Transfer station system and method for secure sharing of audit files

Abstract

The present invention provides a transfer station system and method for secure sharing of audit files. Based on the analytical capabilities of a large model, the application can effectively classify the sensitivity and confidentiality of audit files entering the transfer station system; using blockchain technology and smart contract mechanisms, the approval management workflow for secure shared access and operation of audit files is automatically implemented, while ensuring transparent encryption and strict access control to ensure that only authorized personnel can access the files, thereby solving the problems in the prior art and improving the efficiency and confidentiality of audit file sharing.

Description

Technical Field

[0001] The present application relates to the technical field of data transmission and processing, and in particular to a transfer station system and method for realizing secure sharing of audit files. Background Art

[0002] The shared access and shared operation of audit files is often a necessary process in audit work. However, there are some major challenges and problems in the existing technology.

[0003] First, audit documents often contain sensitive and confidential information. How to achieve correct hierarchical and classified management of audit documents based on the sensitivity and confidentiality of the information contained is often a difficult problem in practice. The existing technology uses specific rules to name, number, and label audit documents to perform hierarchical and classified management of sensitivity and confidentiality. However, it usually lacks in-depth analysis of audit data attribute descriptions, which can easily lead to inaccurate and inappropriate hierarchical and classified management, resulting in the improper sharing and operation of sensitive information, or causing unnecessary burdens on transfer and sharing due to excessive strictness, thereby reducing efficiency.

[0004] Secondly, the access and operation of audit files requires the necessary permission approval control for the visitor, which currently usually relies on a centralized management system, which may lead to single point failures and security vulnerabilities. In addition, the workflow of permission approval control relies on manual operations, which reduces the efficiency of file access and operation. In addition, due to the lack of an effective log tracing mechanism, it is difficult to track responsibility once a data leak occurs. Summary of the invention

[0005] 1. Purpose of application

[0006] In view of this, the purpose of this application is to provide a transfer station system and method for secure sharing of audit files. Based on the analytical capabilities of large models, this application can effectively classify the sensitivity and confidentiality of audit files entering the transfer station system; using blockchain technology and smart contract mechanisms, it can automatically implement the approval management workflow for secure shared access and operation of audit files, while ensuring transparent encryption and strict access control to ensure that only authorized personnel can access files, thereby solving the problems in the existing technology and improving the efficiency and confidentiality of audit file sharing.

[0007] (II) Technical solution

[0008] The present application discloses a transfer station system for secure sharing of audit files, including:

[0009] The audit file receiving unit is used to receive the file stream uploaded by the user and store it in a temporary cache area; and to perform verification on the audit file; the audit file after verification is encrypted using a symmetric encryption algorithm and transmitted to the black box unit for storage through a secure channel;

[0010] A black box unit is used to write and store the encrypted audit file in a secure storage layer; through the secure storage layer, only access requests to the encrypted audit file that have passed the authority authentication are allowed;

[0011] A large model grading unit is used to obtain the audit file after authority authentication, and use the trained special large model to identify and analyze sensitive information in the audit file in a secure computing environment, obtain a graded comprehensive score containing the sensitive information identified in the audit file and its confidentiality attributes, and feed back to the black box unit;

[0012] The blockchain contract management unit, based on the smart contract mechanism written into the blockchain, automatically performs operations related to shared access and operation of audit files based on the graded comprehensive score.

[0013] Preferably, the audit file after verification by the audit file receiving unit is encrypted using a symmetric encryption algorithm and transmitted to the black box unit for storage through a secure channel.

[0014] Preferably, the black box unit requires the initiator of the access request to provide a digital certificate for the access request to the audit file; based on a predefined access control list, the digital certificate is checked to determine whether the access request has the authority to access the specific audit file.

[0015] Preferably, the large model grading unit is trained on the basis of the general large model, takes the audit file as input, performs semantic understanding and feature extraction on the file content through natural language processing technology, identifies sensitive information in the audit data and its corresponding confidentiality attributes; and then forms a label set for the sensitivity and confidentiality of the audit file, and then calculates a graded comprehensive score for the sensitivity and confidentiality of the audit file based on the label set.

