Data intelligent classification driven safe transmission shared storage collaborative control method and system

Abstract

The application discloses a data intelligent classification driving safe transmission shared storage cooperative control method and system, and belongs to the technical field of data security management and data management. In view of the problems that classification and grading in the existing data security scheme depend on manual work, transmission and shared storage link control is fragmented, and static strategy is difficult to adapt to dynamic scenes, the application extracts multi-dimensional features of target data and business context information for classification and grading, generates a cooperative control strategy containing transmission control parameters, shared control parameters and storage control parameters according to the classification and grading results, respectively executes safe transmission control, safe sharing control and safe storage control, and triggers dynamic updating of the classification and grading results and the cooperative control strategy when an abnormal event is detected. The application can improve the data classification and grading efficiency and accuracy, enhance the consistency and dynamic adaptation capability of each link control in the whole life cycle of data, and is suitable for government affairs, medical treatment, finance, industrial internet and enterprise data management scenes.

Description

Technical Field

[0001] This invention relates to the field of data security governance and data management technology, and in particular to a data intelligent classification-driven secure transmission, shared storage, and collaborative control method and system. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] With the continuous advancement of digital transformation, a large amount of structured, semi-structured, and unstructured data has been generated in scenarios such as government affairs, healthcare, finance, industrial internet, and enterprise collaborative office. During the collection, transfer, sharing, archiving, and long-term storage of this type of data, it is generally necessary to balance data availability, sharing efficiency, and security. Therefore, higher requirements are placed on the collaborative protection of data throughout its entire lifecycle of transmission, sharing, and storage.

[0004] In existing technologies, data security control typically consists of multiple independent stages. For example, the transmission stage mainly uses link encryption or channel authentication to protect data transmission; the sharing stage relies primarily on manual approval, role-based access control, or static access control policies; and the storage stage typically uses disk encryption, file encryption, or access isolation for protection. While these solutions can provide some protection at a single stage, the lack of unified judgment criteria and linkage mechanisms between different stages can easily lead to inconsistent security policies being applied to the same target data at different processing stages, making it difficult to achieve true end-to-end collaborative control.

[0005] Furthermore, existing solutions still rely heavily on manual labeling, static rules, or fixed directory structures for identifying data categories and sensitivity levels. This approach suffers from low classification efficiency, high subjectivity, delayed updates, and insufficient dynamic adaptability when dealing with large-scale, diverse, and complex data scenarios. Especially during data sharing, the data's access subjects, usage purposes, terminal environments, and network environments may change. If static security policies configured at the time of data creation are still used, insufficient or excessive security protection can easily occur, impacting data flow efficiency and management effectiveness. Therefore, how to generate collaborative control strategies adapted to different stages of transmission, sharing, and storage based on the intelligent classification results of target data, and dynamically adjust these strategies when business environments or access behaviors change, has become a pressing issue in this field. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a data intelligent classification-driven secure transmission, sharing, and storage collaborative control method and system. By extracting multi-dimensional features, intelligently classifying and grading target data and its business context information, and conducting risk assessments, it automatically generates transmission control parameters, sharing control parameters, and storage control parameters adapted to the classification and grading results, achieving coordinated consistency of security strategies across transmission, sharing, and storage. By extracting multi-dimensional features from target data and business context information for classification and grading, it achieves automatic identification of data categories and sensitivity levels, overcoming the problems of low efficiency and strong subjectivity associated with manual classification. Furthermore, by triggering dynamic updates to the classification and grading results and collaborative control strategies upon detecting abnormal events, it addresses the inability of static strategies to adapt to dynamic business environments, thereby improving the security, adaptability, and scientific management of collaborative control throughout the data lifecycle.

[0007] On the one hand, a data intelligent classification-driven secure transmission and shared storage collaborative control method is provided, including: Obtain the target data and the corresponding business context information; Extract multidimensional features from the target data and the business context information, and classify and grade the target data based on the multidimensional features to obtain classification and grading results; Based on the classification and grading results, a collaborative control strategy for the target data is generated from a preset strategy mapping relationship. The collaborative control strategy includes transmission control parameters, sharing control parameters, and storage control parameters. Based on the transmission control parameters, secure transmission control is performed on the target data; based on the sharing control parameters, secure sharing control is performed on the target data; based on the storage control parameters, secure storage control is performed on the target data, and when an abnormal event is detected, updates to the classification and grading results and the collaborative control strategy are triggered.

[0008] Furthermore, the multidimensional features include data content features and metadata features extracted from the target data, as well as source features, access subject features, and access environment features extracted from the business context information.

[0009] Furthermore, the target data is classified and graded based on the multidimensional features to obtain the classification and grading results, specifically as follows: Based on preset sensitivity rules, the target data is matched to obtain an initial classification result; Based on the trained classification model, the target data is classified and risk assessed to obtain the model classification result. The initial classification result is fused with the model classification result to obtain data category labels, sensitivity level labels, and risk scores, which serve as the classification and grading results.

[0010] Furthermore, the formula for calculating the risk score is as follows: ; in, For the first Risk contribution value corresponding to class feature For the first The weights corresponding to the class features The number of feature categories participating in the risk assessment, and .

