Trust evaluation improvement method and platform for regional health data

Abstract

The invention discloses a regional health data-oriented trust evaluation improvement method and platform, and belongs to the technical field of data analysis. A health data set containing regional identification information is collected; extracting a regional feature parameter set of each record based on a preset rule; constructing a quality parameter set reflecting data semantic consistency and traceability of a generation process; using a multi-dimensional weighted mapping function to fuse the region and quality features, and constructing a region perception type trust factor model; a trust evaluation value is generated for each piece of data, and data with low credibility is marked; generating a regional data quality thermodynamic diagram based on a scoring result, and feeding back the regional data quality thermodynamic diagram to a governance platform for data quality intervention and collaborative decision support; according to the method, the precision, interpretability and regional adaptability of health data trust evaluation are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and particularly to an improved method and platform for trust evaluation of regional health data. Background Art

[0002] With the continuous development of regional health information systems and the increasing improvement of medical data collection technologies, various types of health data are increasingly widely used in fields such as medical diagnosis, public health monitoring, and health management. Regional health data usually includes sensitive information such as individual physical sign indicators, medical records, disease distributions, and health archives, and its accuracy and credibility directly affect the effectiveness of data applications and the scientific nature of decisions.

[0003] However, due to differences in data collection technologies, medical levels, information update mechanisms, and regulatory standards among regions, there are significant differences in the authenticity, integrity, and consistency of health data from different sources. Such differences easily lead to difficulties in evaluating data trustworthiness, bringing great challenges in the processes of data fusion, sharing, and mining, especially having a more significant impact in applications such as cross-regional health collaborative analysis and disease early warning.

[0004] Existing health data trust evaluation methods mostly analyze based on static rules or single features, and fail to fully consider the coupling relationship between regional attributes and dynamic data quality features, resulting in limited representativeness and adaptability of evaluation results. Therefore, there is an urgent need for an improved trust evaluation method that can integrate the characteristics of regional health data and dynamically adapt to multi-dimensional influencing factors to improve the accuracy and robustness of evaluation results, and further enhance the trustworthiness of health data in actual application scenarios. Summary of the Invention

[0005] The purpose of the present invention is to provide an improved method and platform for trust evaluation of regional health data to solve the deficiencies in the background art.

[0006] To achieve the above purpose, the present invention provides the following technical solutions: An improved method for trust evaluation of regional health data, including:

[0007] S100. Collect a health data set D containing regional identification information, where the data set D includes multiple health data records;

[0008] S200. Based on a preset regional classification rule, extract the regional feature parameter set R corresponding to each record in the health data set D;

[0009] S300. Based on the quality features of each health data record, construct a data quality parameter set Q, including but not limited to field context consistency score and record generation trajectory transparency score;

[0010] S400. Use the multi-dimensional weighted feature mapping function f to fuse R and Q into the joint vector feature group V, and construct a region-aware trust factor model F through training samples;

[0011] S500. Input the health data set D into the trained trust factor model F to generate the trust evaluation value T for each record, and mark the data with a trust evaluation value T lower than the threshold to form a marked data set ;

[0012] S600. Based on the confidence results of the marked data set further generate a regional health data quality heat map, and feedback the heat map for regional-level data quality governance or collaborative decision-making support.

[0013] Preferably, the method for obtaining the field context consistency score is as follows:

[0014] Extract the keyword field combinations in each health data record;

[0015] Construct a semantic rule library or context rule matrix to define reasonable field combination logic;

[0016] Match and verify the field combinations of each record, and count the items with consistent and inconsistent logics;

[0017] According to the proportion of consistent items, calculate the consistency score, and the expression is: ; represents the field context consistency score, ranging from [0, 1]; represents the number of field combinations that do not conform to the context semantic rules; represents the total number of field combinations included in the consistency verification.

[0018] Preferably, the method for obtaining the record generation trajectory transparency score is as follows:

[0019] Define the key metadata fields in the health data generation chain;

[0020] Check the existence, format integrity, and content validity of the above fields for each record;

[0021] Assign a basic weight to each field and calculate deductions according to the number of missing / failed items;

[0022] The scoring formula expression is: ; represents the record generation trajectory transparency score, ranging from [0, 1], is the importance weight of the i-th meta-field; is the missing / format error mark of the i-th field, and n is the total number of metadata fields.

