment factor. Since the dynamic adjustment factor is a positive number, this operation slightly increases the evaluation value. However, given the small size of the factor itself, the increase is limited, and the final optimized evaluation value of medical information remains low, objectively reflecting the poor information quality at that node. This process ensures that even if some information is initially poorly evaluated, the system will fine-tune it by examining its information dimensions and update potential, thereby outputting a more refined and accurate optimized evaluation value for subsequent search result ranking.

[0031] Example 2: See Figure 3This demonstrates the method for generating the initial search scope and constructing the target search set. A real-world user query scenario illustrates this, assuming a user enters the search query: "Inhaled treatment methods for childhood asthma." Upon receiving this request, the natural language processing module first parses it, dividing it into medical concept entities and related descriptive entities. In this request, "asthma" is identified as the core medical concept entity, while "children" and "inhaled treatment methods" are categorized as related descriptive entities, with the latter specifying the core concept in terms of age and treatment method.

[0032] Based on the medical concept entity "asthma," the system begins searching within a pre-defined medical knowledge graph to generate an initial search scope. This knowledge graph consists of numerous medical concept nodes and semantic relationship edges between nodes. The system first employs a graph query algorithm to precisely locate the master node that perfectly matches "asthma" within the graph. After locating the master node, the system extracts all first-level related nodes directly connected to it and their attribute labels. These first-level related nodes may include therapeutic drugs such as "bronchodilators," "glucocorticoids," and "allergens," as well as pathological factors and related symptom nodes. Each connection edge carries an attribute label describing the relationship type, such as "treatment method," "triggering factor," or "accompanying symptoms." Subsequently, the system initiates a traversal process, starting from these first-level related nodes and continuing to explore along their interaction paths. Using a breadth-first search algorithm, the system obtains all second-level related nodes directly connected to the first-level related nodes and their relationship types. For example, the first-level node "glucocorticoids" can be traced to the second-level node "budesonide" (a specific drug), with a relationship type of "is a"; similarly, the first-level node "allergens" can be traced to the second-level node "dust mites," with a relationship type of "belongs to." This process collects all relevant nodes within the two-step relationships. Finally, the system merges the main node "asthma," all acquired first-level and second-level related nodes, and automatically removes duplicate nodes, forming a rich and broad initial search scope. This scope covers numerous medical concepts directly and indirectly related to asthma.

[0033] The system then performs semantic filtering on this large initial search scope based on the associated descriptive entities "children" and "inhalation therapy" to construct a precise target search set. The system first parses the grammatical structure of these two associated descriptive entities. "Children" is a noun qualifier, and "inhalation therapy" is a noun phrase, where "inhalation" is an adverbial of manner, and "therapy" is the core noun. Through dependency parsing, the system extracts the keywords "children" and "inhalation," determining their logical connection to be parallel qualifiers, both modifying the core concept.

[0034] The system assigns weight distributions to these keywords. In this query context, "inhalation," as a specific treatment method, may have a higher weight than the broader age limit of "children." The system then calculates the semantic similarity value between each node within the initial search range and this set of keyword weight distributions. This calculation is performed using a word embedding model, which vectorizes the text content of each node (such as node name, descriptive text, attribute tags) and keywords, and measures their cosine similarity in the vector space.

[0035] A node about "management of acute asthma attacks in adults" has a content vector that deviates significantly from the vector directions of "children" and "inhaled," resulting in a lower calculated semantic similarity value. Conversely, a node about "guidelines for the use of metered-dose inhalers for childhood asthma" has a content vector that closely matches the direction of the keyword vector, thus achieving a high semantic similarity value. The system presets a dynamically adjusted similarity threshold, which is optimized based on feedback from historical search data. Of all nodes, only those with semantic similarity values ​​greater than this threshold are filtered out and marked as semantically matched nodes.