[0011] Furthermore, based on the classification and grading results, a collaborative control strategy for the target data is generated from the preset strategy mapping relationship, specifically as follows: The control level to which the target data belongs is determined based on the risk score in the classification and grading results; According to the control level, a target policy template is selected from the preset policy template library. If the control level is low, the standard encrypted transmission template, standard shared approval template, and ordinary encrypted storage template are selected. If the control level is medium, the enhanced encrypted transmission template, domain-limited sharing template, and isolated encrypted storage template are selected. If the control level is high, the high-strength encrypted transmission template, strictly time-limited and frequency-limited sharing template, and fragmented isolated storage template are selected. Based on the access subject attribute, terminal environment attribute, and business purpose attribute in the business context information, the target policy template is modified to obtain the collaborative control policy.

[0012] Furthermore, based on the transmission control parameters, secure transmission control is performed on the target data; based on the sharing control parameters, secure sharing control is performed on the target data; and based on the storage control parameters, secure storage control is performed on the target data, specifically as follows: Based on the transmission control parameters, two-way authentication is performed on the sending end and the receiving end, a session key is generated and the target data is symmetrically encrypted using the session key, the session key is encapsulated using the receiving end's public key, an integrity check value is generated for the encrypted target data and sent, and the target data is restored after the receiving end completes authentication, session key decapsulation and integrity check. Based on the shared control parameters, the system obtains the request subject's identity information, terminal information, access time information, and business purpose information corresponding to the shared request. The sharing conditions are divided into mandatory conditions and compensable conditions, and matched with the obtained information. When both mandatory and compensable conditions are met, an authorization token is generated, and a time limit or access limit is imposed on the shared data. When the mandatory conditions are met, one or more compensable conditions are not met, and the control level of the target data is not higher than the medium-level control and conforms to the preset downgraded sharing rules, the system switches to a de-identified sharing mode, performing dynamic de-identification and watermark embedding on the shared data. When any mandatory condition is not met, the target data belongs to the high-level control, or does not conform to the preset downgraded sharing rules, the sharing request is rejected. Based on the storage control parameters, differentiated storage is performed according to the sensitivity level of the target data. For low sensitivity levels, encrypted storage is performed; for medium sensitivity levels, encrypted storage and logically isolated storage are performed; and for high sensitivity levels, fragmented storage, independent key domain management, and physical or namespace isolated storage are performed.

[0013] Furthermore, the dynamic desensitization includes at least one of field masking processing, data range generalization processing, identifier replacement processing, and sensitive segment deletion processing; the watermark embedding includes at least one of user identifier watermark, time identifier watermark, and scene identifier watermark.

[0014] Furthermore, upon detecting an abnormal event, an update to the classification and grading results and the collaborative control strategy is triggered, specifically as follows: An audit log records the transmission, sharing, access, and decryption behaviors related to the target data; Detect abnormal events, which include at least one of the following: the number of authentication failures exceeds a threshold within a preset time window, the frequency of target data downloads exceeds a threshold, the scope of shared objects expands, the trustworthiness of the access terminal is lower than a threshold, and the business scenario changes. After the abnormal event is detected, the risk score is updated, and the updated risk score satisfies: ; Where R is the risk score before the abnormal event occurs, R′ is the updated risk score, E is the intensity of the abnormal event, B is the deviation of the access behavior, and α and β are adjustment coefficients; the values ​​of R, R′, E, B, α and β all range from 0 to 1. The classification and grading and collaborative control strategy generation are re-executed based on the updated risk score to update the classification and grading results and the collaborative control strategy.

[0015] On the other hand, a data intelligent classification-driven secure transmission, shared storage, and collaborative control system is provided, including: The data acquisition module is configured to acquire target data and business context information corresponding to the target data. The classification and grading module is configured to: extract multi-dimensional features of the target data and the business context information, classify and grade the target data based on the multi-dimensional features, and obtain classification and grading results; The strategy generation module is configured to: generate a collaborative control strategy for the target data from a preset strategy mapping relationship based on the classification and grading results; the collaborative control strategy includes transmission control parameters, sharing control parameters, and storage control parameters. The transmission control module is configured to perform secure transmission control on the target data based on the transmission control parameters. The shared control module is configured to perform secure shared control on the target data based on the shared control parameters. The storage audit module is configured to: perform secure storage control on the target data based on the storage control parameters, and trigger updates to the classification and grading results and the collaborative control strategy when an abnormal event is detected.

[0016] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the method described in the first aspect.

[0017] The above technical solution has the following advantages or beneficial effects: (1) This invention uses the results of intelligent data classification and grading as a unified entry point, and integrates data content features, metadata features, source features, access subject features and access environment features for risk assessment. This can reduce the subjectivity and lag of manual classification and improve the efficiency and accuracy of data classification and grading.

[0018] (2) The present invention establishes a collaborative mapping relationship from classification and grading results to transmission control parameters, sharing control parameters and storage control parameters, so that the target data adopts a consistent and interconnected security control strategy in the transmission, sharing and storage links, avoiding the problem of inconsistent protection caused by the fragmented configuration of different security links.

[0019] (3) The present invention introduces mechanisms such as dynamic desensitization, watermark embedding, authorization token, time limit and access number limit in the sharing stage, which can controllably send target data out according to the sharing subject, access environment and business purpose, thereby improving the security and traceability of the data sharing process.

