Data management method based on industrial big data platform

Abstract

The invention relates to the technical field of data governance, and discloses a data governance method based on an industrial big data platform, which comprises the following steps: processing an organizational structure document and a business process definition through a text semantic analysis technology, and generating a responsibility allocation matrix; analyzing the historical data access behavior of the user, and depicting the data use characteristics of the user; according to the user model and the responsibility matrix, in combination with data sensitivity scoring, scene feature extraction and multi-dimensional risk calculation, dynamic permission decision is realized; for the decision-making mechanism, verifying the security influence of the authorization strategy through a simulation test; performing adaptive adjustment on model parameters according to a test result, and dynamically updating the initial responsibility matrix to form closed-loop management; through organizational structure intelligent analysis, user behavior modeling, risk awareness permission decision, digital twin verification and adaptive optimization, accurate division and dynamic adjustment of data governance responsibilities are realized, and the intelligence and dynamic adaptability of data permission management are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data governance, and more specifically, to a data governance method based on an industrial big data platform. Background Art

[0002] With the rapid development of industry and intelligent manufacturing, industrial big data platforms have become important infrastructure for enterprise digital transformation, which contain a large amount of sensitive data such as production process parameters, R & D experimental data, etc. Existing data governance methods mainly rely on role-based access control (RBAC) or attribute-based access control (ABAC) models to manage permissions through predefined static rules.

[0003] However, these traditional methods have various technical problems: the responsibility assignment mechanism relies on manual definition, making it difficult to reflect organizational structure changes in a timely manner, resulting in unclear responsibility division; static permission rules cannot cope with dynamic access scenarios and risk situations, either overly restricting and affecting data usage efficiency, or providing insufficient protection and leading to security risks; there is a lack of a pre-assessment mechanism for the impact of authorization decisions, and unexpected security risks or efficiency losses may occur after permission adjustment.

[0004] The above technical problems seriously restrict the secure and efficient use of data on industrial big data platforms, and there is an urgent need for a new data governance method that can intelligently analyze organizational structures, dynamically perceive access risks, pre-assess security impacts, and adaptively optimize. Summary of the Invention

[0005] The present invention provides a data governance method based on an industrial big data platform to solve the technical problems of unclear responsibility assignment, lack of adaptability in static permission management, and lack of security impact assessment for authorization decisions in related technologies.

[0006] The present invention provides a data governance method based on an industrial big data platform, including:

[0007] Processing organizational structure documents and business process definitions through text semantic analysis technology to generate a responsibility assignment matrix;

[0008] Analyzing the historical data access behavior of users to construct a time series model of user operation patterns to characterize user data usage characteristics;

[0009] Based on the user model and the responsibility matrix, combining data sensitivity scoring, scenario feature extraction, and multi-dimensional risk calculation to achieve dynamic permission decisions;

[0010] Based on the dynamic permission decision, constructing a virtual data interaction environment to verify the security impact of the authorization policy through simulation testing;

[0011] Based on the simulation test results, the model parameters are adaptively adjusted, and the initial responsibility matrix is dynamically updated to form a closed-loop management.

[0012] Furthermore, the steps of generating a responsibility assignment matrix by processing the organizational structure document and business process definition through text semantic analysis technology include:

[0013] Process unstructured text to extract role entities, responsibility entities, and relationships; the unstructured text includes organizational structure documents, job descriptions, and business process documents;

[0014] Construct an organizational structure knowledge graph and calculate node centrality and influence indicators;

[0015] Combine the graph analysis results and business process definition to generate a RACI responsibility matrix.

[0016] Furthermore, the steps of analyzing the user's historical data access behavior and constructing a time series model of the user operation mode include:

[0017] Collect the user data access logs, and generate a structured user behavior data set after cleaning and preprocessing;

[0018] Organize the preprocessed behavior data into a time series sequence, and use a deep learning model combining long short-term memory network and attention mechanism for modeling;

[0019] Conduct cluster analysis on users with similar behavior patterns to identify typical behavior types and patterns.

[0020] Furthermore, the steps of realizing dynamic permission decision-making based on data sensitivity scoring, scenario feature extraction, and multi-dimensional risk calculation include:

[0021] Classify and evaluate the sensitivity of data assets, and construct a sensitivity scoring model that comprehensively considers the data content type, value, and influence scope;

[0022] Extract access scenario features in real time;

[0023] Multi-dimensional risk calculation and decision-making, based on the user behavior model, data sensitivity, and scenario features, calculate the access risk score, and determine authorization according to the comparison result of the score and the threshold.

[0024] Furthermore, the steps of constructing a data interaction virtual environment and verifying the security impact of the authorization policy through simulation tests include:

[0025] Create virtual copies of data assets, user simulation agents, and interaction behavior simulators to form a virtual environment;

[0026] Simulate the user's access behavior to data with different sensitivities in the virtual environment;

[0027] Analyze the impact of different permission policies on data security and business efficiency through causal inference methods.

[0028] Furthermore, the steps for adaptively adjusting the model parameters based on the governance effect monitoring results include:

[0029] Construct a data governance effect evaluation model that includes basic metrics, process metrics, and outcome metrics;

[0030] Adopt a reinforcement learning algorithm to adaptively adjust the parameters of the risk assessment model, and model its adjustment process as a Markov decision process.

[0031] Furthermore, the steps for dynamically updating the responsibility matrix include:

[0032] Force an update based on organizational structure changes to ensure that the responsibility matrix is consistent with the current organizational structure;

[0033] Optimize and update based on the governance effect evaluation results, and optimize the responsibility assignment plan by analyzing the relationship between responsibility assignment and governance effects.

[0034] Furthermore, the calculation process of the long short-term memory network includes:

[0035]

[0036] f t = σ(W f ·(h t-1 , x t ) + b f );

[0037]

[0038]

[0039] o t = σ(W o ·(h t-1 , x t ) + b o );

[0040] h t = o t ⊙ tanh(c t );

[0041] Where f t and o t represent the values of the input gate, forget gate, and output gate respectively, h t-1 and h t represent the hidden states at time t - 1 and time t respectively, X tDenote the input at time t as c t and c t-1 represent the cell states at time t and t-1 respectively, is the candidate cell state, ⊙ represents element-wise product, σ is the sigmoid activation function, W i 、W f 、W c 、W o represent the weight matrices of the input gate, forget gate, cell state, and output gate respectively, b i 、b f 、b c 、b o represent the bias terms of the input gate, forget gate, cell state, and output gate respectively, tanh represents the hyperbolic tangent activation function.

