A medical institution name matching method, system, device and storage medium

Abstract

This invention discloses a method, system, and storage medium for matching medical institution names. The method includes receiving a source medical institution name string and parsing its semantic components to obtain the geographical coordinates of the source medical institution. A search radius is dynamically determined based on at least one entity attribute of the source medical institution. A geofence is constructed centered on the geographical coordinates and using the search radius to obtain a set of candidate medical institutions. A weighted similarity score is calculated between the source medical institution and any candidate medical institution in the set, and it is determined whether the score falls within a preset fuzzy interval. If so, the semantic relationship between the source medical institution and the candidate medical institution is judged, and a structured arbitration result is output. The weighted similarity score and the structured arbitration result are fused to calculate the final matching confidence score, and a matching decision is made based on the final matching confidence score. This invention can achieve high-precision, high-recall, semantically aware, and adaptive automated matching of medical institution names.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and specifically to a method, system, device, and storage medium for matching medical institution names. Background Technology

[0002] In the development of modern medical information technology, data interconnection is the core foundation for achieving intelligent medical insurance cost control, public health monitoring, clinical research, and improved medical service efficiency. Medical data sources are diverse, covering electronic medical record (EHR) systems, hospital information systems (HIS), regional health information platforms, medical insurance settlement databases, and commercial map point of interest (POI) databases, among others.

[0003] However, a key technical bottleneck in the fusion analysis of these heterogeneous data sources is the inconsistent naming of the same medical institution entity across different databases. For example, a hospital may have multiple forms of expression, including its official full name, commonly used abbreviation, local colloquial name, and names that include the hospital area or department. This naming diversity makes it impossible to effectively link data records pointing to the same entity, creating "data silos" that severely hinder the release of the value of medical big data.

[0004] Existing technologies employ fuzzy matching methods based on string similarity algorithms. The general implementation involves: first, preprocessing the two name strings to be matched (e.g., removing punctuation, converting to equal case), then calculating their similarity score using algorithms such as Levenshtein Distance, Jaro-Winkler distance, or N-gram, and finally comparing it with a preset threshold (e.g., 0.85) to determine if the two names match. However, existing technologies have the following inherent drawbacks:

[0005] Semantic confusion and low accuracy: Existing methods are usually based on character matching and cannot understand the semantic connotation of words. Therefore, they are prone to incorrectly judging names such as "Z City First People's Hospital" and "Z Province First People's Hospital" as highly similar strings but belonging to different entities, resulting in low matching accuracy.

[0006] Ignoring aliases results in low recall: When there is a significant difference between the official full name and the commonly used abbreviation, the string similarity score is extremely low, leading to a low recall rate.

[0007] Lack of context awareness: Traditional methods isolate crucial contextual information such as geographical location. Two organizations with similar names that are geographically far apart are almost certainly not the same entity in reality, but current technology cannot use this prior knowledge to assist in the judgment.

[0008] Lack of relationship identification capability: This method can usually only provide a binary conclusion of "match" or "not match", and cannot identify more granular hierarchical relationships such as "headquarters-branch" or "institution-subordinate department".

[0009] The model is static and has poor adaptability: its matching logic and parameters are usually fixed, lacking the ability to learn from new data and optimize itself, and it is difficult to adapt to new naming methods and complex scenarios. Summary of the Invention

[0010] To address the aforementioned problems, this invention provides a method, system, and storage medium for matching medical institution names, aiming to overcome the technical shortcomings of existing technologies, such as low precision and recall, inability to understand semantic relationships, lack of contextual information utilization, and static and poor adaptability of the model, which rely solely on surface string similarity.

[0011] According to a first aspect of the present disclosure, a method for matching medical institution names is provided, the method comprising the following steps:

[0012] The system receives the name string of the source medical institution, uses a pre-trained large language model to perform semantic component parsing on the name string, identifies the identification name that represents the core identity of the entity, and at least one general name that describes the attributes of the entity; and assigns different preset weights to the identification name and the general name.

[0013] Obtain the geographic coordinates of the source medical institution, and dynamically determine the search radius based on at least one entity attribute of the source medical institution. Construct a geofence centered on the geographic coordinates and with the search radius, and retrieve the candidate medical institution set within the geofence.

[0014] Calculate the weighted similarity score between the source medical institution and any candidate medical institution in the candidate medical institution set. When the weighted similarity score falls within a preset fuzzy range, determine the semantic relationship between the source medical institution and the candidate medical institution, and output a structured arbitration result that includes the relationship type and relationship confidence.

