Administrative institution internal control information processing method and system based on big data analysis

Abstract

The invention discloses an administrative institution internal control information processing method and system based on big data analysis, and relates to the technical field of big data information processing. And furthermore, by capturing dynamic association characteristics generated by interaction of local contents in the business field and the information, internal control information cross-dimension semantic joint coding representation with context sensing capability is generated, and cross-dimension semantic interaction analysis between the business field and information content semantics is further realized by applying a chain reasoning model. The system can automatically identify potential logical association between the business field and the information content, so that the risk of data leakage can be greatly reduced, and the scientificity of desensitization processing can be improved, thereby enhancing the effectiveness and efficiency of internal control of administrative institutions.

Description

Technical Field

[0001] The present application relates to the field of big data information processing, and more specifically, to a method and system for internal control information processing of administrative institutions based on big data analysis. Background Art

[0002] In the context of the current accelerated digital transformation, the security management of internal control information of administrative institutions has become particularly important. With the development of information technology, big data analysis has gradually become a key tool for improving the efficiency and effectiveness of internal control. However, existing technologies still face many challenges and shortcomings in processing internal control information. Traditional identity authentication methods are often unable to meet the increasingly complex information security needs, and are prone to authentication loopholes or misjudgments, resulting in unauthorized access requests being approved, thereby increasing the risk of data leakage.

[0003] In addition, in terms of sensitive information identification, most existing methods are based on single-dimensional feature extraction, such as only considering one aspect of the business field or information content, while ignoring the deep interactive relationship that may exist between the two. This approach not only limits the accuracy of judging the sensitivity of internal control information, but may also lead to inappropriate desensitization strategies, resulting in important information not being effectively protected or unnecessary information being overly desensitized, affecting the normal use and value of information.

[0004] Therefore, we look forward to an optimized internal control information processing method for administrative institutions based on big data analysis. Summary of the invention

[0005] In order to solve the above technical problems, this application is proposed.

[0006] According to one aspect of the present application, a method for processing internal control information of administrative institutions based on big data analysis is provided, which includes:

[0007] In response to receiving a request initiated by a user to access the first internal control information, performing identity authentication on the user to obtain an identity authentication result;

[0008] If the identity authentication result is verification failure, access is denied;

[0009] If the identity access result is that the verification is passed, the sensitivity level of the first internal control information is identified to obtain a sensitivity level identification result;

[0010] Based on the sensitivity identification result, desensitizing the first internal control information to obtain desensitized first internal control information;

[0011] Start the encrypted channel to transmit the first internal control information after desensitization.

[0012] According to another aspect of the present application, there is provided an internal control information processing system for administrative institutions based on big data analysis, which includes:

[0013] an identity authentication module, configured to, in response to receiving a request from a user to access the first internal control information, authenticate the user to obtain an identity authentication result;

[0014] A first verification result module, used to deny access if the identity authentication result is verification failure;

[0015] A second verification result module is used to identify the sensitivity of the first internal control information to obtain a sensitivity identification result if the identity access result is that the verification is passed;

[0016] An information desensitization module, configured to perform desensitization processing on the first internal control information based on the sensitivity identification result to obtain desensitized first internal control information;

[0017] The information transmission module is used to start the encrypted channel to transmit the desensitized first internal control information.

[0018] Beneficial effect: Compared with the prior art, the present application provides a method and system for processing internal control information of administrative institutions based on big data analysis. It adopts a neural network model based on deep learning to respectively mine the semantic embedding coding features of business fields and information content, and then captures the dynamic correlation features generated by the interaction between business fields and local content in information to generate a cross-dimensional semantic joint coding representation of internal control information with context-awareness. It further uses a chain reasoning model to realize cross-dimensional semantic interaction analysis between business fields and information content semantics, so as to achieve more accurate sensitivity judgment. The system can automatically identify the potential logical relationship between business fields and information content, which can not only greatly reduce the risk of data leakage, but also improve the scientific nature of desensitization processing, thereby enhancing the effectiveness and efficiency of internal control of administrative institutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0020] Figure 1 It is a flowchart of the internal control information processing method of administrative institutions based on big data analysis according to an embodiment of the present application;

[0021] Figure 2A data flow diagram of an internal control information processing method for administrative institutions based on big data analysis according to an embodiment of the present application;

[0022] Figure 3 It is a flowchart of sub-step S3 of the internal control information processing method of administrative institutions based on big data analysis according to an embodiment of the present application;

[0023] Figure 4 It is a flowchart of sub-step S31 of the internal control information processing method of administrative institutions based on big data analysis according to an embodiment of the present application;

[0024] Figure 5 It is a flowchart of sub-step S312 of the internal control information processing method of administrative institutions based on big data analysis according to an embodiment of the present application;

[0025] Figure 6 It is a block diagram of an internal control information processing system for administrative institutions based on big data analysis according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described therein.

[0027] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0028] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0029] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0030] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described therein.

[0031] It should be noted that the acquisition and processing of all information or data in this application are carried out in compliance with the relevant national data protection laws and policies and with the authorization of the administrator.

[0032] In the technical solution of the present application, a method for processing internal control information of administrative institutions based on big data analysis is proposed. Figure 1 A flowchart of an internal control information processing method for administrative institutions based on big data analysis according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the internal control information processing method of administrative institutions based on big data analysis according to the embodiment of the present application. Figure 1 and Figure 2 As shown, according to the embodiment of the present application, the internal control information processing method of administrative institutions based on big data analysis includes the following steps: S1, in response to receiving a request from a user to access first internal control information, authenticating the user to obtain an authentication result; S2, if the authentication result is a verification failure, denying access; S3, if the identity access result is a verification success, performing sensitivity identification on the first internal control information to obtain a sensitivity identification result; S4, based on the sensitivity identification result, desensitizing the first internal control information to obtain the desensitized first internal control information; S5, starting an encrypted channel to transmit the desensitized first internal control information.

