Mobile nursing decision support method and device integrated with nursing knowledge base

Abstract

The invention belongs to the technical field of reasoning decision, and particularly relates to a mobile nursing decision support method and device of an integrated nursing knowledge base. The mobile nursing decision support method integrated with the nursing knowledge base comprises the following steps: firstly, constructing the nursing knowledge base, wherein the ontology relationship of the nursing knowledge base is represented by a tetrad containing confidence and a time decay function; acquiring patient data, mapping the patient data into an instance graph, and carrying out concurrent search in the knowledge base to generate a plurality of candidate reasoning paths; calculating a comprehensive semantic vector and a comprehensive confidence coefficient of each path; performing density clustering on the paths based on semantic vector similarity to form path beams, screening the path beams of which the aggregation confidence is higher than a threshold value and the beam endpoint dispersion is lower than the threshold value, and determining the central endpoints of the path beams as high-priority nursing decisions; and analyzing the decision, combining semantic information of the path bundle, generating a customized nursing instruction from the knowledge base, and issuing the customized nursing instruction to the mobile terminal.

Description

Technical Field

[0001] This invention belongs to the field of reasoning and decision-making technology, specifically a mobile nursing decision support method and device integrating a nursing knowledge base. Background Technology

[0002] Clinical decision support systems (CMS) can help nurses make quick and accurate judgments in complex and ever-changing clinical environments. Nursing CMS systems are mostly based on rule engines or expert systems, using pre-defined "if-then" rules for reasoning. However, the construction and maintenance of knowledge bases are costly, the rules are rigid, and they struggle to adapt to rapidly evolving medical knowledge and individualized patient conditions; moreover, they lack deep semantic understanding capabilities.

[0003] In recent years, knowledge graph-based technologies have begun to be applied to the field of decision support. Knowledge graphs, through entities and relationships, structurally store massive amounts of medical knowledge, enabling them to better express complex relationships between concepts. However, most knowledge graphs employ a standard triple (subject, predicate, object) structure, making it difficult to address the inherent uncertainty of the knowledge itself. For example, the reliability of certain nursing knowledge can vary, or a patient's symptom may be time-sensitive; these information cannot be effectively represented in traditional triples. Secondly, existing reasoning methods often rely on single-path search, making them susceptible to interference from noisy data or isolated evidence, leading to unstable reasoning results. When multiple chains of evidence point to similar but not identical conclusions, comprehensive judgment is difficult. Therefore, how to construct a nursing knowledge base that integrates knowledge confidence and, based on this, achieves aggregated reasoning across multiple evidence chains to generate highly reliable and consensus-based decision recommendations is a pressing technical problem in the field of mobile nursing decision support. Summary of the Invention

[0004] To improve the reliability of mobile nursing decision-making, this invention provides a mobile nursing decision support method integrating a nursing knowledge base, comprising the following steps: An integrated nursing knowledge base is constructed, in which entities and relations within the knowledge base have entity embedding vectors and relation embedding vectors, respectively. Ontology relations are represented by quadruples (subject, predicate, object, Ψ), where Ψ is an attribute set. The attribute set includes a confidence parameter for representing the reliability of the relation and a time decay function for calculating timeliness. The system acquires real-time vital signs data and historical nursing records of patients and maps them to a patient-specific instance graph. Starting from one or more nodes in the instance graph, it performs a multi-path concurrent search in the knowledge base to generate a set of candidate inference paths pointing to potential nursing events. For each candidate inference path, it combines the relationship embedding vectors of all relations on the path level by level, and combines them with the entity embedding vector of the starting node to calculate the comprehensive semantic vector of the path. Based on the confidence parameters of each relation on the path and the time decay function, it calculates the comprehensive confidence of the path. Based on the cosine similarity between the comprehensive semantic vectors, density clustering is performed on all candidate inference paths to obtain at least one inference path bundle; the aggregate confidence and bundle endpoint dispersion of each inference path bundle are calculated; when the aggregate confidence of a path bundle is higher than a preset confidence threshold and the bundle endpoint dispersion is lower than a preset dispersion threshold, the event corresponding to the central endpoint node of the path bundle is determined as a high-priority nursing decision. The high-priority nursing decisions are analyzed, and combined with the average comprehensive semantic vector of the path bundle that triggers the high-priority nursing decisions, customized nursing intervention instructions are retrieved from the knowledge base and generated, and the instructions are sent to the mobile nursing terminal.

[0005] Preferably, the aggregate confidence is the sum of the comprehensive confidence of all paths within the bundle; the bundle endpoint dispersion is the average pairwise Euclidean distance between the endpoint nodes of all paths within the bundle in the embedding space.

[0006] Preferably, for each candidate inference path, the step of combining the relation embedding vectors of all relations along the path level by level, and combining them with the entity embedding vector of the starting node, to calculate the comprehensive semantic vector of the path includes: path The comprehensive semantic vector S(P) is calculated using vector addition: ,in Embed the vector for the starting node entity. Let be the embedding vector of the i-th relation on the path.

