Government affair data sharing system based on data security law risk control mode

Abstract

The invention discloses a government affair data sharing system based on a data security law risk control mode, and relates to the technical field of data management. According to the method, full-process closed-loop management from data classification and grading, differential desensitization, law and regulation compliance review, access authority precise control to behavior trajectory real-time analysis and risk trend prediction is realized, and the risk trend prediction is realized by introducing key technologies such as a context sensitive desensitization algorithm, legal knowledge graph reasoning, behavior portraits, semantic trajectory similarity scoring and risk rise rate monitoring. The method effectively improves the recognition and response capability of the hidden data re-recognition risk, and achieves the safe, controllable, compliant and ordered intelligent risk control management of government affair data in the sharing process.

Description

Technical Field

[0001] The present invention relates to the field of data management technology, and in particular to a government data sharing system based on a data security legal risk control model. Background Art

[0002] Government data sharing based on the data security legal risk control model refers to the construction of a management model based on laws and regulations and centered on risk prevention and control in order to ensure data security and compliance in the process of government data sharing. This model identifies and evaluates the legal risks that may be involved in data sharing (such as privacy leakage, unauthorized use, etc.), formulates corresponding institutional norms and technical measures, and ensures that government data can be efficiently circulated during the sharing process without violating relevant data security and legal provisions, thereby achieving the goal of "shareable, controllable, and traceable".

[0003] The prior art has the following deficiencies: In the existing government data sharing practice, there is a hidden re-identification risk problem caused by the aggregation of multi-source heterogeneous data. This problem manifests itself in that when the desensitized data provided by different departments are used together in a sharing platform, attackers can perform aggregate calculations and cross-validation through dimensions such as time, space, and behavioral characteristics, thereby reversely inferring the identity or sensitive characteristics of a specific individual. This type of re-identification behavior does not violate the data provision rules of a single department, but breaks through the desensitization protection boundary at the system level, resulting in the illegal restoration of data in a seemingly safe state, seriously threatening the privacy of the data subject and constituting a potential compliance violation. Summary of the invention

[0004] The purpose of the present invention is to provide a government data sharing system based on a data security legal risk control model to address the deficiencies in the background technology.

[0005] In order to achieve the above-mentioned purpose, the present invention provides the following technical solutions: a government data sharing system based on a data security legal risk control model, comprising a data preprocessing module, a compliance review module, an authority control module, a heterogeneous aggregation protection module, a similarity calculation module and a risk control management module; The data preprocessing module classifies and grades the data provided by various government departments, and uses a context-sensitive desensitization algorithm to perform differential processing on high-risk fields to generate a shareable data copy; The compliance review module, based on the legal knowledge graph, identifies applicable regulations and potential violation risks in the data sharing process and generates a sharing approval strategy; The permission control module sets the access subject permissions based on the principle of least authorization and binds the access behavior to the organizational entity through a cross-domain identity authentication mechanism; The heterogeneous aggregation protection module, in data sharing, performs risk scoring on potential data re-identification behaviors based on behavior profiling and access pattern recognition, and triggers access interruption processing in high-risk situations; Similarity calculation module, for low-risk situations, continuously tracks the behavior trajectory of the access subject, constructs a real-time behavior trajectory vector, and calculates the similarity score between it and the historical access trajectory vector; The risk control management module predicts whether future access behaviors will have high-risk tendencies based on the calculated similarity score and the rate of increase of the risk score, and triggers the risk control response mechanism in advance.

[0006] Preferably, the desensitizing algorithm specifically includes: grouping the fields in the data set according to their sensitivity level and context combination, using differentiated algorithms for different levels, performing reverse identification risk analysis on the desensitized field combination, and automatically triggering a secondary disturbance or field abandonment strategy if a combination is found to still have a high recognition probability after desensitization.

[0007] Preferably, the structural overlap ratio of the fields requested by the visitor in different data sources is analyzed to generate a structural overlap ratio factor: For each data request operation, extract its field set and corresponding data source identifier, and set the user's requested field set in data source A as: ; n is the total number of fields requested by data source A; the set of fields requested by the user in data source B is: ; m is the total number of fields requested by data source B; Standardize or semantically map the field names, use the pre-built field semantic alignment table for matching, and obtain a unified field representation set. and , calculate the structural coincidence scale factor.

