Hospital financial data mining system and method based on cloud platform

Abstract

The invention relates to the field of intelligent data mining, in particular to a hospital financial data mining system and method based on a cloud platform. The method comprises the steps of collecting evaluation system data of a to-be-evaluated doctor object; the evaluation system data of the to-be-evaluated doctor object is uploaded to a cloud platform; on the cloud platform, performing multi-source data collaborative analysis based on embedded semantics on business indexes, economic indexes and patient evaluation indexes in the evaluation system data of the to-be-evaluated doctor object to obtain multi-source index joint coding features of the to-be-evaluated doctor object; at the cloud platform, extracting a set of doctor object type clustering centers from a database; and on the cloud platform, performing dynamic semantic query processing on a set of the multi-source index joint coding features of the to-be-evaluated doctor object and the doctor object type clustering center to obtain query response semantic representation of the to-be-evaluated doctor object. The method is helpful for improving the intelligent level of classification management of the doctor objects.

Description

Technical Field

[0001] The present application relates to the field of intelligent data mining, and specifically to a hospital financial data mining system and method based on a cloud platform. Background Art

[0002] In the modern medical environment, hospital management not only needs to focus on patient treatment and service quality, but also faces the challenges of optimizing resource allocation, improving operational efficiency, and ensuring economic sustainability. Traditional hospital financial management methods often focus on post-assessment and simple data analysis, which limits managers' real-time insight into financial conditions and makes it difficult to provide forward-looking decision support. In addition, traditional hospital financial management methods usually rely on manual processing of large amounts of data, which is prone to information lag and inaccuracy, thus affecting the hospital management's understanding and evaluation of doctors' work performance and financial income.

[0003] Chinese patent CN118484720A provides a financial data mining method and system based on the financial income of doctors' hospitals, which can mine financial data for evaluation indicators such as the financial income of doctors' hospitals and manage according to the mining results, which can improve the efficiency of hospital financial income management for doctors. However, in the above-mentioned financial data mining method based on the financial income of doctors' hospitals, although a clustering algorithm is introduced to classify and manage doctors, the whole process still lacks sufficient intelligent support. For example, when setting the assessment and evaluation system, it relies more on expert experience and fixed rule definitions, which may lead to the evaluation criteria being inflexible and unable to dynamically adapt to the changing medical environment and business needs. In addition, in the above-mentioned patent scheme, more emphasis is placed on data analysis and decision support at the overall level, and specific guidance and support for individual doctors is relatively insufficient. This means that it is difficult for doctors to obtain effective improvement suggestions based on their own characteristics, which is not conducive to personal career development and personal performance improvement.

[0004] Therefore, an optimized hospital financial data mining solution is desired. Summary of the invention

[0005] This application is made in consideration of the above problems. One purpose of this application is to provide a hospital financial data mining system and method based on a cloud platform.

[0006] The embodiment of the present application provides a cloud platform-based hospital financial data mining method, which includes:

[0007] Collecting evaluation system data of the doctor to be evaluated, wherein the evaluation system data includes business indicators, economic indicators and patient evaluation indicators;

[0008] Uploading the evaluation system data of the doctor to be evaluated to the cloud platform;

[0009] On the cloud platform, a multi-source data collaborative analysis based on embedded semantics is performed on the business indicators, economic indicators and patient evaluation indicators in the evaluation system data of the doctor object to be evaluated to obtain a joint coding feature of the multi-source indicators of the doctor object to be evaluated;

[0010] On the cloud platform, extracting a set of cluster centers of doctor object types from a database;

[0011] On the cloud platform, dynamic semantic query processing is performed on the set of multi-source indicator joint coding features of the doctor object to be evaluated and the doctor object type cluster center to obtain a query response semantic representation of the doctor object to be evaluated, including: calculating the fast query positioning factor of the multi-source indicator joint coding features of the doctor object to be evaluated relative to each doctor object type cluster center in the set of doctor object type cluster centers to obtain a set of fast query positioning factors of the doctor object to be evaluated; based on the set of fast query positioning factors of the doctor object to be evaluated, determining a fine-grained matching search window; based on the fine-grained matching search window, fine-grained query response encoding is performed on the multi-source indicator joint coding features of the doctor object to be evaluated and the set of doctor object type cluster centers to obtain a query response semantic representation of the doctor object to be evaluated;

[0012] On the cloud platform, doctor object classification management is performed based on the query response semantic representation of the doctor object to be evaluated to determine the type label of the doctor object to be evaluated.

[0013] For example, according to the cloud platform-based hospital financial data mining method of the embodiment of the present application, in which, on the cloud platform, the business indicators, economic indicators and patient evaluation indicators in the evaluation system data of the doctor object to be evaluated are subjected to multi-source data collaborative analysis based on embedded semantics to obtain the multi-source indicator joint coding features of the doctor object to be evaluated, including:

[0014] Embedding and coding the business indicators, economic indicators and patient evaluation indicators in the evaluation system data of the doctor object to be evaluated to obtain a business indicator embedded coding vector, an economic indicator embedded coding vector and a patient evaluation indicator embedded coding vector;

[0015] The business indicator embedded coding vector, the economic indicator embedded coding vector and the patient evaluation indicator embedded coding vector are input into a multi-source data collaborative analysis module based on a multi-layer perceptron model to obtain a multi-source indicator joint coding vector of the doctor object to be evaluated as the multi-source indicator joint coding feature of the doctor object to be evaluated.

[0016] For example, according to the cloud platform-based hospital financial data mining method of an embodiment of the present application, the rapid query positioning factor of the multi-source indicator joint coding feature of the doctor object to be evaluated relative to each doctor object type cluster center in the set of doctor object type cluster centers is calculated to obtain a set of rapid query positioning factors of the doctor object to be evaluated, including:

[0017] The cross entropy of the multi-source indicator joint encoding vector of the doctor object to be evaluated relative to each doctor object type cluster center in the set of doctor object type cluster centers is calculated to obtain a set of fast query positioning factors of the doctor object to be evaluated.

[0018] For example, according to the cloud platform-based hospital financial data mining method of an embodiment of the present application, a set of positioning factors for a quick query of the doctor object to be evaluated is used to determine a fine-grained matching search window, including:

[0019] The doctor object type cluster center corresponding to the minimum value in the set of the quick query positioning factors of the doctor object to be evaluated is used as the positioning matching doctor object type cluster center;

[0020] Based on the located matching doctor object type cluster center, the fine-grained matching search window is determined, wherein the vector of the center position of the fine-grained matching search window is the located matching doctor object type cluster center, and each doctor object type cluster center in the fine-grained matching search window is defined as a fine-grained query doctor object type cluster center to obtain a set of fine-grained query doctor object type cluster centers.

[0021] For example, according to the cloud platform-based hospital financial data mining method of an embodiment of the present application, based on the fine-grained matching search window, the set of multi-source indicator joint coding features of the doctor object to be evaluated and the doctor object type cluster center is fine-grained query response encoding to obtain the query response semantic representation of the doctor object to be evaluated, including:

[0022] Performing a linear transformation on the multi-source indicator joint encoding vector of the doctor object to be evaluated to obtain a multi-source indicator joint query vector of the doctor object to be evaluated and a multi-source indicator joint value vector of the doctor object to be evaluated;

[0023] Using each fine-grained query doctor object type cluster center in the set of fine-grained query doctor object type cluster centers as a key vector, fine-grained query encoding is performed on the multi-source indicator joint query vector of the doctor object to be evaluated, the multi-source indicator joint value vector of the doctor object to be evaluated and the key vector to obtain a semantic encoding vector of the doctor object query response to be evaluated as the semantic representation of the doctor object query response to be evaluated.

[0024] For example, according to the cloud platform-based hospital financial data mining method of an embodiment of the present application, each fine-grained query doctor object type cluster center in the set of fine-grained query doctor object type cluster centers is used as a key vector, and the multi-source indicator joint query vector of the doctor object to be evaluated, the multi-source indicator joint value vector of the doctor object to be evaluated, and the key vector are fine-grained query encoded to obtain the doctor object query response semantic encoding vector to be evaluated as the doctor object query response semantic representation to be evaluated, including:

[0025] Taking each fine-grained query doctor object type cluster center in the set of fine-grained query doctor object type cluster centers as a key vector, the multi-source indicator joint query vector of the doctor object to be evaluated, the multi-source indicator joint value vector of the doctor object to be evaluated and the key vector are input into a fine-grained query encoding module based on a heterogeneous converter structure to obtain the query response semantic encoding vector of the doctor object to be evaluated.