[0007] Preferably, calculating the overall confidence of the path based on the confidence parameters of each relation on the path and the time decay function includes: The formula for calculating the overall confidence of a candidate inference path P containing n relations is: ,in Let be the confidence parameter for the i-th relation. The value of the time decay function applied to the i-th relation is calculated.

[0008] Preferably, the step of performing density clustering on all candidate inference paths based on the cosine similarity between the comprehensive semantic vectors includes: The DBSCAN algorithm is used for clustering, and the distance between paths is measured by subtracting the cosine similarity from 1. The neighborhood radius is 0.8, and the minimum number of neighborhood points (MinPts) required to form a cluster is 3.

[0009] Preferably, determining the event corresponding to the central endpoint node of the pathway bundle as a high-priority nursing decision includes: Calculate the arithmetic mean of the embedding vectors of all endpoint node entities within the path bundle in the embedding space to obtain a geometric center coordinate; Traverse all endpoint nodes within the path bundle, calculate the Euclidean distance between the endpoint node embedding vector and the geometric center coordinates, and select the endpoint node with the smallest distance as the central endpoint node.

[0010] Preferably, the step of parsing the high-priority nursing decision and, in conjunction with the average comprehensive semantic vector of the path bundle that triggered the high-priority nursing decision, retrieving and generating customized nursing intervention instructions from the knowledge base includes: Starting with the event corresponding to the high-priority nursing decision, a set of candidate nursing intervention entities associated with the starting point are retrieved from the knowledge base; Calculate the arithmetic mean of the comprehensive semantic vectors of all paths within the path bundle to obtain the average comprehensive semantic vector; calculate the cosine similarity between the average comprehensive semantic vector and the embedding vector of each candidate nursing intervention entity. The candidate nursing interventions were ranked according to cosine similarity, and the top 3 nursing interventions were selected. The corresponding standard operating procedure texts were then combined into instructions for issuance.

[0011] The present invention also provides a mobile nursing decision support device integrating a nursing knowledge base, comprising the following modules: The knowledge base construction module is used to build an integrated nursing knowledge base, in which entities and relations in the knowledge base have entity embedding vectors and relation embedding vectors, respectively. Ontology relations are represented by quadruples (subject, predicate, object, Ψ), where Ψ is an attribute set. The attribute set includes a confidence parameter for representing the reliability of the relation and a time decay function for calculating the timeliness. The path confidence calculation module is used to acquire real-time patient vital signs data and historical nursing records, and map them into a patient-specific instance graph; starting from one or more nodes in the instance graph, it performs a multi-path concurrent search in the knowledge base to generate a set of candidate inference paths pointing to potential nursing events; for each candidate inference path, it combines the relationship embedding vectors of all relations on the path level by level, and combines them with the entity embedding vector of the starting node to calculate the comprehensive semantic vector of the path; based on the confidence parameters of each relation on the path and the time decay function, it calculates the comprehensive confidence of the path; The decision module is used to perform density clustering on all candidate inference paths based on the cosine similarity between the comprehensive semantic vectors to obtain at least one inference path bundle; calculate the aggregate confidence and bundle endpoint dispersion of each inference path bundle; when the aggregate confidence of a path bundle is higher than a preset confidence threshold and the bundle endpoint dispersion is lower than a preset dispersion threshold, the event corresponding to the central endpoint node of the path bundle is determined as a high-priority nursing decision. The nursing instruction generation module is used to parse the high-priority nursing decision, and combine it with the average comprehensive semantic vector of the path bundle that triggers the high-priority nursing decision to retrieve and generate customized nursing intervention instructions from the knowledge base, and then send the instructions to the mobile nursing terminal.

[0012] Preferably, the aggregate confidence is the sum of the comprehensive confidence of all paths within the bundle; the bundle endpoint dispersion is the average pairwise Euclidean distance between the endpoint nodes of all paths within the bundle in the embedding space.

[0013] Preferably, for each candidate inference path, the step of combining the relation embedding vectors of all relations along the path level by level, and combining them with the entity embedding vector of the starting node, to calculate the comprehensive semantic vector of the path includes: path The comprehensive semantic vector S(P) is calculated using vector addition: ,in Embed the vector for the starting node entity. Let be the embedding vector of the i-th relation on the path.

[0014] Preferably, calculating the overall confidence of the path based on the confidence parameters of each relation on the path and the time decay function includes: The formula for calculating the overall confidence of a candidate inference path P containing n relations is: ,in Let be the confidence parameter for the i-th relation. The value of the time decay function applied to the i-th relation is calculated.

[0015] Preferably, the step of performing density clustering on all candidate inference paths based on the cosine similarity between the comprehensive semantic vectors includes: The DBSCAN algorithm is used for clustering, and the distance between paths is measured by subtracting the cosine similarity from 1. The neighborhood radius is 0.8, and the minimum number of neighborhood points (MinPts) required to form a cluster is 3.

[0016] Preferably, determining the event corresponding to the central endpoint node of the pathway bundle as a high-priority nursing decision includes: Calculate the arithmetic mean of the embedding vectors of all endpoint node entities within the path bundle in the embedding space to obtain a geometric center coordinate; Traverse all endpoint nodes within the path bundle, calculate the Euclidean distance between the endpoint node embedding vector and the geometric center coordinates, and select the endpoint node with the smallest distance as the central endpoint node.

