A document privacy and security assessment method, apparatus, device, and medium

By using cross-modal pre-trained models and big data matching technology, the problem of low efficiency in assessing privacy leakage risks in publicly available enterprise documents has been solved, achieving efficient and accurate privacy security assessment and protecting enterprise information security.

CN116680743BActive Publication Date: 2026-11-10SHANDONG INSPUR SCI RES INST CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202310782089.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-06-29
Publication Date
2026-11-10
Estimated Expiration
2043-06-29

AI Technical Summary

Technical Problem

Existing technologies struggle to efficiently identify and assess potential privacy breach risks in publicly available corporate documents, resulting in inadequate security for trade secrets and customer information.

Method used

A cross-modal pre-trained model is used for document parsing and feature encoding. This is combined with a local internet search engine and an online trained large model for data matching and reconstruction, generating a privacy and security assessment report.

Benefits of technology

It improves the efficiency and accuracy of privacy and security assessments of publicly available enterprise documents, enabling timely detection and resolution of potential leakage risks, and protecting trade secrets and customer information security.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116680743B_ABST
    Figure CN116680743B_ABST
Patent Text Reader

Abstract

This application discloses a document privacy and security assessment method, apparatus, device, and medium, relating to the fields of deep learning, cross-modal learning, cue learning, and privacy protection. The method includes parsing publicly available enterprise documents to obtain document data; identifying sensitive data within the document data to obtain sensitive data; masking and replacing the sensitive data to obtain masked de-identified data; matching the masked de-identified data to obtain matching results; restoring the masked de-identified data to obtain restored data; feature encoding the document data to obtain the document to be assessed; generating privacy leakage risk warning information; parsing the document to be assessed to obtain a privacy and security analysis result; and generating a document privacy and security assessment report based on the publicly available enterprise documents, masked de-identified data, matching results, restored data, privacy leakage risk warning information, and privacy and security analysis results. This increases the efficiency and improves the security of document privacy and security assessments.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the fields of deep learning, cross-modal learning, cue learning, and privacy protection, and particularly to a document privacy and security assessment method, apparatus, device, and medium. Background Technology

[0002] In modern business operations, the existence of publicly available documents is unavoidable. With the advent of the digital age, daily business operations involve a large number of publicly available documents, including various information published on company websites, such as company profiles, product manuals, news and announcements, market research reports, white papers, and user manuals. For listed companies, official websites typically also provide investor relations pages, which include important documents such as financial reports, annual reports, quarterly reports, shareholder meeting information, and board resolutions to meet investors' needs for financial and corporate governance information. However, these publicly available documents may contain sensitive trade secrets, customer information, and other confidential data. Once leaked, they can cause serious losses and legal risks to the company. Traditional methods often rely on manual review and rule-making, which are inefficient and prone to overlooking potential privacy breaches.

[0003] As can be seen from the above, how to avoid leaking private information in documents, increase the efficiency of privacy security assessments of publicly available corporate documents, promptly identify and resolve potential leakage risks, protect the security of trade secrets and customer information, and thus improve the security of privacy security assessments of publicly available corporate documents are problems that need to be solved in this field. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a document privacy security assessment method, apparatus, device, and medium that can avoid the leakage of private information in documents, increase the efficiency of privacy security assessment of publicly available enterprise documents, promptly identify and resolve potential leakage risks, protect the security of trade secrets and customer information, and thus improve the security of privacy security assessment of publicly available enterprise documents. The specific solution is as follows:

[0005] Firstly, this application discloses a document privacy and security assessment method, including:

[0006] Obtain publicly available enterprise documents, parse the publicly available enterprise documents to obtain document data, and identify sensitive data from the document data to obtain sensitive data;

[0007] The sensitive data is masked and replaced to obtain masked de-identified data. The masked de-identified data is matched using a local Internet search engine to obtain matching results. Finally, the masked de-identified data is restored using an online trained large model to obtain restored data.

[0008] The document data is feature-encoded to obtain the document to be evaluated, and privacy leakage risk warning information corresponding to the document to be evaluated is generated. Based on the privacy leakage risk warning information and using a preset cross-modal pre-trained model, the document to be evaluated is parsed to obtain privacy and security analysis results.

[0009] A document privacy and security assessment report is generated based on the publicly available enterprise documents, the masked and anonymized data, the matching results, the restored data, the privacy leakage risk warning information, and the privacy and security analysis results.

[0010] Optionally, the step of obtaining publicly available enterprise documents and parsing the publicly available enterprise documents to obtain document data includes:

[0011] The publicly available enterprise documents sent by the client are obtained through a preset secure channel; wherein, the publicly available enterprise documents include product manuals, news and announcements, market research reports, white papers, user manuals, financial reports, and annual reports;

[0012] The publicly available enterprise documents are parsed and multimodal data content is extracted to obtain the document data.