[0015] A confidence fusion model is constructed to fuse weighted similarity scores and structured arbitration results, calculate the final matching confidence, and make matching decisions based on the final matching confidence.

[0016] In some embodiments, the weight assigned to the identification name is higher than the weight assigned to the generic name.

[0017] In some embodiments, the method further includes: when the final matching confidence level does not reach a preset decision threshold, manually reviewing the source medical institution and the candidate medical institution, and retraining the confidence fusion model and / or the large language model based on the matching results obtained from the manual review.

[0018] In some embodiments, the confidence fusion model is a Bayesian inference model, which uses the weighted similarity score as the prior probability and the structured arbitration result as the observation evidence to calculate the posterior probability as the final matching confidence.

[0019] In some embodiments, the confidence fusion model is a supervised learning classifier that inputs the weighted similarity score and the structured arbitration result into a pre-trained supervised learning classifier and outputs the final matching probability.

[0020] In some embodiments, a pre-trained large language model or a medical domain knowledge graph is used to determine the semantic relationship between the source medical institution and the candidate medical institution.

[0021] According to a second aspect of the present disclosure, a medical institution name matching system is provided, the system comprising:

[0022] The name structuring and weighting module receives the name string of the source medical institution, uses a pre-trained large language model to perform semantic component parsing on the name string to identify the identification name that represents the core identity of the entity, as well as at least one general name that describes the attributes of the entity; and assigns different preset weights to the identification name and the general name.

[0023] The dynamic geofence candidate set generation module is used to obtain the geographic coordinates of the source medical institutions, and dynamically determine the search radius based on at least one entity attribute of the source medical institutions. It constructs a geofence with the geographic coordinates as the center and the search radius, and retrieves the candidate medical institution set within the geofence.

[0024] The multi-level matching and semantic relationship determination module is used to calculate the weighted similarity score between the source medical institution and any candidate medical institution in the candidate medical institution set. When the weighted similarity score falls into the preset fuzzy range, the semantic relationship between the source medical institution and the candidate medical institution is determined, and a structured arbitration result containing the relationship type and relationship confidence is output.

[0025] The confidence fusion and decision module is used to build a confidence fusion model, which integrates weighted similarity scores and structured arbitration results to calculate the final matching confidence and make matching decisions based on the final matching confidence.

[0026] In some embodiments, the system further includes a human-machine collaborative feedback and optimization module, which is used to manually review the source medical institution and the candidate medical institution when the final matching confidence does not reach a preset decision threshold, and to retrain the confidence fusion model and / or the large language model based on the matching results obtained from the manual review.

[0027] According to a third aspect of the present disclosure, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the above-described medical institution name matching method.

[0028] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium is provided, the storage medium storing computer instructions that, when executed by a processor, implement the steps of the above-described medical institution name matching method.

[0029] This disclosure provides a method, system, and storage medium for matching medical institution names, aiming to achieve high-precision, high-recall, semantically aware, and adaptive automated matching of medical institution names. Its beneficial effects include:

[0030] Significantly improved accuracy: By deeply integrating the semantic weight of names with geospatial information, mismatches and omissions caused by "similar names but geographically distant" or "significantly different names but geographically close and with consistent core words" are effectively avoided, greatly improving the accuracy and recall rate of matching.

[0031] Achieving semantic relationship understanding: When the weighted similarity score falls within the preset fuzzy range, the semantic relationship between the source medical institution and the candidate medical institution is determined, and a structured arbitration result containing the relationship type and relationship confidence is output. It can go beyond the binary matching of "yes / no" and identify various types of semantic relationships such as "same entity", "alias", "branch / campus", providing richer entity relationship dimension information for downstream data applications.

[0032] Robustness and adaptability: Dynamic geofencing can intelligently adapt to the distribution characteristics of institutions at different levels; the Bayesian inference framework provides a unified probabilistic interpretation framework for the fusion of multi-source evidence and offers a clear path for future integration of more dimensional evidence to achieve model optimization.

[0033] High degree of automation, cost reduction and efficiency improvement: By internalizing complex expert judgment logic into an end-to-end automated process, the workload of manual review and intervention is greatly reduced, and the time and human resources costs of medical data governance are significantly reduced.

