Hospital department intelligent recommendation system based on knowledge graph
By constructing a knowledge graph-based intelligent recommendation system for hospital departments, the problem of existing technologies failing to accurately reflect symptom correlations and individual differences has been solved, enabling precise department recommendations and improving the efficiency and quality of medical services.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHAANXI HUIBIN ELECTRONIC TECH CO LTD
- Filing Date
- 2026-04-07
- Publication Date
- 2026-06-26
AI Technical Summary
Existing healthcare systems struggle to fully utilize hospital diagnostic data, failing to accurately reflect the correlations between symptoms and individual differences, resulting in insufficient accuracy and interpretability of departmental recommendations.
A hospital department intelligent recommendation system based on knowledge graphs is constructed. Through patient information acquisition module, department knowledge graph construction module, symptom-department matching module, and patient-department comprehensive evaluation module, the system realizes structured modeling of patients' natural language symptom descriptions and individual characteristics, and performs matching calculations with department feature vector sets to output a list of recommended departments and reasons.
This improves the relevance and accuracy of departmental recommendations, ensures that the recommendation process conforms to clinical knowledge logic, is traceable and can be personalized, thereby improving patient access efficiency and the quality of the hospital's intelligent triage service.
Smart Images
Figure CN122290926A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical information processing technology, and in particular to a hospital department intelligent recommendation system based on knowledge graphs. Background Technology
[0002] With the development of hospital informatization and intelligent medical services, department recommendations for patients have become an important means to alleviate the difficulty of registration and improve the efficiency of diagnosis and treatment. Existing medical service systems usually rely on manual triage, fixed rule matching, or search methods based on single symptom keywords to guide patients to choose a department. In practical applications, these methods are mostly based on static department classification, simple symptom mapping, or historical experience rules, which makes it difficult to make full use of the long-term accumulated diagnosis and treatment data of hospitals, and also makes it difficult to reflect the correlation structure between symptoms and the differences in the ability of different departments in cross-diagnosis. At the same time, patients often describe their discomfort in natural language, and the expression of symptoms is vague and combinatorial. Traditional methods are unable to accurately convert them into structured information that can be used for calculation, thus affecting the accuracy and interpretability of department recommendation results.
[0003] In response to the above situation, existing technologies still have significant shortcomings in the following aspects: First, they lack structured modeling methods capable of depicting the relationships between symptoms, resulting in symptom information participating in matching only as isolated labels, failing to reflect the true diagnostic and treatment semantics corresponding to symptom combinations; second, they fail to uniformly evaluate patient individual characteristics and symptom matching results, making it difficult for recommendation results to reflect individual differences; third, departmental capability modeling is mostly static or manually set, making it difficult to dynamically update with changes in diagnostic and treatment data, thus affecting the timeliness and reliability of recommendations. Therefore, it is necessary to provide a knowledge graph-based intelligent recommendation system for hospital departments to meet the urgent needs for accurate triage and intelligent recommendations in actual medical scenarios. Summary of the Invention
[0004] To achieve the above objectives, this invention provides a hospital department intelligent recommendation system based on knowledge graphs.
[0005] The hospital department intelligent recommendation system based on knowledge graphs includes a patient information acquisition module, a department knowledge graph construction module, a symptom-department matching module, a patient-department comprehensive assessment module, and an intelligent recommendation output module; among which: Patient information acquisition module: used to receive patients' natural language symptom descriptions and individual characteristic data, and output a patient information package containing a structured symptom set and patient feature vectors; Departmental knowledge graph construction module: Based on the hospital's historical diagnosis and treatment data, a dynamically updated departmental knowledge graph is constructed. The knowledge graph uses departments as nodes and uses the department's specialty diseases, symptom correlation strength, and departmental cross-diagnosis and treatment capabilities as edge attributes, outputting a set of departmental feature vectors with real-time weights. Symptom-department matching module: It receives a set of structured symptoms and connects it to a pre-built symptom association network; it identifies the subgraph topological features corresponding to the structured symptom set in the symptom association network; then, it combines the structured symptom set and its subgraph topological features with the set of department feature vectors to perform matching calculations and output a symptom-department matching degree sequence. Patient-Department Comprehensive Assessment Module: Combining patient feature vectors and symptom-department matching degree sequences, a weighted algorithm is used to generate a comprehensive assessment score for each department and output a department priority sequence; Intelligent recommendation output module: Generates and outputs a list of recommended departments and corresponding reasons for the recommendations based on the department priority sequence.