[0013] Optionally, the step of performing sensitive data identification on the document data to obtain sensitive data includes:

[0014] Configure the local initial traditional sensitive data identification module according to the preset sensitive data identification rules to obtain the traditional sensitive data identification module;

[0015] The document data is identified using the traditional sensitive data identification module to obtain sensitive data.

[0016] Optionally, the step of using a local internet search engine to perform masked data matching on the masked data to obtain matching results includes:

[0017] Establish a first connection relationship between the local initial Internet search engine and each external Internet search engine, obtain a list of application interfaces corresponding to each of the external Internet search engines through the first connection relationship, configure the initial Internet search engine according to the list of application interfaces to obtain the Internet search engine.

[0018] The masked and de-identified data is matched using the internet search engine to obtain matching results.

[0019] Optionally, the step of using an online-trained large model to restore the masked and anonymized data to obtain the restored data includes:

[0020] A second connection relationship is established between the local initial online training large model and each external online training large model. The application programming interface list corresponding to each of the external online training large models is obtained through the second connection relationship. The initial online training large model is configured according to the application programming interface list to obtain the online training large model.

[0021] The masked and anonymized data is restored using the online trained large model to obtain the restored data.

[0022] Optionally, the step of feature encoding the document data to obtain the document to be evaluated includes:

[0023] The document data is encoded using a local feature encoding module according to the document data type to obtain the document to be evaluated; wherein, the document data type includes images and text.

[0024] Optionally, the feature encoding operation includes:

[0025] If the document data type is an image, then the local feature encoding module and the convolutional neural network feature extraction method are used to perform feature encoding extraction on the document data to obtain a feature vector sequence and generate a document to be evaluated; if the document data type is text, then the local feature encoding module and the preset Transformer architecture neural network model are used to perform feature encoding extraction on the document data to obtain a feature vector sequence and generate a document to be evaluated.

[0026] Optionally, generating a document privacy and security assessment report based on the publicly available enterprise document, the masked and anonymized data, the matching results, the restored data, the privacy leakage risk warning information, and the privacy and security analysis results includes:

[0027] The matching results and the restored data are compared with the publicly available enterprise documents. At the same time, the masked and anonymized data, the privacy leakage risk warning information, and the privacy security analysis results are cross-compare and verify to obtain the comparison results and cross-compare and verification results.

[0028] The comparison results and the cross-comparison verification results are evaluated and analyzed to obtain a document privacy and security assessment report.

[0029] Secondly, this application discloses a document privacy and security assessment device, comprising:

[0030] The parsing and identification module is used to obtain publicly available enterprise documents, parse the publicly available enterprise documents to obtain document data, and identify sensitive data from the document data to obtain sensitive data.

[0031] The data matching and restoration module is used to perform mask replacement on the sensitive data to obtain masked desensitized data, use a local Internet search engine to match the masked desensitized data to obtain matching results, and use an online trained large model to restore the masked desensitized data to obtain restored data.

[0032] The feature encoding module is used to encode the document data to obtain the document to be evaluated, generate privacy leakage risk warning information corresponding to the document to be evaluated, and parse the document to be evaluated based on the privacy leakage risk warning information and using a preset cross-modal pre-trained model to obtain privacy and security analysis results.

[0033] The document privacy and security report generation module is used to generate a document privacy and security assessment report based on the publicly available enterprise documents, the masked and anonymized data, the matching results, the restored data, the privacy leakage risk warning information, and the privacy and security analysis results.

[0034] Thirdly, this application discloses an electronic device, including:

[0035] Memory, used to store computer programs;

[0036] A processor is used to execute the computer program to implement the aforementioned document privacy and security assessment method.

[0037] Fourthly, this application discloses a computer storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the aforementioned disclosed document privacy and security assessment method.

[0038] As can be seen, this application provides a document privacy and security assessment method, including: acquiring publicly available enterprise documents; parsing the publicly available enterprise documents to obtain document data; identifying sensitive data in the document data to obtain sensitive data; masking and replacing the sensitive data to obtain masked and de-identified data; matching the masked and de-identified data using a local internet search engine to obtain matching results; and restoring the masked and de-identified data using an online trained large model to obtain restored data; feature encoding the document data to obtain a document to be assessed; generating privacy leakage risk warning information corresponding to the document to be assessed; parsing the document to be assessed based on the privacy leakage risk warning information and using a preset cross-modal pre-trained model to obtain a privacy and security analysis result; and generating a document privacy and security assessment report based on the publicly available enterprise documents, the masked and de-identified data, the matching results, the restored data, the privacy leakage risk warning information, and the privacy and security analysis result. This application leverages the general artificial intelligence capabilities of a cross-modal pre-trained model to better uncover deep semantic information in documents, deeply understand and analyze different types of data, and more accurately determine whether sensitive information exists in documents. Furthermore, the cross-modal pre-trained model employs a private deployment method to protect the security of trade secrets and customer information. By comparing and verifying with local internet search engines and various external internet search engines, the timeliness and diversity of sensitive data identification are improved. Simultaneously, by comparing a large-scale local online training model with various external online training models, masked and anonymized data is restored, reducing the risk of document privacy leakage. This application has high practical value and broad application prospects. It can increase the efficiency of privacy security assessments of publicly available enterprise documents, promptly identify and resolve potential leakage risks, protect the security of trade secrets and customer information, and thus improve the security of privacy security assessments of publicly available enterprise documents. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0040] Figure 1 This application discloses a flowchart for a document privacy and security assessment.