[0034] It possesses continuous learning and self-optimization capabilities: Through a feedback loop of human-machine collaboration, for fuzzy cases where the model struggles to make a decision, the system can submit them for manual review and use the review results as high-quality training data to feed back into the model, enabling it to continuously evolve through ongoing use, dynamically adapt to new naming methods and complex scenarios, and achieve long-term self-optimization of the model.

[0035] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0036] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0037] Figure 1 This is a flowchart of the medical institution name matching method in an embodiment of the present invention;

[0038] Figure 2 This is a schematic diagram of the method flow for name structuring and semantic weighting in an embodiment of the present invention;

[0039] Figure 3 This is a schematic diagram of the method for generating a dynamic geofence candidate set in an embodiment of the present invention;

[0040] Figure 4 This is a schematic diagram of the method flow for multi-level matching, intelligent arbitration, and confidence fusion in an embodiment of the present invention;

[0041] Figure 5 This is a schematic diagram of the feedback and optimization method for human-machine collaboration in an embodiment of the present invention;

[0042] Figure 6 This is a schematic diagram of the medical institution name matching system structure in an embodiment of the present invention;

[0043] Figure 7 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0044] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the present invention are shown in the drawings, not the entire structure.

[0045] Before discussing the exemplary embodiments in more detail, it should be noted that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the steps as sequential processes, many of these steps can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the steps can be rearranged. The process can be terminated when its operation is complete, but may also have additional steps not included in the figures. The process can correspond to a method, function, procedure, subroutine, subroutine, etc.

[0046] This invention provides a method, system, electronic device, and computer-readable storage medium for matching medical institution names by integrating semantic weights, dynamic geofencing, and a probabilistic model. The aim is to achieve high-precision, high-recall, semantically aware, and adaptive automated matching of medical institution names. Specific embodiments are as follows:

[0047] A method for matching medical institution names, the method comprising the following steps:

[0048] S100, Name Structuring and Semantic Weighting Steps: Receive the name string of the source medical institution, use a pre-trained large language model to perform semantic component parsing on the name string to identify the identification name representing the core identity of the entity, and at least one general name describing the attributes of the entity; and assign different preset weights to the identification name and the general name. Preferably, the weight assigned to the identification name is higher than the weight assigned to the general name.

[0049] S200, Dynamic Geofence Candidate Set Generation Steps: Obtain the geographic coordinates of the source medical institution, and dynamically determine the search radius based on at least one entity attribute (preset level) of the source medical institution, construct a dynamic geofence, construct the geofence with the geographic coordinates as the center and the search radius, and retrieve the candidate medical institution set within the geofence;

[0050] S300, Multi-level Matching and Intelligent Arbitration Steps: Calculate the weighted similarity score between the source medical institution and any candidate medical institution in the candidate medical institution set. When the weighted similarity score falls into the preset fuzzy range, the intelligent arbitration module is triggered. Specifically, the semantic relationship between the source medical institution and the candidate medical institution is determined, and a structured arbitration result containing the relationship type and relationship confidence is output.

[0051] S400, Confidence Fusion and Decision-Making Steps: Construct a confidence fusion model to fuse weighted similarity scores and structured arbitration results, calculate the final matching confidence, and make matching decisions based on the final matching confidence.

[0052] In some preferred embodiments, such as Figure 1 As shown, the method also includes S500, a feedback and optimization step of human-machine collaboration: when the final matching confidence does not reach the preset decision threshold, the source medical institution and the candidate medical institution are manually reviewed, and the confidence fusion model and / or the large language model are retrained based on the matching results obtained from the manual review.

[0053] In some preferred embodiments, the confidence fusion model is a Bayesian inference model, which uses the weighted similarity score as the prior probability and the structured arbitration result as the observation evidence to calculate the posterior probability as the final matching confidence.

[0054] In some preferred embodiments, the confidence fusion model is a supervised learning classifier that inputs the weighted similarity score and the structured arbitration result into a pre-trained supervised learning classifier and outputs the final matching probability.

[0055] In some preferred embodiments, a pre-trained large language model or a medical domain knowledge graph is used to determine the semantic relationship between the source medical institution and the candidate medical institution.