[0016] Preferably, the training of the large model grading unit includes: collecting a large number of audit file samples containing sensitive information, labeling different categories of sensitive information and corresponding confidentiality attributes, and creating an audit text annotation data set; selecting a general large model architecture as a training basis, and pre-training the general large model on a large-scale general text corpus; after the pre-training is completed, performing fine-tuning stage training, transferring the model to the audit text annotation data set for fine-tuning, and adjusting the model weights; the output of the large model grading unit is a result containing the sensitive information identified in the audit file and its confidentiality attributes.

[0017] Preferably, the loss function used in the fine-tuning stage training combines the cross entropy loss of the sensitive information classification task and the regression loss of the confidentiality attribute level;

[0018]

[0019] The classification loss is the cross entropy loss of sensitive information classification, and the regression loss is the mean square error of confidentiality attribute prediction. is a balance factor.

[0020] Preferably, the identification and analysis of sensitive information in the audit file by the large model classification unit includes: the dedicated large model performs identification of sensitive information based on text features, through a series of neuron layer analysis and extracted features, involving the following formula:

[0021]

[0022] here, Indicates that given a feature vector, the text information belongs to the category The probability of the category here Indicates the category of sensitive information. The value is 1-n; non-sensitive information is uniformly classified as category 0; after identifying sensitive information, the model will further determine its confidentiality attributes based on the classification of these sensitive information; the output layer of the model will generate and output a set of multiple labels, each label consists of two parts, namely the sensitive information type and the corresponding confidentiality attribute; the specific form is as follows:

[0023]

[0024] here, Indicates Types of sensitive information, Indicates the confidentiality attribute corresponding to the information.

[0025] Preferably, the large model grading unit performs quantitative calculations on the labels in the set according to the amount of sensitive information identified and the confidentiality attribute, and calculates a graded comprehensive score:

[0026]

[0027] here, is the total amount of sensitive information, It is The weight factor of sensitive information, It is the basic score of the sensitive information.

[0028] Preferably, the blockchain contract management unit includes a blockchain network composed of multiple blockchain nodes; wherein the black box unit serves as a blockchain node, and deploys at least one blockchain node for the user end of each participant that has access and operation requirements for audit files; the blockchain network adopts a consortium chain architecture, and the nodes of the blockchain network are interconnected through a P2P (Peer-to-Peer) network to form a decentralized network structure; a special smart contract is deployed in the blockchain network, and the smart contract presets a series of rules. As a self-executing program on the blockchain, it ensures that the classification and archiving of audit files, the execution of approval management workflows, the control of audit file access and operation permissions, and log recording and tracing are automatically managed in accordance with established rules without the need for external intervention.

[0029] The present invention provides a transfer method for secure sharing of audit files, which is characterized by comprising:

[0030] Receive the file stream uploaded by the user and store it in a temporary cache area; and perform verification on the audit file; after the verification, the audit file is encrypted using a symmetric encryption algorithm and transmitted to the black box unit storage through a secure channel;

[0031] Writing and storing the encrypted audit file in a secure storage layer of the black box unit; through the secure storage layer, only access requests to the encrypted audit file that have passed the authority authentication are allowed;

[0032] After the authority is authenticated, the audit file is obtained, and the trained dedicated large model is used to identify and analyze the sensitive information in the audit file in a secure computing environment to obtain a graded comprehensive score that includes the sensitive information identified in the audit file and its confidentiality attributes;

[0033] Based on the smart contract mechanism written into the blockchain, the operations related to shared access and operation of the audit files are automatically executed based on the hierarchical comprehensive score.

[0034] (III) Beneficial effects

[0035] The beneficial effects achieved by this application include: effectively classifying audit documents through the parsing capabilities of large models to ensure that the security level of the documents is correctly identified; utilizing blockchain technology and smart contract mechanisms to automate the approval process and improve the efficiency of audit document sharing; ensuring transparent encryption and strict access control to enhance the confidentiality of documents.

[0036] Other advantages, objectives and features of the present application will be described in the following description to some extent, and will be apparent to those skilled in the art based on the following examination and study, or can be taught from the practice of the present application to some extent. The objectives and other advantages of the present application can be achieved and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain and illustrate the present application, but should not be construed as limiting the scope of protection of the present application.