[0041] Figure 2 This is another document privacy and security assessment flowchart disclosed in this application;

[0042] Figure 3This is a diagram of a privacy and security assessment system for publicly available enterprise documents disclosed in this application;

[0043] Figure 4 This application discloses a workflow diagram of each module in an enterprise public document privacy and security assessment system;

[0044] Figure 5 This application discloses a schematic diagram of a document privacy and security assessment device;

[0045] Figure 6 This application provides a structural diagram of an electronic device. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] In modern business operations, the existence of publicly available documents is unavoidable. With the advent of the digital age, companies' daily operations involve a large number of publicly available documents, including various information published on company websites, such as company profiles, product manuals, news and announcements, market research reports, white papers, and user manuals. For listed companies, official websites typically also provide investor relations pages, including important documents such as financial reports, annual reports, quarterly reports, shareholder meeting information, and board resolutions, to meet investors' needs for financial and corporate governance information. However, these publicly available documents may contain sensitive trade secrets, customer information, and other confidential data. Once leaked, they can cause serious losses and legal risks to the company. Traditional methods often rely on manual review and rule-making, which are inefficient and prone to overlooking potential privacy breaches. Therefore, how to avoid leaking private information in documents, increase the efficiency of privacy security assessments of publicly available corporate documents, promptly identify and resolve potential leakage risks, protect the security of trade secrets and customer information, and thus improve the security of privacy security assessments of publicly available corporate documents are problems that need to be solved in this field.

[0048] This application is applied to an enterprise public document privacy and security assessment system. This system assesses the privacy and security of enterprise documents and outputs an assessment report. The system consists of modules including a document parsing and preprocessing module, a traditional sensitive data identification module, a mask replacement module, a search engine data adaptation module (Internet search engine), an external online pre-trained large model data adaptation module (online pre-trained large model), a feature encoding module (Encoder), a prompt data generator (Prompt Gen), a local document privacy and security assessment module, and a comprehensive privacy and security assessment module.

[0049] See Figure 1 As shown in the figure, this invention discloses a document privacy and security assessment method, which may specifically include:

[0050] Step S11: Obtain publicly available enterprise documents, parse the publicly available enterprise documents to obtain document data, and identify sensitive data from the document data to obtain sensitive data.

[0051] In this embodiment, publicly available enterprise documents sent by the client are acquired through a preset secure channel. These documents include product manuals, news and announcements, market research reports, white papers, user manuals, financial reports, and annual reports. The publicly available enterprise documents are parsed and multimodal data content extracted to obtain document data. Then, sensitive data identification is performed on the document data to obtain sensitive data. Specifically, after acquiring the publicly available enterprise documents, a document parsing preprocessing module is used to parse and extract multimodal data content from the documents to obtain document data. Multimodal data content extraction involves extracting text, images, and other multimodal data content from the documents to prepare for the next step of sensitive data identification.

[0052] The specific process for sensitive data identification is as follows: A local initial traditional sensitive data identification module is configured according to preset sensitive data identification rules to obtain a traditional sensitive data identification module. This module is then used to identify sensitive data in the document data to obtain the sensitive data. The traditional sensitive data identification module is used to identify potentially sensitive data in the document data using traditional sensitive data identification methods, including keyword matching, regular expressions, entity recognition, and data pattern recognition. The sensitive data identification rules are formulated based on prior knowledge.

[0053] Step S12: Mask the sensitive data to obtain masked de-identified data, use a local Internet search engine to match the masked de-identified data to obtain matching results, and use an online trained large model to restore the masked de-identified data to obtain restored data.

[0054] In this embodiment, specific symbols, strings, or other forms of alternative content are used as masks. The mask replacement module is used to replace the identified sensitive data to obtain masked desensitized data. The masked desensitized data is matched with the local Internet search engine to obtain the matching results. Finally, the masked desensitized data is restored using an online trained large model to obtain the restored data.

[0055] Step S13: Perform feature encoding on the document data to obtain the document to be evaluated, generate privacy leakage risk warning information corresponding to the document to be evaluated, and parse the document to be evaluated based on the privacy leakage risk warning information and using a preset cross-modal pre-trained model to obtain privacy and security analysis results.

[0056] In this embodiment, the specific process for determining the document to be evaluated is as follows: using a local feature encoding module and performing corresponding feature encoding on the document data according to the document data type to obtain the document to be evaluated; wherein, the document data type includes images and text; correspondingly, the feature encoding operation includes: if the document data type is an image, then using a local feature encoding module and employing a convolutional neural network (CNN) feature extraction method to perform feature encoding extraction on the document data to obtain a feature vector sequence and generate the document to be evaluated; if the document data type is text, then using a local feature encoding module and a preset Transformer architecture neural network model to perform feature encoding extraction on the document data to obtain a feature vector sequence and generate the document to be evaluated.