[0036] In this case, nodes like "budesonide dosage for children" and "selection of inhaler for pediatric asthma" are successfully filtered out, while irrelevant nodes such as "surgical treatment of asthma" or "occupational asthma" are filtered out. Ultimately, the system aggregates all the labeled semantically matching nodes and retains the interconnected edges between these nodes in the knowledge graph, forming a refined, accurate, and closely related target retrieval set that aligns with the user's query intent. This lays the foundation for subsequent in-depth analysis and ranking output.

[0037] Example 3: See Figure 4 This demonstrates the core steps of multi-dimensional feature extraction and the generation of comprehensive medical information evaluation values. This process involves in-depth analysis of medical information nodes that have passed semantic screening and been included in the target retrieval set, aiming to quantitatively evaluate the information quality and applicability of each node. The following is a specific example: Suppose the target retrieval set contains multiple nodes related to "hypertension drug treatment," and the system needs to extract and evaluate the features of these nodes.

[0038] Multi-dimensional feature extraction begins with the authority feature value of medical information. The system parses the metadata and associated data sources of each node. For a node describing "angiotensin-converting enzyme inhibitors," the system extracts its citation level. By querying integrated external academic databases, it confirms that the node's content has been cited by multiple randomized controlled trials and European Society of Cardiology guidelines. According to preset rules, such high-level citations assign a higher citation score to the node. Simultaneously, the system extracts the source institution level of the node. This information originates from an internationally recognized cardiovascular disease research center, which is ranked at the highest level in its internal authority ranking database. The citation level and source institution level are processed through a standardized mapping function and finally fused to generate the node's medical information authority feature value, which is a normalized numerical value.

[0039] The system extracts timeliness features based on the version management metadata built into the nodes. The system reads the latest revision timestamp of the target node; for example, a node about "Clinical Application Guidelines for New Antihypertensive Drugs" would display its timestamp as the current year. Simultaneously, the system calculates the node's historical update cycle, that is, it traces back all its historical versions and calculates the average time interval between adjacent version releases. A node that updates frequently and has a short historical update cycle usually means that its content keeps pace with medical advancements. The proximity of the latest revision timestamp to the current time, along with the length of the historical update cycle, is used by a predefined timeliness calculation model to jointly generate a timeliness feature value that reflects the freshness of the information.

[0040] For extracting audience suitability features, the system relies on the platform's internal user behavior logs and content analysis module. The system extracts historical user feedback data for semantically matching nodes, including but not limited to the average reading time, user ratings, number of shares, and subsequent consultation behaviors. For example, a node about "daily management of hypertension" will be considered easy to understand and accept if it has a high user rating and high average completion rate. Simultaneously, the system uses natural language processing tools to analyze the node's text content and calculate its reading difficulty coefficient, which comprehensively considers factors such as sentence length, vocabulary specialization, and terminology density. The historical user feedback data and reading difficulty coefficient are then processed by an aggregation algorithm to ultimately generate audience suitability features.

[0041] After extracting features across the three dimensions, the system enters the stage of generating a comprehensive medical information evaluation value. The system assigns a weight coefficient to each feature value. These weight coefficients are not fixed but dynamically generated through an offline machine learning process that trains the system on massive amounts of user interaction data and expert-annotated data. Under the current system configuration, the weight coefficient for the authoritativeness feature value is set to the highest, followed by the timeliness feature value, and then the audience suitability feature value.

[0042] The comprehensive evaluation value of medical information is calculated using a linear weighted summation model, and the calculation process can be expressed by the following formula:

[0043] In this formula: character Represents the final calculated comprehensive evaluation value of medical information; character This represents the authoritative feature value of the medical information extracted and standardized in the previous steps; characters Represents the extracted timeliness feature value; character Represents the calculated audience fit feature value; character Represents the dynamic weighting coefficients assigned to authoritative feature values; characters Represents the dynamic weighting coefficients assigned to timeliness feature values; characters Represents the dynamic weighting coefficients assigned to the audience fit feature values.

