A machine learning-based data governance method and system
By constructing a generalized graph of causal relationships in data and using deep learning and fuzzy neural networks for data repair, the problem of low efficiency in data governance in existing technologies is solved, and efficient and flexible data governance is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- 江苏冉闻信息科技有限公司
- Filing Date
- 2025-01-16
- Publication Date
- 2026-06-05
AI Technical Summary
Existing data governance methods are inefficient and inflexible when dealing with massive and complex data, and human operation is easily affected by subjective factors, making it difficult to meet data governance needs.
Machine learning-based data governance methods construct generalized graphs of data causal relationships and utilize deep learning and fuzzy neural networks for data repair, including point repair and path repair, and automate and intelligently process data.
It improves the flexibility and accuracy of data governance, reduces the need for human intervention, minimizes resource waste and decision-making errors, and enhances data repair capabilities.
Smart Images
Figure CN119988847B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data governance technology, specifically to a data governance method and system based on machine learning. Background Technology
[0002] With the rapid development of information technology, various information systems have accumulated immeasurable amounts of data in their daily operations. In the data-driven era, the quality of the information accumulated in these systems has a crucial impact on the results of data analysis. However, due to various reasons, such as data entry errors, system compatibility issues, and human negligence, this data is often riddled with inaccurate or non-standard information. The presence of this non-standard data significantly increases the difficulty of data mining and analysis. It not only consumes more computing resources, leading to resource waste, but may also mislead the results of data analysis, resulting in decision-making errors and incalculable losses.
[0003] Various data governance methods have been proposed in existing technologies. Among them, a more traditional approach is to manually inspect and modify data. While this method can solve the problem to some extent, it is time-consuming, labor-intensive, and inefficient, especially when dealing with large amounts of data, where manual operation is almost impossible to meet the needs of data governance. In addition, manual inspection is easily affected by subjective factors, leading to inconsistent quality of data governance.
[0004] Another common data governance approach is to automate data processing by writing corresponding computer program scripts. While this method improves efficiency to some extent, it also presents significant challenges. Because non-standardized data varies widely in type and rules, different programs or scripts are often required for different data types. This not only increases workload but also makes the data governance process less flexible. Especially when the data source changes, existing programs or scripts may become unusable and need to be rewritten, further reducing the efficiency of data governance.
[0005] In conclusion, existing data governance methods have limitations in terms of efficiency, flexibility, and applicability when dealing with massive and complex data. Therefore, exploring a more efficient, flexible, and universal data governance method is particularly urgent. Summary of the Invention
[0006] The purpose of this invention is to provide a data governance method and system based on machine learning. There is a certain causal relationship between various types of data in the same system, and the principle of data de-causation can be used to restore the data.
[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:
[0008] A machine learning-based data governance method includes the following steps:
[0009] Obtain initial data to be treated, and perform standardization preprocessing on the initial data to be treated; the standardization preprocessing refers to transforming the initial data to be treated into a unified standard and performing preliminary cleaning on the initial data to be treated;
[0010] The preprocessed data to be governed is classified, and the causal relationships between each category of data to be governed are determined to construct a generalized causal relationship diagram of the data.
[0011] Based on the generalized graph of causal relationships in the data, deep learning is used to construct data simulation models and data reconstruction models;
[0012] The types of data to be addressed are determined, including point repair and path repair.
[0013] For the point repair, a data simulation model is used to repair the preprocessed data to be governed. For the path repair, a data simulation model and a data reconstruction model are used to repair the preprocessed data to be governed, thereby performing data governance.
[0014] According to the above technical solution, the steps for constructing the data causal relationship generalization graph include:
[0015] Each type of data to be managed is treated as a node, and the causal relationship between each node is obtained using the F-GES model to construct a preliminary data causal relationship generalization graph; the preliminary data causal relationship generalization graph is a directed acyclic graph;
[0016] Except for the initial node, the sequence corresponding to each node is regarded as the parent sequence in the gray relational degree. Its corresponding subsequence is the sequence corresponding to the node at the starting position of all directed edges pointing to that node. The correlation degree between each parent sequence and the corresponding subsequence is calculated using gray relational degree. Then, the directed edges between the nodes corresponding to the subsequences with correlation degree less than the threshold α and the nodes corresponding to the parent sequences are eliminated in the preliminary data causal relationship generalization graph to obtain the data causal relationship generalization graph. The data causal relationship generalization graph is a directed acyclic graph.