[0057] In this embodiment, a feature encoding module is used to encode the document data. The core of the feature encoding module is a neural network model. Feature encoding is the transformation of the document data representation, preparing for the subsequent generation of a document privacy and security assessment report. A prompt data generator is used to generate privacy leakage risk warning information corresponding to the document to be evaluated. The prompt data generator generates appropriate privacy leakage risk warning information based on the content and structural characteristics of the document to guide the cross-modal pre-trained model to focus on parts that may have privacy leakage risks.

[0058] In this embodiment, the local document privacy and security assessment module is used to parse the document to be assessed. The core of this module is a cross-modal pre-trained model, a deep learning-based artificial intelligence technique. By pre-training on large-scale datasets, it can learn the semantic relationships and representations between multimodal data (such as text, images, and audio). These models possess strong general artificial intelligence capabilities, enabling in-depth understanding and analysis of different types of data, and the generation of semantically coherent content. The cross-modal pre-trained model allows for the parsing of data corresponding to different document data types, outputting sensitive information or potential privacy leakage risks within the document data, and obtaining privacy and security analysis results. Before obtaining the privacy and security analysis results, the feature encoding module and the prompt data generator need to be adjusted based on the real data of the privacy assessment task to ensure that the local document privacy and security assessment module can output privacy and security analysis results that meet the requirements.

[0059] In this embodiment, before parsing the document to be evaluated using the cross-modal pre-trained model, it is necessary to obtain a publicly available initial cross-modal pre-trained model. Based on the actual needs of enterprise public document privacy and security assessment, the parameters of the initial cross-modal pre-trained model are fine-tuned using real data from the privacy assessment to form the core model of the local document privacy and security assessment module (i.e., the cross-modal pre-trained model). Then, it is deployed privately. The cross-modal pre-trained model is fine-tuned and deployed privately, providing services only internally, thereby ensuring the security of the cross-modal pre-trained model.

[0060] Step S14: Generate a document privacy and security assessment report based on the publicly available enterprise documents, the masked and anonymized data, the matching results, the restored data, the privacy leakage risk warning information, and the privacy and security analysis results.

[0061] In this embodiment, the matching result, the restored data, and the publicly available enterprise document are compared. Simultaneously, the masked and anonymized data, the privacy leakage risk warning information, and the privacy security analysis result are cross-computed and verified to obtain the comparison result and the cross-computed verification result. The comparison result and the cross-computed verification result are evaluated and analyzed to obtain a document privacy security assessment report.

[0062] The comprehensive privacy and security assessment module generates a document privacy and security assessment report. Specifically, it comprehensively considers publicly available enterprise documents, masked and anonymized data, matching results, restored data, privacy leakage risk warnings, and privacy and security analysis results to conduct a comprehensive assessment and judgment of the privacy and security of publicly available enterprise documents and generate a document privacy and security assessment report.

[0063] In addition, the enterprise's public document privacy and security assessment system can continuously collect feedback data, thereby adjusting internal and external rules in a timely manner, continuously optimizing the above model, and improving the accuracy and timeliness of document privacy and security assessment.

[0064] In this embodiment, publicly available enterprise documents are acquired, parsed to obtain document data, and sensitive data is identified within the document data to obtain sensitive data. The sensitive data is then masked and replaced to obtain masked and anonymized data. A local internet search engine is used to match the masked and anonymized data to obtain matching results, and an online trained large-scale model is used to restore the masked and anonymized data to obtain restored data. Feature encoding is performed on the document data to obtain a document to be evaluated, and privacy leakage risk warning information corresponding to the document to be evaluated is generated. Based on the privacy leakage risk warning information and a preset cross-modal pre-trained model, the document to be evaluated is parsed to obtain a privacy and security analysis result. Finally, a document privacy and security assessment report is generated based on the publicly available enterprise documents, the masked and anonymized data, the matching results, the restored data, the privacy leakage risk warning information, and the privacy and security analysis result. This application leverages the general artificial intelligence capabilities of a cross-modal pre-trained model to better uncover deep semantic information in documents, deeply understand and analyze different types of data, and more accurately determine whether sensitive information exists in documents. Furthermore, the cross-modal pre-trained model employs a private deployment method to protect the security of trade secrets and customer information. By comparing and verifying with local internet search engines and various external internet search engines, the timeliness and diversity of sensitive data identification are improved. Simultaneously, by comparing a large-scale local online training model with various external online training models, masked and anonymized data is restored, reducing the risk of document privacy leakage. This application has high practical value and broad application prospects. It can increase the efficiency of privacy security assessments of publicly available enterprise documents, promptly identify and resolve potential leakage risks, protect the security of trade secrets and customer information, and thus improve the security of privacy security assessments of publicly available enterprise documents.