[0044] During the calculation, the system substitutes three feature values ​​of each node into this formula. For example, a node with high authority, moderate timeliness, and high audience suitability will have a comprehensive evaluation value of [value missing]. The value will be higher, while a node that, despite its high authority, has extremely low timeliness and is very difficult to read, will be less relevant due to its... and The value is low, even if the feature value A has high weight. Multiply them together to get the final comprehensive evaluation value. This will also be significantly affected, thus objectively reflecting its overall quality. The evaluation values ​​of all nodes are ultimately normalized to a predetermined range to facilitate subsequent comparison and ranking.

[0045] Example 4 describes an engineering process initiated after the system generates a content credibility warning signal. This process aims to refine the initial assessment results by acquiring specific data and performing calculations to generate dynamic adjustment factors. The implementation is illustrated below with a specific case study.

[0046] Suppose that when the system processes a user query about "Alzheimer's disease drug treatment," a node named "Clinical Efficacy of Drug A" triggers a content credibility warning signal because its information source is relatively singular and lacks recent updates, resulting in a preliminary calculated comprehensive medical information assessment value that is lower than the system's preset lower threshold. The system then initiates the process of Example 4 for this specific node. First, the system needs to obtain the cross-modal association path density value of this node. The system comprehensively scans all data paths connected to the "Clinical Efficacy of Drug A" node in the knowledge graph and classifies and statistically analyzes them according to their modality type. The system accurately counts the number of graphic and textual data paths associated with this node, such as the number of linked drug chemical structure diagrams, mechanism of action diagrams, and detailed textual explanatory documents. Simultaneously, the system counts the number of associated video data paths, such as the number of links to expert interpretation videos, patient medication guidance animations, or related academic conference recordings. Furthermore, the system also counts the number of associated clinical case data paths, i.e., the number of case library entries linking to records of the drug's actual clinical application, patient feedback, and efficacy observation records. The sum of the number of the three types of paths mentioned above, after undergoing a standardized calculation process built into the system, ultimately generates the cross-modal association path density value for that node. This value reflects the diversity and richness of the information presentation methods of that node.

[0047] The system simultaneously monitors and acquires the node information update frequency value for that node. A retrospective time window is set, such as the most recent 24 months. Within this window, the system accesses the node's version control log, a structured file that records the history of all content changes. The system accurately calculates the number of valid version changes recorded in this log. A valid version change may include updates to treatment data, supplementation of side effect information, revision of medication guidelines, or addition of cited literature. Simple text polishing or formatting adjustments are usually not counted. The total number of changes counted during this monitoring period is processed by the system and directly generated as the node information update frequency value for that node. This value directly reflects the activity and timeliness of information content maintenance. To more clearly demonstrate the system's data acquisition when processing multiple nodes, refer to Table 1, which records an example of data collection results from the system at a certain moment for a group of nodes that triggered warnings.

[0048] Table 1: Monitoring table of node cross-modal data and update frequency.

[0049]

[0050] After acquiring the aforementioned raw data, the system enters the coupled calculation phase to generate dynamic adjustment factors. The specific calculation begins with processing the cross-modal association path density value. The system calls a preset path density baseline value. This baseline value is not a fixed value, but a representative value dynamically calculated based on historical data of similar nodes. The calculation method is: Path density baseline value = Sum of cross-modal association path density values ​​of similar nodes / Number of similar nodes. Here, "similar nodes" are categorized according to medical information type (e.g., when the current node is a "Drug A Clinical Efficacy" node, similar nodes are all drug treatment efficacy nodes). The cross-modal association path density value is the sum of the number of text / image paths, video paths, and clinical case paths associated with the node, i.e.: Cross-modal association path density value = Number of text / image paths + Number of video paths + Number of clinical case paths. Furthermore, the system recalculates and updates the path density baseline value daily at midnight based on full data of similar nodes from the past three months to ensure its timeliness and representativeness. The system generates a modal coverage coefficient by calculating the ratio. The calculation method is: Modal coverage coefficient = Cross-modal association path density value of the current node / Path density baseline value. This coefficient reflects the degree of information diversity of the node relative to the average level of similar nodes. If the coefficient is >1, it means that the information diversity (cross-modal resource coverage) of the current node is better than the average level of similar nodes; if the coefficient is =1, it means that its information diversity is on par with the average level of similar nodes; if the coefficient is <1, it means that its information diversity is lower than the average level of similar nodes.