[0017] Preferably, the step of parsing the high-priority nursing decision and, in conjunction with the average comprehensive semantic vector of the path bundle that triggered the high-priority nursing decision, retrieving and generating customized nursing intervention instructions from the knowledge base includes: Starting with the event corresponding to the high-priority nursing decision, a set of candidate nursing intervention entities associated with the starting point are retrieved from the knowledge base; Calculate the arithmetic mean of the comprehensive semantic vectors of all paths within the path bundle to obtain the average comprehensive semantic vector; calculate the cosine similarity between the average comprehensive semantic vector and the embedding vector of each candidate nursing intervention entity. The candidate nursing interventions were ranked according to cosine similarity, and the top 3 nursing interventions were selected. The corresponding standard operating procedure texts were then combined into instructions for issuance.

[0018] Compared with the prior art, the present invention has the following beneficial effects: This invention constructs a knowledge base by introducing quadruples containing confidence and time decay functions, enabling the calculation of the reliability and timeliness of nursing knowledge and improving the accuracy of knowledge. Through multi-path concurrent search and density clustering, multiple semantically similar chains of reasoning evidence are aggregated into a bundle of reasoning paths, avoiding the one-sidedness caused by a single reasoning path. Moreover, by applying dual constraints to the aggregated confidence and the dispersion of the bundle endpoints, the final nursing decision is based on a highly consistent conclusion pointed to by multiple pieces of evidence, thereby improving the effectiveness of guidance. Attached Figure Description

[0019] Figure 1 This is a flowchart of Example 1; Figure 2 This is a patient-specific example image; Figure 3 This is a schematic diagram of path bundle generation for clustering. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention. It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this invention are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of related data must comply with relevant laws, regulations, and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0021] Example 1 like Figure 1 As shown, a mobile nursing decision support method integrating a nursing knowledge base includes the following steps: S1. Construct an integrated nursing knowledge base, in which entities and relations within the knowledge base have entity embedding vectors and relation embedding vectors, respectively. Ontology relations are represented by quadruples (subject, predicate, object, Ψ), where Ψ is an attribute set. The attribute set includes a confidence parameter for representing the reliability of the relation and a time decay function for calculating timeliness. The knowledge sources of the nursing knowledge base mainly include nursing guidelines, medical textbooks, and papers. In one embodiment, a massive number of nursing entities and relationships are extracted from sources including but not limited to clinical nursing guidelines, medical textbooks, papers, and desensitized electronic medical records using named entity recognition and relation extraction techniques. For each extracted relationship in the form of subject-predicate-object, such as patient Zhang San-chief complaint-cough, an attribute set is attached. This attribute set includes a confidence parameter and a time decay function. The confidence parameter can be assigned a value based on the authority of the knowledge source. For example, the confidence of a relationship from a national guideline is 0.95, while that from a non-core journal article is 0.7. The time decay function is a function with time difference as the variable, such as an exponential decay function, used to represent the characteristic that the importance of certain facts decreases over time. For example, the timeliness of a blood routine test result three days ago is lower than that of one hour ago. All four-tuple data are used as input and trained using knowledge graph embedding models such as TransE or RotatE to generate a unique low-dimensional vector representation for each entity and relationship in the knowledge base, namely, an entity embedding vector and a relationship embedding vector. In an alternative embodiment, entity embedding vectors and relation embedding vectors can be generated using an encoded method.

[0022] For example, in a clinical application scenario, an event such as patient A having a heart rate of 120 beats per minute measured at 10:00 AM on March 15, 2023, is represented as a quaternion [patient A, heart rate, 120 beats per minute, Ψ]. Here, Ψ includes the timestamp of the event, i.e., the system time corresponding to 10:00 AM on March 15, 2023, and the initial confidence level of the record.

[0023] Assuming the current system time is 12:00 PM on March 15, 2023, and the aforementioned heart rate measurement event occurred at 10:00 AM on the same day, the time difference is 2 hours. If the decay coefficient λ is set to 0.05 per hour, the time decay function value of this event relationship is calculated as F(t) = exp(-0.05 × 2), approximately equal to 0.905. This means the timeliness weight of the event information decays from an initial 1 to 0.905. In contrast, events occurring more than 24 hours prior exhibit more significant time decay and a lower weight, allowing decisions to focus more on the most recent changes in the patient's vital signs.

[0024] S2, acquire real-time vital signs data and historical nursing records of the patient, and map them into a patient-specific instance graph; starting from one or more nodes of the instance graph, perform multi-path concurrent search in the knowledge base to generate a set of candidate inference paths pointing to potential nursing events; for each candidate inference path, combine the relationship embedding vectors of all relations on the path level by level, and combine them with the entity embedding vector of the starting node to calculate the comprehensive semantic vector of the path; calculate the comprehensive confidence of the path based on the confidence parameters of each relation on the path and the time decay function; The system connects to monitoring devices in real time via an interface to acquire data such as heart rate, blood pressure, and body temperature. It also retrieves the patient's electronic medical record from the hospital information system, including historical diagnoses, medication records, and examination reports. The acquired unstructured or semi-structured data undergoes standardization and semantic mapping. For example, a body temperature of 39 degrees Celsius in the text record is mapped to the defined entity "high fever" in the knowledge base, and a white blood cell count of 18 x 10^9 per liter is mapped to the entity "elevated white blood cells." Newly mapped entities from the patient data, along with their relationships to the patient entity itself (e.g., patient Zhang San - symptoms - high fever), together constitute a local knowledge subgraph. Figure 2 As shown, since each patient's situation is different, the generated patient-specific instance diagrams are also different.