[0065] See Figure 2 As shown in the figure, this invention discloses a document privacy and security assessment method, which may specifically include:

[0066] Step S21: Obtain publicly available enterprise documents, parse the publicly available enterprise documents to obtain document data, and identify sensitive data from the document data to obtain sensitive data.

[0067] Step S22: Mask the sensitive data to obtain masked de-identified data, establish a first connection relationship between the local initial Internet search engine and each external Internet search engine, obtain the application interface list corresponding to each of the external Internet search engines through the first connection relationship, configure the initial Internet search engine according to each application interface list to obtain the Internet search engine, and use the Internet search engine to perform de-identified data matching on the masked de-identified data to obtain the matching result.

[0068] In this embodiment, multiple external internet search engines are selected, and their API (Application Program Interface) lists are obtained. Based on the API list, the initial internet search engine is configured to obtain an internet search engine (i.e., the search engine data adaptation module). This module is then used to match the masked and anonymized data, yielding matching results. External internet search engines are tools for searching and obtaining information on the internet, performing web page searches, document library searches, or social media searches based on keywords or query phrases. The data adaptation module matches the masked and anonymized data with the API interfaces of these external internet search engines. Specifically, the masked and anonymized data is truncated and selected, and the API interfaces of multiple external internet search engines are called to obtain multiple sets of search matching results, ultimately generating the final matching result.

[0069] Step S23: Establish a second connection relationship between the local initial online training large model and each external online training large model. Obtain the application interface list corresponding to each of the external online training large models through the second connection relationship. Configure the initial online training large model according to each application interface list to obtain the online training large model. Use the online training large model to restore the masked and desensitized data to obtain the restored data.

[0070] In this embodiment, the local online pre-trained large model (i.e., the external online pre-trained large model data adaptation module) is used to restore data from the API lists provided by other cloud vendors' external online pre-trained large models. Utilizing its text dialogue generation function or cross-modal text generation capability, it restores the masked and anonymized data to obtain the restored data. Specifically, the restored data obtained by restoring the masked and anonymized data using the online pre-trained large model is identical to the sensitive data obtained before the mask replacement operation. In other words, after obtaining the sensitive data, a mask replacement is performed on the sensitive data to obtain masked and anonymized data. The restored data obtained by restoring the masked and anonymized data using the online pre-trained large model is the sensitive data. Online pre-trained large models are natural language large model services or cross-modal large model services provided by cloud service providers. Natural language large model services usually adopt an architecture similar to GPT (Generative Pre-trained Transformer, Natural Language Processing Model), which can accept text input and generate relevant text output. In addition, cross-modal large models combine multiple modal data, such as text, images, and speech, and can realize cross-modal data processing and generation.

[0071] Step S24: Perform feature encoding on the document data to obtain the document to be evaluated, generate privacy leakage risk warning information corresponding to the document to be evaluated, and parse the document to be evaluated based on the privacy leakage risk warning information and using a preset cross-modal pre-trained model to obtain privacy and security analysis results.

[0072] Step S25: Generate a document privacy and security assessment report based on the publicly available enterprise documents, the masked and anonymized data, the matching results, the restored data, the privacy leakage risk warning information, and the privacy and security analysis results.

[0073] This application is applied to an enterprise open-source document privacy and security assessment system. This system assesses the privacy and security of enterprise documents and outputs an assessment report. The structure of the enterprise open-source document privacy and security assessment system is as follows: Figure 3 As shown, it includes modules such as document parsing preprocessing module, traditional sensitive data identification module, mask replacement module, search engine data adaptation module, external online pre-trained large model data adaptation module, feature encoding module, prompt data generator, local document privacy and security assessment module, and comprehensive privacy and security assessment module.

[0074] The module's functions and processes are as follows: Figure 4As shown: (1) Document parsing preprocessing module, used to parse and extract multimodal data content from publicly available enterprise documents to obtain the document data; (2) Traditional sensitive data identification module, used to identify sensitive data that may exist in the document to obtain sensitive data, mainly using keyword matching, regular expressions, entity recognition, data pattern recognition and other methods; (3) Mask replacement module, used to replace the identified sensitive data with a mask, including but not limited to using specific symbols, strings or other forms of replacement content as a mask to obtain masked desensitized data; (4) Search engine data adaptation module, used to match desensitized data with the API interface list of external Internet search engines to obtain matching results; (5) External online pre-trained large model data adaptation module, used to restore data with the API interface list of external online pre-trained large models provided by other cloud vendors, and use its text dialogue generation function or cross-modal text generation function. (6) Feature encoding module, whose core is a neural network model, is used to encode the features of the document data and convert it into a suitable representation to obtain the document to be evaluated; (7) Prompt data generator, which is used to generate corresponding privacy leakage risk prompt information according to the content and structural characteristics of the document to be evaluated, so as to guide the cross-modal pre-trained model to pay attention to the parts that may have privacy leakage risks; (8) Local document privacy security assessment module, whose core is a cross-modal pre-trained model, realizes the understanding and analysis of the document to be evaluated, outputs sensitive information or potential privacy leakage risks in the document, and obtains privacy security analysis results; (9) Comprehensive privacy security assessment module, which is used to comprehensively consider the enterprise's public documents, masked desensitized data, matching results, restored data, privacy leakage risk prompt information, and privacy security analysis results to comprehensively assess and judge the privacy security of the enterprise's public documents and generate a document privacy security assessment report.