[0026] For example, according to the cloud platform-based hospital financial data mining method of an embodiment of the present application, in which, on the cloud platform, doctor object classification management is performed based on the query response semantic representation of the doctor object to be evaluated to determine the type label of the doctor object to be evaluated, including:

[0027] The query response semantic encoding vector of the doctor object to be evaluated is input into a classifier-based doctor object classification management module to obtain the type label of the doctor object to be evaluated.

[0028] The embodiment of the present application also provides a hospital financial data mining system based on a cloud platform, which includes:

[0029] An evaluation system data collection module is used to collect evaluation system data of the doctor object to be evaluated, wherein the evaluation system data includes business indicators, economic indicators and patient evaluation indicators;

[0030] A data uploading module, used to upload the evaluation system data of the doctor object to be evaluated to the cloud platform;

[0031] A collaborative analysis module is used to perform a multi-source data collaborative analysis based on embedded semantics on the business indicators, economic indicators and patient evaluation indicators in the evaluation system data of the doctor object to be evaluated on the cloud platform to obtain a joint coding feature of the multi-source indicators of the doctor object to be evaluated;

[0032] A database extraction module, used to extract a set of doctor object type cluster centers from the database on the cloud platform;

[0033] A dynamic semantic query processing module is used to perform dynamic semantic query processing on the cloud platform on the set of multi-source indicator joint coding features of the doctor object to be evaluated and the doctor object type clustering centers to obtain a query response semantic representation of the doctor object to be evaluated; wherein the dynamic semantic query processing module includes: a fast query positioning factor calculation unit, which is used to calculate the fast query positioning factor of each doctor object type clustering center in the set of doctor object type clustering centers of the multi-source indicator joint coding features of the doctor object to be evaluated to obtain a set of fast query positioning factors of the doctor object to be evaluated; a fine-grained matching search window determination unit, which is used to determine a fine-grained matching search window based on the set of fast query positioning factors of the doctor object to be evaluated; a fine-grained query response encoding unit, which is used to perform fine-grained query response encoding on the set of multi-source indicator joint coding features of the doctor object to be evaluated and the doctor object type clustering centers based on the fine-grained matching search window to obtain a query response semantic representation of the doctor object to be evaluated;

[0034] A classification management module is used to perform classification management of doctor objects on the cloud platform based on the query response semantic representation of the doctor object to be evaluated to determine the type label of the doctor object to be evaluated.

[0035] For example, according to the cloud platform-based hospital financial data mining system of an embodiment of the present application, the collaborative analysis module includes:

[0036] An embedding coding unit, used for embedding coding the business indicators, economic indicators and patient evaluation indicators in the evaluation system data of the doctor object to be evaluated to obtain a business indicator embedding coding vector, an economic indicator embedding coding vector and a patient evaluation indicator embedding coding vector;

[0037] The analysis unit is used to input the business indicator embedded coding vector, the economic indicator embedded coding vector and the patient evaluation indicator embedded coding vector into a multi-source data collaborative analysis module based on a multi-layer perceptron model to obtain a multi-source indicator joint coding vector of the doctor object to be evaluated as a multi-source indicator joint coding feature of the doctor object to be evaluated.

[0038] For example, according to the cloud platform-based hospital financial data mining system of an embodiment of the present application, the fast query positioning factor calculation unit is used to:

[0039] The cross entropy of the multi-source indicator joint encoding vector of the doctor object to be evaluated relative to each doctor object type cluster center in the set of doctor object type cluster centers is calculated to obtain a set of fast query positioning factors of the doctor object to be evaluated.

[0040] According to the cloud platform-based hospital financial data mining system and method of the embodiments of the present application, the intelligent level of classification management of doctor objects is improved, and the defects such as poor flexibility of the traditional scheme relying on expert experience and fixed rules to define the assessment and evaluation system are avoided, and the dependence on human intervention is reduced. At the same time, this method can also support personalized guidance and management of doctors, which is helpful to improve the personalized characteristics of the doctor objects themselves. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application are briefly introduced below. Obviously, the drawings described below only relate to some embodiments of the present application, and are not intended to limit the present application.

[0042] Figure 1 A flowchart of a cloud platform-based hospital financial data mining method in an embodiment of the present application is shown;

[0043] Figure 2 A flowchart of sub-step S130 of the cloud platform-based hospital financial data mining method in an embodiment of the present application is shown;

[0044] Figure 3 A flowchart of sub-step S150 of the cloud platform-based hospital financial data mining method in an embodiment of the present application is shown;

[0045] Figure 4 A flowchart of sub-step S152 of the cloud platform-based hospital financial data mining method in an embodiment of the present application is shown;

[0046] Figure 5 A flowchart showing sub-step S153 of the cloud platform-based hospital financial data mining method in an embodiment of the present application is shown; and

[0047] Figure 6 A schematic diagram of the structure of a cloud platform-based hospital financial data mining system in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0048] The terms used in this specification are those common terms currently widely used in the art in consideration of the functions of the present application, but these terms may vary according to the intentions of those skilled in the art, precedents, or new technologies in the art. In addition, specific terms may be selected, and in this case, their detailed meanings will be described in the detailed description of the present application. Therefore, the terms used in the specification should not be understood as simple names, but rather as a general description based on the meaning of the terms and the present application.

[0049] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0050] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. At the same time, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0051] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present application.

[0052] In view of the above technical problems, in the technical solution of the present application, a cloud platform-based hospital financial data mining method is proposed, which can collect the evaluation system data of the doctor object to be evaluated, and transmit these data to the cloud platform for comparison query processing with multiple doctor object type clustering centers, that is, in the cloud platform, the evaluation system data of the doctor object to be evaluated is analyzed by using data processing and semantic understanding algorithms based on artificial intelligence and deep learning, so as to capture the embedded semantics of the evaluation system of the doctor object to be evaluated, and use this semantics to perform dynamic semantic queries with each doctor object type clustering center respectively, so as to query the query response semantic representation related to the doctor object to be evaluated, which is used for doctor object classification management and obtains the type label of the doctor object to be evaluated. In this way, the type of the doctor object to be evaluated can be automatically matched based on the query responsiveness information between the embedded semantics of the evaluation system data of the doctor object to be evaluated and the characteristics of each doctor object type clustering center. This method improves the intelligent level of doctor object classification management, avoids the defects of the poor flexibility of the traditional solution relying on expert experience and fixed rules to define the assessment and evaluation system, and reduces the dependence on manual intervention. At the same time, this method can also support the personalized guidance and management of doctors, which is helpful to improve the personalized characteristics of the doctor object itself.

[0053] Figure 1 FIG. 1 is a flowchart of a cloud platform-based hospital financial data mining method in an embodiment of the present application. Figure 1As shown, according to the cloud platform-based hospital financial data mining method of the embodiment of the present application, the steps include: S110, collecting the evaluation system data of the doctor object to be evaluated, the evaluation system data including business indicators, economic indicators and patient evaluation indicators; S120, uploading the evaluation system data of the doctor object to be evaluated to the cloud platform; S130, on the cloud platform, performing multi-source data collaborative analysis based on embedded semantics on the business indicators, economic indicators and patient evaluation indicators in the evaluation system data of the doctor object to be evaluated to obtain the multi-source indicator joint coding features of the doctor object to be evaluated; S140, on the cloud platform, extracting a set of doctor object type clustering centers from the database; S150, on the cloud platform, performing dynamic semantic query processing on the multi-source indicator joint coding features of the doctor object to be evaluated and the set of doctor object type clustering centers to obtain the query response semantic representation of the doctor object to be evaluated; S160, on the cloud platform, performing doctor object classification management based on the query response semantic representation of the doctor object to be evaluated to determine the type label of the doctor object to be evaluated. It is worth mentioning that the doctor object type clustering center here is in vector form, that is, the doctor object type clustering center vector.

[0054] It is worth mentioning that, regarding step S110, in the process of hospital financial data mining, collecting the evaluation system data of the doctor to be evaluated is a crucial task, which directly determines the quality and accuracy of subsequent analysis. This process covers the data collection of business indicators, economic indicators and patient evaluation indicators, aiming to build a comprehensive and detailed work performance portrait for each doctor.