[0008] Preferably, the volatility of the semantic target of the field request in the time series is analyzed to generate a semantic consistency offset factor: Each time the visitor requests a collection of fields Mapping to semantic vectors , perform temporal clustering on the vector sequence and establish the semantic cluster center at the initial moment , each time a new , determine whether it belongs to an existing cluster; if it does, update the center of its cluster; if it does not, create a new cluster; at each time point t, calculate the current request vector and the initial cluster center Distance , according to the distance Calculate the semantic consistency offset factor.

[0009] Preferably, the obtained structural coincidence ratio factor and semantic consistency offset factor are normalized and combined into a two-dimensional comprehensive feature vector; Each set of comprehensive feature vectors is used as input, matching the actual annotated dynamic aggregate risk score value labels to construct a training dataset; The training goal is to minimize the sum of squared prediction errors of all samples; Use the gradient descent method to iteratively train the model until the loss function converges, and then terminate the training process; The trained model is used to predict the newly acquired two-dimensional comprehensive feature vector of the access behavior to obtain the dynamic aggregated risk score value of the access behavior.

[0010] Preferably, the obtained dynamic aggregation risk score value is compared with a predetermined threshold. If the dynamic aggregation risk score value is greater than or equal to the predetermined threshold, it indicates that there is a data aggregation risk in the access behavior, and the response mechanism is triggered immediately; if the dynamic aggregation risk score value is less than the predetermined threshold, it indicates that the aggregation risk of the access behavior is low, and continued access to shared data resources is allowed.

[0011] Preferably, the access behavior characteristics of the access subject within a given time window are recorded in real time to construct the current behavior trajectory vector ; Retrieve the behavior trajectory data of the access subject in the past multiple time periods from the historical behavior database to form a historical behavior trajectory vector set ; The cosine similarity is used to calculate the similarity score between the current trajectory vector and each historical vector .

[0012] Preferably, the risk control management module is used to receive the current behavior similarity score output by the similarity calculation module , and the current risk score and the previous score output by the aggregation risk score model, and calculate the risk score increase rate If the calculated similarity score is greater than the set similarity threshold, and the risk score increase rate is greater than the set risk score increase rate threshold, it is determined that the current behavior has a high risk tendency in the future, and the risk control response mechanism is triggered in advance.

[0013] In the above technical solution, the technical effects and advantages provided by the present invention are: 1. The present invention has built a full-process intelligent risk control system including multiple modules such as data preprocessing, compliance review, authority control, heterogeneous aggregation protection, similarity calculation and risk control management. By introducing context-sensitive differentiated desensitization algorithms, legal knowledge graph compliance reasoning, and calculation models of structural overlap ratio factors and semantic consistency offset factors, the system's ability to identify potential data re-identification behaviors and its compliance control capabilities have been effectively improved.

[0014] 2. The present invention uses a combination of similarity scoring and risk score increase rate to build a dynamic aggregation risk prediction mechanism, which can continuously track the evolution trend of access behavior in low-risk situations, warn of high-risk access in advance, and ensure forward-looking security control of the data sharing process. The system not only realizes accurate authority management and dynamic risk control response for government data in the sharing process, but also significantly enhances the robustness of the data desensitization mechanism in a multi-source environment, and has a high practical application value and promotion prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0016] Figure 1 It is a system module diagram of the present invention.

[0017] Figure 2 This is a schematic diagram of the principle of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] For examples, see Figure 1 As shown, the government data sharing system based on the data security legal risk control model described in this embodiment includes a data preprocessing module, a compliance review module, a permission control module, a heterogeneous aggregation protection module, a similarity calculation module and a risk control management module; The data preprocessing module classifies and grades the data provided by various government departments, and uses a context-sensitive desensitization algorithm to perform differential processing on high-risk fields to generate a shareable data copy; The compliance review module, based on the legal knowledge graph, identifies applicable regulations and potential violation risks in the data sharing process and generates a sharing approval strategy; The permission control module sets the access subject permissions based on the principle of least authorization and binds the access behavior to the organizational entity through a cross-domain identity authentication mechanism; The heterogeneous aggregation protection module, in data sharing, performs risk scoring on potential data re-identification behaviors based on behavior profiling and access pattern recognition, and triggers access interruption processing in high-risk situations; Similarity calculation module, for low-risk situations, continuously tracks the behavior trajectory of the access subject, constructs a real-time behavior trajectory vector, and calculates the similarity score between it and the historical access trajectory vector; The risk control management module predicts whether future access behaviors will have high-risk tendencies based on the calculated similarity score and the rate of increase of the risk score, and triggers the risk control response mechanism in advance.