[0025] Key nodes in the patient-specific instance graph, such as high fever, cough, and leukocytosis, are used as starting nodes for the search. Parallel breadth-first search or random walk algorithms are employed to simultaneously explore paths from the starting nodes to the target type nodes, i.e., nursing event or nursing suggestion nodes, throughout the entire integrated nursing knowledge base. For example, starting from the high fever node, path 1 might be found: high fever - symptom - pneumonia - recommended care - monitoring blood oxygen saturation; simultaneously, starting from the cough node, path 2 might be found: cough - symptom - pneumonia - recommended care - monitoring blood oxygen saturation; and path 3 might be found: cough - symptom - bronchitis - recommended care - nebulized inhalation. A maximum search depth and the number of paths expanded at each step are set, and all found complete paths ending with a nursing event are aggregated to form a candidate inference path set.

[0026] For a candidate inference path, such as high fever -> relation 1 -> pneumonia -> relation 2 -> monitoring blood oxygen saturation, the comprehensive semantic vector is calculated by adding the entity embedding vector of the starting node "high fever" to the relation embedding vectors of relation 1 and relation 2 on the path. In one embodiment, the comprehensive confidence is calculated by multiplying the confidence of all relations on the path. If a relation contains a time decay function, the timeliness value of the relation is first calculated based on the current time and the time of the event, and then this value is multiplied by other confidence values. For example, if the confidence of two relations on the path is 0.9 and 0.8 respectively, and the current timeliness value calculated by the time decay function of one of the relations is 0.95, then the comprehensive confidence of the path is 0.9 × 0.8 × 0.95.

[0027] In an optional embodiment, for each candidate inference path, the step of combining the relation embedding vectors of all relations along the path level by level, and combining them with the entity embedding vector of the starting node, to calculate the comprehensive semantic vector of the path includes: path The comprehensive semantic vector S(P) is calculated using vector addition: ,in Embed the vector for the starting node entity. Let be the embedding vector of the i-th relation on the path. For the relationship, It is an entity.

[0028] In a knowledge graph containing patient information, a reasoning path might be: Patient A → Diagnosed with → Pneumonia → Complications → Respiratory Failure. Patient A, pneumonia, and respiratory failure are entity nodes, while "diagnosed with" and "complications" are relations. Each entity and relation is mapped to a high-dimensional vector; for example, V(Patient A) might be [0.2, 0.9, -0.1], V(diagnosed with) might be [0.5, -0.3, 0.4], and V(complications) might be [0.6, -0.2, 0.5]. The mapping process can be achieved using an encoder-decoder model or other encoding methods; this invention does not specifically limit the mapping method.

[0029] The comprehensive semantic vector S(P) of the above path is [0.2,0.9,-0.1]+[0.5,-0.3,0.4]+[0.6,-0.2,0.5]=[1.3,0.4,0.8]. The resulting vector S(P) becomes the semantic representation of the inference chain "Patient A suffers respiratory failure due to pneumonia" in the vector space. Different inference paths will have different comprehensive semantic vectors due to differences in relation vectors.

[0030] In an optional embodiment, calculating the overall confidence of the path based on the confidence parameters of each relation on the path and the time decay function includes: The formula for calculating the overall confidence of a candidate inference path P containing n relations is: ,in Let be the confidence parameter for the i-th relation. The value of the time decay function applied to the i-th relation is calculated.

[0031] Taking the inference path "Patient B → Chief Complaint → Chest Pain → Recommended Examination → Electrocardiogram" as an example, this path consists of two core relations: r1 is "Patient B's chief complaint is chest pain," and r2 is "Chest pain, recommended examination: electrocardiogram." Assume relation r1 is based on an outpatient record from 24 hours prior, with an initial confidence level C(r1) of 0.9. After time decay calculation, its timeliness function value F1(t) decreases to 0.7. Relation r2 is based on standard clinical guidelines, with a confidence level C(r2) of 0.99, which does not change over time, and its timeliness function value F2(t) remains constant at 1. The overall confidence level of the path is C(P) = [C(r1) × F1(t)] × [C(r2) × F2(t)] = 0.624. If the patient's chest pain symptoms have just occurred, the value of F1(t) will be closer to 1, resulting in a higher final overall confidence level. This ensures that reasoning paths that include both high-confidence facts and are based on the latest information receive higher scores and are therefore prioritized in subsequent decisions. To avoid the product of the overall confidence score approaching zero as the path length increases, in another embodiment… .

