An e-government data full-link quality monitoring method and system

Abstract

The application discloses an e-government data full-link quality supervision method and system, and the method comprises e-government data aggregation, feature space construction, government data quality representation, business category weight construction, quality supervision loss construction, data quality evaluation model design and e-government data quality supervision. The application belongs to the field of data processing, and specifically refers to an e-government data full-link quality supervision method and system. In the index structure of the link flow conversion entropy embedded business situation awareness coefficient, the following is realized: the imbalance compensation strength of the e-government data distribution is dynamically amplified with the process complexity; the weight modulation gradient component and the smooth constraint gradient correction component are divided, the former half encodes the dynamic modulation caused by the business distribution and the abnormal concealment degree, and the gradient gain compensation is performed on the minority business and the scarce government data; the latter half corresponds to the gradient correction of the smooth constraint, and then the e-government data full-link quality supervision effect is improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method and system for end-to-end quality supervision of e-government data. Background Technology

[0002] The full-chain quality supervision method for e-government data is a management mechanism that automates the verification, anomaly identification, and risk warning of multi-source heterogeneous data throughout the entire e-government system chain, from front-end data entry, cross-departmental sharing and exchange, business approval workflow, to archiving and storage. However, general e-government data full-chain quality supervision methods often neglect the complexity of government business processes and are insensitive to the risk differences in the quality status of government data, leading to poor reliability of quality supervision. Furthermore, these methods often suffer from uneven impacts on parameter updates from different government business processes and data with different quality statuses, making it difficult to adapt to the complex judgment requirements of government scenarios, thus resulting in poor quality supervision effectiveness. Summary of the Invention

[0003] To address the above issues and overcome the shortcomings of existing technologies, this invention provides a method and system for end-to-end quality supervision of e-government data. Addressing the problems of general e-government data end-to-end quality supervision methods neglecting the process complexity of e-government business itself and being insensitive to the risk differences in e-government data quality status, leading to poor reliability of quality supervision, this solution introduces link flow entropy embedded in the exponential structure of the business situation awareness coefficient. This enables the compensation intensity for e-government data distribution imbalance to dynamically amplify with process complexity, and the focusing intensity for difficult-to-identify anomalies to adaptively increase with process complexity. Based on the main loss, a probabilistic smoothing constraint term bound to the true category is introduced to constitute the quality supervision loss: the main loss term is dynamically weighted according to the business situation awareness coefficient on the true category, preserving the focus on imbalanced business and difficult-to-identify anomalies; the smoothing constraint term improves the end-to-end... The generalization of quality judgment reduces the risk of misjudgment due to overconfidence, thereby improving the reliability of quality supervision across the entire e-government data chain. Addressing the issue that general e-government data quality supervision methods suffer from uneven impacts on parameter updates due to different government business operations and varying quality states, making them difficult to adapt to the complex judgment requirements of government scenarios and resulting in poor quality supervision effectiveness, this solution divides the modified gradient structure into a weighted modulation gradient component and a smoothing constraint gradient correction component. The first part encodes the dynamic modulation caused by business distribution and anomaly concealment, providing gradient gain compensation for niche businesses and scarce government data. The second part corresponds to the smoothing constraint gradient correction, smoothing and reducing gradient oscillations caused by government labeling noise and blurred boundary samples, while ensuring a balanced gradient contribution, thus improving the overall quality supervision effect of e-government data.

[0004] The technical solution adopted by this invention is as follows: This invention provides a method for end-to-end quality supervision of e-government data, which includes the following steps: Step S1: E-government data aggregation; Step S2: Feature space construction; Step S3: Quality characterization of government data; Step S4: Constructing business category weights; Step S5: Quality Supervision Loss Construction; Step S6: Design of data quality assessment model; Step S7: E-government data quality supervision.

[0005] Furthermore, in step S1, the e-government data aggregation involves collecting historical e-government data; uniformly preprocessing the raw data; and labeling the data, including four categories of quality truth labels: compliant and normal, minor defects, serious anomalies, and link flow violations, to obtain the original e-government dataset.

[0006] Furthermore, in step S2, the feature space construction involves decomposing the original government data set into three categories: field compliance features, business association features, and link time sequence features. At the same time, the effective data volume of each government business category is counted, and the original data is transformed into a feature space. Finally, these are spliced ​​together to form a unified government feature vector.