[0023] Preferably, S400 includes:

[0024] S401. Perform density-aware normalization processing on the region feature parameter set R and the data quality parameter set Q respectively. The normalization processing dynamically adjusts the scaling ratio according to the distribution density of each dimension feature in the historical data;

[0025] S402. Construct a structure-preserving weighted mapping function f. The function f combines a predefined region type weight template and feature information entropy to generate a multi-dimensional mapping matrix for regulating the fusion structure of R and Q, retaining the original dimension structure of the features and embedding the region context label vector;

[0026] S403. Input the normalized R and Q into the function f to generate a joint feature vector group V;

[0027] S404. Perform supervised training on the data samples with artificial trust annotations, and dynamically adjust the learning weights of different feature dimensions by using a guided attention mechanism to train and obtain a region-aware trust factor model F.

[0028] Preferably, S500 includes:

[0029] S501. Group the constructed joint feature vector group V according to the region labels, and perform trust inference on each group respectively based on the region weight mapping mechanism in the model F;

[0030] S502. The model F automatically calls the region context-sensitive factor during the inference process to dynamically adjust the importance ranking of features;

[0031] S503. Generate an initial trust value for each health data record, and perform smoothing adjustment on the values with abnormal fluctuations;

[0032] S504. Output the smoothed trust evaluation value T, and bind T to the original data structure.

[0033] Preferably, S502 includes:

[0034] S5021. Before the model F performs inference, identify the region labels included in the input feature vector V, match them to a preset region context factor library, and extract the corresponding region influence factor matrix Rctx;

[0035] S5022. According to the feature priority template defined for each type of region in Rctx, perform a priority pre-sorting on the feature dimensions in V;

[0036] S5023. Combine the historical performance in the current input data to dynamically adjust the original sorting result to generate a region context-sensitive sorting sequence S;

[0037] S5024. Adjust the activation priority of the features in the attention layer or scoring layer of the model F according to the sorting sequence S to achieve dynamic allocation of feature weights.

[0038] Preferably, S600 includes:

[0039] S601, labeling data set Cluster statistics are performed according to administrative divisions and data types to construct a multidimensional data quality indicator matrix based on regions. The matrix includes the mean trust evaluation value, the proportion of abnormal data, and the fluctuation value of the trust evaluation value;

[0040] S602: Map the matrix into a two-dimensional heat map, color each area block in the map according to the comprehensive score index, and dynamically mark the area with data quality degradation trend;

[0041] S603. Push the heat map results to the health supervision platform, and automatically trigger quality feedback instructions based on preset intervention rules.

[0042] The present invention also provides a trust evaluation improvement platform for regional health data, including:

[0043] Health data collection module: collects a health data set D containing area identification information, wherein the data set D includes multiple health data records;

[0044] Regional feature extraction module: extracts the regional feature parameter set R corresponding to each record in the health data set D based on the preset regional classification rules;

[0045] Data quality modeling module: Based on the quality characteristics of each health data record, a data quality parameter set Q is constructed, including but not limited to the field context consistency score and the record generation trajectory transparency score;

[0046] Trust factor model construction module: using a multi-dimensional weighted feature mapping function f, the R and Q are fused into a joint vector feature group V, and a region-aware trust factor model F is constructed through training samples;

[0047] Trust scoring and marking module: Input the health data set D into the trained trust factor model F, generate a trust evaluation value T for each record, and mark the data with a trust evaluation value T below the threshold to form a marked data set ;

[0048] Regional quality visualization module: based on the labeled dataset The confidence results are used to further generate a regional health data quality heat map, and the heat map feedback is used for regional-level data quality governance or collaborative decision support.

[0049] In the above technical solution, the technical effects and advantages provided by the present invention are as follows:

[0050] 1. The present invention can effectively integrate regional attributes and data quality characteristics, construct a trust factor model with regional perception ability, and realize refined scoring of the credibility of multi-source heterogeneous health data. Compared with the traditional evaluation methods that rely on fixed rules or single indicators, the present invention introduces deep quality parameters such as field semantic consistency and data generation trajectory transparency, and dynamically adjusts the scoring weights in combination with the regional context, significantly improving the adaptability, accuracy, and robustness of the model in a cross-regional data environment.