[0051] After processing the cross-modal path density values, the system synchronously calculates the node information update frequency to generate a timeliness enhancement coefficient. The system compares the node's update frequency with a preset update frequency benchmark: update frequency ratio = current node's node information update frequency / update frequency benchmark. This benchmark is also dynamically generated by the system, for example, taking the median of all node update frequencies. The system performs logarithmic normalization on the ratio of the node information update frequency to the benchmark. This method, through the compression properties of the logarithmic function and linear interval mapping, standardizes the highly discrete original update frequency values ​​(e.g., 0-20 times / 24 months) to a preset reasonable range (0.05-2), eliminating the excessive influence of extreme values ​​on the evaluation results while preserving the relative differences in update activity among different nodes, ultimately generating the timeliness enhancement coefficient. The specific calculation consists of two steps: Basic logarithmic transformation: Based on the smoothing coefficient to eliminate interference from zero and extreme values, the formula is as follows: ,in This is the original update frequency. To update the frequency baseline value, The median value after logarithmic transformation. The smoothing factor (default 0.5); Interval mapping: Scaling the logarithmic transformation result to the target interval, the formula is: ,in , These are the minimum and maximum logarithmic transformation values ​​of nodes of the same type, respectively. This is a timeliness enhancement coefficient. , These represent the lower limit of the time-enhancing coefficient as 0.05 and the upper limit as 2, respectively.

[0052] For example, when the original update frequency benchmark value At that time, the calculated aging strengthening coefficient is approximately 0.618; when , At that time, the timeliness enhancement coefficient was approximately 1.463, achieving smoothing and standardization of the original data. This coefficient reflects the strength of node update activity relative to the baseline level.

[0053] The system multiplies the calculated modal coverage coefficient with the timeliness enhancement coefficient. The product of these two coefficients is the dynamic adjustment factor generated for that specific node. This factor is a comprehensive adjustment parameter that considers both the richness of information dimensions and the timeliness of information maintenance. If a node initially has a low score but is associated with multiple forms of learning materials and is frequently updated, the dynamic adjustment factor calculated through this process will be a large positive number, thus significantly improving its final evaluation value in subsequent corrections. Conversely, a node with monotonous and outdated content will have a very small dynamic adjustment factor, even approaching zero, and its evaluation value will not be effectively improved. This process introduces a more detailed and dynamic dimension of consideration for the evaluation of medical information quality.

[0054] Example 5 describes the final output stage of the entire retrieval method. Its core lies in the precise correction and sorting of the deeply analyzed target retrieval set. This process follows the calculation results, transforming abstract evaluation values ​​into an ordered list of retrieval results directly accessible to users. The following concrete example illustrates its operational mechanism: Suppose the system is processing a query about "interventional treatment for coronary artery disease." After all previous steps, the system has constructed a target retrieval set containing dozens of relevant nodes and calculated a comprehensive medical information evaluation value and a dynamic adjustment factor for each node in the set. At this point, the system finds that the node regarding "long-term efficacy of biodegradable stents" has a low comprehensive medical information evaluation value because some of the cited studies are still in the follow-up stage, resulting in a relatively preliminary level of evidence and triggering an early warning mechanism. However, subsequent analysis shows that this node is associated with abundant imaging data, surgical simulation videos, and early clinical cases from multiple centers (high cross-modal association path density value), and has three data update records within the past six months (high node information update frequency value), thus generating a larger dynamic adjustment factor.