[0007] Furthermore, in step S3, the government data quality characterization involves building a characterization network, inputting government feature vectors, and outputting the probability that a single piece of government data is compliant and normal, slightly flawed, seriously abnormal, or has a violation in the link flow.

[0008] Furthermore, in step S4, the construction of business category weights integrates the effective data distribution of business categories with the confidence level of single data quality prediction, and introduces link flow entropy to construct a business situation awareness coefficient, quantifies the process source of anomaly concealment, and at the same time compensates for the weight of niche business data, suppresses the gradient contribution of simple compliant data, and strengthens the attention to difficult-to-identify abnormal data.

[0009] Furthermore, in step S5, the quality supervision loss is constructed by introducing probabilistic smoothing constraints on the basis of business category weights. This retains the ability to focus on unbalanced businesses and difficult-to-identify anomalies, while also constraining the reasonable distribution of model confidence, thereby improving the generalization of quality judgment across the entire chain and ultimately obtaining the quality supervision loss.

[0010] Furthermore, in step S6, the data quality assessment model design adopts a five-layer cascaded end-to-end model architecture. The overall hierarchy from top to bottom is as follows: e-government data aggregation layer, feature space construction layer, government data quality representation layer, business category weight construction layer, and gradient modulation adaptive training layer. Each layer is connected step by step, and parameters and features flow unidirectionally and progressively to construct the data quality assessment model. The gradient modulation adaptive training layer is used to globally differentiate the quality supervision loss, generating a corrected gradient that integrates business situational awareness, the degree of anomaly data concealment, and smoothing constraints; this drives the neural network to backpropagate and update parameters.

[0011] Furthermore, in step S7, the e-government data quality supervision is based on a trained data quality assessment model. The e-government data generated by the entire e-government link nodes is accessed in real time, preprocessed, and then input into the data quality assessment model. The model outputs the quality status category of each piece of e-government data. If the output quality status category is a serious anomaly or a link flow violation, an early warning is issued to the management personnel.

[0012] The present invention provides an e-government data full-link quality supervision system, including an e-government data aggregation module, a feature space construction module, a government data quality characterization module, a business category weight construction module, a quality supervision loss construction module, a data quality assessment model design module, and an e-government data quality supervision module; The e-government data aggregation module collects historical e-government data, which is then preprocessed to produce the original e-government dataset. The feature space construction module decomposes the original government data set and transforms it into a feature space to construct government feature vectors. The government data quality characterization module constructs a characterization network with a deep fully connected feedforward neural network structure, inputs a standardized government feature vector, and outputs the prediction confidence level of the quality status of the government data. The business category weight construction module integrates the number of effective data for each business category with the confidence level of single data quality prediction to construct a business situation awareness coefficient. The quality supervision loss construction module constructs quality supervision loss based on the prediction confidence level using the business situation awareness coefficient; The data quality assessment model design module constructs a data quality assessment model that includes an e-government data aggregation layer, a feature space construction layer, a government data quality representation layer, a business category weight construction layer, and a gradient modulation adaptive training layer. The e-government data quality supervision module performs quality supervision on real-time e-government data based on the trained data quality assessment model.

[0013] The beneficial effects achieved by the present invention using the above solution are as follows: (1) To address the problem that general e-government data full-link quality supervision methods neglect the process complexity of government business itself and are insensitive to the risk differences in the quality status of government data, thus leading to poor reliability of quality supervision, this solution introduces link flow entropy embedded in the exponential structure of the business situation awareness coefficient: to realize the dynamic amplification of the compensation intensity for the imbalance of e-government data distribution with the process complexity, and the adaptive enhancement of the focus intensity of difficult-to-identify anomalies with the process complexity; on the basis of the main loss, a probability smoothing constraint term bound to the real category is introduced to constitute the quality supervision loss: the main loss term is dynamically weighted on the real category according to the business situation awareness coefficient, retaining the focus on imbalanced business and difficult-to-identify anomalies; the smoothing constraint term improves the generalization of the full-link quality judgment and reduces the risk of misjudgment caused by overconfidence; thereby improving the reliability of e-government data full-link quality supervision.