[0055] First, data collection for business indicators mainly relies on hospital information systems (HIS) and electronic medical records (EMR). These systems automatically record key information about each doctor's daily diagnosis and treatment activities, such as outpatient volume, hospital stay days, and number of surgeries. Through the seamless connection of these systems, the doctor's daily workload and its changing trends can be accurately captured. In addition, in order to evaluate the doctor's professional skills and service quality, it is also necessary to compare the patient's final diagnosis results with the initial diagnosis in the medical records, as well as the follow-up after discharge, to measure the correct diagnosis rate and treatment effect of the disease. At the same time, the quality control department within the hospital will regularly count medical quality control indicators such as infection rate and complication rate to ensure the safety and effectiveness of medical services. In terms of economic indicators, the focus is on extracting various fee details from the financial management information system (FMS), such as drug fees, examination and testing fees, and surgical fees, to reflect the direct economic impact of doctors in their diagnosis and treatment activities. Calculating the cost-effectiveness ratio is another important task, which requires combining financial data with clinical pathways or diagnosis and treatment guidelines to determine a reasonable cost expenditure range, so as to evaluate the cost-benefit ratio of each doctor's case. Medical insurance reimbursement should not be ignored either, because tracking the actual reimbursement amount under different medical insurance policies can help understand the rationality of doctors' use of medical insurance resources during diagnosis and treatment, which is of great significance for optimizing resource allocation. Patient evaluation indicators are collected through multiple channels, including standardized satisfaction questionnaires, complaint handling records, and public opinions on social media and online review platforms. Satisfaction surveys usually cover multiple dimensions such as service attitude, communication skills, and treatment effects, and are distributed to patients through online platforms (such as WeChat applets, official websites) or offline paper forms. Establish an effective complaint feedback mechanism to ensure that all patient complaints are responded to in a timely manner and recorded in detail as an important reference for negative evaluations. For social media and online comments, the use of natural language processing technology (NLP) for monitoring and analysis can capture the public's opinion tendencies towards specific doctors or departments, and provide a supplement to patient evaluations through informal channels.

[0056] After completing the data collection of the above parts, a series of data cleaning, conversion and standardization work is required to ensure the consistency and integrity of the data. For example, operations such as removing duplicates, filling missing values, and unifying date formats can effectively merge heterogeneous data from different systems to form a comprehensive and accurate evaluation system data set for the doctor to be evaluated. The work at this stage is crucial to ensure the effectiveness of subsequent collaborative analysis of multi-source data based on embedded semantics.

[0057] Finally, all the sorted data will be uploaded to the cloud platform and prepared for the next step of in-depth analysis. This method not only improves the speed and efficiency of data processing, but also enhances the objectivity and fairness of the evaluation results, providing hospital management with instant financial insights and forward-looking decision support, while also providing personalized guidance for doctors' career development and personal performance improvement.

[0058] Then, regarding step S120, uploading the evaluation system data of the doctor object to be evaluated to the cloud platform is a key step in the hospital financial data mining method. This process not only ensures the secure transmission and efficient storage of data, but also lays the foundation for subsequent real-time processing. The entire implementation process needs to comprehensively consider technical implementation, privacy protection regulations and security standards to ensure the quality, integrity and security of the data.

[0059] Before starting to upload, the collected data must first be carefully organized and preprocessed. This step is crucial to ensure the consistency and integrity of the data. The staff will remove duplicates, fill in missing values, and unify the data format so that data from different sources can be seamlessly connected. Sensitive information such as patient names, ID numbers, etc. must be anonymized or desensitized to ensure that even if the data is leaked, it will not lead to personal privacy exposure. For multi-source heterogeneous data, necessary conversions are also required so that all data form a structured data set for subsequent analysis.

[0060] It is equally important to choose a reliable cloud service provider. Given the particularity of the medical industry, the selected cloud platform should have high availability, strong security, and comply with relevant industry standards (such as HIPAA, GDPR). When choosing, you also need to consider cost-effectiveness, expansion flexibility, and compatibility with existing IT infrastructure to ensure the best service experience.

[0061] In order to ensure the security of data transmission, encrypted communication protocols such as HTTPS or SFTP (Secure File Transfer Protocol) are usually used. These protocols can establish a secure connection between the client and the server to prevent data from being intercepted or tampered with during transmission. SSL / TLS certificates can further enhance the encryption strength. For batch data transmission, automated tools such as API interfaces or ETL (Extract, Transform, Load) tools allow data to be extracted, transformed, and loaded from local systems into cloud databases, reducing the risk of manual intervention and improving efficiency.

[0062] Once the data is successfully uploaded to the cloud platform, the next step is to verify its completeness and accuracy. The cloud platform should provide corresponding tools and services to check whether the uploaded data is consistent with the original records and report any differences in a timely manner. In order to keep the data up to date, it is very important to set up a regular synchronization mechanism, such as updating the new data generated that day every night, or setting a more frequent synchronization frequency based on actual business needs. This helps ensure that management can obtain the latest insights into financial conditions and make more accurate decisions.

[0063] Finally, user permissions and access control must be strictly managed on the cloud platform to ensure that only authorized personnel can view or operate specific data resources. Through role definition and permission allocation, the operation permissions of different user groups to various types of data can be finely controlled. For example, only managers in a specific department are allowed to access comprehensive financial analysis results, while ordinary employees can only see part of the information related to their work. In addition, all access behaviors should be recorded for future auditing and tracking.

[0064] Next, regarding step S140, extracting a set of doctor object type cluster centers from the database on the cloud platform is a key step in the hospital financial data mining process. This process not only needs to ensure that the extracted data accurately reflects the doctor's work performance characteristics, but also provides a solid foundation for subsequent intelligent management and personalized guidance. To achieve this, the entire process involves the application of multiple technologies and methods to ensure the quality, real-time and security of the data.

[0065] First, it is crucial to understand the concept of "doctor object type cluster center". These cluster centers are representative points or vectors obtained by analyzing a large amount of doctor evaluation system data through specific algorithms (such as K-means, hierarchical clustering, etc.). Each cluster center represents a group of doctors with similar characteristics. For example, some doctors may be good at handling complex cases but are more expensive, while others are known for their high efficiency and low cost. The selection and calculation of cluster centers rely on the evaluation system data that has been previously uploaded to the cloud platform and pre-processed.

[0066] When it is necessary to extract these cluster centers from the database, the first thing to consider is the design and optimization of the database. In order to support efficient query operations, the database on the cloud platform must be carefully designed and optimized. This includes creating index structures to speed up retrieval, and reasonably dividing the table structure to reduce redundancy and improve read and write efficiency. For the data table storing the cluster centers of doctor object types, it should contain sufficient meta information, such as cluster ID, category label, timestamp, etc., in order to track the historical changes and development trends of each cluster center. This meticulous design enables the database to respond quickly to complex query requirements while ensuring the integrity and consistency of the data.

[0067] The next step is to build precise query statements. Based on business needs and technical requirements, the development team writes specialized SQL or other types of query statements to extract the required set of cluster centers. The query conditions can be flexibly adjusted according to the specific application scenario, such as filtering out qualified cluster centers according to date range, specific department or specific evaluation indicators. Considering that the cluster centers may be dynamically updated over time, the query statement also needs to be able to handle version control issues to ensure that the latest and valid results are obtained each time. This step is particularly important to ensure the timeliness and accuracy of the extracted data.

[0068] The selection of clustering algorithms and their parameter settings are equally critical. Different clustering algorithms are suitable for different types of data sets and business scenarios. Selecting a suitable clustering algorithm (such as K-means, DBSCAN, Gaussian Mixture Models, etc.) and correctly setting its parameters (such as the number of clusters K) directly affects the quality of the final generated cluster center. In practical applications, it may be necessary to go through multiple experiments and verifications to find the algorithm and its configuration that best suits the current data characteristics. In this way, it can be ensured that the cluster center can capture the main characteristics of the doctor group without over-generalization, thereby providing strong support for refined management.

[0069] During the implementation process, real-time and performance optimization cannot be ignored. Since data analysis in hospital environments often requires high real-time performance, performance optimization should also be emphasized in the process of extracting cluster centers. On the one hand, frequently accessed data can be temporarily saved through a cache mechanism to reduce repeated calculations; on the other hand, large-scale data processing tasks can be distributed to multiple nodes for parallel execution using a distributed computing framework (such as Apache Spark), significantly improving response speed. Such technical means ensure that the system can maintain efficient operation even in the face of massive data.