[0020] (4) The present invention introduces an abnormal event detection and dynamic update mechanism in the storage and auditing stages. When an authentication failure, download, sharing range or terminal trust is detected, the risk score can be recalculated and the policy upgrade can be triggered, thereby improving the dynamic adaptability of the target data security protection throughout the entire life cycle and the scientific nature of data governance. Attached Figure Description

[0021] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0022] Figure 1 This is a flowchart of the method in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the classification and grading and collaborative strategy generation process in Embodiment 1 of the present invention; Figure 3 This is a flowchart of the abnormal event triggering dynamic adjustment process in Embodiment 1 of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings. Those skilled in the art should understand that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0024] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0025] Example 1 This embodiment provides a data intelligent classification-driven secure transmission, shared storage, and collaborative control method. Figure 1 This is an overall flowchart of the method according to Embodiment 1 of the present invention. The method includes the following steps: S1: Obtain the target data and the corresponding business context information; S2: Extract multi-dimensional features of the target data and business context information, classify and grade the target data based on the multi-dimensional features, and obtain the classification and grading results; S3: Based on the classification and grading results, generate a collaborative control strategy for the target data from the preset strategy mapping relationship. The collaborative control strategy includes transmission control parameters, shared control parameters, and storage control parameters. S4: Based on transmission control parameters, perform secure transmission control on the target data; based on sharing control parameters, perform secure sharing control on the target data; based on storage control parameters, perform secure storage control on the target data, and trigger updates to the classification and grading results and collaborative control strategies when an abnormal event is detected.

[0026] In step S1, the target data and business context information are first acquired and preprocessed. In this embodiment, the target data can be any one or more combinations of structured data, semi-structured data, and unstructured data. For example, the target data can be database records, form data, business messages, electronic documents, images, log files, configuration files, contract texts, medical records, or financial data, etc.

[0027] Business context information may include, but is not limited to: source system identifier, data creator identity, business domain, organizational department, purpose of data use, access terminal type, network environment type, time window information, and historical access behavior information.

[0028] Preprocessing includes at least one of the following: format standardization, field parsing, text segmentation, data deduplication, metadata association, and abnormal character cleaning. For structured data, field extraction and field mapping can be performed; for unstructured text, word segmentation, entity recognition, and keyword extraction can be performed; for file-type data, attributes such as file format, file size, extension, and directory path can be extracted.

[0029] like Figure 2 As shown, in step S2, based on the data acquisition and preprocessing completed in step S1, step S2 extracts multidimensional features from the preprocessed target data and performs intelligent classification and grading.

[0030] Specifically, data content features and metadata features are extracted from the preprocessed target data, while source features, access subject features, and access environment features are extracted from the business context information. For data content feature extraction, differentiated processing methods are used for different data types: for structured data, content features are extracted through field parsing, field type identification, regular expression matching, and field statistics; for unstructured text, content features are extracted through text segmentation, keyword extraction, and entity recognition. Metadata features are obtained through file attributes, database dictionaries, or data management interfaces. Source features, access subject features, and access environment features are obtained through source system logs, identity and permission systems, terminal trust detection results, and network access logs, respectively. The data content features include keyword features, field distribution features, file format features, and data structure features; metadata features include creation time, modification time, file size, business identifier, and storage path; source features include source system identifier and creator identity identifier, used to trace the source of data generation and the responsible party; access subject features include visitor role identifier and permission domain identifier, used to clarify the permission boundaries of data users; access environment features include terminal trustworthiness identifier, network environment identifier, and business purpose identifier, used to assess the security level of the data access scenario.

[0031] The above features together constitute a multi-dimensional feature set for classification and grading, providing a comprehensive data foundation for subsequent risk quantification.

[0032] Based on the extracted features, this embodiment classifies and grades the target data to obtain classification and grading results including data category labels, sensitivity level labels, and risk scores.

[0033] Specifically, the classification and grading process may include the following sub-steps: Step S2-1: Perform rule matching on the target data based on preset sensitivity rules to obtain initial classification results. Preset sensitivity rules may include at least one of the following: field matching rules, keyword dictionary rules, regular expression rules, file format rules, business directory rules, and tag inheritance rules. For example, if the target data matches the ID card number field, bank card number field, patient identification field, or internal project number field, the candidate sensitivity level value of the target data can be increased accordingly.

[0034] Step S2-2 involves classifying and assessing the risk of the target data based on the trained classification model to obtain the model classification result. This classification model can be any of the following: decision tree model, random forest model, gradient boosting model, deep neural network model, or text semantic recognition model; this invention does not specifically limit its application. The model input consists of the multidimensional features extracted and standardized in step S1, and the output includes the data category probability distribution, the sensitivity level probability distribution, and the model risk probability corresponding to each feature category. The model risk probability serves as one of the inputs for subsequent calculations of the risk contribution value.

[0035] Steps S2-3 involve fusing the initial classification results with the model classification results to output data category labels, sensitivity level labels, and risk scores.

[0036] This embodiment employs a weighted fusion strategy to comprehensively determine the two classification results. Before performing classification, the system establishes a candidate data category set based on a preset data classification directory. The data classification directory can be configured by business administrators according to applicable data classification specifications and business domains, with each directory item corresponding to a unique data category identifier. All data category identifiers enabled in the current business scenario constitute the candidate data category set C, where K is the number of candidate data categories. For example, candidate data categories may include personal information, financial, medical, R&D materials, and business management categories.

[0037] The preset set of sensitive rules contains M rules, each of which is pre-configured with a target data category. ∈C, Sensitivity Level ∈G, rule confidence weight ∈[0,1] and the rule matching condition, where G is the set of sensitivity levels defined later. Let h j This indicates the hit flag of the j-th rule for the current target data. When a hit occurs... =1, otherwise =0.