[0042] Furthermore, the multi-dimensional risk calculation and decision-making are implemented by the gradient boosting decision tree model. Whether to grant authorization is determined according to the comparison result between the risk score and the preset threshold. The implementation of the gradient boosting decision tree model includes:

[0043] Construct a combined feature vector by combining the user behavior feature, data sensitivity feature, scenario feature, and the corresponding value of the responsibility matrix into a unified feature vector:

[0044] X = (B u , S(d), C scene , M RACI (u, d));

[0045] where X is the combined feature vector, B u is the behavior feature vector of user u, S(d) is the sensitivity feature vector of data d, C scene is the scenario feature vector, M BACI (u, d) is the responsibility value of user u for data d in the responsibility matrix;

[0046]

[0047] where is the output value of the m1-th tree, is the number of leaf nodes of tree m1, represents the feature space region corresponding to the j1-th leaf node in the m1-th tree, is the output value of the j1-th leaf node in the m1-th tree, I ind (·) is the indicator function, denotes the summation over all leaf nodes of tree m1;

[0048] The final prediction of the model is the sum of all trees;

[0049] The structure of the new tree is determined by minimizing the loss function.

[0050] A data governance system based on an industrial big data platform, which is used to execute the data governance method based on the industrial big data platform as described above, includes:

[0051] An organizational structure intelligent analysis module, which is used to process organizational structure documents and business process definitions through text semantic analysis to generate a responsibility assignment matrix;

[0052] A user behavior modeling module, which is used to analyze the historical data access behavior of users and construct a time series model of user operation patterns;

[0053] A risk perception decision-making module, which is used to perform dynamic permission decisions based on data sensitivity scores, scenario features, and multi-dimensional risk calculations;

[0054] A digital twin verification module, which is used to construct a virtual environment for data interaction and verify the security impact of authorization policies;

[0055] An adaptive optimization module, which is used to adjust model parameters and update the responsibility matrix based on the monitoring results of governance effects;

[0056] A processor and a memory, which are used to execute the operations of the above modules and store data.

[0057] The beneficial effects of the present invention are as follows: Through organizational structure intelligent analysis, user behavior modeling, risk perception-based permission decision-making, digital twin verification, and adaptive optimization, an intelligent data governance method based on an industrial big data platform is realized, and the key technical problems of unclear responsibility assignment, static and rigid permission management, and lack of pre-event security assessment in traditional data governance are solved;

[0058] The precise division and dynamic adjustment of data governance responsibilities are realized, reducing the responsibility assignment error rate by 42% and the responsibility update response time by 65%.

[0059] The intelligence and dynamic adaptability of data permission management are improved. On the premise of maintaining the same security level, 38% of unnecessary permission restrictions are reduced, and the data usage efficiency is improved.

[0060] The pre-event security verification of permission decisions is realized, and 85% of potential permission risk points can be identified in advance. Compared with the traditional post-event monitoring method, the incidence rate of risk events is reduced by 56%. Brief Description of the Drawings

[0061] Figure 1 is a flowchart of the data governance method based on the industrial big data platform in the present invention;

[0062] Figure 2 is a flowchart of the organizational structure intelligent analysis step of the present invention;

[0063] Figure 3 is a flowchart of the user behavior modeling steps of the present invention;

[0064] Figure 4 is a flowchart of the permission decision-making steps of the risk perception of the present invention;

[0065] Figure 5 is a flowchart of the digital twin verification steps of the present invention;

[0066] Figure 6 is a flowchart of the adaptive optimization steps of the present invention. Detailed Implementation Manner

[0067] Now, the subject matter described herein will be discussed with reference to exemplary embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein. Without departing from the scope of protection of the content of this specification, changes can be made to the functions and arrangements of the elements discussed. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described in some examples can also be combined in other examples.

[0068] In at least one embodiment of the present invention, a data governance method based on an industrial big data platform is disclosed, as Figures 1 to 6 shown, including the following steps:

[0069] Step 1, processing the organizational structure document and business process definition through text semantic analysis technology to generate a responsibility assignment matrix;

[0070] In this step, a text semantic analysis algorithm and a graph structure analysis model are used to process the enterprise organizational structure document and business process definition file to generate a structured responsibility assignment matrix; the specific implementation includes the following sub-steps:

[0071] Sub-step 1.1, text semantic analysis;

[0072] Applying natural language processing technology to process unstructured text, extracting key entities and relationships; the unstructured text includes organizational structure documents, job descriptions, and business process documents;

[0073] In this sub-step:

[0074] First, preprocess the text, including word segmentation, stop word removal, part-of-speech tagging, etc.;

[0075] Then, use a named entity recognition algorithm to identify role entities (such as departments, positions, personnel, etc.) and responsibility entities in the text;

[0076] Finally, apply the relationship extraction algorithm to extract management relationships, business collaboration relationships, etc. between entities. The relationship extraction uses a model based on dependency syntactic analysis, which is expressed as:

[0077] R e = f relation (E1, E2, C org );

[0078] where R e represents the relationship type between entity E1 and entity E2, C org represents the context feature vector in the organizational structure analysis, and f relation represents the relationship classification function.

[0079] Sub-step 1.2, organizational graph construction;

[0080] Based on the entities and relationships extracted in the previous step, construct an organizational structure knowledge graph, which is formally represented as a directed graph:

[0081] G o = (V, E, T rel );

[0082] where G o represents the organizational graph, V represents the set of nodes (including entities such as departments, roles, and personnel), E represents the set of edges (representing the relationships between entities), and T rel represents the set of relationship types.

[0083] Apply graph structure analysis algorithms to the constructed organizational graph to calculate key indicators such as the centrality and influence of each node:

[0084]

[0085] where C a represents the centrality of node a, N(a) represents the set of all nodes connected to node a, w ab represents the relationship weight between node a and node b, A inf represents the influence index of the node, and ∑ represents the summation symbol.

[0086] Sub-step 1.3, RACI responsibility matrix generation;

[0087] Combine the analysis results of the organizational graph and the business process definition to generate a RACI (Responsible, Accountable, Consulted, Informed) responsibility matrix; this matrix describes the specific responsibility types of each role in the data governance activity and is formally represented as:

[0088] M RACI = {m ij}, m ij ∈ {R, A, C, I, -};

[0089] Among them, M RACI represents the RACI responsibility matrix, and m ij represents the responsibility type of role i for activity j. R represents responsible, A represents approve, C represents consult, I represents informed, and - represents irrelevant;

[0090] The matrix generation process adopts an algorithm based on rules and graph pattern matching to map the organizational relationship, business process, and data governance activities;

[0091] In this formula, if role i has a direct relationship with activity j, then m ij is one of R, A, C, I; if role i has no direct relationship with activity j, then m ij is -.

[0092] The RACI responsibility matrix M generated in this sub-step RACI is used as the key output, where M RACI is a two-dimensional matrix. The rows of the matrix represent various roles in the organization (such as departments, positions, individuals); the columns of the matrix represent various activities or decisions related to data governance.

[0093] Step 2: Analyze the user's historical data access behavior, construct a time series model of the user operation mode, and depict the user data usage characteristics;

[0094] This step constructs a user behavior pattern model by analyzing the user's historical data access and operation records to accurately depict the user data usage characteristics; it specifically includes the following sub-steps:

[0095] Sub-step 2.1: Data access log collection and preprocessing;

[0096] Collect the user's data access logs from the industrial big data platform, including information such as access time, access data type, access operation type, access result, etc.; clean, standardize, and extract features from the collected original logs to generate a structured user behavior data set; the feature extraction process includes multi-dimensional information such as time features (such as access period, frequency, etc.), operation features (such as operation types and proportions of query, modification, deletion, etc.), and data type features (such as data categories and sensitivity levels accessed).