[0017] Among them, F-GES is a score-based causal detection method. Each addition or removal of an edge (i.e., a causal relationship) generates a score to measure the degree of distribution of the causal graph data. The Bayesian information criterion (BIC) is used for scoring, which can effectively learn the structure of causal relationships in time series. However, methods based on determining the effective component number inevitably exhibit interactions and cancellations between data points, resulting in a certain degree of error in the initial causal relationship generalization graph. Therefore, grey relational analysis can be used to explore the correlations between nodes in the initial causal relationship generalization graph, further refining the graph.
[0018] According to the above technical solution, the causal weight of each subsequence corresponding to each parent sequence in the generalized data causal relationship graph is calculated, and the specific steps include:
[0019] The degree of correlation between each parent sequence and its corresponding child sequence in the generalized causal relationship graph of the data is calculated using grey relational analysis.
[0020] The generalized graph of the causal relationship of the data is used as a theoretical model in the structural equation model (SEM), and the path coefficients corresponding to each directed edge are calculated using statistical software.
[0021] The path coefficients and their corresponding correlation degrees of the directed edges between corresponding nodes in each subsequence and corresponding nodes in the parent sequence are added together to obtain the preliminary weights.
[0022] Based on the initial weights between the parent sequence and each of its corresponding subsequences, the causal weights between the parent sequence and each of its corresponding subsequences are combined to 1, thus obtaining the causal weights between the parent sequence and each of its corresponding subsequences.
[0023] In structural equation modeling, causal relationships are typically represented by path coefficients, which reflect the strength and direction of the direct relationship between variables. Therefore, path coefficients can be used as the degree of causal association between nodes. Grey relational analysis can be used to explore the correlation between node data. After standardizing the causal association degree and the grey relational analysis of the correlation between node data, the corresponding causal weights between nodes can be calculated, making the causal weights more accurate.
[0024] According to the above technical solution, the parent sequence and each of its subsequences multiplied by the corresponding causal weight in the generalized data causal relationship graph are divided into a training set and a validation set. Based on the training set and validation set, a data simulation model is constructed using deep learning. Deep learning can be convolutional neural networks, recurrent neural networks, long short-term memory networks, autoencoders, etc.
[0025] According to the above technical solution, the types of data to be managed include:
[0026] Point repair refers to repairing data at a specific point in the generalized graph of data causality relationships. This can be understood as repairing incomplete data where a node on an edge of the generalized graph of data causality relationships is missing.
[0027] Path repair, which involves repairing data for continuous points in the generalized data causal relationship graph, can be understood as repairing incomplete data where continuous nodes in a certain path of the generalized data causal relationship graph are missing.
[0028] According to the above technical solution, the data reconstruction model execution steps include:
[0029] For a missing parent sequence at a certain point in the generalized data causal relationship graph, the corresponding subsequence and simulated subsequence are obtained using a data simulation model, and the virtual causal weights corresponding to the simulated subsequence are calculated.
[0030] Based on the virtual causal weights corresponding to the simulated subsequences, the weight difference prediction model is used to predict the weight difference of the simulated subsequences.
[0031] The predicted weight difference is used to obtain the weight correction amount using a fuzzy neural network. The simulated subsequence is then corrected based on the weight correction amount. That is, the causal weight corresponding to that point in the simulated subsequence is corrected based on the weight correction amount. The corrected causal weight is multiplied by the simulated subsequence to obtain the corrected simulated subsequence that enters the next convolutional network.
[0032] The subsequences corresponding to the missing parent sequence and the corresponding causal weights, as well as the corrected simulated subsequences, are used to extract feature information through convolutional layers. The extracted feature information is then input into global max pooling for fusion. After passing through two fully connected layers, the feature distribution value is calculated, and the ReLU activation function is used to filter out negative values, thereby obtaining the perceptual weights corresponding to the subsequences and the corrected simulated subsequences.