[0038] Figure 1 This is a schematic diagram of the audit file security sharing system architecture of this application;

[0039] Figure 2 It is a flowchart of the audit file security sharing method of this application. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0041] like Figure 1 As shown, this embodiment provides a transfer station system for secure sharing of audit files, and the system specifically includes:

[0042] The audit file receiving unit is used to receive the file stream uploaded by the user and store it in a temporary cache area; and to perform verification on the audit file; the audit file that passes the verification is encrypted using a symmetric encryption algorithm (such as AES) and transmitted to the black box unit for storage through a secure channel;

[0043] A black box unit is used to write and store the encrypted audit file in a secure storage layer; through the secure storage layer, only access requests to the encrypted audit file that have passed the authority authentication are allowed;

[0044] A large model grading unit is used for the black box unit to send an access request to the audit file, obtain the audit file after authority authentication, and use the trained special large model to identify and analyze sensitive information in the audit file in a secure computing environment, obtain a graded comprehensive score containing the sensitive information identified in the audit file and its confidentiality attributes, and feed back to the black box unit;

[0045] The blockchain contract management unit, based on the smart contract mechanism written into the blockchain, automatically performs operations related to shared access and operation of audit files based on the graded comprehensive score.

[0046] Figure 2 The above audit file security sharing transfer station system audit file security sharing workflow is shown. Figure 2 Make an introduction.

[0047] In this application, the user submits the audit file to the transfer station system, and the system receives the file through the audit file receiving unit and performs preliminary verification. The user selects the audit file to be submitted in the file system of his own terminal through the user interface of a dedicated APP installed on his own terminal, and starts the upload process through the "Submit" button on the interface to submit the audit file; the interface communicates with the audit file receiving unit through an encryption protocol to ensure the confidentiality and integrity of the audit file during transmission.

[0048] The audit file receiving unit receives the file stream uploaded by the user and stores it in a temporary cache area; the audit file receiving unit performs the following file verification process: (1) File integrity check: use a hash algorithm (such as SHA-256) to calculate the hash value of the file and compare it with the hash value submitted by the user to verify that the file has not been tampered with during transmission; (2) File type verification: check the MIME type of the file to ensure that it conforms to the standard format of the audit file (for example, PDF or CSV format); (3) File size check: check whether the file size is within the preset allowable range to prevent overly large files from occupying too many system resources. If the file passes all verification steps, the audit file receiving unit will transfer the audit file to the black box unit and feedback to the user that "file submission is successful"; if the file verification fails, the system will notify the user of the error information and require resubmission of the file.

[0049] The audit file receiving unit uses a symmetric encryption algorithm (such as AES) to encrypt the audit file after verification, and transmits it to the black box unit for storage through a secure channel. Specifically, before the file is transmitted, the audit file receiving unit of the system uses the AES (Advanced Encryption Standard) algorithm to encrypt the audit file. AES is a widely used symmetric encryption algorithm that can ensure the security of data; the audit file receiving unit uses TLS (Transport Layer Security Protocol) to establish a secure channel with the black box unit; the TLS protocol can provide data encryption, data integrity and identity authentication to ensure that the data is not eavesdropped or tampered with during the transmission process.

[0050] After receiving the encrypted file, the black box unit assigns a unique file identifier to each audit file to ensure the uniqueness of the file in the secure storage layer, and then writes and stores the encrypted audit file and its unique identifier in the secure storage layer. The secure storage layer only allows access requests to the encrypted audit file that have passed the authority authentication, and rejects all access requests that have not been authenticated or have failed the authority authentication. The secure storage layer usually uses a hardware security module (HSM) or other physical security measures to protect the storage medium. In addition, the black box unit implements the following authority verification process: for access requests to audit files, the initiator of the access request is required to provide a digital certificate; the black box unit checks the digital certificate based on the predefined access control list (ACL) to determine whether the access request has the authority to access a specific audit file.

[0051] The large model hierarchical unit sends an access request for the audit file to the black box unit; after receiving the access request, the black box unit first feedbacks an authority verification instruction; the large model hierarchical unit responds to the authority verification instruction and provides a valid digital certificate to prove its authority. After the black box system verifies the digital certificate, it transmits the audit file to the secure computing environment where the large model hierarchical unit is located. During this period, the audit file remains encrypted.

[0052] After receiving the file, the large model grading unit uses the corresponding key to decrypt the file in its secure computing environment for subsequent sensitive information identification and analysis. The secure computing environment of the large model grading unit is a physically isolated or virtualized isolated environment. Whether it is physical isolation or virtual isolation, it can ensure that all data is isolated from the external system during decryption and subsequent processing of the audit file, and the data cannot be accessed from the outside, thereby ensuring the security of the data information in the audit file. The large model grading unit identifies and analyzes sensitive information in the audit file in this secure computing environment to ensure that the entire process is isolated from the external network.