[0097] Sub-step 2.2: Construction of the behavior sequence model;

[0098] Organize the preprocessed user behavior data into a time series sequence, and use a sequence model to model the user operation mode; the specific implementation is a time series deep learning model, which combines the long short-term memory network (LSTM) and the attention mechanism, and is expressed as:

[0099] h t =LSMT(xt , h t-1 , c t-1 );

[0100]

[0101] where x t represents the user behavior feature vector at time t, h t and c t are the hidden state and cell state of the LSTM respectively, h t-1 and c t-1 represent the hidden state and cell state at time t - 1 respectively, α att is the attention weight, C ctx represents the organizational context information in user behavior modeling, B user represents the user behavior pattern vector, f represents the generation function of the user behavior pattern vector B user , Attention represents the attention mechanism, LSTM represents the long short - term memory network model, f represents the generation function of the user behavior pattern vector; h1, h2, represent the hidden state vectors of the LSTM network at time step 1, time step 2, and the last time step T seq respectively, T seq represents the length of the time series, and they jointly constitute the time series representation of the user behavior sequence for context information extraction in the attention mechanism calculation;

[0102] The LSTM part of this model consists of an input gate, a forget gate, an output gate, and a cell state. The detailed calculation process is as follows:

[0103]

[0104] f t = σ(W f ·(h t-1 , x t ) + b f );

[0105]

[0106] o t = σ(W o ·(h t-1 , x t ) + b o );

[0107] h t = o t ⊙ tanh(c t );

[0108] where ft and o t represent the values of the input gate, forget gate, and output gate respectively, and h t-1 and h t represent the hidden states at time t-1 and time t respectively, and x t represents the input at time t, and c t and c t-1 represent the cell states at time t and time t-1 respectively. is the candidate cell state, ⊙ represents element-wise multiplication, σ is the sigmoid activation function, and W i , W f , W c , W o represent the weight matrices of the input gate, forget gate, cell state, and output gate respectively, and b o , b f , b c , b o represent the bias terms of the input gate, forget gate, cell state, and output gate respectively. tanh represents the hyperbolic tangent activation function.

[0109] The attention mechanism part is calculated as follows:

[0110] e t = v · tanh(W h · h t + W s · s t-1 );

[0111]

[0112] where e t and e n represent the attention scores at time step t and n respectively, h t represents the hidden state vector at time t, s t-1 represents the decoder state at time t-1, α att represents the attention weight, v represents the weight vector of the attention mechanism, W h represents the weight matrix of the hidden state, W s represents the weight matrix of the decoder state, tanh represents the hyperbolic tangent activation function, exp represents the natural exponential function, and T seq represents the length of the time series.

[0113] Specific application examples of this model in the industrial big data platform environment are as follows: Analyze the data access behavior of a steel plant. By extracting the access sequences of users to different types of data such as production parameters, quality data, and equipment status, construct a user behavior time series model. The model can identify normal access patterns, such as the fixed pattern of quality inspection personnel regularly viewing quality data, and abnormal behaviors, such as production personnel suddenly exporting a large amount of commercially sensitive data. Through this model, the time series patterns, preferences, and abnormal behavior characteristics of user data access can be captured.

[0114] Sub-step 2.3, behavior pattern clustering and portrait generation;

[0115] Based on the constructed user behavior model, conduct cluster analysis on users with similar behavior patterns to identify typical user behavior types and patterns; use an improved K-means clustering algorithm to group user behavior vectors:

[0116]

[0117] Where S clust represents the clustering result, S cl，ic represents the ic-th cluster, μ ic is the center of the ic-th cluster, X jc is the user behavior vector, R reg (S clust ) is a regularization term considering organizational structure information, λ clust is the balance parameter in clustering, k represents the total number of clusters, ||x jc -μ ic || 2 represents the square of the Euclidean distance between the vector x jc and the cluster center μ ic , represents the sum over all k clusters (from ic = 1 to ic = k), represents the sum over all vectors x cl，ic in the cluster S jc , represents the clustering result S clust that minimizes the objective function;

[0118] Based on the clustering results, generate a behavior portrait for each user or user group, including dimension information such as typical operation patterns, data access preferences, and risk tendencies.

[0119] In this method, the general user behavior pattern vector B user generated in step 2.2 is used to represent the overall behavior model, while for a specific user u, its personalized behavior model is denoted as B u , which is based on B userAn individualized model derived from the user's specific behavior characteristics and clustering results.

[0120] The user behavior models generated in this step (including the general model B user and the user individual model B u ) are important inputs for the subsequent steps.

[0121] Step 3: Based on the user model and the responsibility matrix, combined with data sensitivity scoring, scenario feature extraction, and multi-dimensional risk calculation, implement dynamic permission decision-making;

[0122] In this step, based on the responsibility matrix and user behavior models obtained in the previous two steps, combined with data sensitivity and access scenario information, a multi-dimensional risk assessment model is constructed to implement dynamic permission decision-making; specifically, it includes the following sub-steps:

[0123] Sub-step 3.1: Data asset sensitivity assessment;

[0124] Classify and assess the sensitivity of data assets in the industrial big data platform, and construct a data sensitivity scoring model; this model comprehensively considers factors such as data content type (such as personal information, trade secrets, production parameters, etc.), data value, and data impact scope, and calculates the sensitivity score of each type of data. The sensitivity scoring function is expressed as:

[0125] S(d) = w1·S type (d) + w2·S value (d) + w3·S impact (d);

[0126] Where S(d) represents the sensitivity score of data d, and S type (d), S value (d), and S impact (d) represent the scoring functions of the data type, value, and impact scope of data d respectively, and w1, w2, and w3 are the weight parameters of the data type, value, and impact scope.

[0127] Sub-step 3.2: Access scenario feature extraction;

[0128] Extract and analyze the feature information of the data access scenario in real time, including factors such as access time, access location, access device, and network environment; through feature engineering techniques, convert these scenario features into structured feature vectors as the input for risk assessment; the scenario feature vector is expressed as:

[0129]

[0130] Where C scene represents the scenario feature vector, and c1, c2, represent the first, second, nthfeat Values of scenario features, such as time anomaly degree, location anomaly degree, device credibility, etc., n feat represents the scenario feature dimension, and T represents the transpose operator.