[0070] Finally, security and privacy protection are always important considerations. All operations involving sensitive personal information must comply with strict regulatory standards (such as HIPAA, GDPR). This means that encryption technology should be used not only at the transmission level, but also at the storage level to prevent unauthorized access. At the same time, any information that may reveal an individual's identity should be anonymized or desensitized to ensure that it does not pose a privacy risk even when used internally. Strict security measures not only protect the personal information of patients and doctors, but also enhance the confidence of medical institutions in the use of data.

[0071] To further explain, in step S130, Figure 2As shown, on the cloud platform, the business indicators, economic indicators and patient evaluation indicators in the evaluation system data of the doctor object to be evaluated are subjected to multi-source data collaborative analysis based on embedded semantics to obtain multi-source indicator joint coding features of the doctor object to be evaluated, including: S131, embedding coding is performed on the business indicators, economic indicators and patient evaluation indicators in the evaluation system data of the doctor object to be evaluated to obtain business indicator embedded coding vectors, economic indicator embedded coding vectors and patient evaluation indicator embedded coding vectors; S132, the business indicator embedded coding vectors, the economic indicator embedded coding vectors and the patient evaluation indicator embedded coding vectors are input into a multi-source data collaborative analysis module based on a multi-layer perceptron model to obtain multi-source indicator joint coding vectors of the doctor object to be evaluated as multi-source indicator joint coding features of the doctor object to be evaluated.

[0072] Specifically, in the technical solution of the present application, first, the evaluation system data of the doctor object to be evaluated is collected, and the evaluation system data includes business indicators, economic indicators and patient evaluation indicators, and the evaluation system data of the doctor object to be evaluated is uploaded to the cloud platform. On the cloud platform, the business indicators, economic indicators and patient evaluation indicators in the evaluation system data of the doctor object to be evaluated are analyzed to extract the multi-source indicator joint coding feature information about the doctor object to be evaluated. It should be understood that since the evaluation system data of the doctor object to be evaluated includes business indicators, economic indicators and patient evaluation indicators, the business indicators include the average number of outpatient visits per day, the number of discharged patients, the number of surgeries, and the average length of stay of the doctor; the economic indicators include the proportion of hospital financial income to medical income, the proportion of inspection and testing income to medical income, the proportion of drug income to medical income, and the proportion of consumables income to medical income; the patient evaluation indicators include patient satisfaction and patient complaint rate. Therefore, the data types contained in the business indicators, the economic indicators and the patient evaluation indicators are all in unstructured form, which makes it difficult to perform subsequent processing and information fusion analysis. Based on this, in the technical solution of the present application, the business indicators, economic indicators and patient evaluation indicators in the evaluation system data of the doctor object to be evaluated are embedded and coded to obtain business indicator embedded coding vectors, economic indicator embedded coding vectors and patient evaluation indicator embedded coding vectors. Through the embedded coding method, the business indicators, economic indicators and patient evaluation indicators in the evaluation system data of the doctor object to be evaluated can be embedded and mapped into a common space, so as to be converted into vector form to obtain business indicator embedded coding vectors, economic indicator embedded coding vectors and patient evaluation indicator embedded coding vectors, so that the computer can understand and process this information. In addition, embedded coding can also capture the embedded coding semantic feature information in the business indicators, the economic indicators and the patient evaluation indicators respectively, providing a basis for the subsequent multi-source indicator embedded semantic joint analysis and doctor object type judgment.

[0073] Then, considering that each doctor in the medical environment involves data of multiple dimensions, including but not limited to business performance (such as the number of outpatient visits, the number of surgeries), economic benefits (such as the proportion of income, drug costs) and patient feedback (such as satisfaction, complaint rate), these different sources of information are crucial for the comprehensive evaluation of a doctor's work performance and the classification and management of doctor objects. However, processing each dimension separately may ignore the inherent connection and nonlinear complex relationship between them. Therefore, in the technical solution of the present application, the business indicator embedding coding vector, the economic indicator embedding coding vector and the patient evaluation indicator embedding coding vector are further input into the multi-source data collaborative analysis module based on the multi-layer perceptron model to obtain the multi-source indicator joint coding vector of the doctor object to be evaluated. Through the processing of the multi-source data collaborative analysis module based on the multi-layer perceptron model, the business indicator embedded semantics, the economic indicator embedded semantics and the patient evaluation indicator embedded semantics of the doctor object to be evaluated can be unified, and the implicit association and collaborative semantic features between the embedded semantics of these indicators are extracted to ensure that each kind of relevant information about the doctor object to be evaluated can be reflected in the final analysis result.

[0074] Furthermore, in order to be able to identify and judge the type of doctor object to be evaluated, it is necessary to compare its semantic features with the cluster centers of each doctor object type to determine. Based on this, a set of doctor object type cluster centers is further extracted from the database. In particular, the doctor object type cluster center here is in vector form, that is, the doctor object type cluster center vector, which represents the cluster center characteristics of each doctor object type. Different doctor object type cluster centers reflect different types of doctor groups. It is used for subsequent doctor object type judgment and classification management tasks.

[0075] It should be understood that since the multi-source indicator joint coding vector of the doctor object to be evaluated and the set of doctor object type cluster centers respectively contain the multi-source indicator joint embedded semantic coding features related to the doctor object to be evaluated and the cluster center features of various doctor object types, when judging and classifying the type of doctor object to be evaluated, it is necessary to compare and match the multi-source indicator joint embedded semantic representation of the doctor object to be evaluated with the features of each doctor object type cluster center to query the category related to the doctor object to be evaluated, thereby facilitating the subsequent classification and management of the doctor object. However, since the traditional feature matching query method is based on the overall doctor object type database for screening, the response speed and accuracy of the retrieval query are low. Therefore, in the technical solution of the present application, the multi-source indicator joint coding features of the doctor object to be evaluated and the set of doctor object type cluster centers are further subjected to dynamic semantic query processing to obtain the query response semantic representation of the doctor object to be evaluated. In particular, the process of dynamic semantic query processing realizes the process from coarse positioning to fine-grained matching and then to deep semantic encoding by integrating technologies such as cross-entropy calculation, local context awareness, linear transformation and heterogeneous transformer structure, aiming to optimize the query response speed and accuracy of the doctor object category to be evaluated.

[0076] Specifically, the dynamic semantic query processing process introduces a fast positioning mechanism, that is, using cross entropy calculation to evaluate the semantic distance between the joint encoding vector of the multi-source indicators of the doctor object to be evaluated and each of the doctor object type cluster centers, so that the semantic difference between the two can be quantified, thereby quickly narrowing the search range to the most likely related area and improving the query matching response speed. Then, the preliminary screening finds the best matching doctor object type cluster center. This method reduces unnecessary computing resource consumption and accelerates the preliminary screening process. Secondly, in order to further improve the matching query accuracy, the system defines a fine-grained matching search window around the best matching doctor object type cluster center, which focuses on the local area around the best matching point determined by the preliminary screening. This strategy enhances the model's ability to understand the deep association semantics between the joint embedding semantic representation of the multi-source indicators of the doctor object to be evaluated and the cluster centers of each doctor object type, enabling the system to capture more subtle matching query response semantic differences, thereby providing more accurate query response retrieval results for the doctor object category to be evaluated, and providing a basis for the subsequent determination of the type label of the doctor object to be evaluated and the classification management of the doctor object.

[0077] Accordingly, if Figure 3As shown, in step S150, on the cloud platform, dynamic semantic query processing is performed on the set of multi-source indicator joint coding features of the doctor object to be evaluated and the doctor object type clustering centers to obtain a query response semantic representation of the doctor object to be evaluated, including: S151, calculating the fast query positioning factor of the multi-source indicator joint coding features of the doctor object to be evaluated relative to each doctor object type clustering center in the set of doctor object type clustering centers to obtain a set of fast query positioning factors of the doctor object to be evaluated; S152, determining a fine-grained matching search window based on the set of fast query positioning factors of the doctor object to be evaluated; S153, based on the fine-grained matching search window, performing fine-grained query response encoding on the multi-source indicator joint coding features of the doctor object to be evaluated and the doctor object type clustering centers to obtain a query response semantic representation of the doctor object to be evaluated.

[0078] Among them, in step S151, the fast query positioning factor of the multi-source indicator joint coding feature of the doctor object to be evaluated relative to each doctor object type cluster center in the set of doctor object type cluster centers is calculated to obtain a set of fast query positioning factors of the doctor object to be evaluated, including: calculating the cross entropy of the multi-source indicator joint coding vector of the doctor object to be evaluated relative to each doctor object type cluster center in the set of doctor object type cluster centers to obtain a set of fast query positioning factors of the doctor object to be evaluated.