[0020] In the government data sharing system, classifying and grading the data provided by various departments is the basis for ensuring data security and compliant sharing. The steps of data classification and grading include: Field identification and extraction: Parse the data structure provided by each department and extract field metadata (field name, type, source, example value, etc.) as the evaluation object.

[0021] Risk label preset: Build a field label system and preliminarily mark the fields according to national standards (such as GB / T 35273, ISO / IEC27001) as follows: public information (such as department name, business number), internal business (such as budget code, approval flow information), personal information (such as ID number, mobile phone number) and sensitive personal information (such as religious beliefs, health status, location information).

[0022] Impact dimension assessment: For each field, based on its potential impact on individual rights, organizational security, and national interests, it is divided into: Level 1 (highly sensitive), Level 2 (moderately sensitive), Level 3 (generally sensitive), and Level 4 (non-sensitive / public).

[0023] Context rule verification: When there are semantic combinations between fields (such as the combination of "date of birth + postal code + gender" can infer identity), the classification level is re-evaluated and adjusted upward to ensure context sensitivity.

[0024] Grading confirmation and archiving: The grading results of each type of field should be confirmed and recorded by risk control personnel to guide the generation of subsequent desensitization strategies and sharing strategies.

[0025] The specific steps of the desensitization algorithm include: The fields in the dataset are grouped according to their sensitivity level and contextual combination. Joint modeling is performed on field combinations that may result in "re-identification" (such as name + unit + address).

[0026] Differentiated algorithms are used for different levels, for example: Primary fields (such as ID number and mobile phone number): format-preserving encryption (FPE) or homomorphic encryption is used; Secondary fields (such as date of birth): use data generalization (such as only retaining the year of birth); Level 3 and 4 fields: can be processed using lightweight methods such as character masking or data replacement.

[0027] Perform reverse identification risk analysis on the desensitized field combinations. If a combination is found to still have a high identification probability after desensitization (the information entropy is too low), the secondary disturbance or field abandonment strategy will be automatically triggered.

[0028] Generate a desensitized copy of the processed data and attach the desensitization rule metadata (record processing method, field mapping relationship, etc.) for subsequent auditing and restoration tracking.

[0029] Perform business availability testing on desensitized copies to ensure that data can still be used for statistics, analysis, policy modeling, etc. without exposing privacy.

[0030] The desensitized copy is hash-signed and version-marked to ensure its integrity and uniqueness, and finally output to the sharing platform.

[0031] The compliance review module in this system is based on the legal knowledge graph, automatically identifies applicable regulations and potential violation risks in the process of government data sharing, and dynamically generates sharing approval strategies, mainly including: Semantic analysis of data requests: The compliance module performs semantic analysis on the received data sharing requests to extract key elements in the request content, such as field type, data source, target department, purpose of use, data flow, etc.

[0032] Legal knowledge graph call: Map the above elements to the embedded legal knowledge graph. The graph pre-integrates the clause entities, applicable conditions and penalty nodes of various industry regulatory rules, and forms a graph structure of "data types, legal obligations and liability consequences".

[0033] Regulatory matching and conflict determination: The module uses graph path reasoning to determine the set of applicable regulatory clauses involved in the data sharing request and marks whether there is a conflict (for example, the data type is restricted and the recipient does not have the legal processing conditions).

[0034] Risk behavior assessment: Based on the regulatory matching results, combined with data sensitivity and sharing purpose, high-risk behaviors (such as unauthorized cross-border, lack of data minimization principle, irrevocable authorization, etc.) are identified and marked as red alerts.