[0032] S3. Based on the cosine similarity between the comprehensive semantic vectors, perform density clustering on all candidate inference paths to obtain at least one inference path bundle; calculate the aggregate confidence and bundle endpoint dispersion of each inference path bundle; when the aggregate confidence of a path bundle is higher than a preset confidence threshold and the bundle endpoint dispersion is lower than a preset dispersion threshold, the event corresponding to the central endpoint node of the path bundle is determined as a high-priority nursing decision. The cosine similarity between the combined semantic vectors of any two candidate inference paths is calculated to obtain a similarity matrix. Then, the DBSCAN density clustering algorithm is used, setting a neighborhood radius parameter and a minimum sample number parameter. Paths whose semantic vectors are close in distance in the vector space, i.e., have high cosine similarity, are automatically grouped into the same cluster. For example, multiple paths originating from different symptoms but all pointing to pneumonia-related care will be clustered into one class, forming an inference path bundle, due to their similar semantic vectors. Figure 3 As shown; however, the path pointing to bronchitis care, because its semantic vector differs significantly from the former, will form another independent path bundle or be regarded as an outlier. For a given inference path bundle, the aggregate confidence scores of all paths contained within the bundle are summed to obtain the bundle's aggregate confidence score; for example, if a bundle contains three paths with aggregate confidence scores of 0.68, 0.71, and 0.65 respectively, its aggregate confidence score is 2.04; simultaneously, the endpoint nodes of all paths within the bundle are extracted, such as those for monitoring blood oxygen saturation, assessing respiratory status, and recording respiratory rate, and the entity embedding vectors of these endpoint nodes are obtained; the Euclidean distance between the entity embedding vectors of every two endpoint nodes is calculated, and all these paired distance values ​​are averaged to obtain the bundle endpoint dispersion.

[0033] A pre-set aggregation confidence threshold is set, for example, 2.0, and a bundle endpoint dispersion threshold is set to 0.5. If the aggregation confidence of a path bundle is 2.04, which is higher than 2.0, and its bundle endpoint dispersion is 0.3, which is lower than 0.5, then the path bundle meets the conditions. Further, the centroid of the embedding vectors of all endpoint nodes in the path bundle is calculated, i.e., the average vector. Among all endpoint nodes, the node whose embedding vector is closest to the centroid vector is found, and this node is the central endpoint node. For example, if the embedding vector of monitoring blood oxygen saturation is closest to the centroid, then the event of monitoring blood oxygen saturation is determined as a high-priority care decision.

[0034] In an optional embodiment, performing density clustering on all candidate inference paths based on the cosine similarity between the comprehensive semantic vectors includes: The DBSCAN algorithm is used for clustering, and the distance between paths is measured by subtracting the cosine similarity from 1. The neighborhood radius is 0.8, and the minimum number of neighborhood points (MinPts) required to form a cluster is 3.

[0035] Suppose that multiple candidate reasoning paths pointing to different potential problems are generated for a patient, and the comprehensive semantic vector of each path has been calculated. For example, paths A and C both point to lung infection, path B points to viral cold, and path D points to digestive system problems. First, the cosine similarity between each pair of these path vectors is calculated. If the semantic vectors of paths A and C have a small angle, their similarity is 0.9, and the distance between them is 0.1. However, the semantic vectors of paths A and D have very different directions, and their similarity may only be 0.1, resulting in a distance of 0.9. When applying the DBSCAN algorithm, the neighborhood radius Eps is set to 0.8, and the minimum number of neighborhood points MinPts is set to 3. The distance between paths A and C, 0.1, is less than 0.8, so they are neighbors of each other. If there is a third path E also pointing to lung infection and its distance from both A and C is less than 0.8, then when path A is examined, it is found that its neighborhood contains both C and E, reaching a total of 3, satisfying the MinPts requirement. Path A becomes the core object and forms an initial cluster with C and E. The cluster represents a reasoning direction about lung infection supported by multiple chains of evidence. Path D, which is farther away from them, is marked as noise because it does not have enough nearest neighbors in its neighborhood.

[0036] In an optional embodiment, determining the event corresponding to the central endpoint node of the pathway bundle as a high-priority care decision includes: Calculate the arithmetic mean of the embedding vectors of all endpoint node entities within the path bundle in the embedding space to obtain a geometric center coordinate; Traverse all endpoint nodes within the path bundle, calculate the Euclidean distance between the endpoint node embedding vector and the geometric center coordinates, and select the endpoint node with the smallest distance as the central endpoint node.

[0037] Suppose clustering yields a path bundle containing five paths, each pointing to a semantically similar but not identical endpoint event: acute myocardial infarction, angina attack, coronary artery occlusion, myocardial ischemia, and unstable angina. Each endpoint event corresponds to an embedding vector V1, V2, V3, V4, and V5. The arithmetic mean of these five vectors, (V1 + V2 + V3 + V4 + V5) / 5, yields a geometric center coordinate C representing the semantic core of this cluster in vector space. The Euclidean distance from the embedding vector of each endpoint node to this center coordinate C is calculated. For example, the distance d1 between V1 and C, d2 between V2 and C, and so on. Assume the calculated results are d1 = 0.2, d2 = 0.5, d3 = 0.3, d4 = 0.4, and d5 = 0.6. By comparison, the smallest value for d1 is found, indicating that the vector V1 corresponding to acute myocardial infarction is semantically closest to the core that this path bundle points to. Acute myocardial infarction was identified as the central endpoint of this pathway and used as a key event to trigger high-priority care decisions.