[0006] Optionally, the patient information acquisition module includes a symptom analysis unit, a feature extraction unit, and an information package generation unit; wherein: Symptom parsing unit: It is used to receive natural language symptom descriptions input by patients, identify entity items related to clinical symptoms through a named entity recognition model built based on a medical semantic dictionary, and map the entity items to standardized structured symptom labels to form a structured symptom set; Feature extraction unit: Used to receive basic information data related to individual patients, including age, gender, history of chronic diseases and family history of genetic diseases, and map them into a unified numerical form to construct a patient feature vector P with consistent dimensions; Information packet generation unit: used to combine and encode the structured symptom set with the patient feature vector P to generate a standard format patient information packet.
[0007] Optionally, the departmental knowledge graph construction module includes a data parsing unit, a graph generation unit, and a weight update unit; wherein: Data parsing unit: Used to receive historical medical data from the hospital, extract chief symptoms, diagnosis results and department information from structured fields, and form a symptom-disease-department triplet set; Graph generation unit: Constructs a heterogeneous graph structure based on triplet sets, with departments as nodes, and connecting edges containing the association information between departments and symptoms and diseases. At the same time, it introduces cross-diagnosis and treatment records between departments to form edge connections. Weight update unit: Periodically count the frequency and coverage of various edges in the graph, calculate their correlation strength within a set time window as edge weights, and update the feature vectors of each department node accordingly to form a set of department feature vectors with real-time weights.
[0008] Optionally, the weight update unit includes: Sideband statistics subunit: used to perform operations within a set time window Inside, the statistical graph shows each type of edge The frequency of occurrence is denoted as ; Coverage analysis sub-unit: used to calculate the proportion of nodes involved in each type of edge that are covered by different connectivity relationships in the graph, denoted as nodes. To node type coverage ; Correlation strength calculation subunit: used to calculate based on side frequencies and coverage Calculate the edges correlation strength As the edge weight; Feature vector update subunit: used to update the features based on the weights of all associated edges. Extract nodes The feature vector of the corresponding department is obtained by summing the feature codes of all outgoing edges and their corresponding target nodes according to the edge weights. .
[0009] Optionally, the symptom-department matching module includes a network access unit, a symptom localization unit, a subgraph extraction unit, a topology feature generation unit, and a matching calculation unit; wherein: Network access unit: used to access a pre-built symptom association network, wherein the symptom association network uses symptoms as nodes and co-occurrence association relationships between symptoms as edges, and configures association strength parameters for each edge; Symptom localization unit: It is used to receive the structured symptom set output by the patient information acquisition module, and perform standardized localization of symptom nodes in the symptom association network to form a set of target symptom nodes that correspond one-to-one with the structured symptom set; Subgraph extraction unit: Used to extract the corresponding locally connected subgraphs in the symptom association network according to the set expansion rules, using the target symptom node set as seed nodes, thereby obtaining subgraph structure data corresponding to the structured symptom set; Topological feature generation unit: Generates a subgraph topological feature vector based on the subgraph structure data. The subgraph topological feature vector includes the subgraph node size, edge density, number of connected components and the centrality ranking result of the core symptom nodes. The subgraph topological feature vector is aligned and encoded with the structured symptom set to form symptom structure representation data. Matching Calculation Unit: This unit receives the set of department feature vectors output by the department knowledge graph construction module, performs matching calculations on the symptom structure representation data and each department feature vector one by one, obtains the symptom-department matching degree for each department, and outputs the symptom-department matching degree sequence in the order of department index.
[0010] Optionally, the subgraph extraction unit includes: Adjacency filtering subunit: Used as seed nodes, the target symptom node set output by the symptom localization unit is used to identify adjacent nodes that have a direct connection relationship with each seed node in the symptom association network, and to construct an initial set of adjacent nodes as the first layer of nodes for subgraph expansion; Hop count control subunit: Used to set a fixed hop count threshold h, expand adjacent nodes from the seed node outward layer by layer based on breadth-first traversal, and mark the path depth between the current layer node and the previous layer node in each hop until the hop count threshold h is reached and the expansion stops, thereby limiting the subgraph range to a locally connected subgraph within the h-hop neighborhood; Edge weight filtering subunit: Used to perform weight filtering operation on all edges during the expansion process, retaining only edges with edge weights greater than a set threshold, thereby eliminating redundant nodes and edges introduced by weakly associated paths, and finally outputting locally connected subgraph structure data that satisfies the dual constraints of hop count and edge weight.