[0055] The system performs a correction operation to obtain the final optimized medical information assessment value. The correction process is arithmetic; the system directly adds the comprehensive medical information assessment value of each node to the dynamically adjusted factor specifically generated for it. For the node "long-term efficacy of biodegradable stents" mentioned above, its originally low comprehensive assessment value was effectively improved by adding a significant positive dynamic adjustment factor. The significance of this correction process is that it does not negate the cautious judgment of the information evidence level in the initial assessment, but at the same time acknowledges the value of the information in terms of the diversity of its presentation format and the urgency of its updates, thus obtaining a more comprehensive and balanced final score. Conversely, a node about "old-style balloon dilation technology" may contain authoritative but outdated content with very few update records; its dynamic adjustment factor will be very small, so its optimized assessment value basically remains unchanged, objectively reflecting its current limited reference value.

[0056] After calculating the optimization evaluation value for all nodes in the target retrieval set, the system enters the priority sorting stage. The sorting algorithm arranges all nodes in descending order according to the calculated medical information optimization evaluation value. The node with the highest optimization evaluation value represents that its overall quality, timeliness, richness, and user suitability have all received the highest level of recognition from the system, and therefore it is placed at the top of the results list.

[0057] The ranking algorithm also includes pre-defined rules for handling ties. When two or more nodes have identical medical information optimization evaluation values, the system does not perform random ranking. Instead, it uses a secondary metric—the cross-modal association path density value—to break the tie. Nodes with higher cross-modal association path density values ​​will receive a higher ranking. The design philosophy behind this rule is that, when the overall evaluation value of information is comparable, information resources that can satisfy users' different learning preferences and comprehension abilities through various formats such as text, images, videos, and case studies are considered to have a slightly higher priority in presentation.

[0058] The system outputs a rigorously sorted list of medical information retrieval results. Taking the search for "interventional treatment of coronary artery disease" as an example, the top-ranked node might be "Latest Clinical Practice Guidelines," with an extremely high optimization evaluation value; closely following is likely the "Drug-eluting Stent Animation Demonstration" node, with a slightly lower evaluation value but extremely high cross-modal density; while some popular science articles with limited content and no long-term updates will be ranked at the bottom. This ordered list is presented to users through the user interface, effectively reducing the information filtering cost for users and guiding them to prioritize and consume high-quality, highly relevant medical information content.

[0059] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0060] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A medical information retrieval method based on an intelligent medical science popularization platform, characterized in that, Includes the following steps: Collect user search requests and divide the user search requests into medical concept entities and associated descriptive entities; Based on the medical concept entities, core concept nodes and adjacent extended nodes are extracted from the preset medical knowledge graph to generate an initial search range. Based on the associated description entities, semantic filtering is performed on the initial search range to select semantically matching nodes and construct the target search set; The medical information within the target retrieval set is analyzed to generate optimized evaluation values ​​for the medical information. The target retrieval set is prioritized and sorted according to the medical information optimization evaluation value, and the sorted medical information retrieval results are output.

2. The medical information retrieval method based on an intelligent medical science popularization platform according to claim 1, characterized in that, The step of analyzing the medical information within the target retrieval set and generating optimized evaluation values ​​for the medical information includes: Multi-dimensional feature extraction is performed on the medical information within the target retrieval set to obtain the authoritative feature value, timeliness feature value, and audience suitability feature value of the medical information; The authoritativeness, timeliness, and audience suitability of the medical information are weighted and calculated to generate a comprehensive evaluation value for the medical information. The comprehensive evaluation value of the medical information is compared with a preset evaluation threshold range. If the comprehensive evaluation value of the medical information is lower than the lower limit of the evaluation threshold range, a content credibility warning signal is generated. Based on the content credibility warning signal, the cross-modal association path density value and node information update frequency value of the semantic matching node are obtained; The cross-modal associated path density value and the node information update frequency value are coupled and calculated to generate a dynamic adjustment factor; Based on the aforementioned dynamic adjustment factor, the comprehensive evaluation value of medical information is corrected to obtain an optimized evaluation value of medical information.