[0033] In particular, in response to receiving a user's request to access the first internal control information, S1 authenticates the user to obtain an authentication result. Traditional authentication relies on a single static password or fixed authority allocation, which is difficult to deal with situations where internal personnel abuse their authority or external attackers steal credentials through vulnerabilities. By dynamically verifying the authenticity and legitimacy of the user's identity, it is ensured that the subsequent sensitive information processing process is only open to trusted subjects. In this way, the probability of information security incidents caused by false identities or overstepping of authority is reduced. In an example, the specific implementation steps for authenticating the user are as follows: When the system receives an access request, it will synchronously activate the multi-dimensional identity authentication engine, and build a dynamic identity confidence model by real-time parsing the spatiotemporal characteristics (such as geographic location, network environment), device fingerprints (including hardware identifiers, system integrity status) and historical operation trajectories of the user's login scene. For example, for users who frequently switch access terminals but claim to perform regular business, the system will automatically retrieve their past operation modes for comparison. If an abnormal device fingerprint is detected or the operation sequence violates the behavioral baseline, even if the correct static password is entered, the auxiliary verification process will still be triggered. At the same time, the authentication engine will dynamically adjust the verification strength threshold in association with the sensitivity level of the internal control information. For high-level information access requests, the system will embed an implicit verification layer and perform secondary verification by analyzing biometric behavioral characteristics such as user operation delays and input habits. This composite authentication mechanism that combines explicit credentials with implicit behavioral characteristics can not only resist traditional brute force attacks, but also prevent the risk of lateral penetration after legitimate credentials are stolen, forming a three-dimensional identity credibility assessment system.

[0034] In particular, in S2, if the result of identity authentication is verification failure, access is denied. Specifically, a rigid access barrier is constructed under the zero-trust security architecture to avoid the risk of residual information exposure after authentication failure by immediately interrupting illegal access attempts. The attacker's trial space is eliminated through a clear rejection response, and a mandatory access control closed loop is formed to prevent the logical separation of the authentication link and the subsequent processing flow, resulting in the phenomenon of permission escape. In this way, the defense transparency of the system can be enhanced, so that users who have not passed the authentication cannot reversely infer the weak links of the system security mechanism through error feedback.

[0035] In particular, in S3, if the identity access result is verified, the first internal control information is subjected to sensitivity identification to obtain a sensitivity identification result. Specifically, by deconstructing the implicit correlation between information content and business scenarios, a dynamic sensitivity assessment mechanism is established, so that the desensitization strategy can be adaptively adjusted according to the needs of specific scenarios, while ensuring information security and avoiding the loss of information value caused by excessive protection. In a specific example of this application, Figure 3As shown, the S3 includes: S31, extracting the two-dimensional semantic features of the business domain and the information content in the first internal control information, and chain-inferring the cross-dimensional global semantic interaction relationship between the business domain and the information content based on the local content interaction information between the two to obtain the cross-dimensional semantic joint coding features of the internal control information; S32, determining the sensitivity recognition result of the first internal control information based on the cross-dimensional semantic joint coding features of the internal control information.

[0036] Specifically, the S31 extracts the dual-dimensional semantic features of the business domain and the information content in the first internal control information, and based on the local content interaction information between the two, chain-reasons the cross-dimensional global semantic interaction relationship between the business domain and the information content to obtain the cross-dimensional semantic joint coding features of the internal control information. In a specific example of the present application, Figure 4 As shown, the S31 includes: S311, extracting the dual-dimensional semantic features of the business domain and the information content in the first internal control information to obtain the semantic embedding coding features of the content part in the business domain dimension and the semantic embedding coding features of the content part in the information content dimension; S312, performing cross-dimensional content semantic deep collaborative analysis on the semantic embedding coding features of the content part in the business domain dimension and the semantic embedding coding features of the content part in the information content dimension to obtain the cross-dimensional semantic joint coding features of the internal control information.

[0037] More specifically, the S311 extracts the dual-dimensional semantic features of the business domain and the information content in the first internal control information to obtain the semantic embedding coding features of the business domain dimension content part and the semantic embedding coding features of the information content dimension content part. In an embodiment of the present application, first, the first internal control information is input into a business domain dimension content extractor based on a large language model to obtain the business domain dimension content part. Among them, the business domain dimensions include budget management, revenue and expenditure management, asset management, procurement management, contract management, personnel management, scientific research management information disclosure and internal audit, etc. Due to the significant differences in regulatory requirements and data sensitivity thresholds in different business fields such as budget management, procurement management, and scientific research management, it is difficult to accurately determine its risk level by simply relying on the surface features of the information content itself. For example, although the performance bond clauses in the contract management module and the performance appraisal details in the personnel management module are both text data, because they belong to different business fields, their sensitivity judgment logic must strictly follow the policies, regulations and business process characteristics of the corresponding fields. The large language model, through the domain knowledge graph and context understanding capabilities accumulated during the pre-training phase, can penetrate the explicit expression of information text, accurately anchor the business management category to which it belongs, and provide a basic classification framework for subsequent cross-dimensional analysis. Among them, the business domain dimension content extractor maps the original data to the preset business management coordinate system by parsing the implicit domain feature identifiers in the information (such as institutional document references, business process terms, and management subject relevance), so that subsequent sensitivity analysis can be carried out based on a domain-specific risk assessment model rather than a generalized strategy.

[0038] Similarly, the first internal control information is input into the information content dimension content extractor based on the large language model to obtain the content part of the information content dimension. Among them, the information content dimension includes financial data, personnel data, business process data, risk assessment data, audit data and compliance data. Specifically, after the user passes the identity authentication, the system needs to quickly lock the sensitive elements that may exist in the information content. For example, abnormal transaction records in audit reports, undisclosed vulnerability descriptions in risk assessments, or traces of policy circumvention in compliance data may all constitute risk exposure points of different levels. The information content extractor based on the large language model can automatically identify and present these key information points in a structured manner through the domain knowledge base obtained through pre-training, effectively solving the problem of sensitive information leakage caused by low efficiency of manual annotation and delayed rule base update. This dynamic extraction mechanism enables the system to maintain a high content parsing robustness when facing new business data or complex nested information. Through deep semantic mining of information content dimensions, the system can not only identify explicit sensitive fields (such as identity information, bank accounts), but also capture implicit sensitive associations (such as permission loopholes in multi-department collaboration processes, and abnormal trends exposed by cross-year financial data comparisons). This feature extraction of the content dimension and the cross-validation of the subsequent business domain dimension form a dual-channel risk perception network.