[0051] 2. The present invention innovatively introduces a heat map feedback mechanism, structurally presents the trust scoring results and docks with the data governance platform, supports real-time identification and intervention suggestions for low-quality regions, thereby realizing the transformation closed-loop of health data quality from "scoring evaluation" to "visual governance" and "collaborative optimization", enhancing the engineering application value and auxiliary decision-making ability of the system, and being applicable to the data governance and evaluation tasks of the national-level regional health information platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings.

[0053] Figure 1 It is a flowchart of the method of the present invention.

[0054] Figure 2 It is a flowchart of the platform system module of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0056] Example 1. Please refer to Figure 1 As shown, the improved trust evaluation method for regional health data in this embodiment includes:

[0057] S100. Collect a health data set D containing regional identification information, where the data set D includes multiple health data records;

[0058] S200. Extract the regional feature parameter set R corresponding to each record in the health data set D based on a preset regional classification rule.

[0059] S300. Construct a data quality parameter set Q based on the quality characteristics of each health data record, including but not limited to field context consistency score and record generation trajectory transparency score.

[0060] S400. Use a multi-dimensional weighted feature mapping function f to fuse R and Q into a joint vector feature group V, and construct a regional perception-based trust factor model F through training samples.

[0061] S500. Input the health data set D into the trained trust factor model F to generate a trust evaluation value T for each record, and mark the data with an evaluation value lower than the threshold to form a marked data set. ;

[0062] S600. Based on the confidence results of the marked data set further generate a regional health data quality heat map, and feedback the heat map for regional-level data quality governance or collaborative decision-making support.

[0063] In the present invention, the collection of the health data set D adopts a data collaborative marking mechanism based on a distributed health information network, and the specific process is as follows:

[0064] First, multi-source collection devices (such as electronic medical record systems, wearable vital sign monitoring devices, vaccination registration systems, etc.) deployed in regional medical institutions collect raw health data in real time. When each piece of data is generated, the system automatically attaches the device number, collection timestamp, operator identity code, and GPS coordinates or institutional code of the collection location.

[0065] Subsequently, use a regional coding mapping module to perform a standardized conversion on the collection location. This module has pre-constructed a multi-level mapping table that matches the "GB / T 2260 - 2023 Administrative Division Codes" with the national medical institution registration database. This mapping table supports automatically converting GPS coordinates or institutional codes into three-level administrative division codes (province, city, county), and can be further classified into labels such as urban core, suburb, rural, etc. according to the hospital level and type.

[0066] Then, the system merges the above additional information with the health data record to generate a structured data entry, including the original health field and its corresponding regional identification metadata, to form a primary health data set. This structuring process uses a column-oriented database structure (such as Parquet format) to improve batch processing efficiency and subsequent indexability.

[0067] To ensure the time comparability of data within the same analysis period, the system introduces a "one-time time window synchronization mechanism" and sets the time span of a single collection batch not to exceed 24 hours. If the time difference of some records in a batch of data exceeds the set threshold, the system will automatically mark it as "time anomaly" and can perform elimination or isolation analysis according to the preset strategy.

[0068] Through the above implementation methods, the collected health data not only has complete geographical identification and traceability of the collection source, but also improves the comparability and structured quality during regional data fusion, providing high-quality basic data support for subsequent trust modeling.

[0069] To achieve the quantitative extraction of regional features, the system first constructs a set of multi-dimensional regional classification rule bases. This rule base combines the following information dimensions:

[0070] According to the national standard administrative division (such as province / city / district), a three-level coding system is formed to calibrate the geographical attribution of the data.

[0071] Based on the indicators such as the number of medical institutions, the number of beds per capita, and the number of doctors per capita announced by the National Health Commission in the region, interval grading is carried out, for example, divided into "high-resource area", "medium-resource area", and "weak medical resource area".

[0072] Using the medical institution registration information, the institutions corresponding to the data collection points are classified into "Grade III Class A", "secondary hospital", "community health service center", etc. and mapped to categorical variables. [[ID=1,7]]

[0073] The access level of the local health information system is introduced, such as whether it supports electronic medical record sharing, remote diagnosis and treatment, automatic reporting, etc., and is calibrated as "high", "medium", and "low" levels.

[0074] By analyzing the historical upload logs, the average data synchronization frequency in this region is quantified (such as once a day, once a week, irregularly, etc.) and uniformly coded.

[0075] Based on the rule system, the system performs the following processing flow on each record in the health data set D:

[0076] Read the "institution code" or "GPS coordinates" in each record, and use the pre-constructed administrative division mapping table to identify its specific region.