[0079] Specifically, first, the cross entropy of the multi-source indicator joint coding vector of the doctor object to be evaluated relative to each doctor object type cluster center in the set of doctor object type cluster centers is calculated to obtain a set of rapid query positioning factors for the doctor object to be evaluated. That is, by calculating the cross entropy of the multi-source indicator joint coding vector of the doctor object to be evaluated relative to each doctor object type cluster center in the set of doctor object type cluster centers, a numerical set reflecting the semantic distance or mismatch between the two is obtained, that is, a set of rapid query positioning factors for the doctor object to be evaluated. Cross entropy, as an important indicator for measuring the difference between two probability distributions, is used here to quantify the semantic difference between the multi-source indicator joint coding vector of the doctor object to be evaluated and each doctor object type cluster center. A smaller cross entropy value means a more similar semantic expression, so that the search range can be quickly narrowed to the most likely related area.

[0080] The process can be expressed as follows:

[0081] D={d1,d2,...,d i ,...,d n}

[0082]

[0083] F={H(q,d1),H(q,d2),...,H(q,d i ),...,H(q,d n )}

[0084] Where D is the set of cluster centers of the doctor object type, d1, d2, d i ,d n are the first, second, i-th and n-th doctor object type cluster centers in the set of doctor object type cluster centers, q k is the eigenvalue of each position in the multi-source index joint encoding vector of the doctor object to be evaluated, n is the number of eigenvalues, is the eigenvalue of each position in the cluster center of the i-th doctor object type, H(q,d i ) is the cross entropy between the multi-source index joint encoding vector of the doctor object to be evaluated and the cluster center of the i-th doctor object type, that is, the fast query positioning factor of the doctor object to be evaluated, F is the set of fast query positioning factors of the doctor object to be evaluated, H(q,d1),H(q,d2),H(q,d i ),H(q,d n ) are respectively the quick query positioning factors for the 1st, 2nd, i-th and n-th doctor objects to be evaluated.

[0085] In particular, in the technical solution of the present application, the cross entropy is not just a simple distance measurement tool, but a key indicator to measure the difference between two probability distributions. When applied to the matching between the joint encoding vector of the multi-source indicators of the doctor object to be evaluated and the cluster center of the doctor object type, by treating each doctor object type cluster center as a possible probability distribution and using the joint encoding vector of the multi-source indicators of the doctor object to be evaluated to evaluate these distributions, a series of scores reflecting the semantic similarity between the two can be obtained-that is, a set of quick query positioning factors for the doctor object to be evaluated. The cross entropy calculation here not only takes into account the direct semantic similarity, but also implicitly captures the potential contextual relationship and semantic structure. This mechanism helps to improve retrieval efficiency, especially when facing large-scale data sets, and can significantly reduce unnecessary computational burden while maintaining high retrieval accuracy.

[0086] Next, in step S152, Figure 4As shown, based on the set of quick query positioning factors of the doctor object to be evaluated, a fine-grained matching search window is determined, including: S1521, taking the doctor object type cluster center corresponding to the minimum value in the set of quick query positioning factors of the doctor object to be evaluated as the positioning matching doctor object type cluster center; S1522, based on the positioning matching doctor object type cluster center, determining the fine-grained matching search window, wherein the vector of the center position of the fine-grained matching search window is the positioning matching doctor object type cluster center, and each doctor object type cluster center in the fine-grained matching search window is defined as a fine-grained query doctor object type cluster center to obtain a set of fine-grained query doctor object type cluster centers.

[0087] Specifically, the minimum value in the set of positioning factors of the doctor object to be evaluated calculated above is used to determine the best matching doctor object type cluster center, that is, to locate the matching doctor object type cluster center. The basis for this selection is that the doctor object type cluster center corresponding to the minimum value is considered to be the candidate that is semantically closest to the joint encoding vector of the multi-source indicators of the doctor object to be evaluated. This strategy ensures that the doctor object type cluster center initially screened has the highest relevance, providing a good starting point for subsequent more detailed searches. From the perspective of machine learning, this method embodies the principle of the greedy algorithm: each iteration makes the best choice that seems to be the best at the moment, hoping to achieve the overall optimality through a series of such choices.

[0088] Next, based on the location matching doctor object type cluster center, a fine-grained matching search window is delineated around this point. The doctor object type cluster center within this window is designated as the fine-grained query doctor object type cluster center for the next more refined matching operation. The design of the fine-grained matching search window allows the system to concentrate resources on in-depth analysis of a small range of data that may contain precise answers, which not only improves retrieval accuracy but also ensures computational efficiency. The process of defining the fine-grained matching search window introduces an important concept - local context awareness. By building a small search area around the initially selected location matching doctor object type cluster center, the system is able to conduct in-depth analysis on a smaller and more targeted data subset. This not only improves search efficiency, but also enhances the model's ability to understand local semantic patterns. In addition, the design of the fine-grained matching search window allows the model to better handle the subtle differences between the query and the doctor object type cluster center. For example, in natural language processing tasks, even if the overall meaning of two sentences is very close, the choice of certain words may lead to subtle changes in sentiment. By focusing on fine-grained features, the model can capture such details more accurately, thereby improving the quality of the final output.

[0089] The process can be expressed as follows:

[0090]

[0091] w={d j-m / 2 ,d j-m / 2+1 ,...,d * ,...,d j+m / 2-1 ,d j+m / 2}

[0092]

[0093] in, Returns the d corresponding to the minimum value i ,H(q,d i ) is the cross entropy between the multi-source index joint encoding vector of the doctor object to be evaluated and the cluster center of the i-th doctor object type, that is, the fast query positioning factor of the doctor object to be evaluated, d * To locate the cluster center of the matching doctor object type, ||·|| 2 It means calculating the square of the vector norm, log2 means calculating the logarithmic function value with base 2, represents the rounding down process, m is the matching search window factor, (jm / 2) is the starting point of the fine-grained matching search window, (j+m / 2) is the end point of the fine-grained matching search window, and d j-m / 2 ,d j-m / 2+1 ,d * ,d j+m / 2-1 ,d j+m / 2 They are respectively the fine-grained query doctor object type cluster centers in the set of fine-grained query doctor object type cluster centers, q is the multi-source indicator joint encoding vector of the doctor object to be evaluated, and W is the set of fine-grained query doctor object type cluster centers.

[0094] Finally, in step S153, Figure 5As shown, based on the fine-grained matching search window, fine-grained query response encoding is performed on the set of multi-source indicator joint coding features of the doctor object to be evaluated and the doctor object type clustering centers to obtain the query response semantic representation of the doctor object to be evaluated, including: S1531, linearly transforming the multi-source indicator joint coding vector of the doctor object to be evaluated to obtain the multi-source indicator joint query vector of the doctor object to be evaluated and the multi-source indicator joint value vector of the doctor object to be evaluated; S1532, using each fine-grained query doctor object type clustering center in the set of fine-grained query doctor object type clustering centers as a key vector, fine-grained query encoding is performed on the multi-source indicator joint query vector of the doctor object to be evaluated, the multi-source indicator joint value vector of the doctor object to be evaluated and the key vector to obtain the doctor object query response semantic coding vector as the doctor object query response semantic representation.

[0095] Among them, in step S1532, each fine-grained query doctor object type cluster center in the set of fine-grained query doctor object type cluster centers is used as a key vector, and the multi-source indicator joint query vector of the doctor object to be evaluated, the multi-source indicator joint value vector of the doctor object to be evaluated and the key vector are fine-grained query encoded to obtain a semantic encoding vector of the doctor object query response to be evaluated as the semantic representation of the doctor object query response to be evaluated, including: using each fine-grained query doctor object type cluster center in the set of fine-grained query doctor object type cluster centers as a key vector, and inputting the multi-source indicator joint query vector of the doctor object to be evaluated, the multi-source indicator joint value vector of the doctor object to be evaluated and the key vector into a fine-grained query encoding module based on a heterogeneous converter structure to obtain the semantic encoding vector of the doctor object query response to be evaluated.