[0038] S4. The high-priority nursing decision is parsed, and combined with the average comprehensive semantic vector of the path bundle that triggered the high-priority nursing decision, a customized nursing intervention instruction is retrieved from the knowledge base and generated, and the instruction is sent to the mobile nursing terminal.

[0039] The process begins by identifying high-priority nursing decisions, such as monitoring blood oxygen saturation. This involves calculating the average semantic vector of all paths within the triggering pathway bundle. Using this average vector as a retrieval criterion, the system searches the knowledge base for the most similar nursing procedures, precautions, and theoretical justifications based on vector similarity. For example, search results might include knowledge fragments such as standard blood oxygen monitoring procedures, low blood oxygen alarm parameter settings, and key observation points for related complications. The retrieved information is then integrated into a structured natural language instruction, such as a high-priority suggestion: monitor blood oxygen saturation; specific operation: use a finger-clip pulse oximeter for continuous monitoring, ensuring the value is not lower than 93%; supporting evidence: multiple current indicators of the patient point to a risk of lung infection. This structured natural language instruction can be generated using a large language model or a self-trained open-source model. Finally, this instruction is pushed to the nurse's handheld mobile device application interface via a wireless network.

[0040] In an optional embodiment, parsing the high-priority care decision and combining it with the average comprehensive semantic vector of the path bundle that triggered the high-priority care decision to retrieve and generate customized care intervention instructions from the knowledge base includes: Starting with the event corresponding to the high-priority nursing decision, a set of candidate nursing intervention entities associated with the starting point are retrieved from the knowledge base; Calculate the arithmetic mean of the comprehensive semantic vectors of all paths within the path bundle to obtain the average comprehensive semantic vector; calculate the cosine similarity between the average comprehensive semantic vector and the embedding vector of each candidate nursing intervention entity. The candidate nursing interventions were ranked according to cosine similarity, and the top 3 nursing interventions were selected. The corresponding standard operating procedure texts were then combined into instructions for issuance.

[0041] Using the example above, the high-priority decision event has been identified as acute myocardial infarction. A search of the knowledge base for nursing interventions directly associated with acute myocardial infarction yields a candidate list, such as immediate oxygen administration, establishing intravenous access, sublingual nitroglycerin administration, and performing a 12-lead electrocardiogram. Simultaneously, the average comprehensive semantic vector of the pathway bundle triggering this decision is calculated. This average comprehensive semantic vector not only contains the conclusion of acute myocardial infarction but also incorporates information that led to this conclusion, such as key information that the pathway might include low blood pressure or a history of aspirin allergy.

[0042] The cosine similarity of the average comprehensive semantic vector with the embedding vector of each nursing intervention in the candidate list was calculated. It was assumed that the average comprehensive semantic vector had a similarity of 0.95 with establishing intravenous access, 0.92 with immediate oxygen administration, and 0.89 with performing a 12-lead electrocardiogram. The similarity with sublingual nitroglycerin was reduced to 0.5 due to the presence of information about low blood pressure. The top three interventions were selected based on their similarity scores, ranked from highest to lowest. The standard operating procedure texts corresponding to these three interventions were retrieved from the knowledge base and combined into a clear instruction: Immediately establish intravenous access for the patient, administer oxygen via nasal cannula, and perform a 12-lead electrocardiogram.

[0043] Example 2 A mobile nursing decision support device integrating a nursing knowledge base includes the following modules: The knowledge base construction module is used to build an integrated nursing knowledge base, in which entities and relations in the knowledge base have entity embedding vectors and relation embedding vectors, respectively. Ontology relations are represented by quadruples (subject, predicate, object, Ψ), where Ψ is an attribute set. The attribute set includes a confidence parameter for representing the reliability of the relation and a time decay function for calculating the timeliness. The path confidence calculation module is used to acquire real-time patient vital signs data and historical nursing records, and map them into a patient-specific instance graph; starting from one or more nodes in the instance graph, it performs a multi-path concurrent search in the knowledge base to generate a set of candidate inference paths pointing to potential nursing events; for each candidate inference path, it combines the relationship embedding vectors of all relations on the path level by level, and combines them with the entity embedding vector of the starting node to calculate the comprehensive semantic vector of the path; based on the confidence parameters of each relation on the path and the time decay function, it calculates the comprehensive confidence of the path; The decision module is used to perform density clustering on all candidate inference paths based on the cosine similarity between the comprehensive semantic vectors to obtain at least one inference path bundle; calculate the aggregate confidence and bundle endpoint dispersion of each inference path bundle; when the aggregate confidence of a path bundle is higher than a preset confidence threshold and the bundle endpoint dispersion is lower than a preset dispersion threshold, the event corresponding to the central endpoint node of the path bundle is determined as a high-priority nursing decision. The nursing instruction generation module is used to parse the high-priority nursing decision, and combine it with the average comprehensive semantic vector of the path bundle that triggers the high-priority nursing decision to retrieve and generate customized nursing intervention instructions from the knowledge base, and then send the instructions to the mobile nursing terminal.

