Commercial secret document identification method and device, equipment and storage medium

Abstract

The invention discloses a business secret document recognition method, device and equipment and a storage medium, relates to the technical field of computers, is applied to a preset hierarchical architecture comprising an engine layer, a capability layer, a service layer and a product layer, and comprises the following steps: issuing a business secret processing technology determined based on basic capability of a large language model to the service layer; a business secret processing technology is utilized to generate a business secret secret-setting detailed rule making service and a business secret recognition service, the initial enterprise document is preprocessed, and the business secret secret-setting detailed rule making service is utilized to process the obtained to-be-processed enterprise document to obtain a target secret-setting rule; and adjusting the large language model to obtain a target large language model, carrying out association operation on the paragraph and the to-be-recognized document on the target secret-setting detailed rule based on the commercial secret recognition service to obtain an association result, and carrying out security classification judgment on the association result by utilizing the target large language model to obtain a security classification judgment result. Therefore, the accuracy of identifying the commercial secret document can be improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and particularly to a method, apparatus, device and storage medium for identifying trade secret documents. Background Art

[0002] Currently, there are some problems in identifying trade secret documents:

[0003] First, it is difficult for enterprises to formulate detailed secrecy rules. In the process of clearly defining which information belongs to the category of trade secrets, a large amount of professional knowledge and technology is involved, and it is also necessary to have an in-depth understanding of business processes, technical characteristics and market conditions. The existing solutions mainly rely on human experts to formulate, and these knowledge and experiences need to be accumulated over the years, resulting in difficulties in manually formulating perfect detailed secrecy rules.

[0004] Second, the detailed secrecy rules have a high degree of conceptual abstraction. The definition of R & D documents implicitly includes various types of documents such as requirement specifications, design schemes, test reports, etc., and the entire rule also contains the condition: "related to xx project". To understand and process such rules, strong natural language understanding ability and general knowledge understanding ability are required, but keyword matching technology, semantic similarity-based methods, and deep learning-based text classification methods are all difficult to achieve the expected natural language understanding ability.

[0005] Third, the identification process of trade secret documents is opaque. When traditional machine learning models process classification tasks, since the internal complex algorithms and weight adjustments are invisible to users, only the final classification decision is output, and no detailed explanation or reasoning path on how to reach this decision can be provided. In actual application scenarios, users not only hope to obtain accurate classification results through the model, but also hope to deeply understand how the model analyzes document content, which features are considered, and how to comprehensively utilize this information to make decisions.

[0006] Fourth, the specificity of the secrecy system is strong. Under the general definition of trade secrets, different industries and companies have different scopes of information to be protected. The importance of each trade secret also varies from enterprise to enterprise. Even within the same company, with the expansion or adjustment of the business scope, the protection norms for trade secrets need to be updated and changed accordingly. Traditional machine learning models need to be customized for specific data sets and rules. Once the scope of protected matters changes, the model needs to be adjusted accordingly. And the process of model adjustment requires input of manpower, time, resources and professional knowledge.

[0007] As can be seen from the above, how to improve the accuracy of identifying trade secret documents in the process of trade secret document identification is an urgent problem to be solved at present. Summary of the Invention

[0008] In view of this, the purpose of the present invention is to provide a method, device, equipment and storage medium for identifying trade secret documents, which can improve the accuracy of identifying trade secret documents during the identification process. The specific scheme is as follows:

[0009] In a first aspect, the present application provides a method for identifying trade secret documents, which is applied to a preset hierarchical architecture including an engine layer, a capability layer, a service layer and a product layer. The method includes:

[0010] In the engine layer, the basic capabilities corresponding to the large language model, including text classification, text generation, logical reasoning and natural language understanding, are sent to the capability layer, so as to determine the trade secret processing technology based on the basic capabilities in the capability layer, and send the trade secret processing technology to the service layer;

[0011] In the service layer, the trade secret processing technology is used to generate data access services, trade secret classification rule formulation services and trade secret identification services. Then, the data access service is used to receive the initial enterprise documents including enterprise basic situation description information, trade secret documents and historical classification rules, and preprocess the initial enterprise documents to obtain the enterprise documents to be processed. Then, the trade secret classification rule formulation service is used to process the enterprise documents to be processed to obtain the target classification rules;

[0012] In the service layer, the large language model is adjusted using a preset model adjustment rule to obtain a target large language model. Based on the trade secret identification service, an association operation between paragraphs and the document to be identified is performed on the target classification rules to obtain an association result. Then, the target large language model is used to perform a confidentiality level determination on the association result to obtain a confidentiality level determination result, so as to send the confidentiality level determination result to the interaction interface in the product layer.

[0013] Optionally, the step of using the data access service to receive the initial enterprise documents including enterprise basic situation description information, trade secret documents and historical classification rules, and preprocessing the initial enterprise documents to obtain the enterprise documents to be processed includes:

[0014] Receive an initial enterprise document including enterprise basic situation description information, business secret documents, and historical declassification rules using the data access service, and perform a format parsing operation on the initial enterprise document using preset preprocessing rules to obtain a format parsing result. Then, perform table marking, picture marking, and layout marking operations on the format parsing result to obtain a marking result. Next, perform a paragraph marking operation and a metadata annotation operation on the marking result to obtain the mapping relationship between the initial enterprise document and business attributes. The metadata annotation operation includes annotation operations on the company name, company industry type, company main business scope, and company business description.

[0015] Determine an intermediate enterprise document based on the mapping relationship and the initial enterprise document, and use preset task description processing to perform data description on the intermediate enterprise document to obtain a data description result. Then, determine a to-be-processed enterprise document based on the data description result and the intermediate enterprise document.

[0016] Optionally, using the business secret declassification rule formulation service to process the to-be-processed enterprise document to obtain a target declassification rule includes:

[0017] Judge whether the to-be-processed enterprise document is enterprise business scope information. If the to-be-processed enterprise document is the enterprise business scope information, use the business secret declassification rule formulation service to perform word segmentation processing on the to-be-processed enterprise document to obtain a word segmentation processing result, and process the word segmentation processing result based on the structure processing instruction to obtain structured data.

[0018] Use the large language model to process the structured data to generate an initial declassification rule document, and then use the business secret declassification rule formulation service and generate a target declassification rule based on the initial declassification rule document. The initial declassification rule document includes business secret carrier types, classification rules for secrecy levels, and criteria for determining secrecy levels.