[0056] In one specific embodiment, the following steps may be included:

[0057] Step S100: Name structuring and semantic weighting: such as Figure 2 As shown, this step aims to transform unstructured name strings into structured data objects containing semantic and importance information, laying the foundation for subsequent weighted matching. In an exemplary embodiment:

[0058] 301. The system receives a string of the name of a source medical institution to be matched, such as "YY Hospital East Branch";

[0059] In steps 302 and 303, the system's processor invokes a pre-trained language model. This model can be a general-purpose large language model (e.g., but not limited to the GPT series, LLaMA series) or a specialized model fine-tuned for the medical field (e.g., but not limited to PULSE, BioBERT, etc.). Through carefully designed prompt engineering, the model is instructed to perform a semantic component parsing task. For example, the prompt could be designed as: "Please parse the following medical institution name into three parts: 'identification name' (words representing the core identity of the entity), 'generic name' (words describing type, location, department, etc.), and 'administrative division' (geographical location). Input: 'YY Hospital East Branch'". After receiving the instruction, the model analyzes the data and outputs a structured result.

[0060] 304. Name Identification Section: This refers to the core words in the name that are most distinctive and can identify the entity. For "YY Hospital East Branch", the model should identify "YY";

[0061] 305. Generic Name Section: This refers to the generic terms describing the type, level, and nature of the institution. For "YY Hospital East Branch," the model should identify "Hospital" and "East Branch"; Administrative Division Section: This refers to the geographical location information explicitly included in the name. In this example, it is empty, but for "Z City H District People's Hospital," it would be "Z City H District";

[0062] 306. Next, the system assigns different preset weights to the different parts parsed above. For example, a higher weight is assigned to the "identification name part" (e.g., weight value in the range [0.7, 0.9]), and a lower weight is assigned to the "generic name part" (e.g., weight value in the range [0.1, 0.3]). These weights reflect the importance of different words in determining entity identity. The weight values ​​can be learned by training on a labeled validation dataset using grid search or gradient-based optimization methods, with the optimization objective being to maximize the F1 score of the final matching task.

[0063] 307. Finally, a structured object is generated, such as { “Identifier Name”: “YY”, “Common Name”: [“Hospital”, “East Campus”], “Weight”: { “Identifier Name”: 0.8, “Common Name”: 0.2}}.

[0064] Step S200: Dynamic geofence candidate set generation: such as Figure 3 As shown, this step utilizes geospatial proximity as strong prior knowledge, greatly narrowing the matching range and filtering out the vast majority of geographically irrelevant candidates, thereby improving matching efficiency and accuracy.

[0065] 401. The system obtains the geographic coordinates of the source medical institution. The acquisition methods may include: direct extraction: if the source data record already contains latitude and longitude fields, it is used directly; address parsing: if the source data only contains a text address (such as "No. 1, Shuaifuyuan, H District, Z City"), the system calls a geographic information service API (such as the APIs of Gaode Maps and Baidu Maps) to convert the text address into precise latitude and longitude coordinates; keyword search: if the source data only contains a name, the system calls the location search API of a geographic information service to search using the name as a keyword to obtain the most likely geographic coordinates; then, the system constructs a dynamic geofence. The "dynamic" aspect here refers to the search radius not being a fixed value, but rather adjusted according to one or more preset attributes of the source medical institution. These attributes may include:

[0066] 403 and 405, Institutional Level: Determined by keywords in the name (such as "Central Hospital," "Grade A Tertiary Hospital," or "University Affiliated") or relevant external knowledge base information. For institutions determined to be high-level, a larger search radius (e.g., 500 to 2000 meters) is used.

[0067] For 402 and 404, for institutions classified as low-level (such as "XX Community Health Service Station"), use a smaller search radius (e.g., 100 to 300 meters).

[0068] Geographical characteristics: such as population density or medical resource density of the area where the geographic coordinates are located. A smaller radius can be used in densely populated urban centers; a larger radius should be used in sparsely populated suburbs or rural areas.

[0069] The specific value of the search radius can be obtained based on statistical analysis of a large amount of geographic points of interest (POI) data. For example, it can be set as the minimum radius that can cover 99% of the known branches or related facilities of the same medical entity, and adjusted in combination with the region type.

[0070] 406. Finally, the system uses the acquired geographic coordinates as the center and performs a circular area search with a dynamically determined radius. From a standardized geographic information database, it filters out all candidate POIs of the type "healthcare services" to form a geographically highly relevant candidate set.

[0071] Step S300: Multi-level matching and intelligent arbitration: such as Figure 4 As shown, this step is the core matching decision process, which adopts a multi-level sequential strategy from deterministic to fuzzy and from simple to complex, and innovatively introduces an intelligent arbitration mechanism to handle difficult situations.