[0044] In an optional embodiment, the aggregate confidence is the sum of the combined confidence of all paths within the bundle; the bundle endpoint dispersion is the average pairwise Euclidean distance between the endpoint nodes of all paths within the bundle in the embedding space.

[0045] In an optional embodiment, for each candidate inference path, the step of combining the relation embedding vectors of all relations along the path level by level, and combining them with the entity embedding vector of the starting node, to calculate the comprehensive semantic vector of the path includes: path The comprehensive semantic vector S(P) is calculated using vector addition: ,in Embed the vector for the starting node entity. Let be the embedding vector of the i-th relation on the path.

[0046] In an optional embodiment, calculating the overall confidence of the path based on the confidence parameters of each relation on the path and the time decay function includes: The formula for calculating the overall confidence of a candidate inference path P containing n relations is: ,in Let be the confidence parameter for the i-th relation. The value of the time decay function applied to the i-th relation is calculated.

[0047] In an optional embodiment, performing density clustering on all candidate inference paths based on the cosine similarity between the comprehensive semantic vectors includes: The DBSCAN algorithm is used for clustering, and the distance between paths is measured by subtracting the cosine similarity from 1. The neighborhood radius is 0.8, and the minimum number of neighborhood points (MinPts) required to form a cluster is 3.

[0048] In an optional embodiment, determining the event corresponding to the central endpoint node of the pathway bundle as a high-priority care decision includes: Calculate the arithmetic mean of the embedding vectors of all endpoint node entities within the path bundle in the embedding space to obtain a geometric center coordinate; Traverse all endpoint nodes within the path bundle, calculate the Euclidean distance between the endpoint node embedding vector and the geometric center coordinates, and select the endpoint node with the smallest distance as the central endpoint node.

[0049] In an optional embodiment, parsing the high-priority care decision and combining it with the average comprehensive semantic vector of the path bundle that triggered the high-priority care decision to retrieve and generate customized care intervention instructions from the knowledge base includes: Starting with the event corresponding to the high-priority nursing decision, a set of candidate nursing intervention entities associated with the starting point are retrieved from the knowledge base; Calculate the arithmetic mean of the comprehensive semantic vectors of all paths within the path bundle to obtain the average comprehensive semantic vector; calculate the cosine similarity between the average comprehensive semantic vector and the embedding vector of each candidate nursing intervention entity. The candidate nursing interventions were ranked according to cosine similarity, and the top 3 nursing interventions were selected. The corresponding standard operating procedure texts were then combined into instructions for issuance.

[0050] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the mobile nursing decision support method of the integrated nursing knowledge base described in various embodiments or some parts of the embodiments of the present invention.

[0051] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and relevant parts can be referred to the descriptions in the method embodiments. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0052] The method and electronic device for providing product object information provided by the present invention have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A mobile nursing decision support method integrating a nursing knowledge base, characterized in that, Includes the following steps: An integrated nursing knowledge base is constructed, in which entities and relations within the knowledge base have entity embedding vectors and relation embedding vectors, respectively. Ontology relations are represented by quadruples (subject, predicate, object, Ψ), where Ψ is an attribute set. The attribute set includes a confidence parameter for representing the reliability of the relation and a time decay function for calculating timeliness. Acquire real-time vital signs data and historical nursing records of patients, and map them into patient-specific instance graphs; Starting from one or more nodes in the instance graph, a multi-path concurrent search is performed in the knowledge base to generate a set of candidate inference paths pointing to potential care events. For each candidate inference path, the relation embedding vectors of all relations on the path are combined level by level along the path, and combined with the entity embedding vector of the starting node to calculate the comprehensive semantic vector of the path. The comprehensive confidence of the path is calculated based on the confidence parameters of each relation on the path and the time decay function. Based on the cosine similarity between the comprehensive semantic vectors, density clustering is performed on all candidate inference paths to obtain at least one inference path bundle; the aggregation confidence and bundle endpoint dispersion of each inference path bundle are calculated. When the aggregation confidence of a certain path bundle is higher than the preset confidence threshold and the dispersion of the bundle endpoints is lower than the preset dispersion threshold, the event corresponding to the central endpoint node of the path bundle is determined as a high-priority nursing decision. The high-priority nursing decisions are analyzed, and combined with the average comprehensive semantic vector of the path bundle that triggers the high-priority nursing decisions, customized nursing intervention instructions are retrieved from the knowledge base and generated, and the instructions are sent to the mobile nursing terminal.

2. The mobile nursing decision support method based on an integrated nursing knowledge base according to claim 1, characterized in that, The aggregate confidence is the sum of the comprehensive confidence of all paths within the bundle; the bundle endpoint dispersion is the average pairwise Euclidean distance of the endpoint nodes of all paths within the bundle in the embedding space.