[0014] (2) In view of the problem that the general e-government data full-link quality supervision method has an uneven impact on parameter updates due to different government business and different quality status data, which makes it difficult to adapt to the complex judgment requirements of government scenarios and thus leads to poor quality supervision effect, this solution divides the gradient structure into weighted modulation gradient components and smoothing constraint gradient correction components. The first half encodes the dynamic modulation caused by business distribution and anomaly concealment, and performs gradient gain compensation for niche business and scarce government data; the second half corresponds to the smoothing constraint gradient correction, which smooths and reduces the gradient oscillation caused by government label noise and blurred boundary samples, while ensuring the balance of gradient contribution, thereby improving the quality supervision effect of e-government data full-link. Attached Figure Description

[0015] Figure 1 A flowchart illustrating a method for end-to-end quality supervision of e-government data provided by this invention; Figure 2 This is a schematic diagram of an e-government data end-to-end quality supervision system provided by the present invention.

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0017] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0018] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the system or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0019] Example 1, see Figure 1 This invention provides a method for end-to-end quality supervision of e-government data, which includes the following steps: Step S1: E-government data aggregation, collecting historical e-government data, and preprocessing the raw e-government dataset; Step S2: Feature space construction. The original government data set is decomposed and transformed into a feature space to construct government feature vectors. Step S3: Government data quality characterization. Construct a characterization network with a deep fully connected feedforward neural network structure, input standardized government feature vectors, and output the prediction confidence level of the quality status of government data. Step S4: Construct business category weights, integrate the number of effective data for each business category with the confidence level of single data quality prediction, and construct a business situation awareness coefficient; Step S5: Constructing quality supervision loss, constructing quality supervision loss based on the prediction confidence coefficient of the business situation awareness coefficient; Step S6: Design the data quality assessment model and construct a data quality assessment model that includes an e-government data aggregation layer, a feature space construction layer, a government data quality representation layer, a business category weight construction layer, and a gradient modulation adaptive training layer; Step S7: E-government data quality supervision, which involves supervising the quality of real-time e-government data based on the trained data quality assessment model.

[0020] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, e-government data aggregation involves collecting historical e-government data, including front-end form data from departments, structured database table data from various commissions and bureaus, cross-departmental shared exchange interface message data, ETL flow scheduling log data, and government business approval time sequence process log data. The raw data undergoes unified preprocessing, including encoding format normalization, null value marking and filling, special garbled character cleaning, time field standardization, and heterogeneous field dimension alignment. Data is also labeled, including four types of quality truth labels: compliant and normal, minor defects, serious anomalies, and link flow violations, to obtain the original e-government dataset.

[0021] Example 3, see Figure 1This embodiment is based on the above embodiment. In step S2, the feature space construction involves decomposing the original government data set into three categories: field compliance features, business association features, and link time-series features. At the same time, the effective data volume of each government business category is counted to depict the natural category distribution imbalance and transform the original data into a feature space. The structured fields are type-normalized and numerically encoded. Text fragments in interface messages and logs are represented by word segmentation and embedding. The time-series attributes of status jumps and time intervals are extracted from the approval process logs. Finally, a unified government feature vector is formed by splicing them together. Represented as: ; in, It is the standardized feature vector of the i-th piece of government data; It is a feature mapping transformation function that maps structured fields, text information, and process timing to the same high-dimensional feature space. This is the i-th piece of original government data; This represents the number of valid data entries for category c of government affairs, where c is the tag category index. It is the data subset corresponding to the c-th type of government affairs; It is a statistical analysis of the effective data base of the set.

[0022] Example 4, see Figure 1 This embodiment is based on the above embodiment. In step S3, the government data quality characterization is to build a characterization network, input the government feature vector, and output the probability that a single piece of government data belongs to compliance and normal, minor defects, serious anomalies, or link flow violations; thereby realizing the probability discrimination of explicit format errors, implicit logical conflicts, and time sequence flow anomalies. Represented as: ; in, It represents the predictive confidence level that the i-th piece of government data output by the network belongs to the c-th quality status. It represents the network; It is the set of learnable parameters for a neural network; The representation network adopts a deep fully connected feedforward neural network structure, taking the unified government affairs feature vector output in step S2 as input. Through multi-layer nonlinear transformation, it achieves deep semantic fusion and quality discrimination of explicit format errors, implicit logical conflicts, and time-series flow anomalies. The main body of the network consists of multiple fully connected layers connected in series. Each layer is followed by batch normalization and ReLU activation functions to enhance the model's expressive power and alleviate gradient vanishing. At the end of the network, a normalized output layer is set to output the probability that the current government affairs data belongs to one of four quality states: compliant and normal, slightly flawed, seriously abnormal, or link flow violation.