[0019] Optionally, using the business secret declassification rule formulation service to process the to-be-processed enterprise document to obtain a target declassification rule includes:

[0020] Judge whether the to-be-processed enterprise document is enterprise business scope information. If the to-be-processed enterprise document is not the enterprise business scope information, set the to-be-processed enterprise document as a collection of business secret sample documents.

[0021] Use the large language model and perform multi-round semantic analysis on the business secret sample documents in the collection of business secret sample documents based on preset multi-round semantic analysis rules to obtain a corresponding collection of business secret carrier types.

[0022] Generate a declassification rules document including specific declassification rules and classification level correlation relationships based on the set of commercial secret carrier types, perform a logical matching check on the declassification rules document and the historical declassification rules to obtain a check result, and then adjust the conflicting clauses in the declassification rules document based on the check result to obtain the target declassification rules.

[0023] Optionally, the adjusting the large language model by using a preset model adjustment rule in the service layer to obtain a target large language model includes:

[0024] Use a preset fine-tuning technique to adjust the text classification accuracy parameters of the large language model by using historical documents and case data in the field of commercial secrets in the service layer to obtain a first large language model;

[0025] Construct a prompt template library including commercial secret feature description information to adjust the domain relevance parameters corresponding to the first large language model by using the prompt template library to obtain a second large language model; the domain relevance parameter is a parameter for making a relevance connection to the domain corresponding to each output result of the model;

[0026] Adjust the logical consistency parameters of the second large language model based on the revision record obtained by the user's revision of the target declassification rules and the feedback information for the recognition result to obtain a target large language model; the logical consistency parameter is a parameter for performing logical consistency processing on each of the output results of the model and the target declassification rules.

[0027] Optionally, the associating the target declassification rules with the paragraphs and the document to be recognized by the commercial secret recognition service to obtain an association result, and then using the target large language model to determine the classification level of the association result to obtain a classification level determination result, so as to send the classification level determination result to the interactive interface in the product layer includes:

[0028] Use the preset paragraph segmentation rule in the commercial secret recognition service to perform a semantic paragraph segmentation operation on the target declassification rules to obtain a paragraph segmentation result;

[0029] Use the preset keyword matching rule and the preset context analysis and extraction rule to perform a content fragment extraction operation on each of the paragraph segmentation results to obtain a content fragment extraction result; the content fragment extraction result includes the classification level determination criteria;

[0030] The content fragment extraction result and the document to be recognized are contextually associated using a preset context association encoding to obtain a context association result. Then, the target large language model is used to process the context association result to obtain a confidentiality level determination result and the determination basis corresponding to the confidentiality level determination result, and the confidentiality level determination result and the determination basis are sent to the interaction interface in the product layer so that the user can view the confidentiality level determination result and the determination basis through the interaction interface.

[0031] Optionally, sending the confidentiality level determination result to the interaction interface in the product layer includes:

[0032] Configure a document upload entry in the product layer to receive several types of historical sample documents, the historical classification rules, the target classification rules, and the confidentiality level determination result using the document upload entry, and set permissions for each user using a preset permission hierarchical mechanism;

[0033] Design and display the generation process corresponding to the confidentiality level determination result through the interaction interface using a preset visualization logic chain so that the user can perform an information viewing operation on the generation process of the confidentiality level determination result based on permissions; wherein, the interaction interface has an interface editing function, a document editing function, and a process traceability function.

[0034] In a second aspect, the present application provides a trade secret document recognition device, which is applied to a preset hierarchical architecture including an engine layer, a capability layer, a service layer, and a product layer. The device includes:

[0035] A processing technology determination module, configured to send the basic capabilities corresponding to the large language model, including text classification, text generation, logical reasoning, and natural language understanding, to the capability layer in the engine layer, so as to determine a trade secret processing technology based on the basic capabilities in the capability layer and send the trade secret processing technology to the service layer;

[0036] A classification rule generation module, configured to generate a data access service, a trade secret classification rule formulation service, and a trade secret recognition service using the trade secret processing technology in the service layer, then receive an initial enterprise document including enterprise basic situation description information, trade secret documents, and historical classification rules using the data access service, preprocess the initial enterprise document to obtain a to-be-processed enterprise document, and then process the to-be-processed enterprise document using the trade secret classification rule formulation service to obtain a target classification rule;

[0037] The confidentiality level determination result determination module is used to adjust the large language model in the service layer by using a preset model adjustment rule to obtain a target large language model, and perform an association operation between paragraphs of the target classification rules and the document to be recognized based on the trade secret recognition service to obtain an association result, and then use the target large language model to perform a confidentiality level determination on the association result to obtain a confidentiality level determination result, so as to send the confidentiality level determination result to the interaction interface in the product layer.

[0038] In a third aspect, the present application provides an electronic device, including:

[0039] A memory for storing a computer program;

[0040] A processor for executing the computer program to implement the foregoing trade secret document recognition method.

[0041] In a fourth aspect, the present application provides a computer-readable storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the foregoing trade secret document recognition method.

[0042] As can be seen from the above, before the present application performs trade secret document recognition, it is necessary to send the basic capabilities corresponding to the large language model, including text classification, text generation, logical reasoning, and natural language understanding, to the capability layer, so as to determine trade secret processing technologies based on the basic capabilities, and use the trade secret processing technologies to generate data access services, trade secret classification rule formulation services, and trade secret recognition services. Then, use the data access service to receive the initial enterprise documents including enterprise basic situation description information, trade secret documents, and historical classification rules, and preprocess the initial enterprise documents to obtain the enterprise documents to be processed. Then, use the trade secret classification rule formulation service to process the enterprise documents to be processed to obtain target classification rules; use a preset model adjustment rule to adjust the large language model to obtain a target large language model, and perform an association operation between paragraphs of the target classification rules and the document to be recognized based on the trade secret recognition service to obtain an association result, and then use the target large language model to perform a confidentiality level determination on the association result to obtain a confidentiality level determination result, so as to send the confidentiality level determination result to the interaction interface in the product layer.

[0043] It can be seen that, first, the basic capabilities of the large language model corresponding to text classification, text generation, logical reasoning, and natural language understanding need to be distributed to the capability layer in this application, so as to determine the trade secret processing technology based on the basic capabilities; subsequently, the trade secret processing technology is used to generate data access services, trade secret classification rule formulation services, and trade secret identification services. Then, the data access service is used to receive the initial enterprise documents including enterprise basic situation description information, trade secret documents, and historical classification rules, and preprocess the initial enterprise documents to obtain the enterprise documents to be processed. Then, the trade secret classification rule formulation service is used to process the enterprise documents to be processed to obtain the target classification rules; furthermore, the large language model is adjusted using the preset model adjustment rules to obtain the target large language model. Finally, based on the trade secret identification service, an association operation between the paragraphs of the target classification rules and the documents to be identified is performed to obtain the association result, and then the target large language model is used to perform a classification level determination on the association result to obtain the classification level determination result, so as to distribute the classification level determination result to the interaction interface in the product layer. In this way, the accuracy of identifying trade secret documents is improved during the trade secret document identification process, thereby enhancing the security of the enterprise production process. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.