[0039] Next, the business domain embedding matrix is ​​used to perform semantic embedding encoding on the content part of the business domain dimension to obtain the semantic embedding encoding vector of the content part of the business domain dimension as the semantic embedding encoding feature of the content part of the business domain dimension. Specifically, the business management of administrative institutions involves highly specialized fields such as budget preparation, procurement management, and scientific research project establishment, each of which has its own unique institutional norms, business processes, and risk thresholds. For example, although the performance terms in contract management and the depreciation rules in asset management are both text information, there are essential differences in their semantic structures and regulatory requirements. Therefore, in the technical solution of the present application, the business domain embedding matrix is ​​used to perform semantic embedding encoding on the content part of the business domain dimension to realize the dynamic association between business rules and data sensitivity, and obtain the semantic embedding encoding vector of the content part of the business domain dimension. Among them, the business domain embedding matrix can map the domain features scattered in the text (such as bidding terms in procurement and achievement transformation terms in scientific research management) into a vector space with domain differentiation through the domain knowledge topology learned in the pre-training stage, thereby breaking through the limitations of the traditional bag of words model or general semantic coding for insufficient domain specificity. This deep encoding mechanism enables the system to capture the implicit semantics of the business domain dimension, providing an accurate domain feature base for subsequent cross-dimensional analysis. Taking the internal audit report as an example, the business domain embedding matrix can map different paragraphs involved in the report, such as asset management loopholes and personnel attendance anomalies, to the corresponding business domain vector clusters, so that subsequent sensitivity analysis can automatically associate the corresponding risk weights based on the domain feature vectors.

[0040] Similarly, the information content embedding matrix is ​​used to perform semantic embedding encoding on the content part of the information content dimension to obtain the semantic embedding encoding vector of the content part of the information content dimension as the semantic embedding encoding feature of the content part of the information content dimension. Specifically, the traditional recognition method based on keyword matching or regular expression is difficult to penetrate the surface semantics of the text, and cannot capture the liability avoidance statement in the contract clause or the compliance hint in the audit report. Therefore, in the technical solution of the present application, the information content embedding matrix is ​​used to perform semantic embedding encoding on the content part of the information content dimension to obtain the semantic embedding encoding vector of the content part of the information content dimension as the semantic embedding encoding feature of the content part of the information content dimension. The information content embedding matrix can convert the content elements scattered in text, tables and even semi-structured data (such as the description of the intellectual property boundary in scientific research data and the potential loss quantification indicator in risk assessment) into a vector space distribution with risk perception ability through the semantic association pattern learned in the pre-training stage, breaking through the limitations of traditional methods for parsing unstructured data. By converting information content into high-dimensional semantic vectors, the system can capture new risk patterns that are difficult to cover by traditional rule bases.

[0041] More specifically, the S312 performs cross-dimensional content semantic deep collaborative analysis on the semantic embedded coding features of the content part of the business domain dimension and the semantic embedded coding features of the content part of the information content dimension to obtain the cross-dimensional semantic joint coding features of the internal control information. Specifically, the traditional method has a split dilemma in the cognition of the relevance between business and content. For example, under the traditional single-dimensional analysis mode, the system may regard the "single-source procurement" clause in the procurement contract as only a business process compliance indicator, but ignore its cross-risk with the contract amount, supplier affiliates and other contents; for example, when a high-value procurement contract adopts a single-source method and the supplier has an associated relationship, the synergy of its business domain characteristics (procurement specifications) and content characteristics (abnormal fund flow) will significantly increase the information sensitivity level. Therefore, in order to mine the implicit coupling relationship between business rules and information entities in administrative data, so as to realize the dynamic coupling of business logic and content information, and realize the composite sensitivity modeling of business scenarios and semantic content, in the technical solution of the present application, the semantic embedded coding features of the content part of the business domain dimension and the semantic embedded coding features of the content part of the information content dimension are subjected to cross-dimensional content semantic deep collaborative analysis to obtain the cross-dimensional semantic joint coding features of the internal control information. Specifically, by deconstructing global interactions into local units, using one-dimensional convolution to capture local semantic patterns within each dimension (such as the "abnormal bidding process" feature segment in the business domain encoding vector and the "payment node deviation" feature segment in the content encoding vector), and then using the single-body interaction engine to mine the fine-grained correlation between the two (such as the causal relationship between the missing bidding link and the loose payment terms), it breaks through the traditional method of splitting the cross-dimensional risk clues. Taking the salary adjustment data in personnel management as an example, the "performance salary adjustment pilot" feature extracted from the business domain dimension and the "executive salary increase" feature extracted from the content dimension, after the local interaction engine identifies the potential violation pattern of "abnormal salary changes of personnel outside the pilot scope", the chain reasoning attention weight will strengthen the contribution of the interaction node, drive the LSTM reasoning engine to trace the historical approval records along the time series, and finally determine its sensitivity level. This analysis mechanism can not only identify explicit data anomalies (such as excessive amounts), but also capture the deep correlation between system implementation deviations (such as pilot policy abuse) and data entity anomalies (such as sudden changes in salary structure). Through the generalization capability of cross-dimensional local interaction, the system can map the characteristics of unknown fields to the interactive space of existing business rules and content models, achieve reasonable inference of sensitivity, and ensure that security policies can both adapt to business innovation needs and maintain the bottom line of risk prevention and control. In a specific example of this application, Figure 5As shown, the S312 includes: S3121, calculating a set of business domain-information content local implicit feature interaction response coding vectors based on the content local implicit coding features of the business domain dimension content part semantic embedding coding vector and the information content dimension content part semantic embedding coding vector; S3122, based on the feature distribution characteristics of each business domain-information content local implicit feature interaction response coding vector in the set of business domain-information content local implicit feature interaction response coding vectors, performing feature modulation based on the chain reasoning attention mechanism on the set of business domain-information content local implicit feature interaction response coding vectors to obtain a set of modulated business domain-information content local implicit feature interaction response coding vectors; S3123, performing internal control information chain reasoning on the set of modulated business domain-information content local implicit feature interaction response coding vectors to obtain an internal control information cross-dimensional semantic joint coding vector as an internal control information cross-dimensional semantic joint coding feature.