[0096] Specifically, a linear transformation is performed on the joint encoding vector of the multi-source indicators of the doctor object to be evaluated to generate a joint query vector of the multi-source indicators of the doctor object to be evaluated and a joint value vector of the multi-source indicators of the doctor object to be evaluated. Subsequently, the cluster centers of each fine-grained query doctor object type in the fine-grained matching search window are combined, and each fine-grained query doctor object type cluster center is input into the fine-grained query encoding module based on the heterogeneous transformer structure. In this module, the interaction between the joint query vector of the multi-source indicators of the doctor object to be evaluated and the joint value vector of the multi-source indicators of the doctor object to be evaluated simulates the attention mechanism, captures the complex relationship between features through the multi-head self-attention layer, and finally outputs the query response semantic encoding vector. The heterogeneous transformer structure allows the model to process different types of input data and introduces additional flexibility and expressiveness in the encoding process, thereby producing a richer and more accurate query response representation. Linear transformation is one of the key steps to convert the original joint encoding vector of the multi-source indicators of the doctor object to be evaluated into a representation form that is more suitable for downstream tasks. In this process, the multi-source index joint encoding vector of the doctor object to be evaluated is decomposed into the multi-source index joint query vector of the doctor object to be evaluated and the multi-source index joint value vector of the doctor object to be evaluated. These two vectors are used to guide different aspects of the attention mechanism: the multi-source index joint query vector of the doctor object to be evaluated is used to determine which parts should be paid attention to; while the multi-source index joint value vector of the doctor object to be evaluated carries the actual information content. This design enables the attention mechanism to operate more flexibly and targetedly. The application of heterogeneous transformers is to meet the challenge of interaction between different types of data. Traditional transformer structures usually assume that the input data has the same modality or format, but in the real world, the multi-source index joint query features of the doctor object to be evaluated and the cluster center features of the doctor object type often come from different sources or representations. The heterogeneous transformer introduces additional flexibility, such as support for variable-length input and multi-modal data processing, to ensure that the model can still maintain good performance in complex environments. The multi-head self-attention layer in the fine-grained query encoding module allows parallel processing of feature associations from multiple different angles, thereby generating a richer and more comprehensive semantic encoding vector of the query response of the doctor object to be evaluated as the semantic representation of the query response of the doctor object to be evaluated.

[0097] The process can be expressed as follows:

[0098] v q =qW q +b q

[0099] v v =qW v +b v

[0100]

[0101] Where q is the multi-source index joint encoding vector of the doctor object to be evaluated, W q and b q are the query weight matrix and query bias vector, respectively, W v and b v are the value weight matrix and value bias vector, respectively, v q and v v are the joint query vector of the multi-source indicators of the doctor object to be evaluated and the joint value vector of the multi-source indicators of the doctor object to be evaluated, respectively. T represents the transpose operation, d l is the lth fine-grained query doctor object type cluster center in the set of fine-grained query doctor object type cluster centers, S is the length of the lth fine-grained query doctor object type cluster center, is matrix multiplication, softmax(·) is the softmax function, v r A query response semantic encoding vector is generated for the doctor object to be evaluated.

[0102] Then, the query response semantic encoding vector of the doctor object to be evaluated is input into the classifier-based doctor object classification management module to obtain the type label of the doctor object to be evaluated. In other words, the doctor object is classified and managed by using the query responsiveness information between the multi-source indicator embedding semantic joint feature representation of the doctor object to be evaluated and the cluster center features of each doctor object type, so as to obtain the type label of the doctor object to be evaluated. This method of automatically matching doctor object types improves the level of intelligence in the classification management of doctor objects, avoids the defects of traditional solutions such as the poor flexibility of relying on expert experience and fixed rules to define the assessment and evaluation system, and reduces the dependence on manual intervention. At the same time, this method can also support personalized guidance and management of doctors, and help to improve the personalized characteristics of the doctor objects themselves.

[0103] Accordingly, in step S160, on the cloud platform, doctor object classification management is performed based on the semantic representation of the query response of the doctor object to be evaluated to determine the type label of the doctor object to be evaluated, including: inputting the semantic encoding vector of the query response of the doctor object to be evaluated into a classifier-based doctor object classification management module to obtain the type label of the doctor object to be evaluated.

[0104] In the process of hospital financial data mining, inputting the semantic encoding vector of the query response of the doctor object to be evaluated into the classifier-based doctor object classification management module to obtain the type label of the doctor object to be evaluated is a key step in realizing intelligent doctor performance management and personalized guidance. This process not only improves the intelligence level of doctor object classification management, but also reduces the dependence on manual intervention and supports a more flexible and personalized assessment and evaluation system.

[0105] In one example, first, when entering step S160, on the cloud platform, the system has obtained the multi-source indicator embedded semantic joint feature representation of the doctor object to be evaluated through the previous steps (such as S130 and S150), as well as the query responsiveness information between these features and the cluster centers of each doctor object type. This query responsiveness information reflects how the performance of the doctor object to be evaluated in different dimensions matches the existing cluster centers, providing a solid foundation for the subsequent classification. Next, these query response semantic encoding vectors are input into a classifier-based doctor object classification management module. The classifier can be a machine learning model, such as a support vector machine (SVM), a random forest, a neural network, etc., or a pre-trained model in a deep learning framework. In order to ensure the accuracy of the classification results, the classifier needs to be trained with a large amount of historical data to learn the performance patterns of different types of doctors and their corresponding feature distributions. During the training process, the classifier gradually learns to identify which feature combinations best represent a certain type of doctor group, so that it can make accurate predictions when facing new doctor objects to be evaluated.

[0106] In practical applications, the working principle of the classifier can be understood from the following aspects. First, the classifier receives the semantic encoding vectors of the doctor objects to be evaluated from the cloud platform as input. These vectors contain information describing the doctor's work performance from multiple perspectives, including business indicators, economic indicators, and patient evaluation indicators. Due to the use of embedded semantics, these vectors not only retain the statistical characteristics of the original data, but also capture the correlation and potential patterns between different indicators. This enables the classifier to understand the doctor's work performance at a higher level, rather than simply processing isolated data points. Then, the classifier begins to analyze the input semantic encoding vector and compares it with the knowledge base previously established through training. This knowledge base contains characteristic patterns shared by various types of doctor groups, and each pattern corresponds to a specific type label. The task of the classifier is to find one or several patterns that are most similar to the input vector and assign the corresponding type label accordingly. In order to improve the accuracy of classification, many modern classifiers adopt a mechanism called "soft voting" or "probabilistic output", that is, instead of directly giving a definite result, the probability distribution of each possible label is calculated, and the one with the highest probability is selected as the final result. This approach not only improves the robustness of classification, but also provides additional confidence measures to help decision makers better understand and interpret classification results. In addition, classifiers can also use techniques such as transfer learning or online learning to continuously adapt to emerging data and changing trends. Transfer learning allows classifiers to apply knowledge learned from other related fields to current tasks, thereby speeding up training and improving performance; while online learning enables classifiers to update their own parameters in real time without interrupting services to cope with the latest changes in the medical environment and personal behavior dynamics. The combination of these two technologies ensures that the classifier is always up to date and can accurately reflect the current actual situation. Finally, the type label obtained in this way not only represents the current work performance characteristics of the doctor to be evaluated, but also provides a basis for subsequent personalized guidance and support. Hospital management can formulate more targeted development plans based on the specific situation of each doctor to help them overcome challenges in their work and improve their personal abilities. At the same time, this automated classification management reduces the need for manual intervention, reduces the influence of subjective factors, and makes the entire evaluation process more objective and fair. In this way, hospitals can not only better allocate resources, but also provide customized improvement suggestions for each doctor, thereby promoting the improvement of overall medical service quality and operational efficiency.