[0045] Figure 1 It is a flowchart of a method for identifying trade secret documents disclosed in this application;

[0046] Figure 2 It is a schematic diagram of the overall architecture for specifically identifying trade secret documents disclosed in this application;

[0047] Figure 3 It is a flowchart of a specific method for identifying trade secret documents disclosed in this application;

[0048] Figure 4 It is a schematic diagram of a specific function for formulating trade secret classification rules disclosed in this application;

[0049] Figure 5 It is a schematic diagram of a solution to the problem of dataset construction disclosed in this application;

[0050] Figure 6 It is a schematic diagram of generating trade secret classification rules according to the carrier type disclosed in this application;

[0051] Figure 7 A schematic diagram of a workflow for identifying trade secrets disclosed in this application;

[0052] Figure 8 Another specific flowchart of a method for identifying trade secret documents disclosed in this application;

[0053] Figure 9 A schematic diagram of the structure of a trade secret document identification device disclosed in this application;

[0054] Figure 10 A structural diagram of an electronic device disclosed in this application. Detailed implementation manners

[0055] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0056] Currently, there are some problems in identifying trade secret documents: First, it is difficult for enterprises to formulate secrecy rules; second, the secrecy rules have a high degree of conceptual abstraction; third, the identification process of trade secret documents is not transparent; fourth, the secrecy system has strong specificity. For this reason, this application provides a method for identifying trade secret documents, which can improve the accuracy of identifying trade secret documents during the identification process.

[0057] See Figure 1 As shown, an embodiment of the present invention discloses a method for identifying trade secret documents, which is applied to a preset hierarchical architecture including an engine layer, a capability layer, a service layer, and a product layer. The method includes:

[0058] Step S11: In the engine layer, the basic capabilities corresponding to the large language model, including text classification, text generation, logical reasoning, and natural language understanding, are sent to the capability layer, so as to determine trade secret processing technologies based on the basic capabilities in the capability layer, and send the trade secret processing technologies to the service layer.

[0059] In this embodiment, the overall architecture for identifying trade secret documents is as Figure 2 shown: from bottom to top are the engine layer, the capability layer, the service layer, and the product layer in sequence.

[0060] Among them, the engine layer is the foundation of the entire architecture, which is used to provide computing resources, storage resources, and resource scheduling capabilities. It is worth mentioning that the core of the engine layer is the large language model, which can provide technical capabilities such as text classification, text generation, logical reasoning, and natural language understanding to the capability layer. Secondly, above the capability layer is the service layer. Among them, the data access service is responsible for accessing user data, including data such as the basic company situation description provided by users, trade secret documents generated during daily operations, and the trade secret classification rules already formulated by the enterprise, and preprocessing the data, and then providing the processed data to subsequent functions. The trade secret classification rule formulation service is used to implement the core function of helping users formulate trade secret classification rules. The trade secret recognition service is used to implement the core function of trade secret recognition. Furthermore, the top layer is the product layer, mainly the trade secret recognition system. It is worth mentioning that the overall architecture for identifying trade secret documents interacts with users in the form of a website and displays the service call results of the service layer, so as to provide an intuitive experience for using the trade secret recognition system.

[0061] Based on the above architecture, users can provide company trade secret samples or company description information to use the trade secret classification rule formulation service. Through the above service, the system can generate a trade secret classification rule document that suits the company's characteristics. Users can modify or confirm the document for subsequent use of the trade secret recognition service. Among them, if users have formulated the company's trade secret classification rules offline, they can directly upload their own trade secret classification rules to use the trade secret recognition service. All uploaded document data needs to be read and preprocessed through the data access service for use as input data for other services. In addition, operations such as user upload, modification, and result viewing are all carried out through the interface of the trade secret recognition system, and the business process is as Figure 3 shown: The document to be recognized corresponding to the red arrow is used to be distributed by the data access service to the trade secret recognition service for processing. The company's trade secret classification rules corresponding to the orange arrow are also distributed by the data access service to the trade secret recognition service for processing. The company business scope description information corresponding to the green arrow is used to be distributed by the data access service to the trade secret classification rule formulation service for processing. The company trade secret sample corresponding to the blue arrow is also distributed by the data access service to the trade secret classification rule formulation service for processing. Subsequently, the trade secret recognition service can process the document to be recognized based on the company's trade secret classification rules formulated by the trade secret classification rule formulation service, so as to obtain the recognition result.

[0062] Step S12: Use the trade secret processing technology in the service layer to generate a data access service, a trade secret classification rule formulation service, and a trade secret identification service. Then, use the data access service to receive the initial enterprise documents including enterprise basic situation description information, trade secret documents, and historical classification rules, and preprocess the initial enterprise documents to obtain the enterprise documents to be processed. Then, use the trade secret classification rule formulation service to process the enterprise documents to be processed to obtain the target classification rules.

[0063] In this embodiment, the workflow for generating trade secret classification rules based on information such as the enterprise's business scope is as follows: First, perform data preprocessing and add task description processing on the company's business scope information provided by the user to form the input data for the large language model. Then, the model performs a text generation task and outputs a set of trade secret classification rules.

[0064] It is worth mentioning that Prompt (i.e., the prompt) is the input or instruction provided to the large language model to guide the large language model to perform specific operations or generate specific outputs. When dealing with relatively complex tasks, the Prompt can provide task-related context information to help the large language model better understand the requirements and goals of the task. In addition, the purpose of prompt engineering is to design and optimize the Prompt to ensure that the large language model can accurately and efficiently execute the user's instructions. Among them, the prompt engineering of this application embodiment has the function of formulating trade secret classification rules related to the enterprise's main business, and the schematic diagram of the function of formulating trade secret classification rules is as Figure 4 shown: First, it is necessary to input the company's business scope description information, specifically including: company name, industry, main business scope, and detailed description of the main business. Among them, the industry and the main business scope are mandatory information, and the others are optional information. It is worth mentioning that regardless of how many types of information the user provides, the Prompt should be able to guide the model to output detailed, practical, and business scope-related trade secret classification rules. In a specific implementation manner, this embodiment designs multiple Prompts to improve the generalization ability of the model. The examples are as follows: "This enterprise is an enterprise in the manufacturing industry, and its main businesses are communication equipment manufacturing and information technology services. Please write a set of trade secret classification rules documents belonging to this enterprise based on the above description."