[0035] Automatic generation of approval strategies: Based on the compliance assessment results, the system generates a personalized approval strategy template, including the approval process chain, required signatures of responsible persons, additional security measures (such as strong desensitization, time-limited access), etc., for system calls or human review intervention.

[0036] The permission control module in this system sets the data permissions of the access subject according to the principle of least authorization, and binds the access behavior to the organizational entity through the cross-domain identity authentication mechanism to achieve trusted identification and access control of data requests, including: Subject identification and role modeling: After receiving a data access request, the system identifies the identity information of the access subject, including organization code, user ID, business system source, etc., and performs role modeling on it to establish a ternary binding relationship of "organization-user-authority role".

[0037] Minimum authorization policy matching: The permission control module calls the permission policy library and dynamically matches the minimum authorization policy based on the subject role and data classification and grading results, and only opens the minimum set of fields, time windows, and operation permissions required for access (such as read-only, desensitized viewing, and audit visibility).

[0038] Cross-domain authentication mechanism: Adopt a cross-domain authentication mechanism based on OAuth 2.0 or SAML protocol to complete the identity authentication of the access subject and generate an access token with a digital signature for subsequent request behavior tracking and permission verification.

[0039] Behavior binding and identifier generation: The system binds identity information, permission scope, and access context to generate a unique trusted access identifier (Trusted Access Token). This identifier uniquely corresponds to an access behavior throughout the entire data sharing lifecycle, ensuring that the calling behavior has a verifiable source.

[0040] Access verification and response control: Each time a data request is triggered, the system verifies whether the access identifier matches the permission policy. If the match is successful, access to the corresponding data copy is allowed; if the permission exceeds the limit or the access is abnormal, the request is automatically rejected and the audit log is recorded.

[0041] The heterogeneous aggregation protection module is used to prevent multi-source data from being illegally spliced ​​and re-identified. During the data sharing process, the access behavior is aggregated and scored for risk through behavioral profiling and access pattern recognition technology, and dynamic blocking is performed when the risk is too high. Specifically, it includes: Collect the historical calling behavior, field request combination, time interval, path frequency and other features of the access subject, and build a multi-dimensional behavior portrait model to identify potential aggregation tendencies.

[0042] The structural overlap ratio of the fields requested by the visitor in different data sources is analyzed to generate a structural overlap ratio factor to determine whether there is an intention to splice information.

[0043] For example, the present invention can use the following method to calculate the structural overlap ratio factor: For each data request operation, extract its field set and corresponding data source identifier, assuming: The user requests the following field set in data source A: ; n is the total number of fields requested by data source A; The user requests the following field set in data source B: ; m is the total number of fields requested by data source B; Standardize or semantically map the field names (e.g., consider "ID Number" vs "ID Number" as equivalent fields), and use the pre-built field semantic alignment table for matching to obtain a unified field representation set. The standardized set is and , calculate the structural overlap scale factor SOF, the calculation expression is: .

[0044] For example, the fields of data source A are: name, ID number, mobile phone number, and home address, a total of 4 fields; the fields of data source B are: name, ID number, marital status, and education level, a total of 4 fields; after standardization, the overlapping fields are: name, ID number (a total of 2); then: SOF==0.5.

[0045] SOF is close to 1: the field structure is highly overlapped, and it is very likely to be used for information splicing; SOF is lower than 0.3: the structure overlap is low, and the risk of splicing is small. A dynamic threshold (such as 0.6) can be set as a risk control interception basis and embedded in the access risk scoring system.

[0046] Analyze the volatility of semantic targets of field requests in time series and generate semantic consistency deviation factors. For example, if access behavior gradually shifts from business approval to demographic attribute information, it reflects the abnormal evolution of data intent. For example, the present invention can calculate the semantic consistency offset factor by the following method: Each time the visitor requests a collection of fields Mapping to semantic vectors , the vector is constructed as follows: Encode fields using pre-trained language models (e.g., BERT, Word2Vec); Take the average or weighted sum of multiple fields in the request to form a single vector .

[0047] Use incremental clustering algorithms (such as online k-means, DBSCAN) to cluster vector sequences Perform temporal clustering: Establishing semantic cluster center at the initial moment , each time a new , determine whether it belongs to an existing cluster; if it does, update the center of its cluster; if it does not, create a new cluster.