[0038] For any candidate category c, accumulate the confidence weights of the rules that hit and point to that category, and obtain the rule category probability through normalization: ; in, The rule matching branch determines the probability of classifying the target data as candidate category c; M is the preset number of sensitive rules. The data category pre-associated for the j-th rule; Let be the confidence weight of the j-th rule; is the hit identifier for the j-th rule; I(·) is the indicator function, which takes the value 1 if the condition in parentheses is true, and 0 otherwise; C is the set of candidate data categories formed by the preset data classification catalog; k is the category index used when summing the set of candidate data categories; ε is a smoothing constant to prevent the denominator from being 0.

[0039] The classification model, based on the multidimensional features extracted and standardized in step S1, outputs an unnormalized class score for each candidate class in the candidate data class set C. .

[0040] When the classification model directly outputs normalized probabilities, use those probabilities directly; when the output is an unnormalized class score, use the Softmax function to convert it into model class probabilities. ; in, The probability of classifying the target data as candidate class c in a classification model; This represents the unnormalized score output by the classification model for the corresponding candidate category c. Let be the unnormalized score of candidate category k; e is the natural constant; the denominator is the sum of all categories in the candidate data category set C.

[0041] After obtaining the rule category probability and the model category probability, the comprehensive category score is obtained by weighting and fusing them as follows: ; in, Give a comprehensive category score to candidate category c; and Let be the probability of the rule category and the probability of the model category, respectively; λ∈[0,1] is the fusion weight of the rule classification result, 1 λ represents the fusion weights of the model's classification results. The fusion weights λ and the rule confidence weights q are... j The smoothing constant ε can be preset or calibrated based on the validation samples.

[0042] For example, λ=0.60 and ε=10 can be chosen. -6 The rule confidence weights for mandatory sensitive rules, high-confidence rules, and general rules can be set to 1.00, 0.80, and 0.50, respectively. When the target data matches multiple rules simultaneously, the results are accumulated and normalized according to the above formula; when no rules are matched, λ is set to 0, and the fusion result is determined by the model's class probability.

[0043] Based on the above weighted fusion formula, the comprehensive category score will be... The category with the largest value is output as the data category label.

[0044] For determining the sensitivity level label, this embodiment converts the rule matching results and classification model output into a unified sensitivity score S. The system pre-establishes a sensitivity level set G={g1, g2, g3}, representing low sensitivity level, medium sensitivity level, and high sensitivity level, respectively, and sets the corresponding numerical mappings v(g1)=0.20, v(g2)=0.60, and v(g3)=1.00.

[0045] Let J h The set of rule numbers that match the current target data, and the rule sensitivity score S. Determined according to the following formula: ; Among them, g j The sensitivity level is pre-associated with the j-th hit rule. Let j be the confidence weight of the j-th rule. This represents the numerical value corresponding to the sensitivity level; when no rule is hit, the rule sensitivity score is... Take 0.

[0046] The classification model simultaneously outputs the probability that the target data belongs to each sensitivity level. ( Model sensitivity score Determined according to the following formula: ; in, ( The classification model determines the sensitivity level of the target data. The probability of is equal to 1 when the summation of all levels in the sensitivity level set G is equal to 1. Sensitive level The corresponding numerical value.

[0047] The sensitivity score S is obtained by fusing the rule sensitivity score and the model sensitivity score according to the following formula: ; Where η∈[0,1] is the fusion weight of the rule sensitivity score, 1 η represents the fusion weights for the model sensitivity scores; the clip function is used to limit the calculation results to the range of 0 to 1. This is the minimum sensitivity constraint value corresponding to the mandatory sensitivity rule. When the mandatory sensitivity rule is not hit... =0, when one or more mandatory sensitive rules are hit. Take the maximum value of the sensitivity level corresponding to each hit mandatory sensitive rule. For example, η can be 0.60. Thus, the sensitivity score S has a clear calculation source and is used to determine the sensitivity level label; the risk score R further integrates multi-dimensional features, historical behavior, and access environment to determine the collaborative control level. The two have different uses.

[0048] Through the above mechanism, this embodiment ensures the reliability and rigid constraints of the rule matching results in high deterministic mandatory rule scenarios, while also taking into account the generalization ability of the model evaluation results in unknown or mutation-sensitive pattern scenarios, thus achieving complementary fusion of the two classification results.

[0049] The formula for calculating the risk score is: ; in, For the first Risk contribution value corresponding to class feature For the first The weights corresponding to the class features The number of feature categories participating in the risk assessment, and .

[0050] The risk contribution value is used to quantify the contribution of the i-th type of feature to the risk of the target data. Assuming that the rule risk score, model risk probability, historical anomaly frequency score, and environmental credibility score corresponding to this type of feature all range from 0 to 1, the risk contribution value is determined according to the following formula: ; in, Let $\frac{i}{i-1}$ be the risk contribution value corresponding to the $i-th type of feature, where $i \in \{1, 2, ..., n}$. , , and These are the sub-item weights for rule-based risk score, model risk probability, historical anomaly frequency score, and environmental unreliability, respectively, with each sub-item weight being no less than 0 and... + + + =1;s i r For the rule risk score of the i-th feature, the maximum risk weight or normalized cumulative value of the hit rules in that category can be taken; if no rule is hit, the score is 0; s i m s represents the risk probability output by the classification model for the i-th class of features; i h The historical anomaly frequency score can be obtained by dividing the number of anomalies of this type within a preset time window by a frequency threshold and truncating it to 1; t iLet be the credibility score corresponding to the i-th feature, with a value ranging from 0 to 1. It can be determined by the weighted result of terminal credibility, network credibility, and identity credibility. Therefore, 1 t i This indicates the degree of environmental unbelievability; `clip(·, 0, 1)` is used to restrict the calculation result to the range of 0 to 1. For inapplicable sub-items, the corresponding weight is set to 0, and the remaining weights are renormalized.