[0023] Example 5, see Figure 1This embodiment is based on the above embodiment. In step S4, the construction of business category weights is to integrate the effective data distribution of business categories with the confidence of single data quality prediction, and introduce the link flow entropy to construct the business situation awareness coefficient, quantify the process source of anomaly concealment, and at the same time compensate for the weight of niche business data, suppress the gradient contribution of simple compliant data, and strengthen the attention to difficult-to-identify abnormal data. Represented as: ; in, It is the business situation awareness coefficient of the c-th quality status of the i-th piece of government data; It is the distribution attenuation control coefficient of government business data, which controls the intensity of compensation for category imbalance, and takes the value [0,1). This refers to the number of valid data entries for Category C government services. It is the government affairs abnormal data control coefficient, which strengthens the weight of difficult-to-identify abnormalities, and takes the value [1,5]. It is a basic weight based on the imbalance in the distribution of valid business data; It is a confidence index modulation term that enables the category weight to change dynamically according to the degree of concealment of data anomalies; It is a weighting item for abnormal data; It is the link flow entropy, which is obtained by statistically analyzing the state transition behavior of government business in the historical full-link approval logs: extract the historical full-link approval logs of the c-th type of government business, obtain the state sequence experienced by each data, construct a state transition matrix, count the transition frequency from any state to the next state and normalize it into a transition probability, calculate the information entropy of the transition distribution for each state, and then take the average of all states to obtain the link flow entropy of the corresponding business.

[0024] Example 6, see Figure 1 This embodiment is based on the above embodiment. In step S5, the quality supervision loss is constructed by introducing a probability smoothing constraint on the basis of the business category weight. This not only retains the ability to focus on unbalanced businesses and difficult-to-identify anomalies, but also constrains the reasonable distribution of model confidence, improves the generalization of the quality judgment of the whole link, and finally obtains the quality supervision loss. Represented as: ;in, It is the quality supervision loss value of the i-th piece of government data; This is a function indicating the quality status of government affairs. If the i-th piece of government data belongs to the c-th category, then... ,on the contrary ; The main loss is determined by the quality of government data; It is a smoothing constraint on the quality of government affairs, which smooths out training and suppresses overconfidence and misjudgment.

[0025] By performing the above operations, this solution addresses the problem that general e-government data end-to-end quality supervision methods neglect the process complexity of government business itself and are insensitive to the risk differences in the quality status of government data, thus leading to poor reliability of quality supervision. This solution introduces link flow entropy embedded in the exponential structure of the business situation awareness coefficient: This enables the compensation intensity for e-government data distribution imbalance to dynamically amplify with process complexity, and the focusing intensity for difficult-to-identify anomalies to adaptively increase with process complexity. Based on the main loss, a probability smoothing constraint term bound to the true category is introduced to constitute the quality supervision loss: the main loss term is dynamically weighted according to the business situation awareness coefficient on the true category, preserving the focus on imbalanced business and difficult-to-identify anomalies; the smoothing constraint term improves the generalization of end-to-end quality judgment and reduces the risk of misjudgment caused by overconfidence; thereby improving the reliability of e-government data end-to-end quality supervision.

[0026] Example 7, see Figure 1 This embodiment is based on the above embodiment. In step S6, the data quality assessment model design adopts a five-layer cascaded end-to-end model architecture. The overall hierarchy from top to bottom is: e-government data aggregation layer, feature space construction layer, government data quality representation layer, business category weight construction layer, and gradient modulation adaptive training layer. Each layer is connected step by step, and parameters and features flow unidirectionally and progressively to construct the data quality assessment model, forming a complete training closed loop of data preprocessing - feature construction - quality representation - weight modulation - loss optimization - model convergence. The functions of each layer are unified and aligned with the overall logic of the scheme. The e-government data aggregation layer, corresponding to step S1, is responsible for collecting multi-source heterogeneous historical data from all nodes in the e-government chain, completing preprocessing such as data cleaning, format normalization, and field alignment, and outputting standardized raw e-government datasets. The feature space construction layer, corresponding to step S2, is responsible for decomposing multi-dimensional government features, characterizing the imbalance of data category distribution, completing the transformation of raw data into a high-dimensional feature space, and outputting standardized feature data to provide input for the data quality assessment model. The government data quality representation layer, corresponding to step S3, relies on the representation network to receive high-dimensional feature data and output multi-category quality state prediction probabilities to achieve preliminary intelligent judgment of data quality. The business category weight construction layer, corresponding to step S4, generates a business situation awareness coefficient based on data distribution and prediction confidence, compensates for business distribution imbalance, focuses on difficult-to-identify abnormal data, provides adaptive weight support for the loss function, and participates in the construction of quality supervision loss. The gradient modulation adaptive training layer is used to globally differentiate the quality supervision loss and generate a corrected gradient that integrates business situation awareness, the degree of anomaly concealment, and smoothing constraints. It drives the backpropagation of the neural network to update parameters, so that the data quality assessment model can adapt to the imbalance of government categories, the difference in the degree of anomaly concealment, and the confidence stability constraint during training. Represented as: ;in, It is the gradient of the loss applied to the neural network parameters; The first part of the gradient is the weight-modulated gradient resulting from the business distribution and the degree of anomaly concealment. The latter half of the gradient is the gradient correction component corresponding to the smoothing constraint term.