[0011] Optionally, the matching calculation unit includes: Feature alignment subunit: Used to receive symptom structure representation data and perform dimensional normalization processing on it to make it consistent with the department feature vector set in terms of vector dimension, value range and feature order, thereby forming a standardized symptom structure vector; Similarity calculation subunit: It is used to calculate the similarity between the standardized symptom structure vector and each department feature vector in the department feature vector set one by one, and obtain the similarity value between the symptom structure vector and each department feature vector. Matching Degree Generation Subunit: This subunit uses similarity values as the symptom-department matching degree for the corresponding department. By normalizing the similarity values, it generates the symptom-department matching degree for each department. ; Sequence output subunit: Used to calculate the symptom-department matching degree for all departments according to the predefined department index order in the system. Arrange them sequentially to generate a symptom-department matching sequence with consistent structure.
[0012] Optionally, the patient-department integrated assessment module includes a weight allocation unit, a weighted calculation unit, and a priority generation unit; wherein: Weighting unit: used to assign calculation weights to the patient feature vector and the symptom-department matching degree sequence respectively. The weight values are set according to the importance of the patient's individual characteristics, symptom coverage and historical assessment accuracy. Weighted Calculation Unit: Used to perform weighted assessment operations for each department. It receives the symptom matching degree and the weighted portion of the patient feature vector for each department, and calculates the comprehensive assessment score of the department through a weighted summation method. ; Priority generation unit: Based on the comprehensive evaluation scores of each department output by the weighted calculation unit, sort them in descending order to generate a structured department priority sequence.
[0013] Optionally, the weight allocation unit includes: Feature Importance Determination Subunit: Used to assign preset importance coefficients to each dimension of the patient feature vector, forming the patient feature importance vector. ; Symptom coverage calculation subunit: Based on the structured symptom set and the locally connected subgraph, calculate the coverage of the current symptom set in the graph. It is defined as the ratio of the number of target symptom nodes to the total number of nodes in the subgraph; Historical accuracy statistics subunit: used to statistically analyze the consistency between system recommendations and actual patient visits within a set historical evaluation period, thus obtaining historical accuracy. ; Weighted fusion subunit: based on symptom coverage With historical accuracy Calculate the symptom matching weight Weights based on patient characteristics Its expression is: ; .
[0014] Optionally, the intelligent recommendation output module includes a recommendation filtering unit, a reason generation unit, and a result output unit; wherein: Recommendation Screening Unit: This unit receives the department priority sequence output by the patient-department comprehensive assessment module. Based on the set maximum number of recommendations, it selects the top N departments from the sequence to form a recommended department list. Reason generation unit: For each selected recommended department, it extracts relevant content from the two components of its comprehensive evaluation score, namely the source of symptom matching degree and the patient characteristic matching factor, and describes them in a structured way to form the text content of the recommendation reason for the department. The result output unit is used to structurally bind the selected list of recommended departments with the reasons for recommendation, and combine them in the order of the original priority sequence to form the final recommendation output data structure, which includes three parts: the name of the recommended department, the evaluation score, and the reasons for recommendation.
[0015] The beneficial effects of this invention are: This invention constructs a heterogeneous knowledge graph oriented towards symptoms, diseases, and departments, integrating natural language symptom parsing, structured subgraph extraction, dynamic graph weight updating, and multi-source vector matching mechanisms. This enables structured modeling of patients' natural language symptom descriptions and individual characteristics, and systematic matching with a set of departmental feature vectors with semantic expression capabilities, thereby improving the relevance and accuracy of recommendation results.
[0016] This invention employs a dynamic evaluation weighting mechanism, incorporating symptom coverage and historical recommendation accuracy into the comprehensive scoring logic, and outputs explanatory recommendation reasons. This ensures that the recommendation process conforms to clinical knowledge logic, is traceable, and has personalized adaptability, effectively improving patient consultation efficiency and the quality of hospital intelligent triage services. Attached Figure Description
[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This is a schematic diagram of a hospital department intelligent recommendation system according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the symptom-department matching module in an embodiment of the present invention. Detailed Implementation
[0019] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.
[0020] It should be noted that the use of terms such as "an embodiment," "an embodiment," "an exemplary embodiment," and "some embodiments" in the specification indicates that the described embodiment may include a specific feature, structure, or characteristic, but not every embodiment necessarily includes that specific feature, structure, or characteristic. Furthermore, when a specific feature, structure, or characteristic is described in connection with an embodiment, implementing such a feature, structure, or characteristic in conjunction with other embodiments (whether explicitly described or not) should be within the knowledge of those skilled in the art.
[0021] Generally, terms can be understood at least partly from their use in context. For example, depending at least partly on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in a singular sense, or a combination of features, structures, or characteristics in a plural sense. Additionally, the term "based on" can be understood not necessarily to convey an exclusive set of factors, but rather, alternatively, depending at least partly on the context, to allow for the presence of other factors that are not necessarily explicitly described.