[0072] 501. For each candidate POI in the source organization and candidate set, the system performs the following determinations in sequence:

[0073] 502. Exact Match Decision: First, a simple string exact match comparison is performed. If the official name of the candidate POI is exactly the same as the source organization name string, it is directly determined as a match, the match confidence is set to the highest value of 1.0, and the subsequent matching process for this candidate POI is terminated;

[0074] 504. Weighted Fuzzy Matching Decision: If exact matching fails, weighted fuzzy matching is initiated. This algorithm (e.g., it can be implemented as a weighted Jaro-Winkler algorithm or a token-based weighted Jaccard similarity algorithm) utilizes the structured names and weights obtained in S100 when calculating similarity. For example, when comparing the source institution "YY Hospital East Branch" and the candidate institution "NN Medical Academy YY Hospital":

[0075] (1) The “identification name” part (“YY”) of the two are completely matched, contributing a score of 1.0 * weight 0.8 = 0.8;

[0076] (2) The "generic name" part of both (["hospital", "East Campus"] vs ["hospital"]) partially matches, contributing 0.5 * weight 0.2 = 0.1 points based on the degree of matching;

[0077] (3) The final weighted similarity score is 0.8 + 0.1 = 0.9;

[0078] 505. Intelligent Arbitration Triggering and Execution: The system presets a fuzzy range, for example, [0.70, 0.95]. The upper and lower limits of this range are the optimal operating points selected to balance the computational overhead of calling the intelligent arbitration module and the final model performance after analyzing the precision-recall curve on the test dataset.

[0079] 506. When the calculated weighted similarity score falls within this range, the intelligent arbitration module is triggered;

[0080] In a preferred embodiment, the semantic relationship determination module is a pre-trained Large Language Model (LLM). The system interacts with the LLM through a carefully designed prompt, assigning it an expert role and providing contextual information. For example:

[0081] Prompt: "You are a healthcare data governance expert. Based on the following information, please determine whether the 'source name' and 'candidate name' point to the same core healthcare entity, and select the most appropriate relationship from ['same entity', 'alias', 'subordinate institution', 'unrelated entity'], while providing a confidence level of 0 to 1. Please return the results in JSON format. Source name: 'YY Hospital East Branch', Candidate name: 'NN Academy of Medical Sciences YY Hospital'"

[0082] 507. LLM, based on its vast knowledge base, performs reasoning and may return the following structured arbitration result: {"relation": "subordinate organization", "confidence": 0.98}.

[0083] It should be noted that the large language model described in this invention is preferably used as a semantic reasoning component. Its output structured arbitration result serves as intermediate evidence and is integrated with other evidence, such as weighted similarity, within a unified fusion framework. This approach leverages the reasoning capabilities of LLM while avoiding the risks associated with its potential instability as the sole decision-maker, thus ensuring the robustness and interpretability of the final decision.

[0084] Step S400: Decision based on confidence fusion model: such as Figure 4 The logical flow shown in this step aims to fuse evidence from different information sources (weighted text similarity, LLM expert judgment) within a unified and well-interpretable probabilistic framework, thereby deriving a more reliable final match confidence than any single piece of evidence.

[0085] like Figure 4 The logical flow shown is as follows: 508. This invention constructs a confidence fusion model to fuse weighted similarity scores and structured arbitration results to calculate the final matching confidence. In a preferred embodiment, this confidence fusion model is a Bayesian inference model. Specifically, it applies the Bayesian formula:

[0086] ,

[0087] The explanations of each part are as follows:

[0088] Prior probability P(Match): Represents the initial belief in the likelihood of a name matching before obtaining strong evidence of intelligent arbitration. This value can be calculated from the weighted similarity score in step S300. It is obtained by mapping through a pre-defined mapping function (such as the Sigmoid function), that is... , where k and b are pre-calibrated mapping function parameters.

[0089] Observational evidence: refers to the structured arbitration result output by the semantic relationship determination module in S300, such as {"relation": "subordinate organization", "confidence": 0.98}.

[0090] Likelihood P (Evidence | Match): Represents "how likely it is that the intelligent arbitration module makes the current judgment assuming that the two names are indeed strongly related (matched)". This value can be obtained directly or by weighting the confidence score returned in the arbitration result.

[0091] Posterior probability P(Match | Evidence): This is the quantity that is ultimately required to be solved, representing the updated belief in the match between the two names after obtaining the expert opinion (evidence) from the intelligent arbitrator.

[0092] 509. After calculating the posterior probability, in practical applications, it is necessary to normalize the posterior probabilities of all possible outcomes (such as "match" and "no match") so that their sum is 1. This normalized posterior probability is used as the final match confidence score.