[0065] In this embodiment, if fine-tuning training is performed using a large language model, it is first necessary to construct a training set and a test set. Specifically, the data set includes two major categories of documents: trade secret classification rules documents and various documents that may be generated during the operation of an enterprise. Among them, to enable the model to have better generalization ability, the data set should cover diverse industries and include a rich variety of document carrier types. In a specific implementation, the data annotation requirements are as follows: for each trade secret classification rule, the corresponding company name, industry, main business scope, and detailed business description need to be annotated; for each document during the operation of an enterprise, the carrier type, and the reasoning process and reasoning conclusion for trade secret identification need to be annotated.

[0066] Specifically, use the data access service to receive the initial enterprise documents including enterprise basic situation description information, trade secret documents, and historical classification rules, and preprocess the initial enterprise documents to obtain the enterprise documents to be processed, which may include: use the data access service to receive the initial enterprise documents including enterprise basic situation description information, trade secret documents, and historical classification rules, and perform a format parsing operation on the initial enterprise documents using a preset preprocessing rule to obtain a format parsing result, and perform table marking, picture marking, and typesetting marking operations on the format parsing result to obtain a marking result, and then perform a paragraph marking operation and a metadata annotation operation on the marking result to obtain the mapping relationship between the initial enterprise document and the business attributes; the metadata annotation operation includes the annotation operations performed on the company name, company industry type, company main business scope, and company business description; determine the intermediate enterprise document based on the mapping relationship and the initial enterprise document, and use the preset task description processing to perform data description on the intermediate enterprise document to obtain a data description result, and determine the enterprise document to be processed based on the data description result and the intermediate enterprise document.

[0067] In addition, the embodiment of this application adopts the method as Figure 5 to solve the problem of data set construction. Among them, the data sources include: publicly available documents on the Internet and documents generated by open-source large language models. The data annotation methods include: pre-annotation by open-source large language models, manual annotation, and pre-annotation using the model in the iteration. All data has been corrected by human experts to ensure the quality of the data. At the same time, this embodiment can use a large language model to perform text rewriting and expansion, supplemented by traditional natural language processing data augmentation techniques to increase the amount of data.

[0068] In this embodiment, the workflow for formulating trade secret classification rules according to the company's trade secret sample document set is divided into two main steps: the first step is to determine the carrier types corresponding to the company's trade secret sample documents to obtain a set of trade secret carrier types. The second step is to generate trade secret classification rules based on the above carrier types, and the above process schematic diagram is as Figure 6Shown as follows: First, use the Prompt designed for the function of formulating commercial secret classification rules based on the collection of enterprise trade secret document samples to guide the model to classify documents from the perspective of carrier types. In a specific implementation, an example of the Prompt is as follows: "Please classify the following text content into the most appropriate category according to your understanding. These categories include but are not limited to [list of carrier categories]. Finally, please directly output the category you think is the most matching, without explanation, just give the category to which it belongs, for example: Notice and Announcement. The given text content is as follows: [text content]". Subsequently, obtain the set of carrier type categories, for example: {Test Report, Strategic Plan, Competitor Analysis}. Then, the embodiments of the present application need to guide the model to generate commercial secret classification rules based on the set of carrier type categories. In a specific implementation, an example of the Prompt is as follows: "Please generate a commercial secret classification rule document, which should stipulate that relevant documents such as test reports, strategic plans, and competitor analyses all belong to the category of commercial secrets. At the same time, please clearly stipulate the corresponding relationship between different confidentiality levels of commercial secrets and document categories in the document to ensure the accurate implementation of information classification management and confidentiality work."

[0069] Specifically, the service for formulating commercial secret classification rules processes the enterprise document to be processed to obtain the target classification rules, which may include: determining whether the enterprise document to be processed is information within the enterprise's business scope. If the enterprise document to be processed is information within the enterprise's business scope, then use the service for formulating commercial secret classification rules to perform word segmentation processing on the enterprise document to be processed to obtain the word segmentation processing result, and process the word segmentation processing result based on the structure processing instruction to obtain structured data; use the large language model to process the structured data to generate an initial classification rule document, and then use the service for formulating commercial secret classification rules and generate the target classification rules based on the initial classification rule document; the initial classification rule document includes commercial secret carrier types, confidentiality level division rules, and confidentiality level determination criteria.

[0070] In this embodiment, after preprocessing the initial enterprise documents to obtain the to-be-processed enterprise documents, it is necessary to use the service for formulating commercial secret classification rules to process the to-be-processed enterprise documents to obtain the target classification rules. Specifically, using the service for formulating commercial secret classification rules to process the to-be-processed enterprise documents to obtain the target classification rules may include: determining whether the to-be-processed enterprise documents are enterprise business scope information. If the to-be-processed enterprise documents are not enterprise business scope information, then set the to-be-processed enterprise documents as the commercial secret sample document set; use the large language model and based on the preset multi-round semantic analysis rules to perform multi-round semantic analysis on the commercial secret sample documents in the commercial secret sample document set to obtain the corresponding commercial secret carrier type set; generate a classification rule document including specific classification rules and classification level correlation relationships based on the commercial secret carrier type set, and perform logical matching verification on the classification rule document and the historical classification rules to obtain the verification result, and then adjust the conflicting clauses in the classification rule document based on the verification result to obtain the target classification rules.

[0071] Step S13: In the service layer, use the preset model adjustment rules to adjust the large language model to obtain the target large language model, and perform an association operation on the paragraphs of the target classification rules and the to-be-identified documents based on the commercial secret recognition service to obtain the association result, and then use the target large language model to perform a classification level determination on the association result to obtain the classification level determination result, so as to send the classification level determination result to the interaction interface in the product layer.

[0072] In this embodiment, since the large language model has extensive general domain knowledge, but its accuracy may be challenged when facing specific domains or tasks that have not been trained. Therefore, the embodiment of the present application adopts the fine-tuning technology, that is, based on the pre-trained large language model, use the data set of a specific domain or task for further training to improve the performance of the model. Among them, one of the basic ways to fine-tune the large model is to fine-tune all the model parameters. However, fine-tuning all the model parameters not only requires a high-quality large-scale data set, but also requires huge computing power support. The embodiment of the present application chooses to use the parameter-efficient fine-tuning technology to fine-tune some of the model parameters.