[0022] like Figures 1-2 As shown, the hospital department intelligent recommendation system based on knowledge graph includes a patient information acquisition module, a department knowledge graph construction module, a symptom-department matching module, a patient-department comprehensive assessment module, and an intelligent recommendation output module; among which: Patient information acquisition module: used to receive patients' natural language symptom descriptions and individual characteristic data, and output a patient information package containing a structured symptom set and patient feature vectors; Departmental Knowledge Graph Construction Module: Based on the hospital's historical diagnosis and treatment data, a dynamically updated departmental knowledge graph is constructed. The knowledge graph uses departments as nodes and departmental specialty diseases, symptom correlation strength, and departmental cross-diagnosis and treatment capabilities as edge attributes, outputting a set of departmental feature vectors with real-time weights. Symptom-department matching module: It receives a set of structured symptoms and connects it to a pre-built symptom association network; it identifies the subgraph topological features corresponding to the structured symptom set in the symptom association network; then, it combines the structured symptom set and its subgraph topological features with the set of department feature vectors to perform matching calculations and output a symptom-department matching degree sequence. Patient-Department Comprehensive Assessment Module: Combining patient feature vectors and symptom-department matching degree sequences, a weighted algorithm is used to generate a comprehensive assessment score for each department and output a department priority sequence; Intelligent recommendation output module: Generates and outputs a list of recommended departments and corresponding reasons for the recommendations based on the department priority sequence.
[0023] The patient information acquisition module includes a symptom analysis unit, a feature extraction unit, and an information package generation unit; among which: Symptom parsing unit: It is used to receive natural language symptom descriptions input by patients, identify entity items related to clinical symptoms through a named entity recognition model built based on a medical semantic dictionary, and map the entity items to standardized structured symptom labels to form a structured symptom set; Feature extraction unit: It is used to receive basic information data related to the individual patient, including age, gender, history of chronic diseases and family history of genetic diseases, and map them into a unified numerical form to construct a patient feature vector P with consistent dimensions, whose dimensions are compatible with the weight structure used by the subsequent evaluation module. The information package generation unit is used to combine and encode the structured symptom set with the patient feature vector P to generate a standard format patient information package. This unit realizes the automatic conversion of patient information from unstructured input to standardized data packages by parsing the natural language symptom input into structured labels and combining them with the embedded vector expression of individual features. This ensures the integrity and consistency of the information required for subsequent matching and evaluation.
[0024] The departmental knowledge graph construction module includes a data parsing unit, a graph generation unit, and a weight update unit; among which: Data parsing unit: Used to receive historical medical data from the hospital, extract chief symptoms, diagnosis results and department information from structured fields, and form a symptom-disease-department triplet set; Graph generation unit: Constructs a heterogeneous graph structure based on triplet sets, with departments as nodes, and connecting edges containing the association information between departments and symptoms and diseases. At the same time, it introduces cross-diagnosis and treatment records between departments to form edge connections. Weight Update Unit: Periodically counts the frequency and coverage of various edges in the graph, calculates their correlation strength within a set time window as edge weights, and updates the feature vectors of each department node accordingly, forming a set of department feature vectors with real-time weights for subsequent modules to use. The above unit, by constructing a heterogeneous graph from historical medical data and introducing a dynamic edge weight update mechanism, can ensure the timeliness of department feature vectors and the accuracy of reflecting correlation strength, providing a reliable foundation for subsequent matching and evaluation.
[0025] The weight update unit includes: Sideband statistics subunit: used to perform operations within a set time window Inside, the statistical graph shows each type of edge The frequency of occurrence is denoted as This indicates the department node within that time period. With the target node The number of times the diagnosis and treatment were linked; Coverage analysis sub-unit: used to calculate the proportion of nodes involved in each type of edge that are covered by different connectivity relationships in the graph, denoted as nodes. To node type coverage Defined as in window The ratio of the number of types of inner edge 𝑖→𝑗 to the total number of types of edges of that type connected to node 𝑖; Correlation strength calculation subunit: used to calculate based on side frequencies and coverage Calculate the edges correlation strength As the edge weight, the calculation formula is as follows: ,in, For the edge The strength of the association; For nodes The total frequency of pointers to all associated nodes; As a weighting balancing factor; Feature vector update subunit: used to update the features based on the weights of all associated edges. Extract nodes The feature vector of a department is obtained by summing the feature codes of all outgoing edges (i.e., those of a specific department) and their corresponding target nodes, weighted by edge weight. The calculation method is as follows: ,in, For the department The updated feature vector; For nodes The set of all adjacent nodes; For nodes The feature vectors; by jointly modeling the edge frequency and coverage as edge weights, and using this to drive the dynamic update mechanism of the feature vectors, the department nodes can form a more realistic representation of the diagnosis and treatment focus and correlation trends within the current time window, thereby enhancing the timeliness and semantic accuracy of the correlation between departments in the graph, which is conducive to improving the judgment accuracy of the subsequent recommendation system.