[0048] At each time point t, calculate the current request vector and the initial cluster center Distance (such as Euclidean distance): ; Record time series , T is the total number of requests in the current time window, and the semantic consistency offset factor SDI is calculated as follows: .

[0049] If SDI < 0.2: the semantic behavior is stable and the consistency is strong; 0.2 ≤ SDI < 0.5: there is a slight semantic deviation, which may be a normal business expansion; SDI ≥ 0.5: the semantic target drifts significantly, which may involve implicit splicing or data exploration behavior; a composite scoring mechanism can be constructed in conjunction with the structural overlap factor (SOF).

[0050] The obtained structural coincidence ratio factor and semantic consistency offset factor are normalized and combined into a two-dimensional comprehensive feature vector; Take each set of comprehensive feature vectors as input, match the actual annotated dynamic aggregate risk score value label (which can be obtained from historical events or manual annotation), and construct a training data set; The training goal is to minimize the sum of squared prediction errors of all samples; The model is iteratively trained using the gradient descent method until the loss function converges (that is, the sum of the prediction errors reaches the set threshold), and the training process is terminated; The trained model is used to predict the newly acquired two-dimensional comprehensive feature vector of the access behavior to obtain the dynamic aggregated risk score value of the access behavior.

[0051] The obtained dynamic aggregation risk score value is compared with the preset threshold. If the dynamic aggregation risk score value is greater than or equal to the preset threshold, it means that the access behavior has a significant data aggregation risk, and the system will trigger response mechanisms such as access interruption, risk control alarm or dynamic permission downgrade; if the dynamic aggregation risk score value is less than the preset threshold, it means that the access behavior has a low aggregation risk and continued access to shared data resources is allowed.

[0052] The similarity calculation module is used to continuously monitor access subjects in a low-risk state during the government data sharing process, and to identify potential repeated calls or abnormal behavior evolution by analyzing their access behavior trajectories.

[0053] The system records the access behavior characteristics of the access subject in a given time window in real time, including the access field set, data source path, request frequency, semantic category, etc.; vectorizes the above information to construct the current behavior trajectory vector .

[0054] The behavior trajectory data of the access subject in the past multiple time periods are retrieved from the historical behavior database, and normalized in the same dimension to form one or more historical behavior trajectory vector sets. .

[0055] Use cosine similarity (or inverse Euclidean distance transformation) to calculate the similarity score between the current trajectory vector and each historical vector : .

[0056] like If the similarity is higher than the set threshold (such as 0.85), the module will mark the access behavior as a risk candidate for repeated behavior and send it to the aggregation risk assessment module for further comprehensive judgment based on structural overlap and semantic deviation trends to determine whether to trigger warnings or restrictive measures.

[0057] The risk control management module is used to receive the current behavior similarity score output by the similarity calculation module , and the current risk score output by the aggregate risk scoring model and the previous moment score , calculate the risk score increase rate , the expression is: ; The risk control management module sets joint judgment conditions: If the calculated similarity score is greater than the set similarity threshold, and the risk score increase rate is greater than the set risk score increase rate threshold, it is determined that the current behavior has a high-risk tendency in the future, that is, there may be behavioral risks such as data duplication, re-identification, and abuse.

[0058] Once the prediction result is a tendency towards high-risk behavior, the system will suspend data access channels in advance, downgrade access permissions, and notify risk control personnel for review. The current access behavior will be marked as a predictive risk control event and written into the behavior log and audit system to provide data support for subsequent risk model training and tuning.