[0077] Take the identified administrative code as the key value, query the feature information of each dimension of the corresponding region in the rule base, and convert it into a standardized structured parameter vector. For example:

[0078] Administrative level: municipal level → 2; medical resource level: high resource → H; medical institution level: secondary hospital → 2; information system capacity: medium → 1; update frequency: weekly → 2.

[0079] For the missing items in the rule base, the system uses interpolation in the neighboring area or supplements them with institutional registration metadata; if there are still outliers (such as positioning errors or unknown institutions), they are automatically marked as "areas to be verified" and incorporated into the low-confidence processing logic.

[0080] Combine the multi-dimensional feature values extracted above into a vector as the regional feature parameter set R of this record and bind it to the metadata structure of this record.

[0081] The purpose of this step is to achieve a refined characterization of the potential influencing factors of the regional environment on the quality of health data, enabling the subsequent trust evaluation model to adjust the weight strategy according to local conditions. For example, the same missing rate may have different explanatory contexts in high-resource cities and grass-roots communities.

[0082] Through the preset rule extraction method, this step avoids the problems of relying on model training, being opaque or unexplainable, improves the controllability, credibility and cross-regional generalization ability of the model, and at the same time has good updateability and engineering deployment adaptability.

[0083] This step aims to quantify the quality level of each health data record and construct a multi-dimensional and structured quality parameter set Q. The composition of Q not only includes common general quality indicators such as integrity and conflict, but also particularly introduces two custom indicators that reflect the deep semantics and the credibility of the data generation process: the field context consistency score ( ), and the record generation trajectory transparency score ( ).

[0084] The system performs standardized parsing on each health data record and extracts structured fields including: basic demographic information, diagnostic information, treatment plans, medication records, timestamps, collection device numbers, etc. If there are missing fields or coding errors, they are marked as the "preprocessing exception" status in advance.

[0085] Missing rate ( ): Calculate the proportion of missing items in the field; Conflict rate ( ): Check whether the logical relationship of the field is contradictory (such as female + prostate disease); Redundancy rate ( ): Analyze the proportion of duplicate records in the sample set.

[0086] The method for obtaining the field context consistency score ( ) is as follows:

[0087] Extract the combination of key fields in each health data record (such as: gender, age, diagnostic code, treatment method, medication record);

[0088] Construct a semantic rule library or a context rule matrix, and define reasonable field combination logics (e.g., when Gender = male, the diagnosis of "cervical cancer" cannot appear);

[0089] Match and verify the field combinations of each record, and count the items with consistent and inconsistent logics;

[0090] Calculate the consistency score based on the proportion of consistent items. The expression is: ; represents the field context consistency score, ranging from [0, 1], and the higher the score, the more consistent; represents the number of field combinations that do not conform to the context semantic rules; represents the total number of field combinations included in the consistency verification. For example, in a record: the number of checked combined fields: 6 groups; there is 1 group inconsistent (e.g., "Age = 2 years old" + "Diagnosis = Alzheimer's disease"); then the score is Q4 == 0.833.

[0091] The method for obtaining the record generation traceability transparency score ( ) is as follows:

[0092] Define the key metadata fields in the health data generation chain, such as: device number (DeviceID), operator identifier (OperatorID), original timestamp (T collection), storage time (T storage), edited record / log entry (LogID), etc.;

[0093] Check the existence, format integrity, and content validity of the above fields for each record;

[0094] Assign a basic weight to each field, and calculate deductions according to the number of missing / failed fields.

[0095] The scoring formula expression is: ; represents the record generation traceability transparency score, ranging from [0, 1], and the higher the score, the stronger the traceability; is the importance weight of the i-th meta-field (e.g., timestamp > log number); is the missing / format error flag of the i-th field (1 indicates abnormality, 0 indicates normal); n is the total number of metadata fields, such as 5 or 6 key fields.

[0096] Combine the above scoring indicators ( ) into a structured parameter set Q;

[0097] Each health data record corresponds to a Q, which is used as the input for the subsequent trust scoring model.

[0098] By expanding the assessment of data quality from "field-level integrity" to "cross-field semantic consistency" and "generative process traceability", this step significantly improves the recognition granularity of data authenticity and credibility. Compared with the traditional method of only scoring based on the missing rate / conflict rate, the Q constructed by this method is more discriminative and context-aware, especially suitable for sensitive or high-risk health data environments.