[0073] Specifically, adjusting the large language model using a preset model adjustment rule in the service layer to obtain a target large language model may include: adjusting the text classification accuracy parameters of the large language model using historical documents and case data in the field of trade secrets in the service layer using a preset fine-tuning technique to obtain a first large language model; constructing a prompt template library including trade secret feature description information to adjust the domain relevance parameters corresponding to the first large language model using the prompt template library to obtain a second large language model; the domain relevance parameter is a parameter for establishing a relevance connection for each domain corresponding to the output results of the model; adjusting the logical consistency parameters of the second large language model based on the revision record obtained by the user's revision of the target classification rule and the feedback information for the recognition result to obtain the target large language model; the logical consistency parameter is a parameter for performing logical consistency processing on each output result of the model and the target classification rule.

[0074] Further, the workflow for trade secret recognition in the embodiments of the present application is as Figure 7 shown: Since the target classification rules may include not only descriptions of which documents belong to trade secrets, methods for discriminating the levels of trade secrets, etc., but also descriptions of how to manage trade secrets at different levels, penalties that may be faced for disclosing the company's trade secrets, etc., and this content is not relevant to the trade secret recognition task, and excessive long and irrelevant information may also interfere with trade secret recognition. Therefore, in the embodiments of the present application, the paragraphs most relevant to trade secret classification are first extracted from the trade secret classification rule document, and then this content and the document to be recognized are input into the large language model together for the large language model to perform inference and recognition.

[0075] In the embodiments of the present application, it is necessary to design a Prompt based on the trade secret recognition function service, and the specific process schematic diagram is as Figure 7Shown as follows: The first Prompt of the trade secret identification function service is applied to guide the model to output the paragraphs most relevant to the classification of trade secrets. In a specific implementation, the Prompt example is shown as follows: "Given a trade secret classification rule document, please extract the key information in the document, including: 1. What specific categories does the trade secret include, 2. How many levels does the trade secret specifically include, 3. The corresponding relationship between the trade secret categories and levels. The document given to you is as follows: [input text]". Subsequently, guide the model to perform trade secret identification and output the reasoning process. The Prompt example is shown as follows: "Please perform a reasoning task to determine whether the given text contains trade secrets. Reasoning premises: 1. Categories of trade secrets: Technical: Technical planning, production process, patent application, etc. Business management: Business plan, financial statements, employee contracts, salary system, etc. 2. Levels of trade secrets: Core level, ordinary level. 3. The corresponding relationship between the trade secret categories and levels: The core level includes: Technical planning, production process, business plan. The ordinary level includes: Trade secrets other than the core level. If the text to be reasoned contains trade secrets, please elaborate on the reasoning process and indicate the categories and levels of the trade secrets involved. The text to be reasoned is as follows: [input text]".

[0076] Specifically, based on the trade secret identification service, perform an association operation between the target classification rules and the document to be identified to obtain an association result. Then, use the target large language model to perform a classification determination on the association result to obtain a classification determination result, so as to send the classification determination result to the interaction interface in the product layer. This may include: performing a semantic paragraph segmentation operation on the target classification rules using the preset paragraph segmentation rules in the trade secret identification service to obtain a paragraph segmentation result; performing a content fragment extraction operation on each paragraph segmentation result using the preset keyword matching rules and preset context analysis and extraction rules to obtain a content fragment extraction result; the content fragment extraction result includes classification determination criteria; performing a context association on the content fragment extraction result and the document to be identified using the preset context association encoding to obtain a context association result. Then, use the target large language model to process the context association result to obtain a classification determination result and the determination basis corresponding to the classification determination result, and send the classification determination result and the determination basis to the interaction interface in the product layer for the user to view the classification determination result and the determination basis through the interaction interface.

[0077] In this embodiment, after obtaining the classification determination result and the corresponding determination basis for the classification determination result, the embodiment of the present application needs to send the classification determination result to the interaction interface in the product layer for the user to view. Specifically, sending the classification determination result to the interaction interface in the product layer may include: configuring a document upload entry in the product layer to receive several types of historical sample documents, historical classification rules, target classification rules, and classification determination results through the document upload entry, and setting permissions for each user using a preset permission hierarchical mechanism; designing and sequentially displaying the generation process corresponding to the classification determination result through the interaction interface using a preset visualization logic chain, so that the user can perform information viewing operations on the generation process of the classification determination result based on the permissions; where the interaction interface has an interface editing function, a document editing function, and a process traceability function.

[0078] As can be seen from the above, the embodiment of the present application first needs to send the basic capabilities corresponding to the large language model, including text classification, text generation, logical reasoning, and natural language understanding, to the capability layer, so as to determine the trade secret processing technology based on the basic capabilities; subsequently, use the trade secret processing technology to generate a data access service, a trade secret classification rule formulation service, and a trade secret identification service, then use the data access service to receive the initial enterprise documents including enterprise basic situation description information, trade secret documents, and historical classification rules, and preprocess the initial enterprise documents to obtain the enterprise documents to be processed, and then use the trade secret classification rule formulation service to process the enterprise documents to be processed to obtain the target classification rules; furthermore, use the preset model adjustment rules to adjust the large language model to obtain the target large language model, and finally, based on the trade secret identification service, perform an association operation between the target classification rules and the document to be identified to obtain an association result, and then use the target large language model to perform a classification determination on the association result to obtain a classification determination result, so as to send the classification determination result to the interaction interface in the product layer. In this way, the accuracy of identifying trade secret documents is improved during the trade secret document identification process, thereby enhancing the security of the enterprise production process.

[0079] As can be seen from the above embodiments, the present application identifies trade secret documents based on the functions corresponding to the large model. Therefore, in order to more systematically illustrate the process of identifying trade secret documents based on the functions corresponding to the large model, see Figure 8 As shown, the embodiment of the present invention discloses a method for identifying trade secret documents, including:

[0080] First, the function of formulating the rules for classifying trade secrets includes formulating the rules for classifying trade secrets according to the company's business scope and formulating the rules for classifying trade secrets according to the company's trade secret sample documents. Among them, the process of formulating the rules for classifying trade secrets according to the company's business scope is as follows: First, design reasonable prompt words to use the above prompt words to guide the model to generate the rules for classifying trade secrets based on the company's business information. Subsequently, construct a training set and a test set, and correct the training set and the test set through artificial experts; finally, use the corrected training set and test set to fine-tune the large language model to make the model meet the expectations, so as to use the fine-tuned model to generate the rules for classifying trade secrets. In addition, the process of formulating the rules for classifying trade secrets according to the company's trade secret sample documents is as follows: First, design reasonable prompt words to use the above prompt words to guide the model to generate the rules for classifying trade secrets according to the company's trade secret sample documents. Subsequently, classify the carrier types of the sample documents to formulate the rules for classifying trade secrets according to the carrier types and the sensitive level labels of the sample documents:

[0081] It is worth mentioning that the specific process of the trade secret recognition function is as follows: Design reasonable prompt words to extract the paragraphs most relevant to the trade secret recognition criteria from the long trade secret classification rules document using the guiding words, and then infer the trade secret categories and levels from the document to be recognized based on the extracted paragraphs. The specific process of the data access function is as follows: Read the content of documents in multiple formats to solve key problems using the key content read. In a specific implementation, the document formats include but are not limited to Microsoft Word (i.e., Microsoft word processing software) documents, PDF (Portable Document Format) documents, and TXT (Text File) documents, and the components for reading the document content include but are not limited to Apache POI (Apache Poor Obfuscation Implementation, i.e., Apache document operation interface), PDF plumber (i.e., PDF parsing tool), and PaddleOCR (PArallel Distributed Deep LEarning Distributed Deep Learning OCR, i.e., Paddle optical character recognition). The specific working process of the trade secret recognition system is as follows: Use the interface for user-system interaction to solve problems including but not limited to system prototype design problems, website page UI (User Interface) design problems, and website front-end development problems, and the available tools include Axure (Axure Rapid Prototyping), Sketch (i.e., Sketch vector design tool), and VUE (i.e., progressive JavaScript framework).

[0082] It can be seen that the embodiment of this application adopts a technical route with fine-tuning the large language model as the core, improves the capabilities of the large language model in the field of trade secret recognition, and realizes the two key functions of formulating trade secret classification rules and trade secret recognition. It uses open-source components with high maturity to read the document content to implement the data access function, and then uses mainstream technical frameworks to build the system, and finally realizes trade secret recognition.

[0083] Correspondingly, as shown in Figure 9 this application also provides a trade secret document recognition device, which is applied to a preset hierarchical architecture including an engine layer, a capability layer, a service layer, and a product layer. The device includes:

[0084] A processing technology determination module 11, configured to issue basic capabilities corresponding to a large language model, including text classification, text generation, logical reasoning, and natural language understanding, to the capability layer in the engine layer, so as to determine a trade secret processing technology based on the basic capabilities in the capability layer, and issue the trade secret processing technology to the service layer;

[0085] A classification rule generation module 12, configured to generate a data access service, a trade secret classification rule formulation service, and a trade secret identification service by using the trade secret processing technology in the service layer, then receive an initial enterprise document including enterprise basic situation description information, trade secret documents, and historical classification rules by using the data access service, preprocess the initial enterprise document to obtain a to-be-processed enterprise document, and then process the to-be-processed enterprise document by using the trade secret classification rule formulation service to obtain a target classification rule;

[0086] A classification level determination result determination module 13, configured to adjust the large language model by using a preset model adjustment rule in the service layer to obtain a target large language model, perform an association operation on paragraphs and documents to be identified for the target classification rules based on the trade secret identification service to obtain an association result, then perform a classification level determination on the association result by using the target large language model to obtain a classification level determination result, so as to issue the classification level determination result to an interaction interface in the product layer.

[0087] As can be seen from the above, before identifying trade secret documents in the embodiments of the present application, it is first necessary to issue basic capabilities corresponding to the large language model, including text classification, text generation, logical reasoning, and natural language understanding, to the capability layer, so as to determine a trade secret processing technology based on the basic capabilities; subsequently, use the trade secret processing technology to generate a data access service, a trade secret classification rule formulation service, and a trade secret identification service, then receive an initial enterprise document including enterprise basic situation description information, trade secret documents, and historical classification rules by using the data access service, preprocess the initial enterprise document to obtain a to-be-processed enterprise document, and then process the to-be-processed enterprise document by using the trade secret classification rule formulation service to obtain a target classification rule; furthermore, adjust the large language model by using a preset model adjustment rule to obtain a target large language model. Finally, perform an association operation on paragraphs and documents to be identified for the target classification rules based on the trade secret identification service to obtain an association result, and then perform a classification level determination on the association result by using the target large language model to obtain a classification level determination result, so as to issue the classification level determination result to an interaction interface in the product layer. In this way, the accuracy of identifying trade secret documents is improved during the process of identifying trade secret documents, thereby enhancing the security of the enterprise production process.

[0088] In some specific embodiments, the classified information rule generation module 12 may specifically include:

[0089] A mapping relationship determination unit, configured to receive, by using the data access service, an initial enterprise document including enterprise basic situation description information, business secret documents, and historical classified information rules, perform a format parsing operation on the initial enterprise document by using a preset preprocessing rule to obtain a format parsing result, perform table marking, picture marking, and typesetting marking operations on the format parsing result to obtain a marking result, and then perform a paragraph marking operation and a metadata annotation operation on the marking result to obtain the mapping relationship between the initial enterprise document and the business attributes; the metadata annotation operation includes annotation operations performed on the company name, company industry type, company main business scope, and company business description;

[0090] A data description unit, configured to determine an intermediate enterprise document based on the mapping relationship and the initial enterprise document, perform data description on the intermediate enterprise document by using a preset task description process to obtain a data description result, and determine a to-be-processed enterprise document based on the data description result and the intermediate enterprise document.

[0091] In some specific embodiments, the classified information rule generation module 12 may specifically include:

[0092] A structured data determination unit, configured to determine whether the to-be-processed enterprise document is enterprise business scope information. If the to-be-processed enterprise document is the enterprise business scope information, perform a word segmentation process on the to-be-processed enterprise document by using the business secret classified information rule formulation service to obtain a word segmentation process result, and process the word segmentation process result based on a structure processing instruction to obtain structured data;

[0093] A classified information rule generation subunit, configured to process the structured data by using the large language model to generate an initial classified information rule document, and then generate a target classified information rule by using the business secret classified information rule formulation service and based on the initial classified information rule document; the initial classified information rule document includes business secret carrier types, classification rules for secret levels, and secret level determination criteria.

