Enterprise digital transformation decision optimization method based on multi-agent collaboration

Abstract

The invention relates to the technical field of digital transformation, in particular to an enterprise digital transformation decision optimization method based on multi-agent collaboration. The method comprises the following steps: performing step-by-step encryption processing according to the sensitivity level of a contract; and establishing an independent file storage block chain, and performing matched storage according to the encryption degree of the contract file version. According to the method, step-by-step encryption processing is carried out according to the sensitivity level of the contract, the contract file versions with different encryption degrees are obtained, adaptive encryption processing is carried out on the contract files with different sensitivity degrees, the files with different encryption degrees are obtained, and the diversified encryption processing on the transformed files is realized; the method is advantaged in that contract file calling under different conditions can be adapted, risk of contact of a user with file core content is reduced, matching storage is carried out according to an encryption degree of a contract file version in cooperation with the established independent file storage block chain, access authority distribution is carried out on various encrypted contract files, and file security after file encryption is improved.

Description

Technical Field

[0001] This invention relates to the field of digital transformation technology, and more specifically, to a method for optimizing enterprise digital transformation decisions based on multi-agent collaboration. Background Technology

[0002] Nowadays, with the continuous improvement of technology, intelligent agent technology is constantly developing and its application scope is constantly expanding, including enterprise digital transformation. From the perspective of technology and industry trends, intelligent agent technology, as a core direction of artificial intelligence, has cross-domain and high adaptability characteristics and broad application prospects. Digital transformation is a key path for improving the quality and efficiency of industries and optimizing and upgrading. The deep integration of the two is an inevitable choice for technological laws and high-quality development.

[0003] During the digital transformation process within enterprises, one of the most direct challenges is the privacy of uploaded files. Because some files contain sensitive information and are processed by different personnel, issues such as file content leakage, tampering, and loss frequently occur during the digitization process. Traditional methods involve standardized encryption of sensitive files using a key-and-key combination. However, since digitization involves multiple staff members with varying tasks and responsibilities, including changes to file format and content, this single-method encryption cannot adapt to the sensitivity of the file. In other words, anyone with the key can access the file content, increasing the risk of leakage during digitization.

[0004] To address the aforementioned issues, there is an urgent need for a decision optimization method for enterprise digital transformation based on multi-agent collaboration. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-agent collaborative decision-making optimization method for enterprise digital transformation. This method involves step-by-step encryption processing based on the sensitivity level of contracts to obtain contract document versions with different encryption levels. Adaptive encryption processing is then applied to these contract documents with varying sensitivity levels, resulting in different encryption versions of the files. This achieves diverse encryption processing for transformed files, adapting to different scenarios of contract document retrieval and reducing the risk of users accessing core file content. Simultaneously, an independent file storage blockchain is established to match and store contracts based on their encryption levels, and access permissions are allocated for various encrypted contract documents. This addresses the problems mentioned in the background art, namely:

[0006] A single encryption method cannot provide adaptive encryption based on the sensitivity of the file, increasing the risk of leakage during the file digitization process.

[0007] To achieve the above objectives, a decision optimization method for enterprise digital transformation based on multi-agent collaboration is provided, including the following steps:

[0008] S1. The intelligent agent collects contracts signed by actual enterprises and identifies and processes the key contents of the contracts.

[0009] S2. Establish an internal database for storing sensitive words, identify sensitive words in the collected contract content, and define the sensitivity level of each contract;

[0010] S3. Perform step-by-step encryption based on the sensitivity level of the contract to obtain contract document versions with different encryption levels;

[0011] S4. Establish an independent document storage blockchain and match and store documents according to the encryption level of the contract document version;

[0012] S5. Authorize access permissions for the corresponding independent file storage blockchain based on the administrator's security level, and assign the corresponding contract file version.

[0013] First, since most of the company's contracts are signed offline, the signed contracts are paper documents. In order to digitize them, it is necessary to first collect the contracts actually signed by the company through an intelligent agent, and then identify and process the key content in the contracts. The corresponding key content will serve as a reference factor for subsequent assessment of the sensitivity level of the contracts.