[0131] Sub-step 3.3, multi-dimensional risk calculation and decision-making;

[0132] Based on the user behavior model, data sensitivity, and scenario features, construct a risk assessment model. The input for this step includes the RACI responsibility matrix M generated in sub-step 1 RACI , the user behavior model B generated in step 2 u and the behavior portrait, the sensitivity score S(d) generated in step 3.1, and the scenario feature vector C generated in step 3.2 scene ;

[0133] Calculate the risk score of the user's access to specific data in real time:

[0134] R(u, d, c) = f risk (B user , S(d), C scene , M RACI );

[0135] where R(u, d, c) represents the risk score of user u accessing data d in scenario c, B u is the behavior model of user u, S(d) is the data sensitivity of data d, C scene is the scenario feature vector; M RACI is the responsibility matrix, and f risk is the risk score function, implemented as a gradient boosting decision tree model.

[0136] The specific implementation of this gradient boosting decision tree model is as follows:

[0137] Construct a combined feature vector by combining the user behavior features, data sensitivity features, scenario features, and corresponding values of the responsibility matrix into a unified feature vector:

[0138] X = (B u , S(d), C scene , M RACI (u, d));

[0139] where X is the combined feature vector, B u is the behavior feature vector of user u, S(d) is the sensitivity feature vector of data d, C scene is the scenario feature vector, and M RACI (u, d) is the responsibility value of user u for data d in the responsibility matrix.

[0140] Use multiple decision trees for ensemble learning, and the output of each tree is an estimated value of the risk score;

[0141] The m1-th tree can be expressed as:

[0142]

[0143] where is the output value of the m1-th tree, is the number of leaf nodes of tree m1, represents the feature space region corresponding to the j1-th leaf node in the m1-th tree, is the output value of the j1-th leaf node in the m1-th tree, I ind (·) is the indicator function (taking the value of 1 when the condition holds, otherwise 0), denotes the sum over all leaf nodes of tree m1 (from j1 = 1 to ).

[0144] The final prediction of the model is the sum of all trees:

[0145]

[0146] where f risk (X) is the final predicted value of the model, m1 is the index of the current tree, M trees is the total number of decision trees, denotes the sum over the outputs of all M trees trees (from m1 = 1 to m1 = M trees ).

[0147] Train the model using the forward stepwise additive modeling approach:

[0148]

[0149] where is the predicted value of the m1-th tree, is the predicted value of the first m1 - 1 trees, is the predicted value of the m1-th tree.

[0150] For each step, determine the structure of the new tree by minimizing the loss function:

[0151]

[0152] where is the predicted value of the m1-th tree, L is the loss function, and in this embodiment, the mean squared error is used as the loss function, n sample is the number of training samples, is the predicted value of the i1-th sample, is the true label of the i1-th sample, is the feature vector of the i1-th sample, denotes the summation over all n sample training samples (from i1 = 1 to i1 = n sample ), and argmin h denotes finding the function h that minimizes the loss function.

[0153] In practical applications, the model sets different weights for access behaviors of different risk levels. For example, higher weights are assigned to high-risk operations that may lead to data leakage, ensuring that the model is more sensitive to high-risk behaviors.

[0154] For example, in the industrial big data platform of a chemical enterprise, access requests for production formula data are analyzed through this risk assessment model. When a R & D personnel accesses the formula data he is responsible for during normal working hours using the enterprise's secure terminal, the model gives a low-risk score; while when the same R & D personnel tries to download a large amount of sensitive formula data outside his scope of responsibility during non-working hours using non-enterprise equipment, the model will give a high-risk score, triggering permission denial or multi-factor authentication requirements. During the trial operation period of the enterprise, the model successfully identified and blocked 12 potential data leakage risk events, while controlling the false interception rate during daily work below 5%.

[0155] Based on the calculated risk score and the preset risk threshold θ, dynamic permission decisions are made:

[0156]

[0157] where R(u, d, c) is the risk score, A(u, d, c) = 1 indicates granting permission, and A(u, d, c) = 0 indicates denying permission; the risk threshold θ can be adjusted according to organizational security policies and data importance.

[0158] The permission decision result A(u, d, c) and the risk score R(u, d, c) of this step are important outputs. This step also receives the responsibility matrix M RACI provided by step 1 (intelligent analysis of organizational structure) u and the user behavior model B

[0159] provided by step 2 (user behavior modeling)

[0160] as inputs, forming the upstream and downstream dependency relationships of the data flow.

[0159] Step 4, based on the dynamic permission decision, construct a virtual environment for data interaction, and verify the security impact of the authorization policy through simulation testing;

[0160] In this step, a virtual environment for data interaction is constructed, and the effectiveness and security impact of the authorization policy are verified through simulation technology to achieve pre-verification of authorization decisions; specifically, it includes the following sub-steps:

[0161] Sub-step 4.1, construction of the virtual environment for data interaction;

[0162] Based on the real data structure and access pattern of the industrial big data platform, construct a virtual environment for data interaction.

[0163] Input: the RACI responsibility matrix M generated in step 1 RACI , which is used to construct the permission mapping relationship of users in the virtual environment; the user behavior model generated in step 2, which provides the behavior pattern B for generating the virtual user agent U virtual ; class ;

[0164] This environment includes components such as virtual copies of data assets, user simulation agents, and interaction behavior simulators, and can conduct permission policy tests without affecting the actual production system; the formal representation of the virtual environment is:

[0165] E virtual ={D virtual , U virtual , I virtual , P test};

[0166] Where E virtual represents the virtual environment, D virtual represents the set of virtual data assets, U virtual represents the set of virtual user agents, I virtual represents the set of virtual interaction behaviors, P test represents the set of permission policies to be tested; the construction of the virtual environment adopts a representation method based on domain-specific languages, mapping key elements in the real system to the virtual environment.

[0167] Specifically, the specific implementation of the virtual environment includes the following core components:

[0168] Data asset virtualization module, which is responsible for creating virtual copies of real data assets, retaining their structural features and sensitivity attributes, but replacing or masking the actual content to protect privacy; the implementation method is to use data desensitization technology to transform the original data while maintaining the data distribution characteristics:

[0169]

[0170] Where D virtual is the set of virtual data assets, is the virtual data asset, i2 is the index of the virtual data asset, Anonymize() is the data desensitization function, α dsm is the desensitization parameter corresponding to the data type, D real is the set of real data;

[0171] For different types of data, different desensitization strategies are adopted. For example, noise addition processing is applied to numerical data, value mapping replacement is performed on categorical data, and synonym replacement is used for text data, etc.

[0172] The user behavior simulator, based on the user behavior model constructed previously, generates virtual user agents that conform to statistical characteristics and simulates the data access behaviors of different types of users:

[0173]

[0174] Among them, U virtual represents the set of virtual user agents, GenerateAgent() is the agent generation function, B class is the behavior model of the user category, P cfg is the permission configuration of this type of user, and P set is the set of permission configurations.

[0175] The simulator adopts a policy-based behavior generation algorithm to generate access sequences according to the historical behavior patterns of user types, and at the same time adds random mutations to simulate the uncertainties in the real environment.