[0093] Finally, the system compares this final match confidence score with a final decision threshold (e.g., 0.90). If the confidence score is higher than or equal to the threshold, the system makes a final decision automatically. If the confidence score is lower than the threshold, the next step is triggered.

[0094] Step S500: Feedback and optimization of human-machine collaboration: such as Figure 5As shown, this step aims to handle complex cases that the automated model cannot be certain of, and to enable the system to learn continuously by introducing expert knowledge.

[0095] 601. Triggering and Presentation: When the final matching confidence level calculated in step S400 is lower than the decision threshold, the system automatically triggers the manual review process, pushing the matching pairs to be reviewed and related auxiliary information to a graphical user interface (GUI). For example... Figure 5 The interactive interface clearly displays information such as 603, source organization name, 604, candidate organization name, geographical location, 605, weighted similarity score, and suggestions from the intelligent arbitration module to assist experts in making judgments.

[0096] 606. Manual review and adjudication: Review experts make a final adjudication on the relationship between the two organizations based on the interface information and their own knowledge.

[0097] 607 and 608, Model Update and Optimization: After receiving authoritative matching results from experts, the system uses them as a high-quality, labeled training sample. For example... Figure 5 As shown in the data flow, once a sufficient number of manually reviewed samples have been accumulated, the system can periodically initiate the model retraining task to fine-tune the semantic analysis model in step S100 or retrain the confidence fusion model in step S400.

[0098] Through this closed-loop feedback mechanism, the method of this invention can learn from uncertainty and achieve continuous self-iteration and optimization of the model.

[0099] For some steps of the above specific embodiments, the following alternatives exist:

[0100] An alternative to step S100 (name structuring and semantic weighting): This step can also be implemented using a rule-based and dictionary-based approach. The system pre-constructs a "general name dictionary" (containing words such as "hospital," "center," "outpatient department," "branch," and "campus") and an "administrative division dictionary" (containing the names of provinces, cities, and districts nationwide). Through string matching and elimination algorithms, words found in the dictionaries are removed from the original names; the remaining parts are considered the identified names. This approach is simple to implement and has low computational cost.

[0101] An alternative to step S300 (multi-level matching and intelligent arbitration) is that the semantic relationship determination module can also be a medical domain knowledge graph. When arbitration is triggered, the system queries the graph to see if there are predefined relationship edges such as sameAs, alias, or subOrganizationOf between entity nodes representing two names, and uses this as the arbitration result. This solution has extremely high accuracy.

[0102] An alternative to step S400 (decision based on the confidence fusion model): The confidence fusion module can also be a supervised learning classifier (such as logistic regression, gradient boosting tree XGBoost, etc.). Weighted string scores, geographical distance, relationship type output by the LLM (after one-hot encoding), and confidence scores are used as features and input into a pre-trained classifier, which directly outputs the final matching probability. Although this method requires a large amount of labeled data for training, it may achieve higher performance.

[0103] Another embodiment illustrates a medical institution name matching system, such as... Figure 6 As shown, the system includes:

[0104] The name structuring and weighting module 10 is used to receive the name string of the source medical institution, use a pre-trained large language model to perform semantic component parsing on the name string to identify the identification name that represents the core identity of the entity, as well as at least one general name that describes the attributes of the entity; and assign different preset weights to the identification name and the general name.

[0105] The dynamic geofence candidate set generation module 20 is used to obtain the geographic coordinates of the source medical institution, and dynamically determine the search radius based on at least one entity attribute of the source medical institution. It constructs a geofence with the geographic coordinates as the center and the search radius, and retrieves the candidate medical institution set within the geofence.

[0106] The multi-level matching and semantic relationship determination module 30 is used to calculate the weighted similarity score between the source medical institution and any candidate medical institution in the candidate medical institution set. When the weighted similarity score falls into the preset fuzzy range, the semantic relationship between the source medical institution and the candidate medical institution is determined, and a structured arbitration result containing the relationship type and relationship confidence is output.

[0107] The confidence fusion and decision module 40 is used to construct a confidence fusion model, which fuses the weighted similarity score and the structured arbitration result, calculates the final matching confidence, and makes a matching decision based on the final matching confidence.

[0108] In a preferred embodiment, the system further includes a human-machine collaborative feedback and optimization module 50, which is used to manually review the source medical institution and the candidate medical institution when the final matching confidence level does not reach the preset decision threshold, and to retrain the confidence fusion model and / or the large language model based on the matching results obtained from the manual review.