[0014] After completing the identification of key contract content, different companies have different areas of expertise and corresponding confidentiality rules. Therefore, in order to adapt to the confidentiality rules of different companies, it is necessary to establish an internal sensitive word storage database in advance during the sensitivity level assessment of the collected contracts. The internal sensitive word storage database of this solution adopts a Tric tree data structure, which efficiently stores sensitive words and supports complexity retrieval. At the same time, a failure pointer is added to the Tric tree, which, together with the AC automaton, scans the text and supports multi-mode string parallel matching, thereby improving the efficiency of subsequent sensitive word traversal.

[0015] To further improve the sensitivity word recognition performance, the sensitivity words are converted into vectors using the BERT model:

[0016] (Sensitive words);

[0017] in For vectors, For sensitive word models, each sensitive word model is converted into a vector and stored in a vector database. A dual-channel parallel detection is used, including an exact matching channel: the AC automaton scans the text and outputs words that directly match the sensitive words, and a semantic matching channel: after text segmentation, the text is vectorized → the vector database is queried → words with similarity greater than a threshold are returned. It is worth noting that the similarity threshold here is determined by the current enterprise. For contracts with a high degree of confidentiality, the corresponding similarity threshold is lowered, which means that the number of approximately related sensitive words increases, and the number of sensitive words that need to be encrypted in the future increases.

[0018] After establishing the enterprise's internal sensitive word storage database, in order to facilitate the classification and identification of subsequent contract content, sensitive word identification was performed on the collected contract content, and the sensitivity level of each contract was defined. The sensitivity range was divided according to the number of sensitive words in each contract, namely low-risk sensitivity range, medium-risk sensitivity range, and high-risk sensitivity range. Based on the risk range, the current contract was classified into sensitivity levels. Contracts in the low-risk sensitivity range were encrypted with files, contracts in the medium-risk sensitivity range were encrypted with files and sensitive word masking, and contracts in the high-risk sensitivity range were encrypted with files, masking, and sensitive paragraphs.

[0019] Furthermore, after encryption, an independent file storage blockchain is established to match and store contract file versions based on their encryption levels. Access permissions for the corresponding independent file storage blockchain are authorized according to the administrator's security level, and corresponding contract file versions are allocated. This enables diverse encryption processing of transformed files during enterprise digital transformation, adapting to contract file access in different scenarios, reducing the risk of users accessing the core content of files, and allocating access permissions for various encrypted contract files to further improve the security of encrypted files.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0021] This enterprise digital transformation decision optimization method based on multi-agent collaboration performs step-by-step encryption processing according to the sensitivity level of the contract, obtaining contract document versions with different encryption levels. Adaptive encryption processing is then applied to contract documents with different sensitivity levels to obtain file versions with different encryption levels. This achieves diversified encryption processing of transformed files to adapt to contract document access in different situations, reducing the risk of users accessing the core content of the files. At the same time, it is combined with the established independent file storage blockchain, which matches and stores the contract document versions according to their encryption levels, and assigns access permissions to various encrypted contract documents to improve the security of encrypted files. Attached Figure Description

[0022] Figure 1 This is a diagram illustrating the overall method steps of the present invention. Detailed Implementation

[0023] The technical solutions in 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.

[0024] Please see Figure 1 As shown, a method for optimizing enterprise digital transformation decisions based on multi-agent collaboration is provided, including the following steps:

[0025] S1. The intelligent agent collects contracts signed by actual enterprises and identifies and processes the key contents of the contracts.

[0026] S2. Establish an internal database for storing sensitive words, identify sensitive words in the collected contract content, and define the sensitivity level of each contract;

[0027] S3. Perform step-by-step encryption based on the sensitivity level of the contract to obtain contract document versions with different encryption levels;

[0028] S4. Establish an independent document storage blockchain and match and store documents according to the encryption level of the contract document version;

[0029] S5. Authorize access permissions for the corresponding independent file storage blockchain based on the administrator's security level, and assign the corresponding contract file version.