[0033] The subsequence and the corrected simulated subsequence are multiplied by the corresponding perceptual weights and then fed into the recurrent neural network and attention mechanism for processing. Finally, the softmax function is used for normalization and output.
[0034] Since predictions made using data simulation models contain errors, fuzzy neural networks can be used to correct these predictions by utilizing causal weights between nodes. When the virtual causal weights in the prediction results are too large, a weight correction amount is obtained using the fuzzy neural network. The virtual causal weights are then corrected based on this correction amount, and the prediction results of the data simulation model are further corrected based on the corrected virtual causal weights. The corrected prediction results are then fed into the next lower-level network for learning.
[0035] Since the number of simulated subsequences corresponding to different parent sequences is not consistent, the error in the learning process of the neural network may be large. Therefore, the accuracy of prediction can be improved by learning the subsequences and the perceptual weights corresponding to the corrected simulated subsequences.
[0036] According to the above technical solution, the construction steps of the weight difference prediction model include:
[0037] For the parent sequence corresponding to a point in the generalized graph of the data causal relationship, obtain the causal weight of each subsequence corresponding to the parent sequence;
[0038] Based on all the subsequences corresponding to the parent sequence, the simulated subsequences are obtained using the data simulation model described above;
[0039] Calculate the virtual causal weights between the parent sequences corresponding to the simulated subsequences;
[0040] Calculate the difference between the virtual causal weights corresponding to the simulated subsequences and the causal weights of the corresponding subsequences;
[0041] A weighted difference model is constructed using a neural network model based on simulated subsequences and their corresponding differences.
[0042] According to the above technical solution, a data simulation model is used to repair the sequence corresponding to the initial point in the path repair, and a data reconstruction model is used to repair the sequence corresponding to the points other than the initial point in the path repair.
[0043] In another embodiment, a machine learning-based data governance system includes:
[0044] The preprocessing module acquires the initial data to be treated and performs standardized preprocessing on the initial data to be treated.
[0045] The generalization graph construction module classifies the preprocessed data to be governed and determines the causal relationships between each type of data to be governed, thus constructing a generalization graph of causal relationships between the data.
[0046] The governance model building module uses deep learning to construct data simulation models and data reconstruction models based on the generalized graph of causal relationships in the data.
[0047] The data type determination module determines the types of data to be treated, including point repair and path repair. The types of data to be treated include:
[0048] Point repair refers to repairing data at a specific point in the generalized data causal relationship graph.
[0049] Path repair involves repairing data for continuous points in the generalized data causal relationship graph.
[0050] The data governance module uses a data simulation model to repair the preprocessed data to be governed for the point repair, uses a data simulation model to repair the sequence corresponding to the initial point in the path repair, and uses a data reconstruction model to repair the sequence corresponding to the points in the path repair other than the initial point, thereby performing data governance.
[0051] Another embodiment is provided, a storage medium for storing computer-executable instructions, characterized in that: when the computer-executable instructions are executed, they implement a data governance method based on machine learning as described in any of the above schemes.
[0052] Compared with existing technologies, the beneficial effects achieved by this invention are as follows: This invention reveals the inherent connections and patterns between data by constructing a generalized graph of data causal relationships considering the causal relationships between data, and adopts different repair strategies based on the generalized graph of data causal relationships (point repair and path repair), thereby improving the flexibility of data governance. During the path repair process, the data entering the next network is corrected by considering the weights between nodes, improving the accuracy of network prediction. By learning the perceptual weights corresponding to subsequences and corrected simulated subsequences, the ability to extract data features is improved, thereby enhancing the data repair capability. Simultaneously, the automated and intelligent data governance process reduces the need for manual intervention, thus lowering the cost of data governance. Attached Figure Description
[0053] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:
[0054] Figure 1 This is a flowchart of the steps of a data governance method based on machine learning according to the present invention;
[0055] Figure 2 This is a flowchart of the calculation steps for the perceived weight in the embodiment. Detailed Implementation
[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0057] Example 1
[0058] This invention provides a technical solution: a data governance method based on machine learning, the steps of which include:
[0059] S1. Obtain initial data to be treated and perform standardized preprocessing on the initial data to be treated.