[0051] For example, , , and The values ​​were successively set to 0.35, 0.35, 0.20, and 0.10; when the rule risk score... =0.9, Model Risk Probability =0.8, Historical Anomaly Frequency Score =0.2, credibility score When the probability is 0.7, the environmental unreliability is 1. =0.3, Risk Contribution Value The risk score is 0.665. The feature category weights in the risk score can be preset based on the business scenario, or obtained through model training or historical data statistics. For example, the weights of data content, metadata, source, access subject, and access environment features can be set to 0.35, 0.10, 0.15, 0.20, and 0.20 respectively; when the corresponding risk contribution values ​​are 0.9, 0.3, 0.4, 0.8, and 0.7 respectively, the risk score R is 0.705.

[0052] In step S3, based on the target data classification and grading completed in step S2 and the output of data category labels, sensitivity level labels and risk scores, step S3 generates a collaborative control strategy corresponding to the target data from the preset strategy mapping relationship according to the above classification and grading results. This strategy includes three dimensions: transmission control parameters, sharing control parameters and storage control parameters, providing a unified strategy basis for the security control of subsequent links.

[0053] Specifically, the process of generating the collaborative control strategy includes the following sub-steps: Step S3-1: Determine the control level of the target data based on the risk score R. In this embodiment, the risk score... The target data is compared with a preset level threshold to determine its control level, which satisfies the following: ; In the formula, , and These represent low-level control, medium-level control, and high-level control, respectively. and These represent the first control threshold and the second control threshold, respectively, and 0 ≤ < ≤1. For example, =0.35, =0.70, that is, when R < 0.35. When 0.35≤R<0.70, it is When R≥0.70, it is The above thresholds can be adjusted based on business risk tolerance, validation sample distribution, and security policy requirements.

[0054] Step S3-2: Select the corresponding target policy template from the preset policy template library according to the control level determined in step S3-1. When the control level is When selecting a template, choose either the standard encrypted transmission template, the standard shared approval template, or the ordinary encrypted storage template. When the control level is When selecting a template, choose either the enhanced encrypted transmission template, the domain-restricted sharing template, or the isolated encrypted storage template. When the control level is At that time, select a high-strength encrypted transmission template, a strictly time-limited and time-limited sharing template, and a fragmented isolated storage template.

[0055] Each template records control parameters and their specific values ​​in machine-readable fields, rather than just abstract control methods. Transmission control parameters include authentication mechanism identifiers, encryption algorithm identifiers, session key validity period, key encapsulation algorithm identifiers, and integrity verification algorithm identifiers. Sharing control parameters include allowed subject role sets, allowed organization scopes, terminal trust thresholds, allowed network type sets, access start and end times, maximum access counts, de-identification policy identifiers, and watermark type identifiers. Storage control parameters include storage encryption algorithm identifiers, isolation levels, fragment quantity, fragment location constraints, and key field identifiers. Among these, mechanism identifiers and algorithm identifiers are enumerated fields, thresholds, durations, counts, and fragment quantity are numerical fields, subject scope and network type are set fields, and the key field is an identifier field.

[0056] For example, the medium-level control template can set the transmission field to two-way certificate authentication, AES-256-GCM, session key validity period of 30 minutes, RSA-OAEP, and HMAC-SHA256; set the sharing field to allow organization whitelisting, terminal trust level not lower than 0.70, weekdays from 8:00 to 20:00, maximum 5 accesses, field mask, and user time watermark; and set the storage field to logical isolation, two fragments, and independent key fields. After selecting the template, the system modifies the field values ​​according to the business context and finally outputs three types of control parameters.

[0057] Step S3-3: Based on the selected target policy template, further modify the parameters of the target policy template by combining the access subject attribute, terminal environment attribute, and business purpose attribute to obtain the final collaborative control policy adapted to the current business scenario.

[0058] Specifically, the access subject attribute reflects the identity type and trust level of the data recipient, the terminal environment attribute reflects the trustworthiness of the accessing device and network conditions, and the business purpose attribute reflects the specific scenario and operation type of data use. These attributes collectively constitute the basis for dynamic adjustment of policy parameters. For example, when the access subject is an external collaborating entity, the accessing terminal is a low-trust terminal, or the network environment is a non-dedicated network, adjustments such as shortening the authorization token validity period, reducing the number of accesses, enabling data masking, increasing the transmission encryption level, or raising the audit level can be triggered. When the target data is medium-level control data, and the access subject is an external collaborating entity, the accessing terminal has low trustworthiness, or the current network environment is a non-dedicated network, the sharing constraint level can be further increased and the audit recording time extended.

[0059] Through the above process, this embodiment enables the target data to share the same policy source in the transmission, sharing and storage stages, thereby achieving consistency and linkage of multi-stage control.

[0060] In step S4, based on the collaborative control strategy generated in step S3, which includes transmission control parameters, sharing control parameters, and storage control parameters, step S4 executes secure transmission control, secure sharing control, and secure storage control based on the above three types of control parameters, and triggers dynamic updates to the classification and grading results and collaborative control strategy when an abnormal event is detected. The three control links share the same strategy source, ensuring the consistency and linkage of security strategies throughout the entire lifecycle of data transmission, sharing, and storage.