[0109] In addition to the modules mentioned above, the system may also include other components, such as Figure 6As shown, the specific implementation method is as described in the above method embodiment. In addition, some components are not related to the content of this disclosure embodiment, so their illustrations and descriptions are omitted here.

[0110] Other specific working processes of the medical institution name matching system are described in the above-described embodiments of the medical institution name matching method, and will not be repeated here.

[0111] Another embodiment illustrating that the system of the present invention can also be achieved by means of... Figure 7 The architecture of the computing device shown is used to implement this. Figure 7 The architecture of the computing device is shown. For example... Figure 7 As shown, the computer system 710 includes a system bus 730, one or more CPUs 740, input / output 720, and memory 750. The memory 750 can store various data or files used for computer processing and / or communication, as well as program instructions executed by the CPU, including the medical institution name matching method of the embodiment. Figure 7 The architecture shown is merely exemplary and should be adjusted according to actual needs when implementing different devices. Figure 7 One or more components in the system. The memory 750, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the medical institution name matching method in this embodiment of the invention (e.g., the name structuring and weighting module 10, the dynamic geofence candidate set generation module 20, the multi-level matching and semantic relationship determination module 30, the confidence fusion and decision module 40, and the human-machine collaboration feedback and optimization module 50 in the medical institution name matching system). One or more CPUs 740 execute various functional applications and data processing of the system of the present invention by running the software programs, instructions, and modules stored in the memory 750, thereby implementing the above-mentioned medical institution name matching method, which includes the following steps:

[0112] The system receives the name string of the source medical institution, uses a pre-trained large language model to perform semantic component parsing on the name string, identifies the identification name that represents the core identity of the entity, and at least one general name that describes the attributes of the entity; and assigns different preset weights to the identification name and the general name.

[0113] Obtain the geographic coordinates of the source medical institution, and dynamically determine the search radius based on at least one entity attribute of the source medical institution. Construct a geofence centered on the geographic coordinates and with the search radius, and retrieve the candidate medical institution set within the geofence.

[0114] Calculate the weighted similarity score between the source medical institution and any candidate medical institution in the candidate medical institution set. When the weighted similarity score falls within a preset fuzzy range, determine the semantic relationship between the source medical institution and the candidate medical institution, and output a structured arbitration result that includes the relationship type and relationship confidence.

[0115] A confidence fusion model is constructed to fuse weighted similarity scores and structured arbitration results, calculate the final matching confidence, and make matching decisions based on the final matching confidence.

[0116] Of course, the processor of the server provided in the embodiments of the present invention is not limited to performing the method operations described above, but can also perform related operations in the medical institution name matching method provided in any embodiment of the present invention.

[0117] The memory 750 may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a given function; the data storage area may store data created based on terminal usage. Furthermore, the memory 750 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory, or other non-volatile solid-state storage device. In some instances, the memory 750 may further include memory remotely configured relative to one or more CPUs 740, which can be connected to the device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0118] Input / output 720 can be used to receive input digital or character information, and to generate key signal inputs related to user settings and function control of the device. Input / output 720 may also include a display device such as a display screen.

[0119] This invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, this computer program implements the medical institution name matching method described in the above embodiments. The computer-readable storage medium of this invention can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0120] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0121] The program code contained on the storage medium can be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0122] Furthermore, other specific operating procedures of a non-transitory computer-readable storage medium are described in the above-described embodiments of the medical institution name matching method, and will not be repeated here.

[0123] In this document, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a step or method that comprises a list of elements includes not only those elements but also other elements not expressly listed or inherent to such a step or method.

[0124] The above is a further detailed description of the present invention in conjunction with specific preferred embodiments, and the specific implementation of the present invention should not be considered to be limited to these descriptions. For those skilled in the art of the present invention, without departing from the concept of the present invention, several simple deductions or substitutions can be made, which should be considered to fall within the scope of protection of the present invention.