[0030] The details are as follows:

[0031] Firstly, since most enterprise contracts are signed offline, these are paper documents. To digitize them, the first step is to utilize intelligent agents (in this solution, the intelligent agents primarily employ the DEEPSEEK+DIFY workflow and the COZE knowledge base intelligent agent). The core capability of the DEEPSEEK+DIFY workflow is its drag-and-drop intelligent agent pipeline, supporting text extraction and structured processing of PDFs, images, and scanned documents. Its built-in OCR engine and natural language understanding module can understand and recognize keywords in documents. The COZE knowledge base intelligent agent's key advantage is its ability to support one-click uploading of local files and web page data to the database, and its automatic... (Using knowledge base content for question answering) Collects contracts signed by actual enterprises and identifies key content within the contracts. During the identification process, contract documents (e.g., PDF, DOC, or scanned copies) are submitted through the agent's dialogue interface or a dedicated upload function. After receiving the input, the agent initiates the collection workflow. Once the file upload is complete, the agent uses NLP tools (e.g., Spacy or Jieba) to parse the text, performing word segmentation and syntactic analysis to extract key content. This key content includes entities (e.g., "Party A," "Amount," or "Effective Clause") and key clauses (e.g., "Confidentiality Clause" or "Breach of Contract Clause"), all of which serve as the basis for subsequently defining the sensitivity level. After completing the contract text parsing, the extracted content is structured and stored in the agent's built-in database, binding the contract documents with the extracted features. The corresponding key features serve as reference factors for subsequent assessment of the contract's sensitivity level.

[0032] After completing the identification of key contract content, due to the different fields involved by different enterprises, the corresponding confidentiality rules also differ. Therefore, in order to adapt to the confidentiality rules of different enterprises, it is necessary to establish an internal sensitive word storage database in advance during the sensitivity level assessment of the collected contracts. The internal sensitive word storage database of this solution adopts the data structure of Tric tree, which efficiently stores sensitive words and supports complexity retrieval. At the same time, failure pointers are added on the Tric tree, and in conjunction with the AC automaton to scan the text, it supports multi-mode string parallel matching and improves the efficiency of subsequent sensitive word traversal.

[0033] To further improve the sensitivity word recognition performance, the sensitivity words are converted into vectors using the BERT model:

[0034] (Sensitive words);

[0035] in For vectors, For sensitive word models, each sensitive word model is converted into a vector and stored in a vector database, such as the Chroma and FAISS vector databases. This vector database can support synonym matching, such as "account" ≈ "account number".

[0036] In the specific implementation process, the first step is to collect sensitive words from enterprise compliance documents, historical audit records, and industry standard dictionaries. The raw sensitive words undergo special character cleaning (e.g., * and spaces) and case standardization, while synonym merging. The collected raw sensitive words are then input into a Tric tree, constructing a tree structure with characters as nodes. For example, root: account (its sensitive word ID is 101). Subsequently, the sensitive words are processed using a vectorization engine, that is, using an ERT model to convert the sensitive words into vectors, obtaining the word vectors for each sensitive word, handling polysemy, and merging context-related vectors. The processed sensitive word model is stored in a vector database, and an index road is constructed. Sensitive words of different categories are sharded and stored according to a partitioned index. A vector database storage structure is established, which serves as the enterprise's internal sensitive word storage database. This includes sensitive word IDs, categories, and sensitive word vectors. Finally, a unified retrieval structure is implemented, employing a dual-channel parallel detection approach. This includes an exact matching channel: the AC automaton scans the text and outputs words that directly match sensitive words (e.g., "amount" matches ID 103, while "structure" does not match any sensitive words in the vector database). It also includes a semantic matching channel: after text segmentation and vectorization, the vector database is queried, returning words with similarity greater than a threshold (e.g., "price" and "single item" have a similarity of 0.95). It's worth noting that this similarity threshold is determined by the enterprise itself. For contracts with a high degree of confidentiality, the corresponding similarity threshold is lowered, meaning the number of approximately related sensitive words increases, and the number of sensitive words requiring subsequent encryption increases.