[0027] By performing the above operations, this solution addresses the problem that general e-government data end-to-end quality supervision methods suffer from uneven impacts on parameter updates due to different government business and different quality statuses, making it difficult to adapt to the complex judgment requirements of government scenarios and thus leading to poor quality supervision results. Instead, this solution divides the gradient structure into a weighted modulation gradient component and a smoothing constraint gradient correction component. The first part encodes the dynamic modulation caused by business distribution and anomaly concealment, providing gradient gain compensation for niche businesses and scarce government data. The second part corresponds to the smoothing constraint gradient correction, smoothing and reducing gradient oscillations caused by government labeling noise and blurred boundary samples, while ensuring a balanced gradient contribution, thereby improving the end-to-end quality supervision effect of e-government data.

[0028] Example 8, see Figure 1 This embodiment is based on the above embodiment. In step S7, the e-government data quality supervision is based on the trained data quality assessment model. The e-government data generated by the entire e-government link node is accessed in real time. After preprocessing, it is input into the data quality assessment model. The model outputs the quality status category of each piece of e-government data. If the output quality status category is serious abnormality or link flow violation, the management personnel will be warned.

[0029] Example 9, see Figure 2 Based on the above embodiments, this embodiment provides an e-government data full-link quality supervision system, including an e-government data aggregation module, a feature space construction module, a government data quality characterization module, a business category weight construction module, a quality supervision loss construction module, a data quality assessment model design module, and an e-government data quality supervision module; The e-government data aggregation module collects historical e-government data, which is then preprocessed to produce the original e-government dataset. The feature space construction module decomposes the original government data set and transforms it into a feature space to construct government feature vectors. The government data quality characterization module constructs a characterization network with a deep fully connected feedforward neural network structure, inputs a standardized government feature vector, and outputs the prediction confidence level of the quality status of the government data. The business category weight construction module integrates the number of effective data for each business category with the confidence level of single data quality prediction to construct a business situation awareness coefficient. The quality supervision loss construction module constructs quality supervision loss based on the prediction confidence level using the business situation awareness coefficient; The data quality assessment model design module constructs a data quality assessment model that includes an e-government data aggregation layer, a feature space construction layer, a government data quality representation layer, a business category weight construction layer, and a gradient modulation adaptive training layer. The e-government data quality supervision module performs quality supervision on real-time e-government data based on the trained data quality assessment model.

[0030] It should be noted that, in this document, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0031] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0032] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for end-to-end quality supervision of e-government data, characterized in that: The method includes the following steps: Step S1: E-government data aggregation, collecting historical e-government data, and preprocessing the raw e-government dataset; Step S2: Feature space construction. The original government data set is decomposed and transformed into a feature space to construct government feature vectors. Step S3: Government data quality characterization. Construct a characterization network with a deep fully connected feedforward neural network structure, input standardized government feature vectors, and output the prediction confidence level of the quality status of government data. Step S4: Construct business category weights, integrate the number of effective data for each business category with the confidence level of single data quality prediction, and construct a business situation awareness coefficient; Step S5: Constructing quality supervision loss, constructing quality supervision loss based on the prediction confidence coefficient of the business situation awareness coefficient; Step S6: Design the data quality assessment model and construct a data quality assessment model that includes an e-government data aggregation layer, a feature space construction layer, a government data quality representation layer, a business category weight construction layer, and a gradient modulation adaptive training layer; Step S7: E-government data quality supervision, which involves supervising the quality of real-time e-government data based on the trained data quality assessment model.