[0026] The symptom-department matching module includes a network access unit, a symptom localization unit, a subgraph extraction unit, a topology feature generation unit, and a matching calculation unit; among which: Network access unit: used to access the pre-built symptom association network. The symptom association network uses symptoms as nodes and co-occurrence relationships between symptoms as edges, and configures association strength parameters for each edge to support subsequent subgraph extraction and topological feature calculation. Symptom localization unit: It is used to receive the structured symptom set output by the patient information acquisition module, and perform standardized localization of symptom nodes in the symptom association network to form a set of target symptom nodes that correspond one-to-one with the structured symptom set; Subgraph extraction unit: Used to extract the corresponding locally connected subgraphs in the symptom association network according to the set expansion rules, using the target symptom node set as seed nodes, thereby obtaining subgraph structure data corresponding to the structured symptom set; Topological feature generation unit: Generates subgraph topological feature vectors based on subgraph structure data. The subgraph topological feature vectors include the subgraph node size, edge density, number of connected components, and centrality ranking results of core symptom nodes. The subgraph topological feature vectors are then aligned and encoded with the structured symptom set to form symptom structure representation data. The matching calculation unit receives the set of department feature vectors output by the department knowledge graph construction module, performs matching calculations on the symptom structure representation data with each department feature vector one by one, obtains the symptom-department matching degree for each department, and outputs the symptom-department matching degree sequence in the order of department index for use by the patient-department comprehensive assessment module. The above unit completes the node localization, local subgraph extraction and topological feature construction of the structured symptom set in the symptom association network, so that the symptom information is expanded from single-point labels to symptom structure representation data containing association structure, and performs consistent matching calculations on this structure representation with the set of department feature vectors, thereby ensuring that the output symptom-department matching degree sequence can simultaneously reflect the symptom ontology information and the relationship between symptoms, providing a stable and aligned input basis for subsequent comprehensive assessment.
[0027] The subgraph extraction unit includes: Adjacency filtering subunit: Used as seed nodes, the target symptom node set output by the symptom localization unit is used to identify adjacent nodes that have a direct connection relationship with each seed node in the symptom association network, and to construct an initial set of adjacent nodes as the first layer of nodes for subgraph expansion; Hop count control subunit: Used to set a fixed hop count threshold h, expand adjacent nodes from the seed node outward layer by layer based on breadth-first traversal, and mark the path depth between the current layer node and the previous layer node in each hop until the hop count threshold h is reached and the expansion stops, thereby limiting the subgraph range to a locally connected subgraph within the h-hop neighborhood; Edge weight filtering subunit: This subunit performs weight filtering on all edges during the expansion process, retaining only edges with weights greater than a set threshold. This eliminates redundant nodes and edges introduced by weakly associated paths, ultimately outputting locally connected subgraph structure data that satisfies both hop count and edge weight constraints. By employing a dual-constraint expansion strategy based on hop count control and edge weight filtering, this subunit can efficiently extract high-quality local subgraphs closely related to the target symptoms in the symptom association network, avoiding ineffective expansion and noise interference, and providing a compact and accurate structural foundation for subsequent topology extraction and matching.
[0028] The matching calculation unit includes: Feature alignment subunit: It is used to receive symptom structure representation data and perform dimensional normalization processing on it to make it consistent with the department feature vector set in terms of vector dimension, value range and feature order, thereby forming a standardized symptom structure vector, which serves as the unified input basis for subsequent matching calculations; The similarity calculation subunit is used to calculate the similarity between the standardized symptom structure vector and each department feature vector in the department feature vector set, obtaining a similarity value between the symptom structure vector and each department feature vector. The similarity value is used to characterize the degree of matching between the symptom structure and the corresponding department feature. The calculation method is as follows: ;in, For the symptom structure vector and the first Similarity values between feature vectors of each department; This is the standardized symptom structure vector; For the department The updated feature vector; Let L be the L2 norm of the vector; Matching Degree Generation Subunit: This subunit uses similarity values as the symptom-department matching degree for the corresponding department. By normalizing the similarity values, it generates the symptom-department matching degree for each department. Its expression is as follows: ,in, For the first Symptoms of each department - department matching degree; For the first Similarity values of individual departments; This represents the minimum similarity value within the current calculation period. This represents the maximum similarity value within the current calculation period. Sequence output subunit: Used to calculate the symptom-department matching degree for all departments according to the predefined department index order in the system. The data are arranged sequentially to generate a consistent symptom-department matching degree sequence. The above sub-units perform one-to-one matching calculations on the symptom structure representation data and the department feature vectors with consistent dimensions, and organize the calculation results into a matching degree sequence according to a unified index order. This ensures the consistency of the matching results of each department in terms of numerical scale and data structure, thereby providing a stable, comparable and directly callable input data foundation for the subsequent comprehensive evaluation process.