[0059] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0060] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0061] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A government data sharing system based on a data security legal risk control model, characterized by: It includes data preprocessing module, compliance review module, permission control module, heterogeneous aggregation protection module, similarity calculation module and risk control management module; The data preprocessing module classifies and grades the data provided by various government departments, and uses a context-sensitive desensitization algorithm to perform differential processing on high-risk fields to generate a shareable data copy; The compliance review module, based on the legal knowledge graph, identifies applicable regulations and potential violation risks in the data sharing process and generates a sharing approval strategy; The permission control module sets the access subject permissions based on the principle of least authorization and binds the access behavior to the organizational entity through a cross-domain identity authentication mechanism; The heterogeneous aggregation protection module, in data sharing, performs risk scoring on potential data re-identification behaviors based on behavior profiling and access pattern recognition, and triggers access interruption processing in high-risk situations; Similarity calculation module, for low-risk situations, continuously tracks the behavior trajectory of the access subject, constructs a real-time behavior trajectory vector, and calculates the similarity score between it and the historical access trajectory vector; The risk control management module predicts whether future access behaviors will have high-risk tendencies based on the calculated similarity score and the rate of increase of the risk score, and triggers the risk control response mechanism in advance.

2. According to claim 1, a government data sharing system based on a data security legal risk control model is characterized by: The desensitization algorithm specifically includes: grouping the fields in the data set according to their sensitivity level and context combination, using differentiated algorithms for different levels, and performing reverse identification risk analysis on the desensitized field combination. If it is found that a combination still has a high recognition probability after desensitization, a secondary disturbance or field abandonment strategy is automatically triggered.

3. According to claim 1, a government data sharing system based on a data security legal risk control model is characterized by: Analyze the structural overlap ratio of the fields requested by the visitor in different data sources and generate the structural overlap ratio factor: For each data request operation, extract its field set and corresponding data source identifier, and set the user's requested field set in data source A as: ; n is the total number of fields requested by data source A; the set of fields requested by the user in data source B is: ; m is the total number of fields requested by data source B; Standardize or semantically map the field names, use the pre-built field semantic alignment table for matching, obtain a unified field representation set, and calculate the structural overlap ratio factor based on the standardized set.

4. A government data sharing system based on a data security legal risk control model according to claim 3, characterized in that: The volatility of the semantic target of field requests in the time series is analyzed to generate the semantic consistency offset factor: Each time the visitor requests a collection of fields Mapping to semantic vectors , perform temporal clustering on the vector sequence and establish the semantic cluster center at the initial moment , each time a new , determine whether it belongs to an existing cluster; if it does, update the center of its cluster; if it does not, create a new cluster; at each time point t, calculate the current request vector and the initial cluster center Distance ; According to the distance Calculate the semantic consistency offset factor.

5. A government data sharing system based on a data security legal risk control model according to claim 4, characterized in that: The obtained structural coincidence ratio factor and semantic consistency offset factor are normalized and combined into a two-dimensional comprehensive feature vector; Each set of comprehensive feature vectors is used as input, matching the actual annotated dynamic aggregate risk score value labels to construct a training dataset; The training goal is to minimize the sum of squared prediction errors of all samples; Use the gradient descent method to iteratively train the model until the loss function converges, and then terminate the training process; The trained model is used to predict the newly acquired two-dimensional comprehensive feature vector of the access behavior to obtain the dynamic aggregated risk score value of the access behavior.

6. A government data sharing system based on a data security legal risk control model according to claim 5, characterized in that: Compare the obtained dynamic aggregation risk score with the preset threshold. If the dynamic aggregation risk score is greater than or equal to the preset threshold, it means that the access behavior has data aggregation risk, and the response mechanism is triggered immediately. If the dynamic aggregate risk score value is less than the predetermined threshold, it indicates that the aggregate risk of the access behavior is low, and access to the shared data resources is allowed to continue.

7. The government data sharing system based on the data security legal risk control model according to claim 1 is characterized by: Record the access behavior characteristics of the access subject in a given time window in real time and construct the current behavior trajectory vector; The behavior trajectory data of the access subject in the past multiple time periods are retrieved from the historical behavior database to form a set of historical behavior trajectory vectors; the cosine similarity is used to calculate the similarity score between the current trajectory vector and each historical vector.

8. A government data sharing system based on a data security legal risk control model according to claim 7, characterized in that: The risk control management module is used to receive the current behavior similarity score output by the similarity calculation module, as well as the current risk score and the score at the previous moment output by the aggregated risk scoring model, and calculate the risk score rising rate. If the calculated similarity score is greater than the set similarity threshold, and the risk score rising rate is greater than the set risk score rising rate threshold, it is determined that the current behavior has a high risk tendency in the future, and the risk control response mechanism is triggered in advance.

Images