[0042] More specifically, the S3121 calculates a set of business domain-information content local implicit feature interactive response coding vectors based on the content local implicit coding features of the semantic embedding coding vector of the business domain dimension content part and the semantic embedding coding vector of the information content dimension content part. In an embodiment of the present application, first, the business domain dimension content part semantic embedding coding vector and the information content dimension content part semantic embedding coding vector are respectively subjected to local implicit feature extraction based on one-dimensional convolutional coding to obtain a set of business domain local implicit feature coding vectors and a set of information content local implicit feature coding vectors. Since the business domain and information content have multi-level and nonlinear semantic structures in text expression, the global coding of a single dimension is prone to ignore local patterns such as word order dependency and context association, resulting in difficulty in accurately modeling key clues of sensitive information during cross-dimensional collaborative analysis. For example, in an internal control document involving a financial approval process, the expression of "approval authority" may be scattered in the business domain description and specific operation terms. If only relying on the overall representation of the semantic embedding vector, it may not be possible to identify the sensitive authority rules formed by a specific field combination. Through one-dimensional convolutional coding, it is possible to capture implicit associations within a local window by sliding along the feature vector sequence. This processing method is closer to the local distribution characteristics of phrases and syntactic structures in natural language. In this process, the one-dimensional convolution performs local scanning on the semantic encoding vectors of the business domain and information content dimensions through a parameter sharing mechanism to extract local implicit feature sets of different scales. These local feature sets not only retain the fine-grained information at the vocabulary and phrase levels in the original semantic embedding, but also strengthen the contextual relevance through the sliding window operation of the convolution kernel. For example, in the business domain dimension, the local features related to "budget preparation" may cover sensitive elements such as responsible departments and time nodes; while in the information content dimension, the approval records of the same business process may contain sensitive fields such as amount and signatory. By extracting local implicit features of the two dimensions separately, the model can more accurately anchor the sensitive information anchor points that need to be focused on in cross-dimensional semantic interactions, avoiding the loss of details caused by global feature fusion. In a specific example, the following one-dimensional convolution formula is used to perform local implicit feature extraction based on one-dimensional convolution coding on the semantic embedding coding vector of the content part of the business domain dimension and the semantic embedding coding vector of the content part of the information content dimension to obtain a set of local implicit feature coding vectors of the business domain and a set of local implicit feature coding vectors of the information content; wherein the one-dimensional convolution formula is:

[0043]

[0044]

[0045] in, is the semantic embedding coding vector of the content of the business domain dimension, represents a one-dimensional convolution operation, are respectively the first, second, and third in the set of local implicit feature coding vectors of the business field. and The local implicit feature encoding vector of each business domain, is the semantic embedding coding vector of the content part of the information content dimension, are respectively the first, second, and third in the set of local implicit feature encoding vectors of the information content. and The information content of the local implicit feature encoding vector.

[0046] Then, each set of corresponding business domain local implicit feature coding vectors and information content local implicit feature coding vectors in the set of business domain local implicit feature coding vectors and information content local implicit feature coding vectors are respectively input into the content monomer feature interaction engine to obtain the set of business domain-information content local implicit feature interaction response coding vectors. Specifically, each unit in the set of business domain local implicit feature coding vectors may correspond to abstract semantic fragments such as process nodes and system clauses, while the local features of the information content dimension focus on specific data fields or operation instructions. Traditional methods often use simple splicing or weighted average when merging features, which cannot capture the nonlinear and dynamic association patterns between local features of different dimensions. For example, in the internal control documents involving the funding approval process, the business domain dimension may contain the stage division features of "budget preparation process", while the information content dimension may involve local features of fields such as "approver position" and "amount threshold". The two are difficult to reveal the sensitive combination relationship of "specific position and high approval authority" in an isolated state. By introducing the monomer feature interaction engine, it is possible to conduct targeted analysis on each set of corresponding local feature pairs to simulate the potential association logic between business rules and information entities. In a specific example, each corresponding set of business domain local implicit feature coding vectors and information content local implicit feature coding vectors in the set of business domain local implicit feature coding vectors and the set of information content local implicit feature coding vectors are respectively input into the content monomer feature interaction engine to obtain a set of business domain-information content local implicit feature interaction response coding vectors; wherein the feature interaction formula is:

[0047]

[0048] in, Indicates cascade, It means point multiplication by position. It means adding by position. It means subtracting by position. and denote the weight matrix and bias vector respectively, express and The corresponding business domain-information content local implicit feature interactive response encoding vector.