[0037] After establishing the enterprise's internal sensitive word storage database, in order to facilitate the subsequent classification and identification of contract content, sensitive word identification is performed on the collected contract content, and the sensitivity level of each contract is defined. The specific definition steps are as follows:

[0038] First, the contract content is entered into the company's internal sensitive word storage database. Then, sensitive words are identified through precise matching and semantic matching channels to obtain the number of sensitive words contained in the contract content. And define the threshold for the number of sensitive words. Sensitive areas are divided into low-risk sensitive areas. Medium-risk sensitive areas and high-risk sensitive areas ,in The threshold multiple for the number of sensitive words is defined internally by the company. The maximum number of sensitive words is the total number of sensitive words stored in the company's internal sensitive word storage database, which is compared with the number of sensitive words contained in the contract content. Obtain the corresponding risk range and classify the sensitivity level of the current contract based on the risk range.

[0039] Furthermore, since the sensitivity level of the contract content directly determines the subsequent encryption method, this solution requires step-by-step encryption based on the contract's sensitivity level to obtain contract file versions with different encryption levels. This solution performs step-by-step encryption for contracts with different sensitivity levels. For contracts in the low-risk sensitivity range, file encryption is used, i.e., the file storing the contract content is encrypted. AES symmetric encryption is used to generate a key and a password for the contract file. Authorized administrators will obtain the corresponding password and can directly extract the contract from the file. For contracts in the medium-risk sensitivity range, file encryption and sensitive word masking encryption are used. Sensitive word characters are replaced with placeholders. For example, in the phrase "product unit price is **", the specific amount is a sensitive word and is therefore encrypted. The specific masking encryption program code is as follows:

[0040] def mask_sensitive(text, words):

[0041] for word in words:

[0042] if word in text:

[0043] masked = word[0] + "*"*(len(word)-2) + word[-1] if len(word) > 1 else "*" # Keep first and last characters

[0044] text = text.replace(word, masked)

[0045] return text.

[0046] For contracts involving high-risk and sensitive areas, file encryption, mask encryption, and encryption of sensitive paragraphs are employed. The methods for encrypting sensitive paragraphs are as follows:

[0047] First, sensitive words in each paragraph are masked and encrypted, and the masking rate of each paragraph is obtained. ,in The number of masked sensitive words in the paragraph. Define the total number of words (including characters) for the current paragraph and set a mask rate threshold. and the mask rate in each paragraph Perform a comparison;

[0048] When the mask rate of the current paragraph Exceeding the mask rate threshold When this happens, the current paragraph is marked as a sensitive paragraph, and placeholders are used to replace all characters in the sensitive paragraph.

[0049] When the mask rate of the current paragraph Not exceeding the mask rate threshold If this happens, the current paragraph is marked as a regular paragraph, and placeholders are used to replace sensitive words in the regular paragraph.

[0050] It is worth noting that contracts in the low-risk sensitive range are encrypted only, resulting in a single encrypted file;

[0051] Contracts falling within the medium-risk sensitive range are encrypted using both file encryption and sensitive word masking, generating encrypted files and sensitive word masked files.

[0052] Contracts in high-risk and sensitive areas are encrypted using file encryption, mask encryption, and sensitive paragraph encryption to generate encrypted files, sensitive word mask files, and sensitive paragraph mask files, which serve as the basis for subsequent adaptive allocation.

[0053] Furthermore, after completing the encryption process, an independent file storage blockchain is established to match and store contracts based on their encryption level. In the specific establishment process, a blockchain platform is first selected. In this solution, an enterprise-grade private blockchain (Hyperledger Fabric) is used to provide flexible permission management and deploy nodes, i.e., multiple authorized nodes (such as servers or dedicated devices) are set up. Each node needs to be authenticated and an independent internal network is set up to avoid direct connection with the external Internet.

[0054] To achieve physical isolation and improve the encryption effect of various types of files, the blockchain needs to be partitioned and divided according to the sensitivity of the files (for example, partition A stores sensitive paragraph mask files, partition B stores sensitive word mask files, and partition C stores encrypted files). Different partitions map to files of different sensitivity levels (encrypted files, sensitive word mask files, and sensitive paragraph mask files). Each partition corresponds to an independent blockchain channel to achieve physical isolation between files.

[0055] When files of different types are uploaded to the blockchain, the system allocates them to the corresponding partitions based on their sensitivity level and mapping relationship. Meanwhile, time files are stored off-chain (such as IPFS), generating a unique hash value. The hash value, file metadata (such as file owner, creation time, and confidentiality level), and access policies are stored on-chain.