[0060] S2. Classify the preprocessed data to be treated and determine the causal relationships between each category of data to be treated to construct a generalized causal relationship graph. Calculate the causal weight of each subsequence corresponding to each parent sequence in the generalized causal relationship graph.
[0061] The steps for constructing a generalized graph of causal relationships in data include:
[0062] Each type of data to be governed is treated as a node, and the causal relationships between each node are obtained using the F-GES model to construct a preliminary generalized graph of data causal relationships; the preliminary generalized graph of data causal relationships is a directed acyclic graph;
[0063] Except for the initial node, the sequence corresponding to each node is regarded as the parent sequence in the grey relational analysis. Its corresponding subsequence is the sequence corresponding to the node at the starting position of all directed edges pointing to that node. The grey relational analysis is used to calculate the correlation degree between each parent sequence and the corresponding subsequence. Then, in the preliminary data causal relationship generalization graph, the directed edges between the nodes corresponding to the subsequences with correlation degree less than the threshold α and the nodes corresponding to the parent sequences are eliminated, resulting in the data causal relationship generalization graph. The data causal relationship generalization graph is a directed acyclic graph.
[0064] The specific steps for calculating the causal weight of each child sequence corresponding to each parent sequence in the generalized data causal relationship graph include:
[0065] The degree of correlation between each parent sequence and its corresponding child sequence in the generalized causal relationship graph of the data is calculated using grey relational analysis.
[0066] The generalized graph of the causal relationship of the data is used as a theoretical model in the structural equation model (SEM), and the path coefficients corresponding to each directed edge are calculated using statistical software.
[0067] The path coefficients and their corresponding correlation degrees of the directed edges between corresponding nodes in each subsequence and corresponding nodes in the parent sequence are added together to obtain the preliminary weights.
[0068] Based on the initial weights between the parent sequence and each of its corresponding subsequences, the causal weights between the parent sequence and each of its corresponding subsequences are combined into 1 to obtain the causal weights between the parent sequence and each of its corresponding subsequences.
[0069] S3. Based on the generalized graph of causal relationships in the data, use deep learning to construct data simulation models and data reconstruction models.
[0070] The parent sequence and each of its subsequences multiplied by the corresponding causal weight in the generalized graph of data causality are divided into a training set and a validation set. Based on the training set and the validation set, the data simulation model is constructed using deep learning.
[0071] The data reconstruction model execution steps include: for a missing parent sequence at a certain point in the generalized data causal relationship graph, obtaining the corresponding subsequence and simulated subsequence, and calculating the virtual causal weights corresponding to the simulated subsequence;
[0072] Based on the virtual causal weights corresponding to the simulated subsequences, the weight difference prediction model is used to predict the weight difference of the simulated subsequences.
[0073] The predicted weight difference is used to obtain the weight correction amount using a fuzzy neural network. The simulated subsequence is then corrected based on the weight correction amount. That is, the causal weight corresponding to that point in the simulated subsequence is corrected based on the weight correction amount. The corrected causal weight is multiplied by the simulated subsequence to obtain the corrected simulated subsequence for entering the next convolutional network.
[0074] The subsequence corresponding to the missing parent sequence multiplied with the corresponding causal weight, and the corrected simulated subsequence, are each processed using three convolutional layers to extract feature information. The extracted feature information is then fused into a global max pooling layer. Next, after passing through two fully connected layers, the feature distribution values Q1 and Q2 of the simulated subsequence are calculated. Negative values are filtered out using the ReLU activation function. Finally, a weight allocation mechanism is used to obtain the perceptual weight W2 of the subsequence and the perceptual weight W1 corresponding to the corrected simulated subsequence, respectively. Figure 2 As shown
[0075] in,
[0076] The subsequence and the corrected simulated subsequence are multiplied by the corresponding perceptual weights and then fed into the recurrent neural network and attention mechanism for processing. Finally, the softmax function is used for normalization and output.