[0029] The patient-department integrated assessment module includes a weight allocation unit, a weighted calculation unit, and a priority generation unit; among which: Weighting unit: Used to assign calculation weights to the patient feature vector and the symptom-department matching degree sequence respectively. The weight values are set according to the importance of the individual patient characteristics, symptom coverage and historical assessment accuracy. Weighted Calculation Unit: Used to perform weighted assessment operations for each department. It receives the symptom matching degree and the weighted portion of the patient feature vector for each department, and calculates the comprehensive assessment score of the department through a weighted summation method. This indicates the priority of recommendations made by the corresponding department based on the current patient characteristics and symptoms. Priority generation unit: Based on the comprehensive evaluation scores of each department output by the weighted calculation unit, the departments are sorted in descending order to generate a structured department priority sequence, which is then output to the intelligent recommendation output module. By introducing patient characteristics and symptom matching information into a unified weighted scoring framework and performing independent comprehensive scoring and priority ranking for each department, this unit can generate a targeted department recommendation order based on individual differences and symptom characteristics, providing a stable and differentiated decision support basis for the intelligent recommendation output module.
[0030] The weight allocation unit includes: Feature Importance Determination Subunit: Used to assign preset importance coefficients to each dimension of the patient feature vector, forming the patient feature importance vector. ,in, This represents the total number of feature dimensions. Symptom coverage calculation subunit: Based on the structured symptom set and the locally connected subgraph, calculate the coverage of the current symptom set in the graph. It is defined as the ratio of the number of target symptom nodes to the total number of nodes in the subgraph, and the expression is: ,in, Symptom coverage; This represents the number of nodes in the subgraph that belong to the target symptom set. This represents the total number of nodes in the local subgraph. Historical accuracy statistics subunit: used to statistically analyze the consistency between system recommendations and actual patient visits within a set historical evaluation period, thus obtaining historical accuracy. Its expression is: ,in, To ensure the accuracy of historical assessments; To recommend the correct number of samples; This represents the total number of samples included in the statistics. Weighted fusion subunit: based on symptom coverage With historical accuracy Calculate the symptom matching weight Weights based on patient characteristics Its expression is: ; The aforementioned sub-units construct a dynamic and adaptive weight fusion mechanism by incorporating the importance of patient characteristics, the coverage of symptom maps, and the accuracy of historical recommendations into a unified weight setting process. This ensures that the weight allocation of symptom and patient information in the comprehensive assessment is targeted, timely, and supported by data.
[0031] The formula for calculating the comprehensive assessment score is as follows: ,in, For the first The overall evaluation score of each department; For the first Symptoms corresponding to each department - department matching degree; The dimension of the patient feature vector; For the first In the department, the first Weighted values for matching based on patient characteristics; For the first Internal normalized weights for each patient characteristic dimension.
[0032] The intelligent recommendation output module includes a recommendation filtering unit, a reason generation unit, and a result output unit; wherein: Recommendation Screening Unit: This unit receives the department priority sequence output by the patient-department comprehensive assessment module. Based on the set upper limit for the number of recommendations, it selects the top N departments from the sequence to form a list of recommended departments, ensuring that the recommendation results have the highest assessment score within the calculation limit. Reason generation unit: For each selected recommended department, it extracts relevant content from the two components of its comprehensive evaluation score, namely the source of symptom matching degree and the patient characteristic matching factor, and describes them in a structured way to form the text content of the recommendation reason for the department. The recommendation reason includes, but is not limited to, a combination of factors such as high correlation with the chief complaint symptoms, specific disease expertise, conformity with the patient's age and gender characteristics, and high historical recommendation accuracy. The results output unit is used to structurally bind the selected list of recommended departments with the reasons for recommendation, and combine them in the original priority sequence to form the final recommendation output data structure, which includes three parts: the name of the recommended department, the evaluation score, and the reason for recommendation. By converting the sorted priority sequence into a list of recommended departments with controlled quantity, and generating an explainable reason for recommendation for each recommendation, the unit ensures that the final recommendation results are both based on evaluation and readable by users, which helps to improve users' trust in the accuracy and rationality of the recommendations.
[0033] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.