[0094] In some specific embodiments, the classified information rule generation module 12 may specifically include:

[0095] An enterprise document judgment unit, configured to determine whether the to-be-processed enterprise document is enterprise business scope information. If the to-be-processed enterprise document is not the enterprise business scope information, set the to-be-processed enterprise document as a business secret sample document set;

[0096] A sample document semantic analysis unit, configured to perform multiple rounds of semantic analysis on the trade secret sample documents in the trade secret sample document set by using the large language model and based on preset multiple rounds of semantic analysis rules, to obtain a corresponding trade secret carrier type set;

[0097] The verification result determination unit is used to generate a confidentiality details document including specific confidentiality rules and confidentiality level association relationships based on the set of trade secret carrier types, and to perform a logical matching verification on the confidentiality details document and the historical confidentiality details to obtain a verification result, and then to adjust the conflicting clauses in the confidentiality details document based on the verification result to obtain a target confidentiality rule.

[0098] In some specific implementations, the confidentiality level determination result determination module 13 may specifically include:

[0099] A first model parameter adjustment unit, configured to use a preset fine-tuning technology to adjust the text classification accuracy parameters of the large language model in the service layer using historical documents and case data in the field of trade secrets to obtain a first large language model;

[0100] A second model parameter adjustment unit is used to construct a prompt word template library including the characteristic description information of the trade secret, so as to adjust the domain relevance parameter corresponding to the first language model by using the prompt word template library to obtain a second language model; the domain relevance parameter is a parameter for correlating the domains corresponding to each output result of the model;

[0101] The third model parameter adjustment unit is used to adjust the logical consistency parameters of the second large language model based on the revision record obtained by the user to revise the target confidentiality rule and the feedback information on the recognition result, so as to obtain the target large language model; the logical consistency parameters are parameters for performing logical consistency processing on each output result of the model and the target confidentiality rule.

[0102] In some specific implementations, the confidentiality level determination result determination module 13 may specifically include:

[0103] A paragraph segmentation result determination unit, configured to perform a semantic paragraph segmentation operation on the target confidentiality details by using a preset paragraph segmentation rule in the trade secret identification service to obtain a paragraph segmentation result;

[0104] A content segment extraction unit, configured to perform a content segment extraction operation on each of the paragraph segmentation results using a preset keyword matching rule and a preset context analysis and extraction rule to obtain a content segment extraction result; the content segment extraction result includes a confidentiality level determination standard;

[0105] An association result determination unit is configured to perform context association on the content segment extraction result and the document to be recognized by using a preset context association encoding to obtain a context association result, and then use the target large language model to process the context association result to obtain a classification determination result and a determination basis corresponding to the classification determination result, and send the classification determination result and the determination basis to the interaction interface in the product layer so that the user can view the classification determination result and the determination basis through the interaction interface.

[0106] In some specific embodiments, the classification determination result determination module 13 may specifically include:

[0107] A permission setting unit is configured to configure a document upload entry in the product layer to receive several types of historical sample documents, the historical classification rules, the target classification rules, and the classification determination result by using the document upload entry, and perform permission setting for each user by using a preset permission hierarchical mechanism;

[0108] A generation process display unit is configured to design and display the generation process corresponding to the classification determination result through the interaction interface in sequence by using a preset visualization logic chain so that the user can perform an information viewing operation on the generation process of the classification determination result based on the permission; wherein, the interaction interface has an interface editing function, a document editing function, and a process traceability function.

[0109] Furthermore, an embodiment of the present application also discloses an electronic device, Figure 10 It is a structural diagram of an electronic device 20 shown according to an exemplary embodiment, and the content in the figure cannot be considered as any limitation to the scope of use of the present application. 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. Among them, the memory 22 is used to store a computer program, and the computer program is loaded and executed by the processor 21 to implement the relevant steps in the commercial secret document recognition method disclosed in any of the foregoing embodiments. In addition, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0110] In this embodiment, the power supply 23 is used to provide a working 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 is any communication protocol applicable to the technical solution of the present application, and no specific limitation is imposed on it here; the input / output interface 25 is used to obtain external input data or output data to the outside, and its specific interface type can be selected according to specific application needs, and no specific limitation is made here.

[0111] In addition, as a carrier for storing resources, the memory 22 can be a read-only memory, a random access memory, a magnetic disk, an optical disc, etc. The resources stored thereon can include an operating system 221, a computer program 222, etc. The storage method can be transient storage or permanent storage.

[0112] Among them, the operating system 221 is used to manage and control each hardware device and the computer program 222 on the electronic device 20, and it can be Windows Server, Netware, Unix, Linux, etc. In addition to the computer program that can be used to complete the trade secret document recognition method executed by the electronic device 20 disclosed in any of the foregoing embodiments, the computer program 222 can further include computer programs that can be used to complete other specific tasks.

[0113] Furthermore, the present application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the trade secret document recognition method disclosed above is implemented. For the specific steps of this method, reference can be made to the corresponding content disclosed in the foregoing embodiments, and details will not be repeated here.

[0114] In this specification, the various embodiments are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts between the various embodiments, reference can be made to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and reference can be made to the description of the method part for related parts.

[0115] Those skilled in the art can further realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0116] The steps of the method or algorithm described in combination with the embodiments disclosed herein can be directly implemented by hardware, a software module executed by a processor, or a combination of the two. The software module can be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, register, hard disk, removable disk, CD-ROM, or any other form of storage medium well-known in the technical field.

[0117] Finally, it should also be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or device comprising said element.

[0118] The technical solutions provided in this application have been introduced in detail above. Specific examples are used in this text to elaborate on the principle and implementation manner of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, according to the idea of this application, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. A method for identifying trade secret documents, characterized in that, Applied to a preset hierarchical architecture including an engine layer, an ability layer, a service layer, and a product layer, the method includes: In the engine layer, the basic capabilities corresponding to the large language model, including text classification, text generation, logical reasoning, and natural language understanding, are sent down to the ability layer, so that based on the basic capabilities in the ability layer, a trade secret processing technology is determined and sent down to the service layer; In the service layer, data access services, trade secret classification rule formulation services, and trade secret identification services are generated using the trade secret processing technology. Then, the initial enterprise documents including enterprise basic situation description information, trade secret documents, and historical classification rules are received using the data access services, and the initial enterprise documents are preprocessed to obtain enterprise documents to be processed. Then, the enterprise documents to be processed are processed using the trade secret classification rule formulation services to obtain target classification rules; In the service layer, the large language model is adjusted using preset model adjustment rules to obtain a target large language model. Based on the trade secret identification services, an association operation between paragraphs and documents to be identified is performed on the target classification rules to obtain an association result. Then, the target large language model is used to perform a classification level determination on the association result to obtain a classification level determination result, so as to send down the classification level determination result to the interaction interface in the product layer.