3. The mobile nursing decision support method based on an integrated nursing knowledge base according to claim 1, characterized in that, For each candidate reasoning path, the relation embedding vectors of all relations along the path are combined level by level, and combined with the entity embedding vector of the starting node to calculate the comprehensive semantic vector of the path, including: path The comprehensive semantic vector S(P) is calculated using vector addition: ,in Embed the vector for the starting node entity. Let be the embedding vector of the i-th relation on the path.

4. The mobile nursing decision support method based on an integrated nursing knowledge base according to claim 1, characterized in that, The step of calculating the overall confidence of the path based on the confidence parameters of each relation on the path and the time decay function includes: The formula for calculating the overall confidence of a candidate inference path P containing n relations is: ,in Let be the confidence parameter for the i-th relation. The value of the time decay function applied to the i-th relation is calculated.

5. The mobile nursing decision support method based on an integrated nursing knowledge base according to claim 1, characterized in that, The density clustering of all candidate reasoning paths based on the cosine similarity between the comprehensive semantic vectors includes: The DBSCAN algorithm is used for clustering, and the distance between paths is measured by subtracting the cosine similarity from 1. The neighborhood radius is 0.8, and the minimum number of neighborhood points (MinPts) required to form a cluster is 3.

6. The mobile nursing decision support method based on an integrated nursing knowledge base according to claim 1, characterized in that, The process of identifying the event corresponding to the central endpoint node of the pathway bundle as a high-priority nursing decision includes: Calculate the arithmetic mean of the embedding vectors of all end-point node entities within the path bundle in the embedding space to obtain a geometric center coordinate; Traverse all endpoint nodes within the path bundle, calculate the Euclidean distance between the endpoint node embedding vector and the geometric center coordinates, and select the endpoint node with the smallest distance as the central endpoint node.

7. The mobile nursing decision support method based on an integrated nursing knowledge base according to claim 1, characterized in that, The process involves parsing the high-priority nursing decisions and, in conjunction with the average comprehensive semantic vector of the path bundles that triggered these decisions, retrieving and generating customized nursing intervention instructions from the knowledge base, including: Starting with the event corresponding to the high-priority nursing decision, a set of candidate nursing intervention entities associated with the starting point are retrieved from the knowledge base; Calculate the arithmetic mean of the comprehensive semantic vectors of all paths within the path bundle to obtain the average comprehensive semantic vector; calculate the cosine similarity between the average comprehensive semantic vector and the embedding vector of each candidate nursing intervention entity. The candidate nursing interventions were ranked according to cosine similarity, and the top 3 nursing interventions were selected. The corresponding standard operating procedure texts were then combined into instructions for issuance.

8. A mobile nursing decision support device integrating a nursing knowledge base, characterized in that, Includes the following modules: The knowledge base construction module is used to build an integrated nursing knowledge base, in which entities and relations in the knowledge base have entity embedding vectors and relation embedding vectors, respectively. Ontology relations are represented by quadruples (subject, predicate, object, Ψ), where Ψ is an attribute set. The attribute set includes a confidence parameter for representing the reliability of the relation and a time decay function for calculating the timeliness. The path confidence calculation module is used to acquire real-time patient vital signs data and historical nursing records, and map them into a patient-specific instance graph; Starting from one or more nodes in the instance graph, a multi-path concurrent search is performed in the knowledge base to generate a set of candidate inference paths pointing to potential care events. For each candidate inference path, the relation embedding vectors of all relations on the path are combined level by level along the path, and combined with the entity embedding vector of the starting node to calculate the comprehensive semantic vector of the path. The comprehensive confidence of the path is calculated based on the confidence parameters of each relation on the path and the time decay function. The decision module is used to perform density clustering on all candidate inference paths based on the cosine similarity between the comprehensive semantic vectors to obtain at least one inference path bundle; and to calculate the aggregate confidence and bundle endpoint dispersion of each inference path bundle. When the aggregation confidence of a certain path bundle is higher than the preset confidence threshold and the dispersion of the bundle endpoints is lower than the preset dispersion threshold, the event corresponding to the central endpoint node of the path bundle is determined as a high-priority nursing decision. The nursing instruction generation module is used to parse the high-priority nursing decision, and combine it with the average comprehensive semantic vector of the path bundle that triggers the high-priority nursing decision to retrieve and generate customized nursing intervention instructions from the knowledge base, and then send the instructions to the mobile nursing terminal.

9. The mobile nursing decision support device with an integrated nursing knowledge base according to claim 8, characterized in that, The aggregate confidence is the sum of the comprehensive confidence of all paths within the bundle; the bundle endpoint dispersion is the average pairwise Euclidean distance of the endpoint nodes of all paths within the bundle in the embedding space.

10. The mobile nursing decision support device with an integrated nursing knowledge base according to claim 8, characterized in that, For each candidate reasoning path, the relation embedding vectors of all relations along the path are combined level by level, and combined with the entity embedding vector of the starting node to calculate the comprehensive semantic vector of the path, including: path The comprehensive semantic vector S(P) is calculated using vector addition: ,in Embed the vector for the starting node entity. Let be the embedding vector of the i-th relation on the path.

Images