[0075] In this embodiment, publicly available enterprise documents are acquired, parsed to obtain document data, and sensitive data is identified within the document data to obtain sensitive data. The sensitive data is then masked and replaced to obtain masked and anonymized data. A local internet search engine is used to match the masked and anonymized data to obtain matching results, and an online trained large-scale model is used to restore the masked and anonymized data to obtain restored data. Feature encoding is performed on the document data to obtain a document to be evaluated, and privacy leakage risk warning information corresponding to the document to be evaluated is generated. Based on the privacy leakage risk warning information and a preset cross-modal pre-trained model, the document to be evaluated is parsed to obtain a privacy and security analysis result. Finally, a document privacy and security assessment report is generated based on the publicly available enterprise documents, the masked and anonymized data, the matching results, the restored data, the privacy leakage risk warning information, and the privacy and security analysis result. This application leverages the general artificial intelligence capabilities of a cross-modal pre-trained model to better uncover deep semantic information in documents, deeply understand and analyze different types of data, and more accurately determine whether sensitive information exists in documents. Furthermore, the cross-modal pre-trained model employs a private deployment method to protect the security of trade secrets and customer information. By comparing and verifying with local internet search engines and various external internet search engines, the timeliness and diversity of sensitive data identification are improved. Simultaneously, by comparing a large-scale local online training model with various external online training models, masked and anonymized data is restored, reducing the risk of document privacy leakage. This application has high practical value and broad application prospects. It can increase the efficiency of privacy security assessments of publicly available enterprise documents, promptly identify and resolve potential leakage risks, protect the security of trade secrets and customer information, and thus improve the security of privacy security assessments of publicly available enterprise documents.

[0076] See Figure 5 As shown in the embodiment of the present invention, a document privacy and security assessment device is disclosed, which may specifically include:

[0077] The parsing and identification module 11 is used to obtain publicly available enterprise documents, parse the publicly available enterprise documents to obtain document data, and identify sensitive data from the document data to obtain sensitive data.

[0078] The data matching and restoration module 12 is used to perform mask replacement on the sensitive data to obtain masked desensitized data, use a local Internet search engine to perform desensitized data matching on the masked desensitized data to obtain matching results, and use an online trained large model to restore the masked desensitized data to obtain restored data.

[0079] The feature encoding module 13 is used to encode the document data to obtain the document to be evaluated, generate privacy leakage risk warning information corresponding to the document to be evaluated, and parse the document to be evaluated based on the privacy leakage risk warning information and using a preset cross-modal pre-trained model to obtain privacy and security analysis results.

[0080] The document privacy and security report generation module 14 is used to generate a document privacy and security assessment report based on the publicly available enterprise documents, the masked and anonymized data, the matching results, the restored data, the privacy leakage risk warning information, and the privacy and security analysis results.

[0081] In this embodiment, publicly available enterprise documents are acquired, parsed to obtain document data, and sensitive data is identified within the document data to obtain sensitive data. The sensitive data is then masked and replaced to obtain masked and anonymized data. A local internet search engine is used to match the masked and anonymized data to obtain matching results, and an online trained large-scale model is used to restore the masked and anonymized data to obtain restored data. Feature encoding is performed on the document data to obtain a document to be evaluated, and privacy leakage risk warning information corresponding to the document to be evaluated is generated. Based on the privacy leakage risk warning information and a preset cross-modal pre-trained model, the document to be evaluated is parsed to obtain a privacy and security analysis result. Finally, a document privacy and security assessment report is generated based on the publicly available enterprise documents, the masked and anonymized data, the matching results, the restored data, the privacy leakage risk warning information, and the privacy and security analysis result. This application leverages the general artificial intelligence capabilities of a cross-modal pre-trained model to better uncover deep semantic information in documents, deeply understand and analyze different types of data, and more accurately determine whether sensitive information exists in documents. Furthermore, the cross-modal pre-trained model employs a private deployment method to protect the security of trade secrets and customer information. By comparing and verifying with local internet search engines and various external internet search engines, the timeliness and diversity of sensitive data identification are improved. Simultaneously, by comparing a large-scale local online training model with various external online training models, masked and anonymized data is restored, reducing the risk of document privacy leakage. This application has high practical value and broad application prospects. It can increase the efficiency of privacy security assessments of publicly available enterprise documents, promptly identify and resolve potential leakage risks, protect the security of trade secrets and customer information, and thus improve the security of privacy security assessments of publicly available enterprise documents.