[0061] S4-1 executes secure transmission control. Based on the transmission control parameters, it performs authentication, key negotiation, data encryption, and integrity verification on the target data to complete the secure transmission of the target data.

[0062] In this embodiment, secure transmission control includes the following process: First, perform two-way authentication on the sending and receiving ends to verify the legitimacy of both parties' identities; Subsequently, a session key is generated, and the target data is symmetrically encrypted using the session key; Next, the session key is encapsulated using the receiver's public key or private key encryption key; Then, an integrity check value is generated for the encrypted target data, and the integrity check value is sent together with the encrypted data; Finally, the target data is restored after the receiving end completes identity authentication, session key decryption, and integrity verification.

[0063] In a preferred embodiment, for high-level control data, shorter session key lifecycles, stricter certificate verification, and stronger integrity verification strategies can be further enabled to reduce the risk of highly sensitive data being intercepted, tampered with, or replayed during transmission.

[0064] S4-2, Implement secure sharing control. When target data needs to be shared externally, based on the sharing control parameters, the target data is subjected to sharing request determination, dynamic desensitization, watermark embedding, authorization token generation, and sharing time limit to complete the secure sharing of the target data.

[0065] In this embodiment, the shared control process includes the following sub-steps: First, obtain the requester's identity information, terminal information, access time information, and business purpose information corresponding to the sharing request. The requester's identity information may include user identity, organization affiliation, role type, and existing permissions; the terminal information may include terminal trustworthiness, device fingerprint, and network environment information; and the business purpose information may include the purpose type such as downloading, browsing, forwarding, collaborative editing, or statistical analysis.

[0066] Secondly, the requester's identity information, terminal information, access time information, and business purpose information are matched with the sharing control parameters. Specifically, the sharing control parameters are converted into a set of sharing conditions, which are divided into mandatory conditions and compensable conditions. Mandatory conditions include at least a legitimate requester identity, a subject role within the permitted scope, and a business purpose within the permitted scope; compensable conditions include at least a terminal trustworthiness reaching a threshold, a network type meeting requirements, and an access time within the permitted time window. The request information is compared item by item with the corresponding conditions: when both mandatory and compensable conditions are met, an authorization token is generated, and the access validity period and number of accesses are set according to the sharing control parameters; when the mandatory conditions are met, one or more compensable conditions are not met, and the control level of the target data is not higher than the medium-level control and meets the preset downgraded sharing rules, the system switches to de-identified sharing mode, and the generated authorization token only allows online browsing and prohibits downloading or forwarding the original text, while performing dynamic de-identification and watermark embedding on the shared data; when any mandatory condition is not met, the target data belongs to the high-level control, or does not meet the preset downgraded sharing rules, the sharing request is rejected.

[0067] Finally, dynamic desensitization and watermark embedding are performed on the shared data. Dynamic desensitization includes at least one of field masking, data range generalization, identifier replacement, and sensitive segment deletion. Watermark embedding includes at least one of user identifier watermark, time identifier watermark, and scene identifier watermark.

[0068] Through the above mechanism, this embodiment can generate a shared version suitable for the current subject and the current environment in real time according to the sharing scenario without changing the original storage state of the target data, thereby balancing data security and sharing efficiency.

[0069] S4-3 executes secure storage controls and performs auditing and dynamic adjustments. Based on storage control parameters, it implements encrypted storage, tiered storage, and isolated storage for target data, and triggers reclassification and grading of target data and updates to collaborative control strategies when anomalies are detected based on audit results.

[0070] In this embodiment, storage control includes: When the sensitivity level label corresponding to the target data is low sensitivity level, the target data is stored in encrypted form. When the sensitivity level label corresponding to the target data is medium sensitivity level, the target data is stored in encrypted form and logically isolated. When the sensitivity level label corresponding to the target data is high sensitivity level, the target data is stored in fragments, managed by independent key domains, and physically isolated or namespace isolated.

[0071] The sensitivity level is determined as follows: The sensitivity score S is calculated using the fusion formula of the rule sensitivity score and the model sensitivity score in steps S2-3, and its value ranges from 0 to 1. The first sensitivity threshold is θ1, and the second sensitivity threshold is θ2, where 0 ≤ θ1 < θ2 ≤ 1. When S < θ1, it is determined to be a low sensitivity level; when θ1 ≤ S < θ2, it is determined to be a medium sensitivity level; and when S ≥ θ2, it is determined to be a high sensitivity level. Simultaneously, when the target data matches mandatory sensitivity rules such as ID card numbers, financial accounts, medical records, or core business identifiers, the minimum sensitivity constraint value is used. The sensitivity level is ensured to be no lower than the level corresponding to the mandatory sensitivity rule. In this embodiment, θ1 is exemplarily preset to 0.35, and θ2 is exemplarily preset to 0.70. That is, a sensitivity score below 0.35 is a low sensitivity level, between 0.35 and 0.70 is a medium sensitivity level, and no lower than 0.70 is a high sensitivity level. The specific thresholds can be adjusted by those skilled in the art according to the security policy requirements of the application scenario.

[0072] like Figure 3 As shown, the system records audit logs of transmission, sharing, access, and decryption behaviors related to the target data, and monitors various abnormal events in real time. Abnormal events include at least one of the following: the number of authentication failures exceeds the threshold within a preset time window, the frequency of target data downloads exceeds the threshold, the scope of shared objects expands, the trustworthiness of the access terminal is lower than the threshold, and the business scenario changes.