[0049] More specifically, the S3122, based on the characteristic distribution characteristics of each business domain-information content local implicit feature interaction response coding vector in the set of business domain-information content local implicit feature interaction response coding vectors, performs feature modulation based on the chain reasoning attention mechanism on the set of business domain-information content local implicit feature interaction response coding vectors to obtain a set of modulated business domain-information content local implicit feature interaction response coding vectors. In an embodiment of the present application, first, based on the characteristic distribution characteristics of each business domain-information content local implicit feature interaction response coding vector in the set of business domain-information content local implicit feature interaction response coding vectors, the business domain-information content chain reasoning attention weights of each business domain-information content local implicit feature interaction response coding vector are determined to obtain a set of initial business domain-information content chain reasoning attention weights. Specifically, when processing multi-dimensional feature interactions, traditional methods often adopt static weight allocation or homogenization processing modes, ignoring the differential contributions of different local interaction units in the overall sensitivity reasoning chain. For example, in a business process involving asset management, the interaction between certain business domain features (such as "fixed asset scrapping approval process") and specific information content features (such as "level of the signatory of the assessment report") may constitute a high-risk sensitive node, while other interactions (such as "conference room use registration" and "appointment time") may belong to low-sensitivity routine operations. If the importance of these local interactions is not dynamically evaluated, the model may reduce the accuracy of sensitive identification due to excessive attention to non-critical features, or cause misjudgment due to information overload in complex document processing. Therefore, in the technical solution of the present application, based on the characteristic distribution characteristics of each business domain-information content local implicit feature interaction response coding vector in the set of business domain-information content local implicit feature interaction response coding vectors, the business domain-information content chain reasoning attention weights of each business domain-information content local implicit feature interaction response coding vector are determined to obtain a set of initial business domain-information content chain reasoning attention weights. Specifically, the decision logic for sensitivity determination is optimized by quantifying the semantic value of local interaction units. Business Domain - Information Content Chain Reasoning The generation of attention weights is essentially to model the importance of each local interaction response vector based on feature distribution characteristics (such as vector entropy, norm, or high-order statistics). For example, when the distribution of an interaction response vector shows high discreteness (which may reflect the unconventional association between business rules and information entities), its attention weight will be dynamically increased, thereby dominating the subsequent chain reasoning. The attention weight mechanism can autonomously identify sensitive focus in different business scenarios and form a flexible reasoning chain by quantifying the feature distribution characteristics of each interaction unit.For example, when processing a budget report for cross-departmental collaboration, the model may automatically weaken the weight of the "document format specification" related features by marking the interaction between "fund allocation rules" and "department head approval records" with high weights. This dynamic focusing capability not only improves the detection accuracy of complex sensitive information (such as the cross-correlation of system loopholes and data leakage points), but also provides an explainable weight basis for audit traceability, so that the formulation of desensitization strategies can accurately match the actual business risk level. In a specific example, the business domain-information content chain reasoning attention weights of the interaction response encoding vectors of the local implicit features of each business domain-information content are determined by the following calculation formula to obtain a set of initial business domain-information content chain reasoning attention weights; wherein, the calculation formula is:.

[0050]

[0051] in, express function, The first in the set of interactive response encoding vectors representing the business domain and information content local implicit features The feature value of the jth position of the business domain-information content local implicit feature interaction response encoding vector, represents the square of the norm, Indicates the length of the business domain-information content local implicit feature interaction response encoding vector, express Corresponding initial business domain-information content chain reasoning attention weight.

[0052] Then, the set of initial business domain-information content chain reasoning attention weights is spatially corrected based on the interaction specification to obtain the set of business domain-information content chain reasoning attention weights. Specifically, when dealing with cross-dimensional semantic associations, the traditional attention mechanism often ignores the differences in the spatial role of interaction specifications. For example, stable constraints (such as rigid requirements of institutional clauses) and dynamic fluctuations (such as abnormal operations of temporary approval authority) between business rules and information entities may exist at the same time, but the initial weight calculation is difficult to effectively distinguish the weights of the two on the sensitivity. Taking the financial reimbursement process as an example, the conventional interaction between "approval level" and "reimbursement amount" may follow a fixed spatial translation specification (gradient direction consistency), while the sudden interaction between "emergency approval" and "extra-budgetary expenditure" may introduce fluctuations (gradient orthogonality). If these two types of interaction specifications are not spatially decomposed, the model may not be able to accurately capture the sensitive risks implicit in abnormal operations. Therefore, in the preferred example of the present application, the set of initial business domain-information content chain reasoning attention weights is spatially corrected based on the interaction specification to obtain the set of business domain-information content chain reasoning attention weights. By introducing spatial correction based on interaction norms, the system can break through the limitations of static weight allocation. Specifically, through spatial decomposition based on translation and fluctuation, the multimodal analysis of the interaction relationship between business domains and information content dimensions is carried out. Translation corresponds to the stable association between business rules and information entities (such as the positive coupling of "procurement process" and "contract amount threshold"), and its spatial correction strengthens the transmission of institutional constraints through local rule compensation; while fluctuation maps the potential risks in unconventional interactions (such as the abnormal combination of "temporary authorization" and "high payment"), and its periodic phase change correction can capture sensitive signals at the edge of the rules. For example, in the asset management scenario, when the local features of "asset scrapping process" interact with the information features of "missing evaluation report", the spatial correction can dynamically increase the attention weight of such unconventional interactions by weighted fusion of translation and fluctuation, thereby identifying the risk of sensitive information exposure caused by institutional execution loopholes. By decomposing and integrating the rule transmission of translation and the risk perception of fluctuation, the corrected attention weight can more accurately portray the semantic value of cross-dimensional interactions. In the budget preparation scenario of cross-departmental collaboration, the model can distinguish the stable interaction (high translation weight) between "department budget allocation rules" and "historical execution data" and the dynamic interaction (high fluctuation weight) between "special project appropriations" and "emergency approval records" through spatial correction, thereby hierarchically locating core sensitive nodes in complex documents. By enhancing the correlation between spatial auxiliary specifications, an explainable decision path is provided for multi-level sensitive judgments, effectively avoiding the rigid rules or abnormal missed detection problems caused by static weight allocation in traditional methods, ensuring that internal control information is always under precise and controllable desensitization protection in a dynamic business environment.

[0053] As mentioned above, when calculating each business domain-information content local implicit feature interaction response coding vector, the interaction features between the corresponding business domain local implicit feature coding vector and the information content local implicit feature coding vector are introduced, for example , , Etc., specifically, they essentially correspond to different interaction norms, thus having different spatial auxiliary constraint associations within the interaction space.

[0054] Therefore, in order to enhance the norm variability assistance in chain reasoning attention weights, it is preferred to modify the set of initial business domain-information content chain reasoning attention weights based on the spatial action decomposition of the interaction norms. , is regarded as a translation effect, specifically, the gradient direction is consistent with the spatial interaction direction, and It is regarded as a wave action, that is, the gradient direction is orthogonal to the spatial interaction direction. In this way, for vector statistics of different actions, for example , , , it is believed that the wave action leads to a localized periodic effect in the translation direction, and thus the periodic local regular compensation is calculated as:

[0055]

[0056] Specifically, As a localized representation of translation, the wave action It increases with the logarithmic size of the translation domain.