[0082] In some specific embodiments, the parsing and identification module 11 may specifically include:

[0083] The document acquisition module is used to acquire publicly available enterprise documents sent by the client through a preset secure channel; wherein, the publicly available enterprise documents include product manuals, news and announcements, market research reports, white papers, user manuals, financial reports, and annual reports;

[0084] The parsing and content extraction module is used to parse the publicly available enterprise documents and extract multimodal data content to obtain the document data.

[0085] In some specific embodiments, the parsing and identification module 11 may specifically include:

[0086] The first configuration module is used to configure the local initial traditional sensitive data identification module according to the preset sensitive data identification rules to obtain the traditional sensitive data identification module;

[0087] A sensitive data identification module is used to identify sensitive data in the document data using the traditional sensitive data identification module, so as to obtain sensitive data.

[0088] In some specific embodiments, the data matching and restoration module 12 may specifically include:

[0089] The second configuration module is used to establish a first connection relationship between the local initial Internet search engine and each external Internet search engine, obtain a list of application interfaces corresponding to each of the external Internet search engines through the first connection relationship, and configure the initial Internet search engine according to the list of application interfaces to obtain the Internet search engine.

[0090] The de-identified data matching module is used to perform de-identified data matching on the masked de-identified data using the Internet search engine to obtain matching results.

[0091] In some specific embodiments, the data matching and restoration module 12 may specifically include:

[0092] The third configuration module is used to establish a second connection relationship between the local initial online training large model and each external online training large model, obtain the application interface list corresponding to each of the external online training large models through the second connection relationship, and configure the initial online training large model according to the application interface list to obtain the online training large model.

[0093] The data restoration module is used to restore the masked and anonymized data using the online trained large model to obtain the restored data.

[0094] In some specific embodiments, the feature encoding module 13 may specifically include:

[0095] The feature encoding module is used to perform corresponding feature encoding on the document data according to the document data type using a local feature encoding module, so as to obtain the document to be evaluated; wherein, the document data type includes images and text;

[0096] The document generation module is used to extract features from the document data by using a local feature encoding module and a convolutional neural network feature extraction method if the document data type is an image, so as to obtain a feature vector sequence and generate the document to be evaluated; if the document data type is text, it uses a local feature encoding module and a preset Transformer architecture neural network model to extract features from the document data, so as to obtain a feature vector sequence and generate the document to be evaluated.

[0097] In some specific embodiments, the document privacy and security report generation module 14 may specifically include:

[0098] The cross-comparison and verification module is used to compare the matching results, the restored data and the publicly available enterprise documents, and to perform cross-comparison and verification on the masked and anonymized data, the privacy leakage risk warning information and the privacy security parsing results to obtain the comparison results and the cross-comparison and verification results.

[0099] The evaluation and analysis module is used to evaluate and analyze the comparison results and the cross-comparison verification results to obtain a document privacy and security evaluation report.

[0100] Figure 6 This application provides a schematic diagram of the structure of an electronic device. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the document privacy and security assessment method performed by the electronic device disclosed in any of the foregoing embodiments.

[0101] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0102] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored on it include operating system 221, computer program 222 and data 223, etc., and the storage method can be temporary storage or permanent storage.

[0103] The operating system 221 manages and controls the various hardware devices on the electronic device 20 and the computer program 222 to enable the processor 21 to perform operations and processing on the data 223 in the memory 22. The operating system 221 can be Windows, Unix, Linux, etc. The computer program 222, in addition to including a computer program capable of performing the document privacy and security assessment method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, may further include computer programs capable of performing other specific tasks. The data 223 may include data received by the document privacy and security assessment device from external devices, as well as data collected by its own input / output interface 25.

[0104] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0105] Furthermore, embodiments of this application also disclose a computer-readable storage medium storing a computer program. When the computer program is loaded and executed by a processor, it implements the document privacy and security assessment method steps disclosed in any of the foregoing embodiments.

[0106] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0107] The foregoing has provided a detailed description of a document privacy security assessment method, apparatus, device, and storage medium provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A document privacy and security assessment method, characterized in that, include: Obtain publicly available enterprise documents, parse the publicly available enterprise documents to obtain document data, and identify sensitive data from the document data to obtain sensitive data; The sensitive data is masked and replaced to obtain masked de-identified data. The masked de-identified data is matched using a local Internet search engine to obtain matching results. Finally, the masked de-identified data is restored using an online trained large model to obtain restored data. The document data is feature-encoded to obtain the document to be evaluated, and privacy leakage risk warning information corresponding to the document to be evaluated is generated. Based on the privacy leakage risk warning information and using a preset cross-modal pre-trained model, the document to be evaluated is parsed to obtain a privacy and security analysis result. The privacy leakage risk warning information is used to guide the cross-modal pre-trained model to focus on parts that may have privacy leakage risks. The privacy and security analysis result includes sensitive information or potential privacy leakage risks in the document data. A document privacy and security assessment report is generated based on the publicly available enterprise documents, the masked and anonymized data, the matching results, the restored data, the privacy leakage risk warning information, and the privacy and security analysis results. The process involves generating a document privacy and security assessment report based on publicly available enterprise documents, masked and anonymized data, matching results, restored data, privacy leakage risk warnings, and privacy and security analysis results. This includes comparing the matching results and restored data with publicly available enterprise documents, and simultaneously cross-verifying the masked and anonymized data, privacy leakage risk warnings, and privacy and security analysis results to obtain comparison results and cross-verification results. The comparison results and cross-verification results are then evaluated and analyzed to obtain the document privacy and security assessment report.