[0077] The construction steps of the weighted difference prediction model include:
[0078] For the parent sequence corresponding to a point in the generalized graph of the data causal relationship, obtain the causal weight of each subsequence corresponding to the parent sequence;
[0079] Based on all the subsequences corresponding to the parent sequence, the simulated subsequences are obtained using the data simulation model described above;
[0080] Calculate the virtual causal weights between the parent sequences corresponding to the simulated subsequences;
[0081] Calculate the difference between the virtual causal weights corresponding to the simulated subsequences and the causal weights of the corresponding subsequences;
[0082] A weighted difference model is constructed using a neural network model based on simulated subsequences and their corresponding differences.
[0083] S4. Determine the types of data to be treated, including point repair and path repair; point repair refers to repairing data at a specific point in the generalized causal relationship graph; path repair refers to repairing data at consecutive points in the generalized causal relationship graph.
[0084] S5. For the point repair, a data simulation model is used to repair the preprocessed data to be treated. For the path repair, a data simulation model and a data reconstruction model are used to repair the preprocessed data to be treated, thereby performing data governance. Specifically, for the initial point in the path repair, a data simulation model is used to repair the sequence corresponding to the initial point. For the points in the path repair other than the initial point, the sequence corresponding to the data reconstruction model is used to repair.
[0085] Example 2
[0086] A machine learning-based data governance system includes:
[0087] The preprocessing module acquires the initial data to be treated and performs standardized preprocessing on the initial data to be treated.
[0088] The generalization graph construction module classifies the preprocessed data to be governed and determines the causal relationships between each type of data to be governed, thus constructing a generalization graph of causal relationships between the data.
[0089] The governance model building module uses deep learning to construct data simulation models and data reconstruction models based on the generalized graph of causal relationships in the data.
[0090] The data type determination module determines the types of data to be treated, including point repair and path repair.
[0091] Point repair refers to repairing data at a specific point in the generalized data causal relationship graph.
[0092] Path repair involves repairing data for continuous points in the generalized data causal relationship graph.
[0093] The data governance module uses a data simulation model to repair the preprocessed data to be governed for the point repair, uses a data simulation model to repair the sequence corresponding to the initial point in the path repair, and uses a data reconstruction model to repair the sequence corresponding to the points in the path repair other than the initial point, thereby performing data governance.
[0094] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, 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.
[0095] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A data governance method based on machine learning, characterized in that, The steps include: Obtain initial data to be treated, and perform standardized preprocessing on the initial data to be treated; The preprocessed data to be treated is classified, and the causal relationships between each category of data are determined to construct a generalized causal relationship graph. The steps for constructing the generalized causal relationship graph include: Each type of data to be managed is treated as a node, and the causal relationship between each node is obtained using the F-GES model to construct a preliminary data causal relationship generalization graph; the preliminary data causal relationship generalization graph is a directed acyclic graph; Except for the initial node, the sequence corresponding to each node is considered the parent sequence in the grey relational analysis. Its corresponding child sequences are the sequences corresponding to the starting positions of all directed edges pointing to that node. Grey relational analysis is used to calculate the correlation degree between each parent sequence and its corresponding child sequences. Sequences with correlation degrees less than a threshold are eliminated in the preliminary data causal relationship generalization graph. The directed edges between the nodes corresponding to the corresponding subsequences and the nodes corresponding to the parent sequence are used to obtain a generalized graph of data causal relationships; the causal weight of each subsequence corresponding to each parent sequence in the generalized graph of data causal relationships is calculated, specifically including the following steps: The degree of correlation between each parent sequence and its corresponding child sequence in the generalized causal relationship graph of the data is calculated using grey relational analysis. The generalized graph of the causal relationship of the data is used as a theoretical model in the structural equation model (SEM), and the path coefficients corresponding to each directed edge are calculated using statistical software. The path coefficients and their corresponding correlation degrees of the directed edges between corresponding nodes in each subsequence and corresponding nodes in the parent sequence are added together to obtain the preliminary weights. Based on the initial weights between the parent sequence and each of its corresponding subsequences, the weights between the parent sequence and each of its corresponding subsequences are combined into 1 to obtain the causal weights between the parent sequence and each of its corresponding subsequences. The generalized graph of causal relationships in the data is a directed acyclic graph; Based on the generalized graph of causal relationships in the data, deep learning is used to construct data simulation models and data reconstruction models; The types of data to be addressed are determined, including point repair and path repair. For the point repair, a data simulation model is used to repair the preprocessed data to be governed. For the path repair, a data simulation model and a data reconstruction model are used to repair the preprocessed data to be governed, thereby performing data governance.