[0057] On the other hand, the wave action will also lead to a localized translational phase transition, so the periodic local auxiliary phase transition is:

[0058]

[0059] Therefore, the chain reasoning attention weight is corrected based on the weighted sum of the above two items :

[0060]

[0061] By distinguishing the normative roles of different interaction norms corresponding to local interaction responses of various business fields and information contents in the interaction space, the correlation between different spatial auxiliary constraint norms in the interaction space can be improved, thereby improving the calculation accuracy of the chain reasoning attention weight.

[0062] Then, based on the set of attention weights of the business domain-information content chain reasoning, the set of business domain-information content local implicit feature interaction response encoding vectors is weighted modulated to obtain the set of modulated business domain-information content local implicit feature interaction response encoding vectors. Specifically, the semantic space of local interaction response is reconstructed through dynamic weight allocation to achieve accurate extraction of sensitive information representation. Weighted modulation based on chain reasoning attention weights essentially maps the interaction relationship between business rules and information entities to an interpretable metric space. For example, in a procurement process document, when the business feature of "compliance of the bidding process" interacts with the content feature of "integrity of supplier qualification documents", high attention weights will amplify the semantic strength of their response vectors, making them dominate the decision logic in subsequent sensitive identification; while low-weight interactions (such as the association between "contract number generation rules" and "document page numbers") are suppressed as background noise. The set of weighted modulated business domain-information content local implicit feature interaction response encoding vectors significantly improves the decision accuracy and scenario adaptability of the internal control information processing system. In a specific example, the set of business domain-information content local implicit feature interactive response coding vectors is weighted modulated by the following weighted modulation formula to obtain a set of modulated business domain-information content local implicit feature interactive response coding vectors; wherein the weighted modulation formula is:

[0063]

[0064]

[0065] in, is a set of the modulation service domain-information content local implicit feature interaction response coding vectors, are respectively the first, second, and third in the set of the modulation business field-information content local implicit feature interaction response coding vector. and A modulation business domain - information content local implicit feature interactive response coding vector.

[0066] More specifically, the S3123 performs chain reasoning of internal control information on the set of the interaction response coding vectors of the modulated business domain-information content local implicit features to obtain the internal control information cross-dimensional semantic joint coding vector as the internal control information cross-dimensional semantic joint coding feature. Specifically, when integrating local interaction features, traditional methods often use simple pooling or splicing operations, which makes it difficult to capture the dynamic evolution of the association pattern between business rules and information entities. For example, in the procurement approval process, the interaction between the business domain features in the "tender announcement release" stage and the information content of the "supplier qualification review record" and the interaction with the "expert scoring rules" and "bid document terms" in the subsequent "bid evaluation organization resolution" stage has a progressive relationship in temporal logic. If this process-driven interaction evolution law is not modeled through chain reasoning, the model may misjudge the sensitivity level due to the severance of the contextual connection of the local interaction, such as misjudging isolated high-risk links as routine operations, or ignoring the systemic risks triggered by multiple links in series. Therefore, in the technical solution of the present application, the set of the interaction response coding vectors of the local implicit features of the modulation business domain-information content is input into the internal control information chain reasoning engine based on the forward LSTM model to obtain the cross-dimensional semantic joint coding vector of the internal control information as the cross-dimensional semantic joint coding feature of the internal control information. Specifically, the spatiotemporal evolution law of the cross-dimensional interaction is modeled by serialized reasoning, and a global sensitive information map is constructed. In this process, the chain reasoning engine of the forward LSTM dynamically manages the set of the interaction response coding vectors of the local implicit features of the modulation business domain-information content through the gating mechanism (input gate, forget gate, output gate), simulating the transmission and accumulation effect of sensitive information in the internal control process. Taking the scientific research project management scenario as an example, when the business rules of "topic confidentiality setting" and the information features of "participating personnel confidentiality authority" form a high-risk mark in the initial interaction, the memory unit of the LSTM will continue to transmit the risk signal to the subsequent "results release approval" link, ensuring that even if the "results summary" itself does not trigger the sensitive word match, the model can still identify its potential sensitive attributes based on the full-process risk transmission mechanism. This time-series sensitive modeling capability enables the system to break through the limitations of traditional methods for determining isolated features, thereby building an intelligent and adaptive internal control information protection mechanism for administrative institutions. In a specific example, the following inference formula is used to input the set of modulated business domain-information content local implicit feature interaction response coding vectors into the internal control information chain inference engine based on the forward LSTM model to obtain the cross-dimensional semantic joint coding vector of the internal control information; wherein, the inference formula is:

[0067]

[0068] in, Indicates forward , is a cross-dimensional semantic joint encoding vector for the internal control information.

[0069] Specifically, the S32 determines the sensitivity recognition result of the first internal control information based on the cross-dimensional semantic joint coding feature of the internal control information. Specifically, in the technical solution of the present application, the cross-dimensional semantic joint coding vector of the internal control information is input into the internal control information sensitivity identifier based on the classifier to obtain the recognition result, and the recognition result is used to represent the sensitivity of the first internal control information. Among them, the classifier can dynamically divide the sensitivity threshold according to the spatial position of the cross-dimensional semantic joint coding vector of the internal control information by training and learning the sensitive pattern distribution of cross-dimensional features in different business scenarios. For example, in personnel file management, the interaction between "job responsibilities description" and "salary and benefits data" may present different sensitivity weights due to job level differences. For example, the responsibilities-salary interaction of senior management positions may trigger a high-risk judgment, while the same interaction of grassroots positions may only be marked as moderately sensitive. The nonlinear decision boundary of the classifier can capture such complex associations, avoid the misjudgment problem caused by the solidification of business scenarios in traditional threshold rules, and support the refined output of multi-level sensitive labels (such as public, internal, and confidential). The final output sensitivity label not only provides a quantitative basis for dynamic desensitization, but can also be reversely mapped to the association path between business rules and information entities.