2. The document privacy and security assessment method according to claim 1, characterized in that, The process of obtaining publicly available enterprise documents and parsing those documents to obtain document data includes: The publicly available enterprise documents sent by the client are obtained through a preset secure channel; wherein, the publicly available enterprise documents include product manuals, news and announcements, market research reports, white papers, user manuals, financial reports, and annual reports; The publicly available enterprise documents are parsed and multimodal data content is extracted to obtain the document data.

3. The document privacy and security assessment method according to claim 1, characterized in that, The process of identifying sensitive data in the document data to obtain sensitive data includes: Configure the local initial traditional sensitive data identification module according to the preset sensitive data identification rules to obtain the traditional sensitive data identification module; The document data is identified using the traditional sensitive data identification module to obtain sensitive data.

4. The document privacy and security assessment method according to claim 1, characterized in that, The step of using a local internet search engine to perform masked data matching on the masked data to obtain matching results includes: Establish a first connection relationship between the local initial Internet search engine and each external Internet search engine, obtain a list of application interfaces corresponding to each of the external Internet search engines through the first connection relationship, configure the initial Internet search engine according to the list of application interfaces to obtain the Internet search engine. The masked and anonymized data is matched using the internet search engine to obtain matching results.

5. The document privacy and security assessment method according to claim 1, characterized in that, The step of using an online-trained large model to restore the masked and anonymized data to obtain the restored data includes: A second connection relationship is established between the local initial online training large model and each external online training large model. The application programming interface list corresponding to each of the external online training large models is obtained through the second connection relationship. The initial online training large model is configured according to the application programming interface list to obtain the online training large model. The masked and anonymized data is restored using the online trained large model to obtain the restored data.

6. The document privacy and security assessment method according to claim 1, characterized in that, The process of feature encoding the document data to obtain the document to be evaluated includes: The document data is encoded using a local feature encoding module according to the document data type to obtain the document to be evaluated. The document data types include images and text; Accordingly, the feature encoding operation includes: if the document data type is an image, then using a local feature encoding module and a convolutional neural network feature extraction method to perform feature encoding extraction on the document data to obtain a feature vector sequence and generate a document to be evaluated; if the document data type is text, then using a local feature encoding module and a preset Transformer architecture neural network model to perform feature encoding extraction on the document data to obtain a feature vector sequence and generate a document to be evaluated.

7. A document privacy and security assessment device, characterized in that, include: The parsing and identification module is used to obtain publicly available enterprise documents, parse the publicly available enterprise documents to obtain document data, and identify sensitive data from the document data to obtain sensitive data. The data matching and restoration module is used to perform mask replacement on the sensitive data to obtain masked desensitized data, use a local Internet search engine to match the masked desensitized data to obtain matching results, and use an online trained large model to restore the masked desensitized data to obtain restored data. The feature encoding module is used to encode the document data to obtain the document to be evaluated, generate privacy leakage risk warning information corresponding to the document to be evaluated, and parse the document to be evaluated based on the privacy leakage risk warning information and a preset cross-modal pre-trained model to obtain a privacy and security analysis result. The privacy leakage risk warning information is used to guide the cross-modal pre-trained model to focus on parts that may have privacy leakage risks. The privacy and security analysis result includes sensitive information or potential privacy leakage risks in the document data. The document privacy and security report generation module is used to generate a document privacy and security assessment report based on the publicly available enterprise documents, the masked and anonymized data, the matching results, the restored data, the privacy leakage risk warning information, and the privacy and security analysis results. The process involves generating a document privacy and security assessment report based on publicly available enterprise documents, masked and anonymized data, matching results, restored data, privacy leakage risk warnings, and privacy and security analysis results. This includes comparing the matching results and restored data with publicly available enterprise documents, and simultaneously cross-verifying the masked and anonymized data, privacy leakage risk warnings, and privacy and security analysis results to obtain comparison results and cross-verification results. The comparison results and cross-verification results are then evaluated and analyzed to obtain the document privacy and security assessment report.

8. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the document privacy and security assessment method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, Used to store computer programs; wherein, when the computer programs are executed by a processor, they implement the document privacy and security assessment method as described in any one of claims 1 to 6.

Citation Information

Patent Citations

  • Big data-based structured sensitive data restoration detection method

    CN112417456A

  • Cross-data-center data desensitization method and device, and electronic equipment

    CN114048511A