[0107] In a preferred example, the semantic encoding vector of the query response of the doctor object to be evaluated is passed through a classifier-based doctor object classification management module to obtain a type label of the doctor object to be evaluated, including:

[0108] Calculate the distance between each pair of eigenvalues ​​of the semantic encoding vector of the query response of the doctor object to be evaluated, such as the L2 distance, and take the square root of the distance to obtain the semantic encoding distance representation matrix of the query response of the doctor object to be evaluated, that is, Where V is the query response semantic encoding vector of the doctor object to be evaluated, v i is the ith eigenvalue of the semantic encoding vector of the query response of the doctor object to be evaluated, v j is the jth eigenvalue of the semantic encoding vector of the query response of the doctor object to be evaluated, d(v i ,v j ) represents the calculation of the distance between the i-th eigenvalue and the j-th eigenvalue of the semantic encoding vector of the query response of the doctor object to be evaluated, D i,j is the (i, j)th eigenvalue of the semantic encoding distance representation matrix D of the query response of the doctor object to be evaluated;

[0109] Obtain the semantic encoding auto-association matrix of the query response of the doctor object to be evaluated of the semantic encoding vector of the query response of the doctor object to be evaluated, that is, (V is a row vector); where V is the semantic encoding vector of the query response of the doctor object to be evaluated, (·) T represents the transpose operation, represents vector multiplication, D' is the semantic encoding self-association matrix of the query response of the doctor object to be evaluated;

[0110] The semantic encoding vector of the query response of the doctor object to be evaluated is matrix-multiplied with the semantic encoding distance representation matrix of the query response of the doctor object to be evaluated to obtain the first-level mapping vector of the semantic encoding of the query response of the doctor object to be evaluated, that is, Wherein, V is the semantic encoding vector of the query response of the doctor object to be evaluated, D is the distance representation matrix of the semantic encoding of the query response of the doctor object to be evaluated, represents matrix multiplication, V' is the first-level mapping vector of the query response semantic encoding of the doctor object to be evaluated;

[0111] The first-level mapping vector of the semantic encoding of the query response of the doctor object to be evaluated is matrix-multiplied with the matrix product of the distance representation matrix of the semantic encoding of the query response of the doctor object to be evaluated and the self-association matrix of the semantic encoding of the query response of the doctor object to be evaluated to obtain the multi-level mapping vector of the semantic encoding of the query response of the doctor object to be evaluated, that is, Wherein, D is the distance representation matrix of the semantic encoding of the query response of the doctor object to be evaluated, D' is the self-association matrix of the semantic encoding of the query response of the doctor object to be evaluated, represents matrix multiplication, V' is the semantic encoding primary mapping vector of the query response of the doctor object to be evaluated, and V" is the semantic encoding multi-level mapping vector of the query response of the doctor object to be evaluated;

[0112] The optimized semantic encoding vector of the query response to be evaluated for the doctor object is obtained by performing a dot-splitting operation on the semantic encoding multi-level mapping vector of the query response to be evaluated for the doctor object and the semantic encoding associated eigenvector of the query response to be evaluated for the doctor object composed of the eigenvalues ​​of the semantic encoding auto-association matrix of the query response to be evaluated for the doctor object (if the eigenvalue is insufficient, interpolation or zero filling is performed);

[0113] The optimized query response semantic encoding vector of the doctor object to be evaluated is passed through a classifier-based doctor object classification management module to obtain a type label of the doctor object to be evaluated.

[0114] Taking into account that the multi-source indicator joint coding vector of the doctor object to be evaluated represents the multi-source embedded coding fusion features of the evaluation system data of the doctor object to be evaluated, and the set of doctor object type cluster centers represents multiple doctor object type cluster coding features, when performing dynamic semantic query encoding based on a fast positioning mechanism, inconsistent feature distribution cluster density will also lead to the lack of query coding feature instance determination of the query response semantic coding vector of the doctor object to be evaluated, thereby affecting the accuracy of the type label of the doctor object to be evaluated obtained through the classifier-based doctor object classification management module.

[0115] Therefore, through the linear target mapping representation based on the similarity distance representation matrix of the semantic coding vector of the query response of the doctor object to be evaluated, the self-correlation of the complete similarity instantiation of the semantic coding vector of the query response of the doctor object to be evaluated is instantiated with a secondary target mapping representation based on a multi-level distribution hierarchy, and the association mismatch negative influence factor is compensated by the association fusion kernel bias to improve the degree of instance determination of the eigenvalue of the semantic coding vector of the query response of the doctor object to be evaluated under the similarity constraint, that is, the degree of significance of the eigenvalue as an instance for classification and regression judgment, and improve the accuracy of the type label of the doctor object to be evaluated obtained by the semantic coding vector of the query response of the doctor object to be evaluated through the classifier-based doctor object classification management module.

[0116] Based on the above embodiments, see Figure 6As shown, it is a structural diagram of a cloud platform-based hospital financial data mining system 100 in an embodiment of the present application. The hospital financial data mining system 100 based on the cloud platform includes: an evaluation system data collection module 110, which is used to collect evaluation system data of the doctor object to be evaluated, and the evaluation system data includes business indicators, economic indicators and patient evaluation indicators; a data upload module 120, which is used to upload the evaluation system data of the doctor object to be evaluated to the cloud platform; a collaborative analysis module 130, which is used to perform multi-source data collaborative analysis based on embedded semantics on the business indicators, economic indicators and patient evaluation indicators in the evaluation system data of the doctor object to be evaluated on the cloud platform to obtain multi-source indicator joint coding features of the doctor object to be evaluated; a database extraction module 140, which is used to extract a set of doctor object type clustering centers from the database on the cloud platform; a dynamic semantic query processing module 150, which is used to perform dynamic semantic query processing on the multi-source indicator joint coding features of the doctor object to be evaluated and the set of doctor object type clustering centers on the cloud platform to obtain a query response semantic representation of the doctor object to be evaluated; and a classification management module 160, which is used to perform doctor object classification management on the cloud platform based on the query response semantic representation of the doctor object to be evaluated to determine the type label of the doctor object to be evaluated.

[0117] In one example, the collaborative analysis module 130 includes: an embedded coding unit, which is used to embed the business indicators, economic indicators and patient evaluation indicators in the evaluation system data of the doctor object to be evaluated to obtain a business indicator embedded coding vector, an economic indicator embedded coding vector and a patient evaluation indicator embedded coding vector; an analysis unit, which is used to input the business indicator embedded coding vector, the economic indicator embedded coding vector and the patient evaluation indicator embedded coding vector into a multi-source data collaborative analysis module based on a multi-layer perceptron model to obtain a multi-source indicator joint coding vector of the doctor object to be evaluated as a multi-source indicator joint coding feature of the doctor object to be evaluated.

[0118] In one example, the dynamic semantic query processing module 150 includes: a fast query positioning factor calculation unit, which is used to calculate the fast query positioning factor of the multi-source indicator joint coding feature of the doctor object to be evaluated relative to each doctor object type cluster center in the set of doctor object type cluster centers to obtain a set of fast query positioning factors of the doctor object to be evaluated; a fine-grained matching search window determination unit, which is used to determine the fine-grained matching search window based on the set of fast query positioning factors of the doctor object to be evaluated; and a fine-grained query response encoding unit, which is used to perform fine-grained query response encoding on the multi-source indicator joint coding feature of the doctor object to be evaluated and the set of doctor object type cluster centers based on the fine-grained matching search window to obtain the query response semantic representation of the doctor object to be evaluated. The fast query positioning factor calculation unit is used to: calculate the cross entropy of the multi-source indicator joint coding vector of the doctor object to be evaluated relative to each doctor object type cluster center in the set of doctor object type cluster centers to obtain the set of fast query positioning factors of the doctor object to be evaluated.

[0119] Here, those skilled in the art can understand that the specific functions and operations of each module in the above-mentioned cloud platform-based hospital financial data mining system 100 have been referred to above. Figures 1 to 5 The description of the cloud platform-based hospital financial data mining method has been introduced in detail, and therefore, its repeated description will be omitted.

[0120] Those skilled in the art will appreciate that all or part of the steps in the above method can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a disk or an optical disk. Optionally, all or part of the steps in the above embodiment can also be implemented using one or more integrated circuits. Accordingly, each module / unit in the above embodiment can be implemented in the form of hardware or in the form of software function modules. The present application is not limited to any particular form of combination of hardware and software.

[0121] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those of ordinary skill in the art to which this application belongs. It should also be understood that terms such as those defined in common dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an idealized or extremely formal sense, unless explicitly defined as such herein.

[0122] The above is an explanation of the present application and should not be considered as limiting thereof. Although several exemplary embodiments of the present application have been described, those skilled in the art will readily appreciate that many modifications may be made to the exemplary embodiments without departing from the novel teachings and advantages of the present application.