2. The trade secret document recognition method according to claim 1, wherein The step of using the data access services to receive the initial enterprise documents including enterprise basic situation description information, trade secret documents, and historical classification rules, and preprocessing the initial enterprise documents to obtain enterprise documents to be processed includes: Using the data access services to receive the initial enterprise documents including enterprise basic situation description information, trade secret documents, and historical classification rules, and performing a format parsing operation on the initial enterprise documents using preset preprocessing rules to obtain a format parsing result, and performing table marking, picture marking, and typesetting marking operations on the format parsing result to obtain a marking result. Then, a paragraph marking operation and a metadata annotation operation are performed on the marking result to obtain the mapping relationship between the initial enterprise documents and business attributes; the metadata annotation operation includes annotation operations on the company name, company industry type, company main business scope, and company business description; Based on the mapping relationship and the initial enterprise documents, intermediate state enterprise documents are determined, and data description is performed on the intermediate state enterprise documents using preset task description processing to obtain a data description result. Based on the data description result and the intermediate state enterprise documents, enterprise documents to be processed are determined.

3. The method for identifying a trade secret document according to claim 1, wherein The step of using the trade secret classification rule formulation services to process the enterprise documents to be processed to obtain target classification rules includes: Determine whether the enterprise document to be processed is enterprise business scope information. If the enterprise document to be processed is the enterprise business scope information, use the service for formulating commercial secret classification rules to perform word segmentation processing on the enterprise document to be processed, obtain the word segmentation processing result, and process the word segmentation processing result based on the structure processing instruction to obtain structured data; Use the large language model to process the structured data to generate an initial classification rule document, and then use the service for formulating commercial secret classification rules and generate a target classification rule based on the initial classification rule document; the initial classification rule document includes commercial secret carrier types, classification level division rules, and classification level determination criteria.

4. The trade secret document recognition method according to claim 1, wherein The use of the service for formulating commercial secret classification rules to process the enterprise document to be processed to obtain a target classification rule includes: Determine whether the enterprise document to be processed is enterprise business scope information. If the enterprise document to be processed is not the enterprise business scope information, set the enterprise document to be processed as a commercial secret sample document set; Use the large language model and perform multi-round semantic analysis on the commercial secret sample documents in the commercial secret sample document set based on the preset multi-round semantic analysis rules to obtain the corresponding commercial secret carrier type set; Generate a classification rule document including specific classification rules and classification level association relationships based on the commercial secret carrier type set, perform logical matching verification on the classification rule document and the historical classification rules to obtain a verification result, and then adjust the conflicting clauses in the classification rule document based on the verification result to obtain a target classification rule.

5. The method for identifying a trade secret document according to claim 1, wherein The adjustment of the large language model using the preset model adjustment rules in the service layer to obtain a target large language model includes: Use the preset fine-tuning technology to perform an adjustment operation on the text classification accuracy parameters of the large language model in the service layer using historical documents and case data in the field of commercial secrets to obtain a first large language model; Construct a prompt word template library including commercial secret feature description information to use the prompt word template library to adjust the domain relevance parameters corresponding to the first large language model to obtain a second large language model; the domain relevance parameter is a parameter for performing relevance connection on the fields corresponding to the output results of the model respectively; Adjust the logical consistency parameters of the second large language model based on the revision record obtained by the user's revision of the target classification rule and the feedback information for feedback on the recognition result to obtain a target large language model; the logical consistency parameter is a parameter for performing logical consistency processing on each of the output results of the model and the target classification rule.

6. The method for identifying a trade secret document according to claim 1, wherein The association operation of paragraphs and documents to be recognized on the target classification rules based on the commercial secret recognition service to obtain an association result, and then use the target large language model to perform classification level determination on the association result to obtain a classification level determination result, so as to send the classification level determination result to the interaction interface in the product layer, includes: Perform a semantic paragraph segmentation operation on the target classification rules using the preset paragraph segmentation rules in the commercial secret recognition service to obtain a paragraph segmentation result; Perform a content segment extraction operation on each of the paragraph segmentation results using the preset keyword matching rules and the preset context analysis and extraction rules to obtain a content segment extraction result; the content segment extraction result includes a classification determination criterion; Perform context association on the content segment extraction result and the document to be recognized using the preset context association encoding to obtain a context association result, and then use the target large language model to process the context association result to obtain a classification determination result and a determination basis corresponding to the classification determination result, and send the classification determination result and the determination basis to the interaction interface in the product layer so that the user can view the classification determination result and the determination basis through the interaction interface.

7. The method for identifying a trade secret document according to any one of claims 1 to 6, characterized in that The sending the classification determination result to the interaction interface in the product layer includes: Configure a document upload entry in the product layer to receive several types of historical sample documents, the historical classification rules, the target classification rules, and the classification determination result using the document upload entry, and set permissions for each user using the preset permission hierarchical mechanism; Design and display the generation process corresponding to the classification determination result through the interaction interface using the preset visualization logic chain so that the user can perform an information viewing operation on the generation process of the classification determination result based on the permission; wherein, the interaction interface has an interface editing function, a document editing function, and a process traceability function.

8. A trade secret document recognition device, characterized in that, Applied to a preset hierarchical architecture including an engine layer, an ability layer, a service layer, and a product layer, the device includes: A processing technology determination module for sending the basic capabilities corresponding to the large language model including text classification, text generation, logical reasoning, and natural language understanding to the ability layer in the engine layer so that the commercial secret processing technology can be determined based on the basic capabilities in the ability layer and the commercial secret processing technology can be sent to the service layer; A classification rule generation module for generating a data access service, a commercial secret classification rule formulation service, and a commercial secret recognition service using the commercial secret processing technology in the service layer, then receiving an initial enterprise document including enterprise basic situation description information, commercial secret documents, and historical classification rules using the data access service, and preprocessing the initial enterprise document to obtain a to-be-processed enterprise document, and then processing the to-be-processed enterprise document using the commercial secret classification rule formulation service to obtain target classification rules; The confidentiality level determination result determination module is used to adjust the large language model in the service layer by using a preset model adjustment rule to obtain a target large language model, and perform an association operation between paragraphs of the target confidentiality determination rules and the document to be recognized based on the trade secret recognition service to obtain an association result, and then use the target large language model to perform a confidentiality level determination on the association result to obtain a confidentiality level determination result, so as to send the confidentiality level determination result to the interaction interface in the product layer.

9. An electronic device, characterized in that, It includes: A memory for storing computer programs; A processor for executing the computer program to implement the trade secret document recognition method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, For storing a computer program, wherein the computer program, when executed by the processor, implements the trade secret document recognition method according to any one of claims 1 to 7.

Images