[0070] In particular, the S4, based on the sensitivity identification result, performs desensitization processing on the first internal control information to obtain the desensitized first internal control information. Specifically, traditional desensitization methods often use static rules (such as keyword replacement or fixed field masking), which are difficult to adapt to the distribution law of complex sensitive information in internal control documents. For example, in the approval documents involving personnel transfers, the business process description of "position appointment and removal resolution" may be moderately sensitive, while the associated "salary adjustment details" information content is highly sensitive. If differentiated processing is not implemented according to the dynamically determined sensitivity level, it may lead to the exposure of key salary data or excessive desensitization of process information, thereby affecting the integrity of audit traceability. By combining the sensitivity identification results generated by cross-dimensional semantic joint coding features, the system can accurately locate the sensitivity levels of different information units in the document, providing a scientific basis for the formulation of dynamic desensitization strategies. In one example, the desensitization processing based on sensitivity identification achieves controllable release of information value through multimodal conversion technology (such as conditional mask generation and semantic preservation replacement). This intelligent desensitizing system based on semantic understanding not only solves the dilemma of "over-protection leading to information rigidity" and "insufficient protection leading to leakage" in traditional methods, but also builds a new internal control information governance model for administrative institutions that takes into account compliance, security and business continuity.

[0071] In particular, the S5 starts an encrypted channel to transmit the desensitized first internal control information. The encrypted channel can be implemented using a variety of encryption technologies, such as TLS (Transport Layer Security) or SSL (Secure Sockets Layer), both of which are encryption standards widely used in current Internet communications and can provide strong data encryption and identity authentication functions. In the process of creating an encrypted channel, both parties (i.e., the sender and the receiver) exchange necessary encryption parameters through a handshake protocol and generate a session key for this communication. This process ensures that data can be transmitted in an encrypted form even in an insecure network environment, thereby preventing security risks such as man-in-the-middle attacks. Once the encrypted channel is established, the next step is to encode the desensitized first internal control information into a data format suitable for network transmission. This usually involves converting the data into a byte stream and possibly adding some additional header information, such as a protocol version number, content length, etc., so that the receiving end can correctly parse the received data packet. Then, the prepared data can be securely sent out through the previously established encrypted channel. In this process, all data will be encrypted using the previously generated session key. In summary, the internal control information processing method of administrative institutions based on big data analysis according to the embodiment of the present application is explained, which uses a neural network model based on deep learning to respectively mine the semantic embedding coding features of the business field and information content, and then captures the dynamic correlation features generated by the interaction between the business field and the local content in the information, and generates a cross-dimensional semantic joint coding representation of internal control information with context perception capabilities, and further uses a chain reasoning model to realize the cross-dimensional semantic interaction analysis between the business field and the information content semantics, so as to achieve more accurate sensitivity judgment. In this way, the system can automatically identify the potential logical association between the business field and the information content, which can not only greatly reduce the risk of data leakage, but also improve the scientific nature of the desensitization processing, thereby enhancing the effectiveness and efficiency of internal control of administrative institutions.

[0072] Furthermore, an internal control information processing system for administrative institutions based on big data analysis is also provided.

[0073] Figure 6 FIG. 1 is a block diagram of an internal control information processing system for administrative institutions based on big data analysis according to an embodiment of the present application. Figure 6As shown, according to the embodiment of the present application, the internal control information processing system 300 of administrative and public institutions based on big data analysis includes: an identity authentication module 310, which is used to authenticate the user in response to receiving a request from the user to access the first internal control information to obtain an identity authentication result; a first verification result module 320, which is used to deny access if the identity authentication result is a verification failure; a second verification result module 330, which is used to identify the sensitivity of the first internal control information to obtain a sensitivity identification result if the identity access result is a verification success; an information desensitization module 340, which is used to desensitize the first internal control information based on the sensitivity identification result to obtain the desensitized first internal control information; and an information transmission module 350, which is used to start an encrypted channel to transmit the desensitized first internal control information.

[0074] As described above, the administrative and public institutions internal control information processing system 300 based on big data analysis according to the embodiment of the present application can be implemented in various wireless terminals, such as a server with an administrative and public institutions internal control information processing algorithm based on big data analysis. In one possible implementation, the administrative and public institutions internal control information processing system 300 based on big data analysis according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the administrative and public institutions internal control information processing system 300 based on big data analysis can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the administrative and public institutions internal control information processing system 300 based on big data analysis can also be one of the many hardware modules of the wireless terminal.

[0075] Alternatively, in another example, the administrative and public institution internal control information processing system 300 based on big data analysis and the wireless terminal may also be separate devices, and the administrative and public institution internal control information processing system 300 based on big data analysis may be connected to the wireless terminal through a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

[0076] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method for processing internal control information of administrative institutions based on big data analysis, characterized in that: include: In response to receiving a request initiated by a user to access the first internal control information, performing identity authentication on the user to obtain an identity authentication result; If the identity authentication result is verification failure, access is denied; If the identity access result is that the verification is passed, the sensitivity level of the first internal control information is identified to obtain a sensitivity level identification result, including: extracting the two-dimensional semantic features of the business domain and the information content in the first internal control information, and chain-inferring the cross-dimensional global semantic interaction relationship between the business domain and the information content based on the local content interaction information between the two to obtain the cross-dimensional semantic joint coding feature of the internal control information; based on the cross-dimensional semantic joint coding feature of the internal control information, determining the sensitivity level identification result of the first internal control information; Based on the sensitivity identification result, desensitizing the first internal control information to obtain desensitized first internal control information; Start the encrypted channel to transmit the first internal control information after desensitization.

2. The internal control information processing method for administrative institutions based on big data analysis according to claim 1 is characterized in that: Extracting the dual-dimensional semantic features of the business domain and the information content in the first internal control information, and chain-reasoning the cross-dimensional global semantic interaction relationship between the business domain and the information content based on the local content interaction information between the two to obtain the cross-dimensional semantic joint coding features of the internal control information, including: Extracting dual-dimensional semantic features of the business domain and the information content in the first internal control information to obtain semantic embedding coding features of the content part of the business domain dimension and semantic embedding coding features of the content part of the information content dimension; A cross-dimensional content semantic deep collaborative analysis is performed on the semantic embedding coding features of the content part of the business domain dimension and the semantic embedding coding features of the content part of the information content dimension to obtain the cross-dimensional semantic joint coding features of internal control information.