[0099] In the present invention, to effectively fuse the regional feature parameter set R and the data quality parameter set Q and construct a trust factor model F with regional adaptability, the system performs the following steps:

[0100] First, normalize the data of each dimension of R and Q. Different from the traditional linear scaling method, the system introduces a density-aware normalization strategy. Based on historical data samples, this strategy performs kernel density estimation (such as using a Gaussian kernel function) for each feature dimension to calculate the density of the feature in the data distribution. For features that are sparse in distribution but have an important impact, the system automatically increases their scaling ratio to prevent them from being diluted in subsequent modeling.

[0101] Second, construct a structure-preserving weighted mapping function f. This function not only determines the importance weights of each feature according to information entropy but also incorporates a set of preset regional type templates (such as urban core areas, suburbs, remote areas) to adjust the embedding strength of region-related features. The output of the mapping function is a vector group V that is consistent with the original R+Q structure, and at the same time, an "area context vector" is attached to represent the structural position of this data in the regional background.

[0102] Then, the system inputs the normalized R and Q into the function f to obtain the joint feature group V. Different from common simple concatenation, the fusion method designed in the present invention preserves the semantic structure of the original features, avoids dimensionality reduction information loss, and at the same time establishes a non-linear interaction expression ability between the regional dimension and the quality dimension within V, facilitating subsequent modeling to learn these complex relationships.

[0103] Finally, based on the health data records with manually annotated trust levels, the system performs supervised training on the joint feature group V, adopting a training strategy with a guided attention mechanism. During the training process, the attention mechanism will automatically assign feature weights and focus on the dimensions that have a significant impact on the trust score, such as the combination of "low update frequency + high missing rate", or the situation of "remote area + weak information system access", etc. This mechanism can dynamically adjust the sensitivity of the model to different regional combinations, and the finally obtained trust factor model F has good regional perception ability.

[0104] Through the above embodiments, the present invention not only realizes high-dimensional feature fusion, but also significantly enhances the generalization and interpretability of the model in a cross-regional health data environment, effectively solving the problem of poor generalization ability of traditional scoring models in the face of regional heterogeneity.

[0105] In the present invention, the system uses the trained region-aware trust factor model F to generate a trust evaluation value T for each record in the health data set D, which specifically includes the following operation processes:

[0106] First, the system groups the joint feature vector group V according to the region labels bound to it. Each region group correspondingly calls the region weight mapping module in the model F, which contains multiple sets of region context-sensitive factor sets for adjusting the scoring logic.

[0107] Before the model F performs inference, the system automatically identifies the region label of each input vector (such as "county-level city / secondary hospital / low information access area") and matches it with the predefined region context factor library to extract the corresponding region impact factor matrix Rctx. This matrix contains the characteristic weight tendencies shown by the region in historical scoring. For example, in areas with weak medical resources, the impact of the missing rate and data update time on trust scoring is significantly higher than the impact of device ID information.

[0108] Next, the system preliminarily sorts the feature dimensions in the current data according to the preset feature priority template in Rctx. For example, if a region template indicates that "missing rate > conflict rate > redundancy rate > collection time accuracy", then these dimensions in the input vector are arranged in this order.

[0109] To improve the adaptive ability, the system further combines the current data with the statistical characteristics of the historical model performance to perform a secondary optimization on the preliminary sorting result. The optimization strategy considers the stability and sensitivity of each feature dimension in the scoring history under this region to form the final sorting sequence S.

[0110] Finally, the system applies the sorting sequence S to the attention mechanism control logic of the model F to set higher activation weights for feature dimensions with high importance. During the inference process, the model outputs a preliminary trust score accordingly. .

[0111] To prevent extreme values from interfering or fluctuating abnormally, the system also calls the "scoring robustness calibration mechanism", which compares the scoring value with the historical fluctuation range. If the score of a certain record deviates from the historical mean by more than the set threshold, an adjustment based on exponential weighted moving average (EWMA) is performed to output the final trust evaluation value T. And data with a trust evaluation value T lower than the threshold is marked for credibility to form a marked data set. .

[0112] This method not only realizes the dynamic adaptation of the scoring process to regional differences, but also introduces a scoring calibration mechanism to enhance the stability and interpretability of the model, significantly superior to the existing fixed-weight scoring method or unsupervised modeling method.