Claims

1. A method for matching medical institution names, characterized in that, The method includes the following steps: The system receives the name string of the source medical institution, uses a pre-trained large language model to perform semantic component parsing on the name string, identifies the identification name that represents the core identity of the entity, and at least one general name that describes the attributes of the entity; and assigns different preset weights to the identification name and the general name. Obtain the geographic coordinates of the source medical institution, and dynamically determine the search radius based on at least one entity attribute of the source medical institution. Construct a geofence centered on the geographic coordinates and with the search radius, and retrieve the candidate medical institution set within the geofence. Calculate the weighted similarity score between the source medical institution and any candidate medical institution in the candidate medical institution set. When the weighted similarity score falls within a preset fuzzy range, determine the semantic relationship between the source medical institution and the candidate medical institution, and output a structured arbitration result that includes the relationship type and relationship confidence. A confidence fusion model is constructed to fuse weighted similarity scores and structured arbitration results, calculate the final matching confidence, and make matching decisions based on the final matching confidence. The confidence fusion model is a Bayesian inference model, and its specific expression is as follows: ， Wherein, the prior probability P(Match) represents the initial belief in the likelihood of the two names matching before obtaining the arbitration result, and is determined by the weighted similarity score. It is obtained by mapping through a preset mapping function; The likelihood P (Evidence | Match) represents the probability that the arbitration result will give such a judgment if the two names are indeed strongly related. It is obtained directly or by weighting the confidence level returned in the arbitration result. The posterior probability P(Match | Evidence) represents the updated belief in the match between the two names after obtaining expert opinion from the intelligent arbitrator.

2. The medical institution name matching method according to claim 1, characterized in that, The weight assigned to the identifier name is higher than the weight assigned to the generic name.

3. The medical institution name matching method according to claim 1, characterized in that, The method further includes: when the final matching confidence level does not reach the preset decision threshold, manually reviewing the source medical institution and the candidate medical institution, and retraining the confidence fusion model and / or the large language model based on the matching results obtained from the manual review.

4. The medical institution name matching method according to claim 1, characterized in that, The confidence fusion model is a Bayesian inference model that uses the weighted similarity score as the prior probability and the structured arbitration result as the observation evidence to calculate the posterior probability as the final matching confidence.

5. The medical institution name matching method according to claim 1, characterized in that, The confidence fusion model is a supervised learning classifier that takes the weighted similarity score and the structured arbitration result as inputs into a pre-trained supervised learning classifier and outputs the final matching probability.

6. The medical institution name matching method according to claim 1, characterized in that, The semantic relationship between source medical institutions and candidate medical institutions can be determined by using pre-trained large language models or medical domain knowledge graphs.

7. A medical institution name matching system, characterized in that, The system includes: The name structuring and weighting module receives the name string of the source medical institution, uses a pre-trained large language model to perform semantic component parsing on the name string to identify the identification name that represents the core identity of the entity, as well as at least one general name that describes the attributes of the entity; and assigns different preset weights to the identification name and the general name. The dynamic geofence candidate set generation module is used to obtain the geographic coordinates of the source medical institutions, and dynamically determine the search radius based on at least one entity attribute of the source medical institutions. It constructs a geofence with the geographic coordinates as the center and the search radius, and retrieves the candidate medical institution set within the geofence. The multi-level matching and semantic relationship determination module is used to calculate the weighted similarity score between the source medical institution and any candidate medical institution in the candidate medical institution set. When the weighted similarity score falls into the preset fuzzy range, the semantic relationship between the source medical institution and the candidate medical institution is determined, and a structured arbitration result containing the relationship type and relationship confidence is output. The confidence fusion and decision module is used to build a confidence fusion model, which fuses weighted similarity scores and structured arbitration results, calculates the final matching confidence, and makes matching decisions based on the final matching confidence. The confidence fusion model is a Bayesian inference model, and its specific expression is as follows: ， Wherein, the prior probability P(Match) represents the initial belief in the likelihood of the two names matching before obtaining the arbitration result, and is determined by the weighted similarity score. It is obtained by mapping through a preset mapping function; The likelihood P (Evidence | Match) represents the probability that the arbitration result will give such a judgment if the two names are indeed strongly related. It is obtained directly or by weighting the confidence level returned in the arbitration result. The posterior probability P(Match | Evidence) represents the updated belief in the match between the two names after obtaining expert opinion from the intelligent arbitrator.

8. The medical institution name matching system according to claim 7, characterized in that, The system also includes a human-machine collaborative feedback and optimization module, which is used to manually review the source medical institution and the candidate medical institution when the final matching confidence does not reach the preset decision threshold, and to retrain the confidence fusion model and / or the large language model based on the matching results obtained from the manual review.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the medical institution name matching method as described in any one of claims 1 to 6.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, When the instruction is executed by the processor, it implements the steps of the medical institution name matching method as described in any one of claims 1 to 6.

Images