[0176] The interaction event engine is responsible for simulating the interaction between users and data assets, including operations such as querying, modifying, deleting, and exporting, and records the interaction process and results:

[0177]

[0178] Among them, I virtual represents the set of virtual interaction events, u vir represents the virtual user, d vir represents the virtual data, op represents the operation type, t op represents the operation time, r op represents the operation result, OP represents the set of operation types, T1 represents the time axis, and R result represents the set of operation results;

[0179] The engine adopts the discrete event simulation method to process events in the order of the preset time axis and trigger the corresponding permission checks and operation executions.

[0180] For example, in a data governance system for the power industry, a virtual environment reflecting grid operation data, customer electricity consumption data, and power trading data is constructed. The system simulates the access scenarios of different sensitive-level data by creating user agents with different roles such as virtual dispatchers, analysts, and marketers. For example, the system generates the behavior patterns of dispatchers frequently accessing grid operation parameters during peak load periods, and the behavior patterns of analysts regularly exporting summary reports but not frequently accessing customer detailed information. In this way, new permission policies can be safely tested without exposing the actual data, such as dynamic policies like "temporarily elevating the access rights of dispatchers to specific system parameters during grid emergency situations". The virtual environment successfully simulates 95% of the typical user behavior patterns and 87% of the extreme access scenarios within one year, providing strong support for the optimization of permission policies.

[0181] Sub-step 4.2, permission policy simulation test;

[0182] Simulate the access behaviors of users to data with different sensitivities in the virtual environment, and test the response of the permission decision-making model. This sub-step directly receives the permission decision results A(u, d, c) and risk scores R(u, d, c) in sub-step 3.3 as inputs, and evaluates the model effect by reproducing its operating conditions and parameters. The test scenarios include three categories: regular access scenarios, boundary condition scenarios, and extreme scenarios. Large-scale simulation tests are executed through an automated test framework, and decision results and key metrics are collected. The test process records the logs of each interaction, including information such as access request parameters, decision results, and decision bases, providing data support for subsequent analysis.

[0183] Sub-step 4.3, security impact assessment;

[0184] Based on the simulation test results, construct a security impact assessment model to analyze the impacts of different permission policies on data security and business efficiency. This model uses a method based on causal inference to depict the causal relationships between permission policies, security risks, and business efficiency:

[0185] I impact (P)=I security (P), I efficiency (P);

[0186] I security (P)=E(Y security |do(P));

[0187] I efficiency (P)=E(Y efficiency |do(P));

[0188] where I impact (P) represents the comprehensive impact of the permission policy P, I security(P) and L efficiency (P) represents the impacts on security and efficiency respectively, Y security and Y efficiency are the corresponding result variables, do(P) represents the intervention strategy variable P in the causal model (i.e., setting the variable P to a specific value regardless of its causes), and E(·) represents the expected value operator (calculating the average value of a random variable).

[0189] The specific implementation of this causal inference model is as follows:

[0190] Construct a structural causal model (SCM) to represent the causal relationships between variables:

[0191] SCM = V scm , U scm , F scm , Pr(U scm );

[0192] where SCM represents the structural causal model, V scm is the set of observed variables (including the strategy variable, security indicators, efficiency indicators, etc.), U scm is the set of latent variables (representing unobserved confounding factors), F scm is the set of functions (describing the deterministic relationships between variables), and Pr(U scm ) is the joint distribution of the latent variables.

[0193] Use a causal graph to represent the causal relationships between variables. Nodes in the graph represent variables, and directed edges represent direct causal relationships.

[0194] For evaluating the impact of strategy P on security Y security , calculate through do-calculus (intervention calculus):

[0195]

[0196] where Pr(Y security |do(P)) represents the probability of security Y security after the intervention strategy P, z is the value of Z sec , Z sec represents the set of moderator variables between strategy P and security Y security , such as user behavior characteristics, data sensitivity, etc., and Pr(·) represents probability.

[0197] Similarly, calculate the impact of strategy P on efficiency Y efficiency :

[0198]

[0199] where Pr(Y efficiency|do(P)) represents the efficiency Y after the intervention strategy P efficiency The probability that z′ is Z eff The value of Z eff Represents strategy P and efficiency Y efficiency is the set of adjustment variables between , and Pr(·) represents the probability.

[0200] Calculate the comprehensive impact index of different strategies:

[0201]

[0202] in represents the partial derivative of the comprehensive impact with respect to the strategy (measures the extent to which a small change in the strategy affects the comprehensive impact), and They represent the partial derivatives of security and efficiency impact with respect to the strategy, W security and W efficiency is a weighting factor used to balance the importance of safety and efficiency.

[0203] For example, in the data governance system of a certain intelligent manufacturing enterprise, the model is applied to evaluate the impact of "time-based sensitive formula data access strategy". The system simulates the impact of different policy variables (such as access time limit range, approval process level, etc.) on security indicators (such as data leakage risk rate, number of unauthorized access, etc.) and efficiency indicators (such as data access waiting time, R&D process completion cycle, etc.) in a virtual environment. Through causal analysis, it was found that limiting access to sensitive formula data to working hours on weekdays and cooperating with two-factor authentication can reduce the risk of data leakage by 78%, while the impact on R&D efficiency is only to extend the process time by 8%. This model helps enterprises determine the best balance point and optimize data governance strategies. Through this evaluation model, the risk-benefit ratio of different strategies can be quantified to provide a basis for authority decisions.

[0204] The security impact assessment results of this step I impact (P) is the key output, flowing in two directions:

[0205] As emergency feedback, it flows directly back to sub-step 3.3 in step 3 to adjust the risk threshold θ and decision rules in real time to prevent immediate safety risks;

[0206] As a basis for long-term optimization, it flows to step 5 to evaluate the governance effect. gov Provides reference indicators and the reward function R of the reinforcement learning model t，rl Provides scoring components for security and efficiency.

[0207] This step also receives the responsibility matrix M of step 1 RACI , Step 2 User Behavior Model B userUsing the risk assessment model R(u, d, c) in step 3 as input, synthesize this information to construct a virtual environment and simulation test scenarios.

[0208] Step 5, based on the simulation test results, adaptively adjust the model parameters and dynamically update the initial responsibility matrix to form a closed-loop management;

[0209] This step establishes a feedback mechanism for the effectiveness of the data governance strategy, continuously optimizes the responsibility assignment and permission decision-making model, and realizes the closed-loop management of data governance. It specifically includes the following sub-steps:

[0210] Sub-step 5.1, governance effect monitoring and evaluation;

[0211] Construct a multi-level data governance quality monitoring model, collect and analyze key data governance indicators. The monitoring index system includes three levels: basic indicators (such as the number of permission operations, rejection rate, etc.), process indicators (such as decision response time, user feedback satisfaction, etc.), and result indicators (such as the number of data leakage incidents, compliance score, etc.); based on these indicators, construct a data governance effect evaluation model:

[0212] E gov =f evaluate (I base ,I process ,I result ,W ind );

[0213] Among them, E gov represents the governance effect score, I base 、I process and I result respectively represent the sets of basic indicators, process indicators, and result indicators, and W ind represents the set of indicator weights; f evaluate represents the evaluation function, implemented as a weighted scoring model.