[0073] After an abnormal event is detected, the risk score R is updated. satisfy: ; in, R Risk scoring before an abnormal event occurs. For the updated risk score, E The intensity of the abnormal event. B For the deviation of access behavior, and It is the adjustment coefficient, and R , , E , B , and The values ​​range from 0 to 1. To avoid risk scores exceeding the range, the upper bound truncation update method shown in the above formula is uniformly adopted; for example, =0.30, =0.20.

[0074] The intensity E of an anomaly event is determined by the weighted sum of four normalized indicators: authentication failure frequency, anomaly duration, degree of sharing scope expansion, and degree of terminal trust degradation, satisfying the following: ; Among them, e k For the k-th normalized outlier, Let $k$ be the weight of the $k$-th outlier indicator, where $k = 1, 2, 3, 4$, and all weights are not less than 0 and the sum of the weights is 1. The four normalized outlier indicators are determined according to the following formulas: ; ; in, This represents the number of authentication failures within a preset time window. The threshold for the number of authentication failures; t a The duration of the abnormal behavior, The threshold for the duration of the abnormality; This refers to the number of newly added shared objects relative to the historical normal baseline. Expand the threshold for the number of shared objects; N represents the baseline value of terminal trustworthiness, and τ represents the current terminal trustworthiness. a T a M a and All are preset baseline values ​​greater than 0. t a and All are observations not less than 0, and τ ranges from 0 to 1. e1, e2, e3, and e4 represent the frequency of authentication failure, the duration of anomalies, the degree of expansion of the sharing scope, and the degree of decline in terminal trustworthiness, respectively, and their values ​​are all limited to the range of 0 to 1. For example, γ1, γ2, γ3, and γ4 can be 0.30, 0.20, 0.25, and 0.25, respectively.

[0075] Access behavior deviation B is based on the current access behavior vector Compared with historical normal behavior baseline The Euclidean distance between them is determined, and it satisfies: ; in, The preset distance normalization threshold is greater than 0; x is the current access behavior vector composed of behavioral features such as access time, access frequency, download quantity, access region, terminal type, and business purpose; μ is the baseline vector of the same dimension of behavior calculated based on historical normal access records. Before calculation, the features of each dimension constituting x and μ are normalized to the range of 0 to 1 according to preset maximum values, minimum values, or historical statistical parameters. ||x μ‖2 represents the Euclidean distance between the two. When the distance reaches or exceeds... When B is 1, it is not reached. At that time, B, according to the stated distance, and The ratio is determined. The frequency of authentication failures, duration of anomalies, extent of expansion of sharing scope, and degree of decline in terminal trust are reflected by the anomaly event intensity E; abnormal download frequency and changes in business scenarios are reflected by vector dimensions such as download quantity, access time, access region, and business purpose in the access behavior deviation degree B. Therefore, E and B have clear and complementary calculation sources and are jointly substituted into the risk scoring update formula.

[0076] For example, when high-sensitivity target data is downloaded multiple times across regions in a short period of time, the shared objects are expanded from internal personnel to external entities, or the terminal is switched from a trusted device to a low-trust device, the intensity of the abnormal event E and the deviation of the access behavior B will both increase, which in turn leads to an increase in the updated risk score R′.

[0077] The system is based on the updated risk score. Re-execute the classification, grading, and collaborative control policy generation process to update the data category labels, sensitivity level labels, risk scores, and collaborative control policies for the target data. If necessary, synchronously execute control actions such as shared token invalidation, storage domain switching, access frequency tightening, or audit level upgrades.

[0078] Through the sequential execution and coordinated operation of the above three stages, this embodiment achieves closed-loop control of the entire process, from intelligent data classification and grading, collaborative strategy generation, transmission, shared storage and linkage control, anomaly audit feedback to dynamic strategy updates, effectively improving the security, adaptability and scientific management of collaborative control throughout the data lifecycle.

[0079] Example 2 This embodiment provides a data intelligent classification-driven secure transmission, shared storage, and collaborative control system, including: The data acquisition module is configured to acquire target data and the corresponding business context information. The classification and grading module is configured to: extract multi-dimensional features of target data and business context information, classify and grade the target data based on the multi-dimensional features, and obtain classification and grading results; The strategy generation module is configured to generate a collaborative control strategy for the target data from a preset strategy mapping relationship based on the classification and grading results. The collaborative control strategy includes transmission control parameters, shared control parameters, and storage control parameters. The transmission control module is configured to perform secure transmission control on the target data based on transmission control parameters. The shared control module is configured to perform secure shared control on the target data based on shared control parameters. The storage audit module is configured to: perform secure storage control on the target data based on the storage control parameters, and trigger updates to the classification and grading results and collaborative control strategies when an abnormal event is detected.

[0080] It should be noted that each module in this embodiment corresponds one-to-one with each step in Embodiment 1, and their specific implementation process is the same, so it will not be repeated here.

[0081] Example 3 This embodiment also provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the method described in Embodiment 1.

[0082] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A data intelligent classification-driven secure transmission, shared storage, and collaborative control method, characterized in that: include: Obtain the target data and the corresponding business context information; Extract multidimensional features from the target data and the business context information, and classify and grade the target data based on the multidimensional features to obtain classification and grading results; Based on the classification and grading results, a collaborative control strategy for the target data is generated from a preset strategy mapping relationship. The collaborative control strategy includes transmission control parameters, sharing control parameters, and storage control parameters. Based on the transmission control parameters, secure transmission control is performed on the target data; based on the sharing control parameters, secure sharing control is performed on the target data; based on the storage control parameters, secure storage control is performed on the target data, and when an abnormal event is detected, updates to the classification and grading results and the collaborative control strategy are triggered.