[0053] Sensitive information included in audit documents may include, but is not limited to: (1) Personal or corporate identity information, such as name, address, telephone number, email address, personal identification number, corporate credit code, etc.; (2) Financial information, such as bank account information, credit card number, transaction records, portfolio records, financial statements, tax records, and annual reports of listed companies; (3) Trade secret information, such as internal company strategies, business plans, customer lists, and supplier information; (4) Legal and compliance information, such as contracts, legal documents, litigation records, and compliance reports; (5) Technical secret information, such as product design information, product testing reports, R&D project reports, and R&D plans. For different sensitive information, the confidentiality attributes can include high confidentiality, medium confidentiality, and low confidentiality. For example, for financial information, the confidentiality attributes of financial statements and tax records are high confidentiality, the confidentiality attributes of transaction records, portfolios, etc. are medium confidentiality, and the confidentiality attributes of annual reports of listed companies are low confidentiality.

[0054] The large model grading unit conducts in-depth analysis of the audit data in the audit documents. The large model grading unit is trained on the basis of the general large model, takes the audit documents as input, and uses natural language processing technology to perform semantic understanding and feature extraction on the document content. It can accurately identify the sensitive information in the audit data and its corresponding confidentiality attributes; then, based on the total number of these sensitive information and their confidentiality attributes, a label set for the sensitivity and confidentiality of the audit document is formed, and then a graded comprehensive score for the sensitivity and confidentiality of the audit document is calculated based on the label set.

[0055] The training process of the large model classification unit is as follows. First, perform training data collection and annotation, collect a large number of audit file samples containing sensitive information (such as decrypted and expired audit files), manually annotate the sensitive information in the audit file samples, including annotating different categories of sensitive information and corresponding confidentiality attributes, so as to create a high-quality audit text annotation dataset. Then, select a general large model architecture as the training basis, such as a BERT model, a GPT model, etc., and pre-train the general large model. This step is performed on a large-scale general text corpus, with the purpose of allowing the model to learn the general characteristics and laws of the general language; the pre-training process is as follows: by using a large amount of unlabeled text data, the general large model predicts the next unit of a word or phrase in the text, and in this way learns the statistical information and deep structure of the language. Next, after the pre-training is completed, perform fine-tuning stage training. In this stage, the model is transferred to the audit text annotation dataset for fine-tuning, with the purpose of adjusting the weight of the model so that it can better identify and classify sensitive information in the audit text; during the fine-tuning process, a multi-task learning framework is used, in which the model not only predicts the category of sensitive information, but also predicts the level of confidentiality attributes. The loss function used for fine-tuning combines the cross entropy loss for the sensitive information classification task and the regression loss at the confidentiality attribute level:

[0056]

[0057] The classification loss is the cross entropy loss of sensitive information classification, and the regression loss is the mean square error of confidentiality attribute prediction. is a balancing factor. Through the above fine-tuning process, the large model classification unit learns how to identify and classify sensitive information from the audit text and give corresponding confidentiality attribute levels for this information. Ultimately, the output of the large model classification unit will be a result containing the sensitive information identified in the audit document and its confidentiality attributes, which can be used for further decision support.

[0058] After pre-training and fine-tuning, the dedicated large model formed by the large model grading unit will conduct an in-depth analysis of the audit files provided by the black box system to identify sensitive information and determine the confidentiality of this information. The following is the specific implementation process: the audit file is first sent to the input layer of the model, and after preprocessing steps such as word segmentation, stop word removal and stem extraction, the text is converted into a format suitable for model processing; then it is passed in the hidden layer of the model, where the convolutional neural network layer and the recurrent neural network layer collaborate to extract the features of the text, including local patterns and global semantic information; then the dedicated large model analyzes the extracted features based on the text features through a series of neuron layers to perform sensitive information recognition, involving the following formula:

[0059]

[0060] here, Indicates that given a feature vector, the text information belongs to the category The probability of the category here Indicates the category of sensitive information. The value is 1-n; non-sensitive information can be uniformly classified as category 0; after identifying sensitive information, the model will further determine its confidentiality attributes based on the classification of these sensitive information; the output layer of the model will generate and output a set of multiple labels, each label consists of two parts, namely the sensitive information type and the corresponding confidentiality attribute; the specific form is as follows:

[0061]

[0062] here, Indicates Types of sensitive information, Indicates the confidentiality attribute corresponding to the information.