Claims

1. A cloud platform-based hospital financial data mining method, characterized in that: include: Collecting evaluation system data of the doctor to be evaluated, wherein the evaluation system data includes business indicators, economic indicators and patient evaluation indicators; Uploading the evaluation system data of the doctor to be evaluated to the cloud platform; On the cloud platform, a multi-source data collaborative analysis based on embedded semantics is performed on the business indicators, economic indicators and patient evaluation indicators in the evaluation system data of the doctor object to be evaluated to obtain a joint coding feature of the multi-source indicators of the doctor object to be evaluated; On the cloud platform, extracting a set of cluster centers of doctor object types from a database; On the cloud platform, dynamic semantic query processing is performed on the multi-source indicator joint coding features of the doctor object to be evaluated and the set of doctor object type cluster centers to obtain a query response semantic representation of the doctor object to be evaluated, including: calculating the fast query positioning factor of the multi-source indicator joint coding features of the doctor object to be evaluated relative to each doctor object type cluster center in the set of doctor object type cluster centers to obtain a set of fast query positioning factors of the doctor object to be evaluated; Based on the set of quick query positioning factors of the doctor object to be evaluated, a fine-grained matching search window is determined; based on the fine-grained matching search window, a set of multi-source indicator joint coding features of the doctor object to be evaluated and the doctor object type cluster center is fine-grained query response encoding is performed to obtain a query response semantic representation of the doctor object to be evaluated; On the cloud platform, doctor object classification management is performed based on the query response semantic representation of the doctor object to be evaluated to determine the type label of the doctor object to be evaluated.

2. The cloud platform-based hospital financial data mining method according to claim 1, characterized in that: On the cloud platform, a multi-source data collaborative analysis based on embedded semantics is performed on the business indicators, economic indicators and patient evaluation indicators in the evaluation system data of the doctor object to be evaluated to obtain the multi-source indicator joint coding features of the doctor object to be evaluated, including: Embedding and coding the business indicators, economic indicators and patient evaluation indicators in the evaluation system data of the doctor object to be evaluated to obtain a business indicator embedded coding vector, an economic indicator embedded coding vector and a patient evaluation indicator embedded coding vector; The business indicator embedded coding vector, the economic indicator embedded coding vector and the patient evaluation indicator embedded coding vector are input into a multi-source data collaborative analysis module based on a multi-layer perceptron model to obtain a multi-source indicator joint coding vector of the doctor object to be evaluated as the multi-source indicator joint coding feature of the doctor object to be evaluated.

3. The cloud platform-based hospital financial data mining method according to claim 2 is characterized in that: Calculating the fast query positioning factor of the multi-source indicator joint coding feature of the to-be-evaluated doctor object relative to each doctor object type cluster center in the set of doctor object type cluster centers to obtain a set of fast query positioning factors of the to-be-evaluated doctor object, including: The cross entropy of the multi-source indicator joint encoding vector of the doctor object to be evaluated relative to each doctor object type cluster center in the set of doctor object type cluster centers is calculated to obtain a set of fast query positioning factors of the doctor object to be evaluated.

4. The cloud platform-based hospital financial data mining method according to claim 3 is characterized in that: Based on the set of positioning factors of the to-be-evaluated doctor object, a fine-grained matching search window is determined, including: The doctor object type cluster center corresponding to the minimum value in the set of the quick query positioning factors of the doctor object to be evaluated is used as the positioning matching doctor object type cluster center; Based on the located matching doctor object type cluster center, the fine-grained matching search window is determined, wherein the vector of the center position of the fine-grained matching search window is the located matching doctor object type cluster center, and each doctor object type cluster center in the fine-grained matching search window is defined as a fine-grained query doctor object type cluster center to obtain a set of fine-grained query doctor object type cluster centers.

5. The cloud platform-based hospital financial data mining method according to claim 4 is characterized in that: Based on the fine-grained matching search window, fine-grained query response encoding is performed on the set of multi-source indicator joint coding features of the doctor object to be evaluated and the doctor object type cluster center to obtain the query response semantic representation of the doctor object to be evaluated, including: Performing a linear transformation on the multi-source indicator joint encoding vector of the doctor object to be evaluated to obtain a multi-source indicator joint query vector of the doctor object to be evaluated and a multi-source indicator joint value vector of the doctor object to be evaluated; Using each fine-grained query doctor object type cluster center in the set of fine-grained query doctor object type cluster centers as a key vector, fine-grained query encoding is performed on the multi-source indicator joint query vector of the doctor object to be evaluated, the multi-source indicator joint value vector of the doctor object to be evaluated and the key vector to obtain a semantic encoding vector of the doctor object query response to be evaluated as the semantic representation of the doctor object query response to be evaluated.

6. The cloud platform-based hospital financial data mining method according to claim 5, characterized in that: Using each fine-grained query doctor object type cluster center in the set of the fine-grained query doctor object type cluster centers as a key vector, fine-grained query encoding is performed on the multi-source indicator joint query vector of the doctor object to be evaluated, the multi-source indicator joint value vector of the doctor object to be evaluated, and the key vector to obtain a query response semantic encoding vector of the doctor object to be evaluated as the query response semantic representation of the doctor object to be evaluated, including: Taking each fine-grained query doctor object type cluster center in the set of fine-grained query doctor object type cluster centers as a key vector, the multi-source indicator joint query vector of the doctor object to be evaluated, the multi-source indicator joint value vector of the doctor object to be evaluated and the key vector are input into a fine-grained query encoding module based on a heterogeneous converter structure to obtain the query response semantic encoding vector of the doctor object to be evaluated.

7. The cloud platform-based hospital financial data mining method according to claim 6, characterized in that: On the cloud platform, classifying and managing doctor objects based on the query response semantic representation of the doctor object to be evaluated to determine the type label of the doctor object to be evaluated includes: The query response semantic encoding vector of the doctor object to be evaluated is input into a classifier-based doctor object classification management module to obtain the type label of the doctor object to be evaluated.

8. A hospital financial data mining system based on a cloud platform, characterized in that: include: An evaluation system data collection module is used to collect evaluation system data of the doctor object to be evaluated, wherein the evaluation system data includes business indicators, economic indicators and patient evaluation indicators; A data uploading module, used to upload the evaluation system data of the doctor object to be evaluated to the cloud platform; A collaborative analysis module is used to perform a multi-source data collaborative analysis based on embedded semantics on the business indicators, economic indicators and patient evaluation indicators in the evaluation system data of the doctor object to be evaluated on the cloud platform to obtain a joint coding feature of the multi-source indicators of the doctor object to be evaluated; A database extraction module, used to extract a set of doctor object type cluster centers from the database on the cloud platform; A dynamic semantic query processing module is used to perform dynamic semantic query processing on the cloud platform on the set of multi-source indicator joint coding features of the doctor object to be evaluated and the doctor object type clustering centers to obtain a query response semantic representation of the doctor object to be evaluated; wherein the dynamic semantic query processing module includes: a fast query positioning factor calculation unit, which is used to calculate the fast query positioning factor of each doctor object type clustering center in the set of doctor object type clustering centers of the multi-source indicator joint coding features of the doctor object to be evaluated to obtain a set of fast query positioning factors of the doctor object to be evaluated; a fine-grained matching search window determination unit, which is used to determine a fine-grained matching search window based on the set of fast query positioning factors of the doctor object to be evaluated; a fine-grained query response encoding unit, which is used to perform fine-grained query response encoding on the set of multi-source indicator joint coding features of the doctor object to be evaluated and the doctor object type clustering centers based on the fine-grained matching search window to obtain a query response semantic representation of the doctor object to be evaluated; A classification management module is used to perform classification management of doctor objects on the cloud platform based on the query response semantic representation of the doctor object to be evaluated to determine the type label of the doctor object to be evaluated.

9. The cloud platform-based hospital financial data mining system according to claim 8, characterized in that: The collaborative analysis module includes: An embedding coding unit, used for embedding coding the business indicators, economic indicators and patient evaluation indicators in the evaluation system data of the doctor object to be evaluated to obtain a business indicator embedding coding vector, an economic indicator embedding coding vector and a patient evaluation indicator embedding coding vector; The analysis unit is used to input the business indicator embedded coding vector, the economic indicator embedded coding vector and the patient evaluation indicator embedded coding vector into a multi-source data collaborative analysis module based on a multi-layer perceptron model to obtain a multi-source indicator joint coding vector of the doctor object to be evaluated as a multi-source indicator joint coding feature of the doctor object to be evaluated.

10. The cloud platform-based hospital financial data mining system according to claim 9, characterized in that: The fast query positioning factor calculation unit is used to: The cross entropy of the multi-source indicator joint encoding vector of the doctor object to be evaluated relative to each doctor object type cluster center in the set of doctor object type cluster centers is calculated to obtain a set of fast query positioning factors of the doctor object to be evaluated.

Images