3. The internal control information processing method for administrative institutions based on big data analysis according to claim 2 is characterized in that: Extracting the dual-dimensional semantic features of the business domain and the information content in the first internal control information to obtain the semantic embedding coding features of the content part of the business domain dimension and the semantic embedding coding features of the content part of the information content dimension, including: Inputting the first internal control information into a business domain dimension content extractor based on a large language model to obtain a business domain dimension content part; Inputting the first internal control information into an information content dimension content extractor based on a large language model to obtain an information content dimension content part; Use the business domain embedding matrix to perform semantic embedding coding on the business domain dimension content part to obtain the semantic embedding coding vector of the business domain dimension content part as the semantic embedding coding feature of the business domain dimension content part; The information content dimension content part is semantically embedded and encoded using the information content embedding matrix to obtain the information content dimension content part semantic embedding encoding vector as the information content dimension content part semantic embedding encoding feature.

4. The internal control information processing method for administrative institutions based on big data analysis according to claim 3 is characterized in that: The business area dimension includes budget management, revenue and expenditure management, asset management, procurement management, contract management, personnel management, scientific research management information disclosure and internal audit; the information content dimension includes financial data, personnel data, business process data, risk assessment data, audit data and compliance data.

5. The internal control information processing method for administrative institutions based on big data analysis according to claim 4 is characterized in that: The semantic embedding coding features of the business domain dimension content and the semantic embedding coding features of the information content dimension content are analyzed in depth across dimensions to obtain the cross-dimensional semantic joint coding features of internal control information, including: Based on the content local implicit coding features of the semantic embedding coding vector of the content part of the business domain dimension and the content local implicit coding features of the semantic embedding coding vector of the content part of the information content dimension, a set of business domain-information content local implicit feature interaction response coding vectors is calculated; Based on the characteristic distribution characteristics of each business domain-information content local implicit feature interaction response coding vector in the set of business domain-information content local implicit feature interaction response coding vectors, the set of business domain-information content local implicit feature interaction response coding vectors is subjected to feature modulation based on a chained reasoning attention mechanism to obtain a set of modulated business domain-information content local implicit feature interaction response coding vectors; The internal control information chain reasoning is performed on the set of modulation business domain-information content local implicit feature interaction response coding vectors to obtain the internal control information cross-dimensional semantic joint coding vector as the internal control information cross-dimensional semantic joint coding feature.

6. The internal control information processing method for administrative institutions based on big data analysis according to claim 5 is characterized in that: Based on the content local implicit coding features of the semantic embedding coding vector of the content part of the business domain dimension and the content local implicit coding features of the semantic embedding coding vector of the content part of the information content dimension, a set of business domain-information content local implicit feature interactive response coding vectors is calculated, including: Performing local implicit feature extraction based on one-dimensional convolutional coding on the semantic embedding coding vector of the content part of the business domain dimension and the semantic embedding coding vector of the content part of the information content dimension to obtain a set of local implicit feature coding vectors of the business domain and a set of local implicit feature coding vectors of the information content; Each corresponding group of business domain local implicit feature coding vectors and information content local implicit feature coding vectors in the set of business domain local implicit feature coding vectors and the set of information content local implicit feature coding vectors are respectively input into the content monomer feature interaction engine to obtain a set of business domain-information content local implicit feature interaction response coding vectors.

7. The method for processing internal control information of administrative institutions based on big data analysis according to claim 6 is characterized in that: Based on the characteristic distribution characteristics of each business domain-information content local implicit feature interactive response encoding vector in the set of business domain-information content local implicit feature interactive response encoding vectors, the set of business domain-information content local implicit feature interactive response encoding vectors is subjected to feature modulation based on a chained reasoning attention mechanism to obtain a set of modulated business domain-information content local implicit feature interactive response encoding vectors, including: Based on the characteristic distribution characteristics of each business domain-information content local implicit feature interaction response coding vector in the set of business domain-information content local implicit feature interaction response coding vectors, determine the business domain-information content chain reasoning attention weight of each business domain-information content local implicit feature interaction response coding vector to obtain a set of initial business domain-information content chain reasoning attention weights; Performing spatial correction based on the interaction specification on the set of initial business domain-information content chain reasoning attention weights to obtain a set of business domain-information content chain reasoning attention weights; Based on the set of business domain-information content chain reasoning attention weights, the set of business domain-information content local implicit feature interaction response coding vectors is weighted modulated to obtain a set of modulated business domain-information content local implicit feature interaction response coding vectors.

8. The method for processing internal control information of administrative institutions based on big data analysis according to claim 7 is characterized in that: The internal control information chain reasoning is performed on the set of the modulation business domain-information content local implicit feature interaction response coding vectors to obtain the internal control information cross-dimensional semantic joint coding vector as the internal control information cross-dimensional semantic joint coding feature, including: The set of modulated business domain-information content local implicit feature interaction response coding vectors is input into the internal control information chain reasoning engine based on the forward LSTM model to obtain the cross-dimensional semantic joint coding vector of the internal control information.

9. The internal control information processing method for administrative institutions based on big data analysis according to claim 8 is characterized in that: Based on the cross-dimensional semantic joint coding features of the internal control information, a sensitivity recognition result of the first internal control information is determined, including: The cross-dimensional semantic joint coding vector of the internal control information is input into the classifier-based internal control information sensitivity identifier to obtain a recognition result, and the recognition result is used to represent the sensitivity of the first internal control information.

10. An internal control information processing system for administrative institutions based on big data analysis, characterized in that: include: an identity authentication module, configured to, in response to receiving a request from a user to access the first internal control information, authenticate the user to obtain an identity authentication result; A first verification result module, used to deny access if the identity authentication result is verification failure; A second verification result module is used to identify the sensitivity of the first internal control information to obtain a sensitivity identification result if the identity access result is that the verification is passed; An information desensitization module, configured to perform desensitization processing on the first internal control information based on the sensitivity identification result to obtain desensitized first internal control information; The information transmission module is used to start the encrypted channel to transmit the desensitized first internal control information.

Images