2. The method according to claim 1, characterized in that, The multidimensional features include data content features and metadata features extracted from the target data, as well as source features, access subject features, and access environment features extracted from the business context information.

3. The method according to claim 1, characterized in that, The target data is classified and graded based on the multidimensional features to obtain the classification and grading results, specifically: Based on preset sensitivity rules, the target data is matched to obtain an initial classification result; Based on the trained classification model, the target data is classified and risk assessed to obtain the model classification result. The initial classification result is fused with the model classification result to obtain data category labels, sensitivity level labels, and risk scores, which serve as the classification and grading results.

4. The method according to claim 3, characterized in that, The formula for calculating the risk score is as follows: ； in, For the first Risk contribution value corresponding to class feature For the first The weights corresponding to the class features The number of feature categories participating in the risk assessment, and .

5. The method according to claim 1, characterized in that, Based on the classification and grading results, a collaborative control strategy for the target data is generated from a preset strategy mapping relationship, specifically as follows: The control level to which the target data belongs is determined based on the risk score in the classification and grading results; According to the control level, a target policy template is selected from the preset policy template library. If the control level is low-level control, the standard encrypted transmission template, the standard shared approval template, and the ordinary encrypted storage template are selected. If the control level is medium level, then select the enhanced encrypted transmission template, the domain-limited sharing template, and the isolated encrypted storage template; If the control level is high-level control, then select the high-strength encrypted transmission template, the strict time-limited and time-limited sharing template, and the fragmented isolated storage template; Based on the access subject attribute, terminal environment attribute, and business purpose attribute in the business context information, the target policy template is modified to obtain the collaborative control policy.

6. The method according to claim 1, characterized in that, Based on the transmission control parameters, secure transmission control is performed on the target data; based on the sharing control parameters, secure sharing control is performed on the target data; based on the storage control parameters, secure storage control is performed on the target data, specifically as follows: Based on the transmission control parameters, two-way authentication is performed on the sending end and the receiving end, a session key is generated and the target data is symmetrically encrypted using the session key, the session key is encapsulated using the receiving end's public key, an integrity check value is generated for the encrypted target data and sent, and the target data is restored after the receiving end completes authentication, session key decapsulation and integrity check. Based on the shared control parameters, the system obtains the request subject's identity information, terminal information, access time information, and business purpose information corresponding to the shared request. The sharing conditions are divided into mandatory conditions and compensable conditions, and matched with the obtained information. When both mandatory and compensable conditions are met, an authorization token is generated, and a time limit or access limit is imposed on the shared data. When the mandatory conditions are met, one or more compensable conditions are not met, and the control level of the target data is not higher than the medium-level control and conforms to the preset downgraded sharing rules, the system switches to a de-identified sharing mode, performing dynamic de-identification and watermark embedding on the shared data. When any mandatory condition is not met, the target data belongs to the high-level control, or does not conform to the preset downgraded sharing rules, the sharing request is rejected. Based on the storage control parameters, differentiated storage is performed according to the sensitivity level of the target data. For low sensitivity levels, encrypted storage is performed; for medium sensitivity levels, encrypted storage and logically isolated storage are performed; and for high sensitivity levels, fragmented storage, independent key domain management, and physical or namespace isolated storage are performed.

7. The method according to claim 6, characterized in that, The dynamic desensitization includes at least one of field masking, data range generalization, identifier replacement, and sensitive segment deletion; the watermark embedding includes at least one of user identifier watermark, time identifier watermark, and scene identifier watermark.

8. The method according to claim 1, characterized in that, When an abnormal event is detected, the classification and grading results and the collaborative control strategy are updated, specifically as follows: An audit log records the transmission, sharing, access, and decryption behaviors related to the target data; Detect abnormal events, which include at least one of the following: the number of authentication failures exceeds a threshold within a preset time window, the frequency of target data downloads exceeds a threshold, the scope of shared objects expands, the trustworthiness of the access terminal is lower than a threshold, and the business scenario changes. After the abnormal event is detected, the risk score is updated, and the updated risk score satisfies: ； Where R is the risk score before the abnormal event occurs, R′ is the updated risk score, E is the intensity of the abnormal event, B is the deviation of the access behavior, and α and β are adjustment coefficients; the values ​​of R, R′, E, B, α and β all range from 0 to 1. The classification and grading and collaborative control strategy generation are re-executed based on the updated risk score to update the classification and grading results and the collaborative control strategy.

9. A data intelligent classification-driven secure transmission, shared storage, and collaborative control system, characterized in that: include: The data acquisition module is configured to acquire target data and business context information corresponding to the target data. The classification and grading module is configured to: extract multi-dimensional features of the target data and the business context information, classify and grade the target data based on the multi-dimensional features, and obtain classification and grading results; The strategy generation module is configured to: generate a collaborative control strategy for the target data from a preset strategy mapping relationship based on the classification and grading results; the collaborative control strategy includes transmission control parameters, sharing control parameters, and storage control parameters. The transmission control module is configured to perform secure transmission control on the target data based on the transmission control parameters. The shared control module is configured to perform secure shared control on the target data based on the shared control parameters. The storage audit module is configured to: perform secure storage control on the target data based on the storage control parameters, and trigger updates to the classification and grading results and the collaborative control strategy when an abnormal event is detected.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the data intelligent classification-driven secure transmission shared storage collaborative control method as described in any one of claims 1-8.

Images