[0214] Sub-step 5.2, adaptive adjustment of model parameters;

[0215] Based on the governance effect evaluation results and user feedback, use the reinforcement learning algorithm to adaptively adjust the parameters of the permission decision-making model. By observing the differences between the actual behavior and predicted behavior of users through the reinforcement learning algorithm, dynamically adjust the parameters of the behavior model to improve the prediction accuracy of the model. The specific objects to be optimized include the weight parameters of the risk scoring function and the risk threshold θ in step three.

[0216] The adjustment process is modeled as a Markov decision process:

[0217]

[0218] Among them, π *Represents the optimal policy function, argmax π Represents finding the optimal policy in the policy space, T represents the time step, γ disc Is the discount factor, Is the reward obtained at time t (based on the governance effect score). The adjusted parameters include the weights in the risk assessment model, the risk threshold, and the update frequency of the responsibility matrix, etc. E(·) represents the expected value operator.

[0219] The specific implementation of this reinforcement learning model is as follows:

[0220] State space definition, state vector s t Contains information such as key indicators of data governance, user behavior pattern distribution, current permission rejection rate, security incident occurrence rate, etc.:

[0221]

[0222] Where s t Represents the state vector, I base，t 、I process，t And I result，t Respectively represent the values of the basic indicator, process indicator, and result indicator at time t, U pattern，t Represents the user behavior pattern distribution, Reject r ate t Represents the current permission rejection rate, Security e vent r ate t Represents the security incident occurrence rate,... represents other indicators.

[0223] Action space definition, action vector a t Contains adjustable parameters, such as the weight coefficients of each feature in the risk assessment model, the risk threshold, etc.:

[0224]

[0225] Where a t Represents the action vector, w1, w2, Respectively represent the 1st, 2nd, n weight th weight parameters, θ is the risk threshold, f update Is the update frequency of the responsibility matrix.

[0226] Reward function design, the reward function combines considerations of both security and efficiency:

[0227]

[0228] Where Represents the reward at time t, Security score t Represents the security score, Efficiency s core t Represents the efficiency score, (Change r ate t ) represents the parameter change rate (used to prevent excessive parameter fluctuations), and λ1, λ2, and λ3 are trade-off parameters.

[0229] Implement the reinforcement learning algorithm using the Deep Q-Network (DQN):

[0230]

[0231] Where Q(s t , a t ) represents the expected cumulative reward for taking action a t in state s t , E(·) represents the expected value operator, γ disc is the discount factor, represent the rewards at time t, t + 1, and t + 2 respectively, s t and a t represent the state and action.

[0232] Use a neural network approximation function to represent the Q-value function:

[0233]

[0234] Where Q(s t , a t ) is the expected cumulative reward for taking action a t in state s t , is a neural network approximation function with parameter θ Q .

[0235] Adopt the ε-greedy strategy to balance exploration and exploitation:

[0236]

[0237] Where a t represents the action vector, ∈ exp is the exploration rate, argmax a Q(s t , a) represents choosing the action with the maximum Q-value in state s t , the random action represents randomly selecting an action, which gradually decreases over time to achieve the transition from exploration to exploitation.

[0238] For example, in the data governance system of a financial institution, the reinforcement learning model is applied to dynamically adjust risk decision-making parameters. The system sets different initial risk weights and access thresholds for different types of data such as credit data, customer information, and transaction records. As the business runs, the system observes that the false rejection rate of customer information data is relatively high (25%), while the authorization for transaction record data is too loose (the proportion of security incidents is 8%). Through the reinforcement learning model, the system automatically increases the sensitivity weight of transaction record data and reduces the risk scoring threshold for accessing customer information in specific scenarios. After three months of adaptive adjustment, the false rejection rate of customer information drops to 7%, the security incidents of transaction records drop to 2%, and the overall data governance efficiency increases by 31%. This model is particularly suitable for the financial environment where the business model is constantly changing and can automatically optimize data governance strategies according to the actual operation effect.

[0239] Sub-step 5.3, the responsibility matrix is updated dynamically;

[0240] According to the organizational changes and the evaluation results of governance effects, the RACI responsibility matrix is updated regularly; the update process includes two parts:

[0241] Forced update based on organizational structure changes to ensure that the responsibility matrix is consistent with the current organizational structure;

[0242] Optimized update based on governance effect evaluation, by analyzing the relationship between the current responsibility assignment and governance effects, to optimize the responsibility assignment plan.

[0243] The responsibility matrix is updated dynamically through the Update function to ensure that the responsibility assignment is consistent with organizational changes and governance effects. The responsibility matrix update function is expressed as:

[0244]

[0245] where and represent the responsibility matrix after and before the update respectively, Update represents the update function, ΔG o represents the change in the organizational structure, and E gov represents the governance effect score.

[0246] This step is the optimization center of the entire data governance method, receiving feedback information from all other steps and outputting optimized parameters and models to each step;

[0247] Sub-step 5.1 receives the decision-making records of step 3 and the security impact assessment result I impact (P), and comprehensively generates the governance effect score E gov ;

[0248] Sub-step 5.2 is based on E gov, optimize the behavior model parameters in step 2 and the risk assessment model parameters in step 2 through the reinforcement learning algorithm;

[0249] Sub-step 5.3 then feeds back the updated responsibility matrix to step 1 to form a closed-loop optimization of responsibility assignment.

[0250] This multi-directional feedback and parameter adjustment mechanism ensures that the entire data governance system can continuously self-optimize and adapt to the dynamic evolution of organizational changes and security requirements.

[0251] Through the above five steps, this embodiment realizes an intelligent data governance method based on an industrial big data platform, solves key technical problems such as unclear responsibility assignment, static and rigid permission management, and lack of pre-event security assessment in traditional data governance, and improves the intelligent level and adaptability of data governance.

[0252] A data governance system based on an industrial big data platform is used to execute the data governance method based on the industrial big data platform as described above, including:

[0253] An organizational structure intelligent analysis module, which is used to generate a responsibility assignment matrix by processing organizational structure documents and business process definitions through text semantic analysis;

[0254] A user behavior modeling module, which is used to analyze the historical data access behavior of users and construct a time series model of user operation patterns;

[0255] A risk perception decision-making module, which is used to make dynamic permission decisions based on data sensitivity scores, scenario characteristics, and multi-dimensional risk calculations;

[0256] A digital twin verification module, which is used to construct a virtual environment for data interaction and verify the security impact of authorization policies;

[0257] An adaptive optimization module, which is used to adjust model parameters and update the responsibility matrix based on the monitoring results of governance effects;

[0258] A processor and a memory, which are used to execute the operations and data storage of the above modules.

[0259] Here, the present invention provides an implementation example:

[0260] Applying this embodiment in the industrial big data platform of a large petrochemical group realizes the intelligent governance of various types of data such as production control data, process recipes, and equipment management data. The application scenario, implementation process, and effect verification will be introduced in detail below.