[0063] The large model grading unit quantifies the labels in the set according to the number of sensitive information identified and the confidentiality attributes, and calculates a graded comprehensive score, which reflects the sensitivity and confidentiality level of the entire audit file. Specifically, first, a basic score is calculated based on the number of sensitive information identified in the audit file; each time a sensitive information is identified, a preset score value is added; this basic score reflects the richness of sensitive information in the file; secondly, considering the confidentiality attributes of each sensitive information, different levels of sensitive information contribute differently to the overall score, and a weight factor is set for each confidentiality attribute; the higher the confidentiality requirement corresponding to the confidentiality attribute, the greater its weight factor; thus, the detailed formula for the graded comprehensive score is:

[0064]

[0065] here, is the total amount of sensitive information, It is The weight factor of sensitive information, It is the basic score of the sensitive information.

[0066] After completing the above-mentioned hierarchical comprehensive scoring of the sensitivity and confidentiality of the audit file, the large model grading unit feeds back the hierarchical comprehensive score to the black box unit and deletes the audit file from its own secure computing environment.

[0067] The blockchain contract management unit, based on the smart contract mechanism written into the blockchain and based on the graded comprehensive score, automatically executes the shared access and operation-related operations of the audit files. Specifically, the blockchain contract management unit implements the classification and archiving of audit files according to sensitivity and confidentiality, the execution of approval management workflows, the control of audit file access and operation permissions, and log recording and tracing by deploying smart contracts under the blockchain architecture.

[0068] The blockchain contract management unit includes a blockchain network composed of multiple blockchain nodes; wherein the black box unit, as a blockchain node, deploys at least one blockchain node for each user end of each participant who has access and operation requirements for audit files. The blockchain network adopts a consortium chain architecture, and the nodes of the blockchain network are interconnected through a P2P (Peer-to-Peer) network to form a decentralized network structure. The P2P network ensures high availability and anti-attack resistance of data. A special smart contract is deployed in the blockchain network, which presets a series of rules. As a self-executing program on the blockchain, the smart contract can ensure the automatic management of the classification and archiving of audit files, the execution of approval management workflows, the control of access and operation permissions of audit files, and log recording and tracing according to established rules without external intervention.

[0069] Regarding the blockchain contract management unit realizing the classification and archiving of audit files according to sensitivity and confidentiality through smart contracts, the smart contract contains the logic and rules for audit file classification; when the black box unit receives the hierarchical comprehensive score of the sensitivity and confidentiality of the audit file, the smart contract will be triggered, automatically receive and analyze the score, and according to the score, the smart contract will automatically perform the archiving operation on the audit file and determine that the audit file is stored in the corresponding classified file directory of the secure storage layer of the black box unit. Different classified file directories correspond to the classification of the sensitivity and confidentiality of the audit files stored therein, and their approval management workflows can be designed differently, and different control strictness can be adopted for audit file access and operation permissions; for example, if the first classified file directory of the black box unit stores audit files with high hierarchical comprehensive scores, it indicates that the sensitivity and confidentiality requirements of the audit files in the first classified file directory are high, and a more complex approval management workflow is required, and stricter control is given to access and operation permissions.

[0070] Regarding the blockchain contract management unit, the approval management workflow is implemented through smart contracts. When the user terminal makes any access and operation request to the audit file to the black box unit, the smart contract will be triggered to ensure that only verified and authorized users can access and operate the audit file. Specifically, the user first needs to initiate identity authentication to the black box unit and provide a digital certificate as proof of his or her identity. The black box unit checks the digital certificate based on the predefined access control list (ACL) to determine whether the user has the right to access the audit file; after the identity authentication is passed, the user terminal submits a request for access or operation for a specific audit file. The request contains information such as the user role, request type, and target data category in the audit file; the smart contract automatically determines whether to approve the request based on the user role and request type and the preset permission policy; if the request is approved, the smart contract will authorize the black box unit to allow the user terminal to access the audit file and perform the corresponding operation; if the request is not approved, access is denied and the event is recorded.