[0034] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A hospital department intelligent recommendation system based on knowledge graphs, characterized in that, It includes a patient information acquisition module, a departmental knowledge graph construction module, a symptom-department matching module, a patient-department comprehensive assessment module, and an intelligent recommendation output module; among which: Patient information acquisition module: used to receive patients' natural language symptom descriptions and individual characteristic data, and output a patient information package containing a structured symptom set and patient feature vectors; Departmental knowledge graph construction module: Based on the hospital's historical diagnosis and treatment data, a dynamically updated departmental knowledge graph is constructed. The knowledge graph uses departments as nodes and uses the department's specialty diseases, symptom correlation strength, and departmental cross-diagnosis and treatment capabilities as edge attributes, outputting a set of departmental feature vectors with real-time weights. Symptom-department matching module: It receives a set of structured symptoms and connects it to a pre-built symptom association network; it identifies the subgraph topological features corresponding to the structured symptom set in the symptom association network; then, it combines the structured symptom set and its subgraph topological features with the set of department feature vectors to perform matching calculations and output a symptom-department matching degree sequence. Patient-Department Comprehensive Assessment Module: Combining patient feature vectors and symptom-department matching degree sequences, a weighted algorithm is used to generate a comprehensive assessment score for each department and output a department priority sequence; Intelligent recommendation output module: Generates and outputs a list of recommended departments and corresponding reasons for the recommendations based on the department priority sequence.
2. The hospital department intelligent recommendation system based on knowledge graph as described in claim 1, characterized in that, The patient information acquisition module includes a symptom analysis unit, a feature extraction unit, and an information package generation unit; wherein: Symptom parsing unit: It is used to receive natural language symptom descriptions input by patients, identify entity items related to clinical symptoms through a named entity recognition model built based on a medical semantic dictionary, and map the entity items to standardized structured symptom labels to form a structured symptom set; Feature extraction unit: Used to receive basic information data related to individual patients, including age, gender, history of chronic diseases and family history of genetic diseases, and map them into a unified numerical form to construct a patient feature vector P with consistent dimensions; Information packet generation unit: used to combine and encode the structured symptom set with the patient feature vector P to generate a standard format patient information packet.
3. The hospital department intelligent recommendation system based on knowledge graph as described in claim 1, characterized in that, The departmental knowledge graph construction module includes a data parsing unit, a graph generation unit, and a weight update unit; wherein: Data parsing unit: Used to receive historical medical data from the hospital, extract chief symptoms, diagnosis results and department information from structured fields, and form a symptom-disease-department triplet set; Graph generation unit: Constructs a heterogeneous graph structure based on triplet sets, with departments as nodes, and connecting edges containing the association information between departments and symptoms and diseases. At the same time, it introduces cross-diagnosis and treatment records between departments to form edge connections. Weight update unit: Periodically count the frequency and coverage of various edges in the graph, calculate their correlation strength within a set time window as edge weights, and update the feature vectors of each department node accordingly to form a set of department feature vectors with real-time weights.
4. The hospital department intelligent recommendation system based on knowledge graph as described in claim 3, characterized in that, The weight update unit includes: Sideband statistics subunit: used to perform operations within a set time window Inside, the statistical graph shows each type of edge The frequency of occurrence is denoted as ; Coverage analysis sub-unit: used to calculate the proportion of nodes involved in each type of edge that are covered by different connectivity relationships in the graph, denoted as nodes. To node type coverage ; Correlation strength calculation subunit: used to calculate based on side frequencies and coverage Calculate the edges correlation strength As the edge weight; Feature vector update subunit: used to update the features based on the weights of all associated edges. Extract nodes The feature vector of the corresponding department is obtained by summing the feature codes of all outgoing edges and their corresponding target nodes according to the edge weights. .
5. The hospital department intelligent recommendation system based on knowledge graph as described in claim 1, characterized in that, The symptom-department matching module includes a network access unit, a symptom localization unit, a subgraph extraction unit, a topology feature generation unit, and a matching calculation unit; wherein: Network access unit: used to access a pre-built symptom association network, wherein the symptom association network uses symptoms as nodes and co-occurrence association relationships between symptoms as edges, and configures association strength parameters for each edge; Symptom localization unit: It is used to receive the structured symptom set output by the patient information acquisition module, and perform standardized localization of symptom nodes in the symptom association network to form a set of target symptom nodes that correspond one-to-one with the structured symptom set; Subgraph extraction unit: Used to extract the corresponding locally connected subgraphs in the symptom association network according to the set expansion rules, using the target symptom node set as seed nodes, thereby obtaining subgraph structure data corresponding to the structured symptom set; Topological feature generation unit: Generates a subgraph topological feature vector based on the subgraph structure data. The subgraph topological feature vector includes the subgraph node size, edge density, number of connected components and the centrality ranking result of the core symptom nodes. The subgraph topological feature vector is aligned and encoded with the structured symptom set to form symptom structure representation data. Matching Calculation Unit: This unit receives the set of department feature vectors output by the department knowledge graph construction module, performs matching calculations on the symptom structure representation data and each department feature vector one by one, obtains the symptom-department matching degree for each department, and outputs the symptom-department matching degree sequence in the order of department index.