2. The method for end-to-end quality supervision of e-government data according to claim 1, characterized in that: In step S4, the construction of business category weights integrates the effective data distribution of business categories with the confidence level of single data quality prediction, and introduces link flow entropy to construct a business situation awareness coefficient, quantifies the process source of anomaly concealment, and compensates for the weight of niche business data, suppresses the gradient contribution of simple compliant data, and strengthens the attention to difficult-to-identify abnormal data.

3. The method for end-to-end quality supervision of e-government data according to claim 2, characterized in that: In step S5, the quality supervision loss is constructed by introducing probabilistic smoothing constraints on the basis of business category weights. This retains the ability to focus on unbalanced businesses and difficult-to-identify anomalies, while also constraining the reasonable distribution of model confidence, thereby improving the generalization of quality judgment across the entire chain and ultimately obtaining the quality supervision loss.

4. The method for end-to-end quality supervision of e-government data according to claim 3, characterized in that: In step S6, the data quality assessment model design adopts a five-layer cascaded end-to-end model architecture. The overall hierarchy from top to bottom is as follows: e-government data aggregation layer, feature space construction layer, government data quality representation layer, business category weight construction layer, and gradient modulation adaptive training layer. Each layer is connected step by step, and parameters and features flow unidirectionally and progressively to construct the data quality assessment model. The gradient modulation adaptive training layer is used to globally differentiate the quality supervision loss, generating a corrected gradient that integrates business situational awareness, the degree of anomaly data concealment, and smoothing constraints; this drives the neural network to backpropagate and update parameters.

5. The method for end-to-end quality supervision of e-government data according to claim 4, characterized in that: In step S1, the e-government data aggregation involves collecting historical e-government data; uniformly preprocessing the raw data; and labeling the data, including four categories of quality truth labels: compliant and normal, minor defects, serious anomalies, and link flow violations, to obtain the original e-government dataset.

6. The method for end-to-end quality supervision of e-government data according to claim 5, characterized in that: In step S2, the feature space construction involves decomposing the original government data set into three categories: field compliance features, business association features, and link time series features. At the same time, the effective data volume of each government business category is counted, and the original data is transformed into a feature space. Finally, the data are spliced ​​together to form a unified government affairs feature vector.

7. The method for end-to-end quality supervision of e-government data according to claim 6, characterized in that: In step S3, the government data quality characterization involves building a characterization network, inputting government feature vectors, and outputting the probability that a single piece of government data is compliant and normal, slightly flawed, seriously abnormal, or has a violation in the link flow.

8. The method for end-to-end quality supervision of e-government data according to claim 7, characterized in that: In step S7, the e-government data quality supervision is based on a trained data quality assessment model. It accesses e-government data generated by all nodes in the e-government chain in real time, and after preprocessing, it is input into the data quality assessment model. The model outputs the quality status category of each piece of e-government data. If the output quality status category is serious abnormality or link flow violation, an early warning is issued to the management personnel.

9. A system for end-to-end quality supervision of e-government data, used to implement the method for end-to-end quality supervision of e-government data as described in any one of claims 1-8, characterized in that: It includes an e-government data aggregation module, a feature space construction module, a government data quality characterization module, a business category weight construction module, a quality supervision loss construction module, a data quality assessment model design module, and an e-government data quality supervision module; The e-government data aggregation module collects historical e-government data, which is then preprocessed to produce the original e-government dataset. The feature space construction module decomposes the original government data set and transforms it into a feature space to construct government feature vectors. The government data quality characterization module constructs a characterization network with a deep fully connected feedforward neural network structure, inputs a standardized government feature vector, and outputs the prediction confidence level of the quality status of the government data. The business category weight construction module integrates the number of effective data for each business category with the confidence level of single data quality prediction to construct a business situation awareness coefficient. The quality supervision loss construction module constructs quality supervision loss based on the prediction confidence level using the business situation awareness coefficient; The data quality assessment model design module constructs a data quality assessment model that includes an e-government data aggregation layer, a feature space construction layer, a government data quality representation layer, a business category weight construction layer, and a gradient modulation adaptive training layer. The e-government data quality supervision module performs quality supervision on real-time e-government data based on the trained data quality assessment model.

Images