[0071] Regarding the blockchain contract management unit’s control of audit file access and operation permissions through smart contracts, the details include: (1) Role definition: The smart contract defines different user roles, such as “auditor”, “audit manager”, “file administrator”, etc., and assigns different permission levels to each role; (2) Permission policy formulation: A detailed permission policy is formulated in the smart contract, which clarifies the access and operation permissions of different roles to audit data at various levels. (3) Permission matching: When a user attempts to access or operate an audit file, the smart contract will verify the user’s role and match it with the sensitivity level of the data. (4) Access and operation control: Based on the permission policy, the smart contract determines whether the user has the permission to perform the requested access or operation.

[0072] The blockchain contract management unit performs log recording and tracing through smart contracts to ensure that every operation and access to the audit data has a detailed record for easy tracking and auditing. When the black box unit receives a user's request, it triggers the smart contract to generate access logs and operation logs. The access log is used to record every user's access to the audit file, including detailed information such as access time, visitor identity, and accessed data objects; while the operation log records all operations on the audit data, such as creation, editing, and deletion, and also includes data such as operation time, operator, and operation type. The following are the detailed implementation steps of the log tracing mechanism: When a user accesses or operates an audit file, it is captured in real time and triggered by the smart contract execution; the smart contract matches the behavioral characteristics with the predefined log format to generate the corresponding access log or operation log; the log information will be encrypted and stored in the secure log database of the black box unit to ensure the security and integrity of the log data; for easy tracing, each log will be assigned a unique log number and associated with the corresponding audit data.

[0073] The smart contract unit implants a self-executing code on the blockchain node, and the self-executing code can automatically execute the contract terms when specific conditions are met. The self-executing code is based on the PBFT (Practical Byzantine FaultTolerance) consensus mechanism, shared between the smart contract units of the blockchain nodes and realizes consensus verification. Therefore, once deployed on the blockchain, its code cannot be tampered with, which ensures that the results of contract execution are credible and free from human intervention. Smart contracts are closely integrated with blockchains, and the decentralized, transparent and tamper-proof characteristics of blockchains are used to ensure that the execution and recording of contracts are credible. In addition, the execution results of smart contracts are also registered in the blockchain ledger. Through the sharing and consensus mechanism of the blockchain ledger, the execution results of smart contracts are verified and confirmed by all nodes in the network, which establishes a trust mechanism without the need for a third-party intermediary.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present application and are not intended to limit it. Although the present application has been described in detail with reference to the preferred embodiments, a person of ordinary skill in the art should understand that the technical solution of the present application can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution of the present application, which should be included in the scope of the claims of the present application.

Claims

1. A transfer station system for secure sharing of audit files, characterized in that: include: An audit file receiving unit is used to receive the file stream uploaded by the user and store it in a temporary cache area; and, perform verification of audit documents; After verification, the audit file is encrypted using a symmetric encryption algorithm and transmitted to the black box unit for storage through a secure channel; A black box unit that writes and stores encrypted audit files in a secure storage layer; Through the secure storage layer, only access requests to the encrypted audit files that have passed the authority authentication are allowed; A large model grading unit is used to obtain the audit file after authority authentication, and use the trained special large model to identify and analyze sensitive information in the audit file in a secure computing environment, obtain a graded comprehensive score containing the sensitive information identified in the audit file and its confidentiality attributes, and feed it back to the black box unit; A blockchain contract management unit, based on a smart contract mechanism written into the blockchain, automatically performs operations related to shared access and operation of audit files based on the graded comprehensive score; The loss function used in the fine-tuning phase of the training of the large model classification unit combines the cross entropy loss of the sensitive information classification task and the regression loss of the confidentiality attribute level; The classification loss is the cross entropy loss of sensitive information classification, and the regression loss is the mean square error of confidentiality attribute prediction. is a balance coefficient; The identification and analysis of sensitive information in the audit file by the large model classification unit includes: the dedicated large model extracts features based on text features through a series of neuron layer analysis and performs sensitive information identification, involving the following formula: here, Indicates that given a feature vector, the text information belongs to the category The probability of the category here Indicates the category of sensitive information. The value is 1-n; non-sensitive information is uniformly classified as category 0; after identifying sensitive information, the model will further determine its confidentiality attributes based on the classification of these sensitive information; the output layer of the model will generate and output a set of multiple labels, each label consists of two parts, namely the sensitive information type and the corresponding confidentiality attribute; the specific form is as follows: here, Indicates Types of sensitive information, Indicates the confidentiality attribute corresponding to the information.