2. The data governance method based on machine learning according to claim 1, characterized in that, The parent sequence and each of its subsequences multiplied by the corresponding causal weight in the generalized data causal relationship graph are divided into a training set and a validation set. The data simulation model is then constructed using deep learning based on the training set and the validation set.
3. The data governance method based on machine learning according to claim 1, characterized in that, The types of data to be processed include: Point repair refers to repairing data at a specific point in the generalized data causal relationship graph. Path repair involves repairing data for continuous points in the generalized data causal relationship graph.
4. The data governance method based on machine learning according to claim 1, characterized in that, The data reconstruction model execution steps include: For a missing parent sequence at a certain point in the generalized data causal relationship graph, the corresponding subsequence and simulated subsequence are obtained using a data simulation model, and the virtual causal weights corresponding to the simulated subsequence are calculated. Based on the virtual causal weights corresponding to the simulated subsequences, the weight difference prediction model is used to predict the weight difference of the simulated subsequences. The predicted weight difference is used to obtain the weight correction amount using a fuzzy neural network, and the simulated subsequence is corrected based on the weight correction amount; The subsequences corresponding to the missing parent sequence and the corresponding causal weights, as well as the corrected simulated subsequences, are used to extract feature information through convolutional layers. The extracted feature information is then input into global max pooling for fusion. After passing through two fully connected layers, the feature distribution value is calculated, and the ReLU activation function is used to filter out negative values, thereby obtaining the perceptual weights corresponding to the subsequences and the corrected simulated subsequences. The subsequence and the corrected simulated subsequence are multiplied by the corresponding perceptual weights and then fed into the recurrent neural network and attention mechanism for processing. Finally, the softmax function is used for normalization and output.
5. The data governance method and system based on machine learning according to claim 4, characterized in that, The construction steps of the weighted difference prediction model include: For the parent sequence corresponding to a point in the generalized graph of the data causal relationship, obtain the causal weight of each subsequence corresponding to the parent sequence; Based on all the subsequences corresponding to the parent sequence, the simulated subsequences are obtained using the data simulation model described above; Calculate the virtual causal weights between the parent sequences corresponding to the simulated subsequences; Calculate the difference between the virtual causal weights corresponding to the simulated subsequences and the causal weights of the corresponding subsequences; A weighted difference model is constructed using a neural network model based on simulated subsequences and their corresponding differences.
6. The data governance method based on machine learning according to claim 1, characterized in that, For the initial point in the path repair, a data simulation model is used to repair the sequence corresponding to the initial point. For the points other than the initial point in the path repair, a data reconstruction model is used to repair the sequence corresponding to the initial point.
7. A system employing the machine learning-based data governance method as described in claim 1, characterized in that, include: The preprocessing module acquires the initial data to be treated and performs standardized preprocessing on the initial data to be treated. The generalization graph construction module classifies the preprocessed data to be governed and determines the causal relationships between each category of data to be governed, thus constructing a generalization graph of causal relationships between the data. The governance model building module uses deep learning to construct data simulation models and data reconstruction models based on the generalized graph of causal relationships in the data. The data type determination module determines the types of data to be treated, including point repair and path repair. The types of data to be processed include: Point repair refers to repairing data at a specific point in the generalized data causal relationship graph. Path repair involves repairing data for continuous points in the generalized data causal relationship graph. The data governance module uses a data simulation model to repair the preprocessed data to be governed for the point repair, uses a data simulation model to repair the sequence corresponding to the initial point in the path repair, and uses a data reconstruction model to repair the sequence corresponding to the points in the path repair other than the initial point, thereby performing data governance.
8. A storage medium for storing computer-executable instructions, characterized in that: When the computer-executable instructions are executed, they implement a data governance method based on machine learning as described in any one of claims 1 to 6.