[0113] In the present invention, the system is based on the labeled data set The generated trust evaluation value T is further used to carry out visual modeling and governance feedback on the quality of regional health data, specifically including the following steps:

[0114] First of all, the system will The health data records in are clustered at multiple levels according to administrative regions (such as provinces, cities, and counties) and data types (such as outpatient records, chronic disease follow-up, vaccination, etc.). For each type of combination, the system extracts the following key statistical indicators: average trust score, proportion of low-trust records, standard deviation of trust scores, etc., to form a two-dimensional quality index matrix of region-type.

[0115] Subsequently, the system calls the regional quality visualization module to map the index matrix into a two-dimensional geographical heat map. The map takes the regional boundary as the basic unit, combines the comprehensive score intervals of each unit, and uses a color gradient method for hierarchical coloring (for example, green represents good data quality, and red represents serious problems). At the same time, the system marks the regions where the scoring volatility exceeds the threshold as data anomaly trend warning regions, facilitating decision-makers to quickly identify risk hotspots.

[0116] Furthermore, the system pushes the generated heat map and corresponding data to the connected health information supervision platform through the built-in regional governance interface. The platform automatically analyzes according to the rule base and triggers corresponding intervention instructions, such as: sending a data source verification notice to the region with a trust level lower than the set threshold, or suggesting governance measures such as increasing the data collection frequency and field verification standards.

[0117] Finally, the system supports the comparison and analysis function of historical heat map versions. Supervisors can view the quality change trend map of a certain region at different time nodes to assist in evaluating the actual effects of existing governance means and form a dynamic data quality closed-loop management system.

[0118] Through the above implementation manner, the present invention not only realizes the visualization and spatial presentation of trust scoring, but also constructs a feedback mechanism linked with the governance platform, with strong practical application value and integrated innovation ability.

[0119] Example 2, please refer to Figure 2 As shown, the trust evaluation improvement platform for regional health data described in this embodiment includes:

[0120] Health data collection module: Collect a health data set D containing regional identification information, where the data set D includes multiple health data records;

[0121] Regional feature extraction module: Based on a preset regional classification rule, extract a set of regional feature parameters R corresponding to each record in the health data set D;

[0122] Data quality modeling module: Based on the quality characteristics of each health data record, construct a set of data quality parameters Q, including but not limited to field context consistency score and record generation trajectory transparency score;

[0123] Trust factor model construction module: Use a multi-dimensional weighted feature mapping function f to fuse the R and Q into a joint vector feature group V, and construct a region-aware trust factor model F through training samples;

[0124] Trust score and marking module: Input the health data set D into the trained trust factor model F to generate a trust evaluation value T for each record, and mark the data with a trust evaluation value T lower than the threshold to form a marked data set ;

[0125] Regional quality visualization module: Based on the confidence results of the marked data set further generate a heat map of regional health data quality, and feedback the heat map for regional-level data quality governance or collaborative decision-making support.

[0126] As described above, this is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed by this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application.

Claims

1. An improved method for trust evaluation of regional health data, characterized in that: Including: S100. Collect a health dataset D containing area identification information, where the dataset D includes multiple health data records; S200. Based on a preset area classification rule, extract the area feature parameter set R corresponding to each record in the health dataset D; S300. Based on the quality characteristics of each health data record, construct a data quality parameter set Q, including but not limited to field context consistency score and record generation track transparency score; S400. Use a multi-dimensional weighted feature mapping function f to fuse R and Q into a joint vector feature group V, and construct an area-aware trust factor model F through training samples; S500. Input the health data set D into the trained trust factor model F to generate a trust evaluation value T for each record, and perform a credibility marking on the data with a trust evaluation value T lower than the threshold to form a marked data set ; S600. Further generate a regional health data quality heat map based on the confidence results of the labeled data set, and feedback the heat map for regional-level data quality governance or collaborative decision-making support. ​ 2. The trust evaluation improvement method for regional health data according to claim 1, wherein: The method for obtaining the field context consistency score is as follows: Extract the keyword field combination in each health data record; Construct a semantic rule library or context rule matrix to define reasonable field combination logic; Match and verify the field combination of each record, and count the items with consistent and inconsistent logic; Calculate the consistency score based on the proportion of consistent items, and the expression is: ; represents the field context consistency score, ranging from [0, 1]; represents the number of field combinations that do not conform to the context semantic rules; represents the total number of field combinations included in the consistency verification.