2. The transfer station system for secure sharing of audit files according to claim 1 is characterized in that: in, The audit file receives the audit file verified by the receiving unit and encrypts the file using a symmetric encryption algorithm, and transmits the file to the black box unit for storage through a secure channel.

3. The transfer station system for secure sharing of audit files according to claim 2 is characterized in that: The black box unit requires the initiator of the access request to provide a digital certificate for the access request to the audit file; based on the predefined access control list, the digital certificate is checked to determine whether the access request has the authority to access the specific audit file.

4. The transfer station system for secure sharing of audit files according to claim 3 is characterized in that: The large model grading unit is trained on the basis of the general large model, takes the audit file as input, performs semantic understanding and feature extraction on the file content through natural language processing technology, identifies sensitive information in the audit data and its corresponding confidentiality attributes; and then forms a label set for the sensitivity and confidentiality of the audit file, and then calculates a graded comprehensive score for the sensitivity and confidentiality of the audit file based on the label set.

5. The transfer station system for secure sharing of audit files according to claim 4 is characterized in that: The training of the large model grading unit includes: collecting a large number of audit file samples containing sensitive information, marking different categories of sensitive information and corresponding confidentiality attributes, and creating an audit text annotation data set; selecting a general large model architecture as the basis for training, and pre-training the general large model on a large-scale general text corpus; after the pre-training is completed, performing fine-tuning stage training, transferring the model to the audit text annotation data set for fine-tuning, and adjusting the model's weight; the output of the large model grading unit is a result containing the sensitive information identified in the audit file and its confidentiality attributes.

6. The transfer station system for secure sharing of audit files according to claim 1 is characterized in that: The large model grading unit quantifies the labels in the set based on the amount of sensitive information identified and the confidentiality attributes, and calculates a graded comprehensive score: here, is the total amount of sensitive information, It is The weight factor of sensitive information, It is the basic score of the sensitive information.

7. The transfer station system for secure sharing of audit files according to claim 6 is characterized in that: The blockchain contract management unit includes a blockchain network composed of multiple blockchain nodes; wherein the black box unit serves as a blockchain node, and deploys at least one blockchain node for the user end of each participant that has access and operation requirements for audit files; the blockchain network adopts a consortium chain architecture, and the nodes of the blockchain network are interconnected through a P2P (Peer-to-Peer) network to form a decentralized network structure; a smart contract is deployed in the blockchain network, and the smart contract presets a series of rules. As a self-executing program on the blockchain, it ensures that the classification and archiving of audit files, the execution of approval management workflows, the control of audit file access and operation permissions, and log recording and tracing are automatically managed in accordance with established rules without the need for external intervention.

8. A transfer method for secure sharing of audit files, characterized in that: include: Receive the file stream uploaded by the user and store it in a temporary cache area; and, perform verification of audit documents; After verification, the audit file is encrypted using a symmetric encryption algorithm and transmitted to the black box unit for storage through a secure channel; Write and store the encrypted audit files in the secure storage layer of the black box unit; Through the secure storage layer, only access requests to the encrypted audit files that have passed the authority authentication are allowed; After the authority is authenticated, the audit file is obtained, and the trained dedicated large model is used to identify and analyze the sensitive information in the audit file in a secure computing environment to obtain a graded comprehensive score that includes the sensitive information identified in the audit file and its confidentiality attributes; Based on the smart contract mechanism written into the blockchain, the operations related to shared access and operation of the audit files are automatically executed based on the said graded comprehensive score; The loss function used in the fine-tuning phase of the training of the dedicated large model combines the cross entropy loss of the sensitive information classification task and the regression loss of the confidentiality attribute level; The classification loss is the cross entropy loss of sensitive information classification, and the regression loss is the mean square error of confidentiality attribute prediction. is a balance coefficient; The identification and analysis of sensitive information in the audit file includes: a dedicated large model extracts features based on text features through a series of neuron layer analysis and performs sensitive information identification, involving the following formula: here, Indicates that given a feature vector, the text information belongs to the category The probability of the category here Indicates the category of sensitive information. The value is 1-n; non-sensitive information is uniformly classified as category 0; after identifying sensitive information, the model will further determine its confidentiality attributes based on the classification of these sensitive information; the output layer of the model will generate and output a set of multiple labels, each label consists of two parts, namely the sensitive information type and the corresponding confidentiality attribute; the specific form is as follows: here, Indicates Types of sensitive information, Indicates the confidentiality attribute corresponding to the information.

Images