6. The hospital department intelligent recommendation system based on knowledge graph as described in claim 5, characterized in that, The subgraph extraction unit includes: Adjacency filtering subunit: Used as seed nodes, the target symptom node set output by the symptom localization unit is used to identify adjacent nodes that have a direct connection relationship with each seed node in the symptom association network, and to construct an initial set of adjacent nodes as the first layer of nodes for subgraph expansion; Hop count control subunit: Used to set a fixed hop count threshold h, expand adjacent nodes from the seed node outward layer by layer based on breadth-first traversal, and mark the path depth between the current layer node and the previous layer node in each hop until the hop count threshold h is reached and the expansion stops, thereby limiting the subgraph range to a locally connected subgraph within the h-hop neighborhood; Edge weight filtering subunit: Used to perform weight filtering operation on all edges during the expansion process, retaining only edges with edge weights greater than a set threshold, thereby eliminating redundant nodes and edges introduced by weakly associated paths, and finally outputting locally connected subgraph structure data that satisfies the dual constraints of hop count and edge weight.
7. The hospital department intelligent recommendation system based on knowledge graph as described in claim 5, characterized in that, The matching calculation unit includes: Feature alignment subunit: Used to receive symptom structure representation data and perform dimensional normalization processing on it to make it consistent with the department feature vector set in terms of vector dimension, value range and feature order, thereby forming a standardized symptom structure vector; Similarity calculation subunit: It is used to calculate the similarity between the standardized symptom structure vector and each department feature vector in the department feature vector set one by one, and obtain the similarity value between the symptom structure vector and each department feature vector. Matching Degree Generation Subunit: This subunit uses similarity values as the symptom-department matching degree for the corresponding department. By normalizing the similarity values, it generates the symptom-department matching degree for each department. ; Sequence output subunit: Used to calculate the symptom-department matching degree for all departments according to the predefined department index order in the system. Arrange them sequentially to generate a symptom-department matching sequence with consistent structure.
8. The intelligent hospital department recommendation system based on knowledge graph as described in claim 1, characterized in that, The patient-department integrated assessment module includes a weight allocation unit, a weighted calculation unit, and a priority generation unit; wherein: Weighting unit: used to assign calculation weights to the patient feature vector and the symptom-department matching degree sequence respectively. The weight values are set according to the importance of the patient's individual characteristics, symptom coverage and historical assessment accuracy. Weighted Calculation Unit: Used to perform weighted assessment operations for each department. It receives the symptom matching degree and the weighted portion of the patient feature vector for each department, and calculates the comprehensive assessment score of the department through a weighted summation method. ; Priority generation unit: Based on the comprehensive evaluation scores of each department output by the weighted calculation unit, sort them in descending order to generate a structured department priority sequence.
9. The hospital department intelligent recommendation system based on knowledge graph as described in claim 8, characterized in that, The weight allocation unit includes: Feature Importance Determination Subunit: Used to assign preset importance coefficients to each dimension of the patient feature vector, forming the patient feature importance vector. ; Symptom coverage calculation subunit: Based on the structured symptom set and the locally connected subgraph, calculate the coverage of the current symptom set in the graph. It is defined as the ratio of the number of target symptom nodes to the total number of nodes in the subgraph; Historical accuracy statistics subunit: used to statistically analyze the consistency between system recommendations and actual patient visits within a set historical evaluation period, thus obtaining historical accuracy. ; Weighted fusion subunit: based on symptom coverage With historical accuracy Calculate the symptom matching weight Weights based on patient characteristics Its expression is: ; .
10. The intelligent hospital department recommendation system based on knowledge graphs according to claim 1, characterized in that, The intelligent recommendation output module includes a recommendation filtering unit, a reason generation unit, and a result output unit; wherein: Recommendation Screening Unit: This unit receives the department priority sequence output by the patient-department comprehensive assessment module. Based on the set maximum number of recommendations, it selects the top N departments from the sequence to form a recommended department list. Reason generation unit: For each selected recommended department, it extracts relevant content from the two components of its comprehensive evaluation score, namely the source of symptom matching degree and the patient characteristic matching factor, and describes them in a structured way to form the text content of the recommendation reason for the department. The result output unit is used to structurally bind the selected list of recommended departments with the reasons for recommendation, and combine them in the order of the original priority sequence to form the final recommendation output data structure, which includes three parts: the name of the recommended department, the evaluation score, and the reasons for recommendation.