3. The trust evaluation improvement method for regional health data according to claim 2, wherein: The method for obtaining the record generation track transparency score is as follows: Define the key metadata fields in the health data generation chain; Check the existence, format integrity, and content validity of the above fields for each record; Assign a basic weight to each field and calculate deductions according to the number of missing / failed ones; The scoring formula expression is as follows: ; represents the transparency score of the record generation trajectory, with a range of [0, 1], is the importance weight of the i-th meta-field; is the missing / format error flag of the i-th field, and n is the total number of metadata fields.

4. The trust evaluation improvement method for regional health data according to claim 1, wherein: S400 includes: S401. Perform density-aware normalization processing on the area feature parameter set R and the data quality parameter set Q respectively. The normalization processing dynamically adjusts the scaling ratio according to the distribution density of each dimension feature in the historical data; S402. Construct a structure-preserving weighted mapping function f. The function f combines a predefined area type weight template and feature information entropy to generate a multi-dimensional mapping matrix for regulating the fusion structure of R and Q, retains the original dimension structure of the features, and embeds the area context label vector; S403. Input the normalized R and Q into the function f to generate a joint feature vector group V; S404. Perform supervised training on the data samples with artificial trust annotations, and use a guided attention mechanism to dynamically adjust the learning weights of different feature dimensions to train and obtain an area-aware trust factor model F.

5. The improved method for trust evaluation of regional health data according to claim 1, characterized in that: S500 includes: S501. Group the constructed joint feature vector group V according to the area label, and perform trust reasoning on each group respectively based on the area weight mapping mechanism in the model F; S502. During the reasoning process of the model F, automatically call the area context-sensitive factor to dynamically adjust the importance ranking of features; S503. Generate an initial trust value for each health data record After that, for the value with abnormal fluctuations, perform smoothing adjustment; S504. Output the smoothed trust evaluation value T and bind T to the original data structure.

6. The trust evaluation improvement method for regional health data according to claim 5, characterized in that: S502 includes: S5021. Before the model F performs reasoning, identify the area label included in the input feature vector V, match it to a preset area context factor library, and extract the corresponding area influence factor matrix Rctx; S5022. According to the feature priority template defined for each type of area in Rctx, perform a priority pre-sorting on the feature dimensions in V; S5023. Combine the historical performance in the current input data to dynamically adjust the original sorting result to generate an area context-sensitive sorting sequence S; S5024. Adjust the activation priority of features in the attention layer or scoring layer of model F according to the sorting sequence S to achieve dynamic allocation of feature weights.

7. The trust evaluation improvement method for regional health data according to claim 1, wherein: S600 includes: S601. Perform clustering statistics on the labeled data set according to administrative divisions and data types, and construct a multi-dimensional data quality index matrix with regions as units. The matrix includes the mean value of trust evaluation values, the proportion of abnormal data, and the fluctuation value of trust evaluation values. S602. Map the matrix into a two-dimensional heat map, where each regional block in the map is colored by layer according to the comprehensive scoring index, and the area with the trend of data quality deterioration is dynamically marked. S603. Push the heat map result to the health supervision platform and automatically trigger a quality feedback instruction according to the preset intervention rule.

8. A trust evaluation improvement platform for regional health data, which is used to implement the trust evaluation improvement method for regional health data according to any one of claims 1-7, and is characterized in that: It includes: Health data collection module: Collect a health data set D containing regional identification information, and the data set D includes multiple health data records. Regional feature extraction module: Based on the preset regional classification rules, extract the regional feature parameter set R corresponding to each record in the health data set D. Data quality modeling module: Based on the quality characteristics of each health data record, construct a data quality parameter set Q, including but not limited to the field context consistency score and the record generation track transparency score. Trust factor model construction module: Use the multi-dimensional weighted feature mapping function f to fuse R and Q into a joint vector feature group V, and construct a region-aware trust factor model F through training samples. Trust Score and Labeling Module: Input the health dataset D into the trained trust factor model F to generate a trust score value T for each record, and perform credibility labeling on the data with a trust score value T lower than the threshold to form a labeled dataset ; Region Quality Visualization Module: Based on the confidence results of the labeled dataset to further generate a heat map of regional health data quality and feedback the heat map for regional-level data quality governance or collaborative decision-making support.

Images