3. The medical information retrieval method based on an intelligent medical science popularization platform according to claim 1, characterized in that, The specific method for generating the initial search range is as follows: Identify the main node position of the medical concept entity in the medical knowledge graph, and extract the first-level related nodes directly connected to the main node and their attribute labels; Traverse the interaction paths of the first-level associated nodes to obtain all second-level associated nodes and their relationship types; The main node, first-level associated nodes, and second-level associated nodes are merged to form the initial search range.

4. The medical information retrieval method based on an intelligent medical science popularization platform according to claim 2, characterized in that, The specific method for constructing the target retrieval set is as follows: Parse the syntactic structure of the associated descriptive entities and extract the keyword weight distribution and logical connection relationships; Calculate the semantic similarity value between each node within the initial retrieval range and the keyword weight distribution; Nodes with semantic similarity values ​​greater than a preset similarity threshold are selected and marked as semantically matched nodes; All semantically matching nodes and their associated edges are aggregated to form the target retrieval set.

5. A medical information retrieval method based on an intelligent medical science popularization platform according to claim 4, characterized in that, The specific method for multi-dimensional feature extraction is as follows: Extract the literature citation level and source institution level of the semantic matching node to generate the authority feature value of the medical information; Extract the latest revision timestamp and historical update cycle of the semantic matching node to generate the timeliness feature value; Extract the user's historical feedback data and reading difficulty coefficient of the semantic matching node to generate the audience suitability feature value.

6. A medical information retrieval method based on an intelligent medical science popularization platform according to claim 5, characterized in that, The specific method for generating comprehensive medical information evaluation values ​​is as follows: The first weighting coefficient of the authoritative feature value of the medical information, the second weighting coefficient of the timeliness feature value, and the third weighting coefficient of the audience suitability feature value are set. The comprehensive evaluation value of the medical information is obtained by multiplying the authoritative feature value of the medical information by the first weighting coefficient, the timeliness feature value by the second weighting coefficient, and the audience suitability feature value by the third weighting coefficient, and then summing them up.

7. A medical information retrieval method based on an intelligent medical science popularization platform according to claim 6, characterized in that, The specific method for obtaining cross-modal association path density values ​​and node information update frequency values ​​is as follows: The number of image and text data paths, video data paths, and clinical case data paths associated with the semantic matching nodes is counted, and the cross-modal association path density value is calculated. Monitor the number of version changes of the semantic matching node within a preset time window, and generate the node information update frequency value.

8. A medical information retrieval method based on an intelligent medical science popularization platform according to claim 7, characterized in that, The specific method for generating the dynamic adjustment factor is as follows: The modal coverage coefficient is obtained by calculating the ratio between the cross-modal associated path density value and the path density benchmark value. The node information update frequency value and the update frequency benchmark value are logarithmically normalized to obtain the timeliness enhancement coefficient; The modal coverage coefficient is multiplied by the time-effect enhancement coefficient to generate the dynamic adjustment factor.

9. A medical information retrieval method based on an intelligent medical science popularization platform according to claim 8, characterized in that, The specific method for correcting the comprehensive evaluation value of the medical information to obtain the optimized evaluation value of the medical information is as follows: The comprehensive evaluation value of the medical information is added to the dynamic adjustment factor to obtain the optimized evaluation value of the medical information.

10. A medical information retrieval method based on an intelligent medical science popularization platform according to claim 9, characterized in that, The specific method for prioritizing the target retrieval set based on the medical information optimization evaluation value is as follows: The nodes in the target retrieval set are sorted from high to low according to the medical information optimization evaluation value; When the medical information optimization evaluation values ​​are the same, a secondary sort is performed based on the cross-modal association path density value.

Images