[0261] This petrochemical group has 10 production bases, 3 R & D centers, and 5 sales branches. The data governance faces the following challenges:

[0262] Complex organizational structure: The group has approximately 1,200 employees in total, covering multiple departments such as production, R & D, sales, operation and maintenance, etc. The unclear division of responsibilities leads to a lack of a large amount of data governance responsibilities;

[0263] Diverse data types and varying sensitivity levels: The platform contains 17 core business data types, as shown in Table 1. Among them, highly sensitive data such as formula data and process parameters account for 33%, and more stringent permission control needs to be implemented;

[0264] Changing access scenarios: The access behaviors of employees vary significantly at different times, locations, and devices. The static permission model is difficult to cope with the dynamically changing access requirements and risks;

[0265] Conflict between data sharing and security: Data needs to be shared among production bases and R & D centers to improve efficiency, but at the same time, there is a risk of data leakage. Under the existing mechanism, the authorization cycle is long and there are problems of over - restriction or blind opening.

[0266] The core data types and sensitivity levels of the petrochemical group's industrial big data platform are shown in Table 1:

[0267] Table 1: Core data types and sensitivity levels of the petrochemical group's industrial big data platform

[0268]

[0269] In this environment, traditional static permission management and role - based access control (RBAC) models are difficult to balance data security and efficient sharing. There is an urgent need for a data governance solution that can dynamically perceive organizational changes, user behaviors, and access risks.

[0270] First, collect materials such as the organizational structure documents, job descriptions, and business process documents of the petrochemical group, including a total of 158 unstructured documents. Extract 245 role entities, 187 responsibility entities, and 423 relationships through text semantic analysis to construct an organizational structure knowledge graph.

[0271] Based on the organizational graph analysis and business process definition, generate a RACI responsibility matrix to clarify the specific responsibilities of each role in data governance activities. The responsibility assignments of some key roles for different data types are shown in Table 2:

[0272] Table 2: Example of RACI responsibility matrix (partial)

[0273]

[0274] Among them: R represents responsible, A represents approve, C represents consult, I represents informed, and - represents irrelevant.

[0275] Collect the user data access logs of the group within two years, which contain approximately 28 million records. After cleaning and preprocessing, a structured user behavior dataset is obtained. Use the deep learning model combining LSTM and attention mechanism in this embodiment to model the operation patterns of 1,324 active users.

[0276] The settings of the key parameters in the model are as follows: the number of LSTM layers is 2, the number of hidden units is 128, the batch size is 64, the learning rate is 0.001, and the number of training epochs is 50. Verify the model performance through sequence prediction tasks, and the F1 score reaches 87.3% on the test set.

[0277] Based on the constructed user behavior model, use the improved K-means clustering algorithm (k = 12) to group the user behavior vectors and identify 12 typical user behavior types.

[0278] First, classify and evaluate the sensitivity of the data assets in the group's industrial big data platform, construct a sensitivity scoring model, and show the sensitivity scoring results of the main data types, as shown in Table 3:

[0279] Table 3: Data Sensitivity Scoring Results

[0280]

[0281] Based on the user behavior model, data sensitivity, and real-time extracted scenario features, use the gradient boosting decision tree model in this embodiment to construct a risk assessment model. The number of decision trees is set to 500, the maximum depth of the tree is 6, and the learning rate is 0.05. The model training uses 100,000 risk assessment samples marked by experts.

[0282] Finally, a two-level risk threshold system is constructed, and different risk decision thresholds are set according to the data sensitivity, as shown in Table 4:

[0283] Table 4: Risk Threshold Settings

[0284]

[0285] Construct a data interaction virtual environment and focus on modeling the process recipes and R & D experimental data of key concern. Simulate the typical user access behaviors within one year in the virtual environment, with a total of approximately 5 million interaction events. Based on the simulation test results, use the causal inference model in this embodiment to analyze the security and efficiency impacts of different permission policies.

[0286] A data governance effect monitoring system including 12 groups of core indicators was established, covering basic indicators, process indicators and result indicators. The parameters of the risk assessment model were adaptively adjusted through a reinforcement learning algorithm. The actual deployment adopted a deep Q-network with a state space dimension of 28 and an action space dimension of 16.

[0287] Among them, the adaptive optimization effect after 6 months of system deployment is shown in Table 5:

[0288] Table 5: Comparison of Adaptive Optimization Effects

[0289]

[0290] After this implementation method was implemented in this petrochemical group, a comprehensive evaluation was carried out through two key technical effect indicators: 1) The precise division and dynamic adjustment effect of data governance responsibilities; 2) The intelligence and dynamic adaptation effect of data permission management.

[0291] After implementing this method, a 12-month follow-up evaluation was carried out on the data governance responsibility allocation situation of the group. The quality and responsiveness indicators of the responsibility matrix are shown in Table 6:

[0292] Table 6: Evaluation of Responsibility Matrix Quality and Responsiveness

[0293]

[0294] The responsibility coverage rate increased from the initial 78.4% to 99.7%, the responsibility allocation conflict rate decreased from 14.6% to 1.2%, the number of missing responsibility items decreased from 325 to only 5, the average update response time shortened from 24.5 days to 4.2 days, and the responsibility allocation satisfaction score increased from 62.3 points to 91.4 points. These indicators jointly confirm the advantages of this method in clarifying responsibility allocation and dynamically responding to organizational changes.

[0295] When there was a major change in the organizational structure (the group carried out department integration in the 8th month), the system could automatically adjust and stabilize the responsibility allocation within 2 weeks, while the traditional manual adjustment method took 2 months and had a higher error rate.

[0296] Through setting up a control group for a comparative experiment, the performance differences of this method and the traditional RBAC permission model in various key indicators were evaluated. The control group adopted the traditional RBAC model, and the experimental group adopted this method. They were run in the same data environment for 12 months. The results are shown in Table 7:

[0297] Table 7: Results of Permission Management Comparative Experiment

[0298] Evaluation metrics Traditional RBAC model This method Relative improvement (%) Average authorization decision time (seconds) 47.56 0.38 -99.2％ Data access rejection rate (%) 22.73 8.21 -63.9％ Abnormal authorization rate (%) 8.47 1.36 -83.9％ Excessive permission problems (items) 685 45 -93.4％ Insufficient permission problems (items) 524 67 -87.2％ Data leakage incidents (times) 7 0 -100％ Authorization satisfaction (points) 67.8 92.6 +36.6％

[0299] Compared with the traditional RBAC model, this method reduces the average authorization decision time from 47.56 seconds to 0.38 seconds, the data access rejection rate from 22.73% to 8.21%, the abnormal authorization rate from 8.47% to 1.36%, and the problems of excessive and insufficient permissions are reduced by 93.4% and 87.2% respectively. It is particularly worth noting that during the implementation period, 7 data leakage incidents occurred in the traditional RBAC model, while this method successfully avoided all leakage risks, and the authorization satisfaction score also increased from 67.8 points to 92.6 points.

[0300] To further verify the dynamic adaptation ability of this method, a variety of special scenario tests were designed, including emergency response, off-site collaboration, and personnel transfer scenarios. The test results show that this method shortens the average response time from 38 minutes to 45 seconds in the emergency scenario, increases the authorization rate from 37% to 92% in the cross-site cooperation scenario, and shortens the permission adaptation time from 2 - 3 weeks to within 3 days in the organizational adjustment scenario, enhancing the adaptation ability of data governance in complex scenarios.

[0301] In summary, the application of this implementation method in the industrial big data platform of the petrochemical group fully verifies its technical effects in the precise division and dynamic adjustment of data governance responsibilities, and the intelligentization and dynamic adaptation of permission management, solves the core technical problems faced by traditional data governance, effectively balances the contradiction between data security and efficient use, and has broad application value and promotion significance.

[0302] The above describes the embodiments of the present invention, but these embodiments are not limited to the above specific implementation manners. The above specific implementation manners are merely illustrative and not restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make more equivalent embodiments in various forms, all of which fall within the protection scope of this embodiment.

Claims

1. A data governance method based on an industrial big data platform, characterized in that, Including: Processing the organizational structure document and business process definition through text semantic analysis technology to generate a responsibility assignment matrix; Analyzing the user's historical data access behavior, constructing a time-series model of the user operation mode, and depicting the user data usage characteristics; According to the user model and the responsibility matrix, combined with data sensitivity scoring, scenario feature extraction, and multi-dimensional risk calculation, realizing dynamic permission decision-making; According to the dynamic permission decision-making, constructing a virtual environment for data interaction, and verifying the security impact of the authorization policy through simulation testing; According to the simulation test results, adaptively adjusting the model parameters, and dynamically updating the initial responsibility matrix to form a closed-loop management.

2. The data governance method based on an industrial big data platform according to claim 1, characterized in that, The steps of processing the organizational structure document and business process definition through text semantic analysis technology to generate a responsibility assignment matrix include: Processing unstructured text, extracting role entities, responsibility entities, and relationships; the unstructured text includes organizational structure documents, job descriptions, and business process documents; Constructing an organizational structure knowledge graph, and calculating node centrality and influence indicators; Combining the graph analysis results and the business process definition to generate a RACI responsibility matrix.

3. The data governance method based on an industrial big data platform according to claim 1, characterized in that, The steps of analyzing the user's historical data access behavior and constructing a time-series model of the user operation mode include: Collecting user data access logs, and generating a structured user behavior data set after cleaning and preprocessing; Organizing the preprocessed behavior data into a time series, and using a deep learning model combining long short-term memory network and attention mechanism for modeling; Conducting cluster analysis on users with similar behavior patterns to identify typical behavior types and patterns.

4. The data governance method based on an industrial big data platform according to claim 1, wherein The steps of combining data sensitivity scoring, scenario feature extraction, and multi-dimensional risk calculation to realize dynamic permission decision-making include: Classifying and evaluating the sensitivity of data assets, and constructing a sensitivity scoring model that comprehensively considers data content type, value, and influence scope; Real-time extracting access scenario features; Multi-dimensional risk calculation and decision-making, based on the user behavior model, data sensitivity, and scenario features, calculating the access risk score, and determining authorization according to the comparison result between the score and the threshold.

5. The data governance method based on an industrial big data platform according to claim 1, characterized in that The steps of constructing a virtual environment for data interaction and verifying the security impact of the authorization policy through simulation testing include: Creating virtual copies of data assets, user simulation agents, and interaction behavior simulators to form a virtual environment; Simulating the user's access behavior to data with different sensitivities in the virtual environment; Analyzing the impact of different permission policies on data security and business efficiency through causal inference methods.

6. The data governance method based on an industrial big data platform according to claim 1, wherein, The steps of adaptively adjusting the model parameters include: Constructing a data governance effect evaluation model including basic indicators, process indicators, and result indicators; Using a reinforcement learning algorithm to adaptively adjust the parameters of the risk assessment model, and modeling its adjustment process as a Markov decision process.

7. The data governance method based on an industrial big data platform according to claim 1, wherein The steps of dynamically updating the initial responsibility matrix include: Forced updating based on organizational structure changes to ensure that the responsibility matrix is consistent with the current organizational structure; Optimized updating based on the governance effect evaluation results, and optimizing the responsibility assignment plan by analyzing the relationship between responsibility assignment and governance effect.

8. The data governance method based on an industrial big data platform according to claim 3, wherein The calculation process of the long short-term memory network includes: f t = σ(W f ·(h t-1 , X t ) + b f ) o t = σ(W o ·(h t-1 , x t ) + b o ); h t = o t ⊙tanh(c t ); Among them f t and o t represent the values of the input gate, forget gate, and output gate respectively, h t-1 and h t represent the hidden states at times t - 1 and t respectively, X t represents the input at time t, c t and c t-1 represent the cell states at times t and t - 1 respectively, is the candidate cell state, ⊙ represents element-wise product, σ is the sigmoid activation function, W i 、W f 、W c 、W o represent the weight matrices of the input gate, forget gate, cell state, and output gate respectively, b i 、b f 、b c 、b o represent the bias terms of the input gate, forget gate, cell state, and output gate respectively, and tanh represents the hyperbolic tangent activation function.

9. The data governance method based on an industrial big data platform according to claim 4, wherein The multi-dimensional risk calculation and decision-making are implemented by a gradient boosting decision tree model, and authorization is determined according to the comparison result between the risk score and the preset threshold. The implementation of the gradient boosting decision tree model includes: Construct a combined feature vector by combining the user behavior features, data sensitivity features, scenario features, and corresponding values of the responsibility matrix into a unified feature vector: X = (B u , S(d), C scene , M RACI (u, d)); where X is the combined feature vector, B u is the behavior feature vector of user u, S(d) is the sensitivity feature vector of data d, C scene is the scenario feature vector, M RACI (u, d) is the responsibility value of user u for data d in the responsibility matrix; where is the output value of the m1-th tree, is the number of leaf nodes of tree m1, represents the feature space region corresponding to the j1-th leaf node in the m1-th tree, is the output value of the j1-th leaf node in the m1-th tree, I ind (·) is the indicator function, means to sum over all leaf nodes of tree m1; The final prediction of the model is the sum of all trees; Determine the structure of the new tree by minimizing the loss function.

10. A data governance system based on an industrial big data platform, characterized in that, For implementing the data governance method based on the industrial big data platform as described in any one of claims 1-9, it includes: An organizational structure intelligent analysis module for generating a responsibility assignment matrix by processing organizational structure documents and business process definitions through text semantic analysis; A user behavior modeling module for analyzing the user's historical data access behavior and constructing a user operation mode time series model; A risk perception decision-making module for making dynamic permission decisions based on data sensitivity scores, scenario features, and multi-dimensional risk calculations; A digital twin verification module for constructing a data interaction virtual environment and verifying the security impact of the authorization policy; An adaptive optimization module for adjusting model parameters and updating the responsibility matrix based on the governance effect monitoring results; A processor and a memory for performing the operations of the above modules and storing data.

Images