[0056] Finally, based on the administrator's security level, access permissions for the corresponding independent file storage blockchain are authorized, and the corresponding contract file version is assigned. Access rules are enforced through smart contracts. The administrator submits an access request via the blockchain client, including: the administrator's digital certificate (public key), the target partition (file-sensitive partition), the file number (i.e., the sequence number locating the specific file), and the operation type (e.g., downloading or accessing files). Upon receiving the access request, the blockchain client verifies the administrator's identity using a certificate verification mechanism. This scheme uses an SSL public key mechanism analogous to identify the uploaded administrator's digital certificate and obtain the current administrator's sensitive permissions, meaning they can access files in the corresponding area. For example, if the current administrator is a level A administrator, they can access sensitive paragraph mask files in partition A, sensitive word mask files in partition B, and files in partition C. The encrypted file is accessed through a partition that corresponds to the target partition submitted by the administrator. After the administrator's identity is verified, the node in the corresponding partition locates the target file using the file number in the access request and sends the download path and the file's decryption path back to the current administrator. The administrator downloads the file requested by the access request through the download path and obtains the authorized content of the corresponding file according to the corresponding decryption path (i.e., including the key of the encrypted file and the mask decryptor of the decryption mask). That is, the current contract file generates an encrypted file, a sensitive word mask file, and a sensitive paragraph mask file according to different levels of encryption processing. The administrator with the highest privileges can directly obtain the encrypted file and the corresponding key of the encrypted file to obtain the decrypted file content. The administrator with the lowest privileges can only obtain the sensitive paragraph mask file, that is, some paragraphs in the obtained file content are masked.

[0057] This invention employs a step-by-step encryption process based on the sensitivity level of the contract, obtaining contract document versions with different encryption levels. Adaptive encryption is then applied to these different sensitivity levels, resulting in various encrypted versions of the document. This enables diverse encryption processing of documents during enterprise digital transformation, adapting to different scenarios and reducing the risk of users accessing core document content. Furthermore, an independent file storage blockchain is established to match and store contracts based on their encryption levels, assigning access permissions to different types of encrypted contract documents, further enhancing the security of encrypted files.

[0058] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A decision optimization method for enterprise digital transformation based on multi-agent collaboration, characterized by: Includes the following steps: S1. The intelligent agent collects contracts signed by actual enterprises and identifies and processes the key contents of the contracts. S2. Establish an internal database for storing sensitive words, identify sensitive words in the collected contract content, and define the sensitivity level of each contract; S3. Perform step-by-step encryption based on the sensitivity level of the contract to obtain contract document versions with different encryption levels; S4. Establish an independent document storage blockchain and match and store documents according to the encryption level of the contract document version; S5. Authorize access permissions for the corresponding independent file storage blockchain based on the administrator's security level, and assign the corresponding contract file version.

2. The enterprise digital transformation decision optimization method based on multi-agent collaboration as described in claim 1, characterized in that: The method for identifying and processing key content in the contract in S1 includes the following steps: S1.1 Submit contract documents through the intelligent agent's dialogue interface or dedicated upload function; S1.2 After receiving input, the intelligent agent starts the data acquisition workflow; S1.3 After the file upload is completed, the agent calls NLP tools to parse the text, performs word segmentation and syntactic analysis on the contract text, and extracts key content; S1.

4. The extracted content is structured and stored in the database built into the agent, and the contract documents are bound to the extracted features for storage.

3. The enterprise digital transformation decision optimization method based on multi-agent collaboration as described in claim 2, characterized in that: The key content in S1.3 includes entities and key terms.

4. The enterprise digital transformation decision optimization method based on multi-agent collaboration as described in claim 1, characterized in that: The method for establishing an internal sensitive word storage database in S2 includes the following steps: S2.1 Collect sensitive words for enterprises, clean the original sensitive words with special symbols and unify the capitalization, and merge synonyms at the same time; S2.2 Input the collected raw sensitive words into the Tric tree and construct a tree structure with characters as nodes; S2.

3. Processing through a vectorization engine, sensitive words are converted into vectors using the BERT model: (Sensitive words); in For vectors, For sensitive word models, each sensitive word model is converted into a vector and stored in a vector database; S2.

4. Store the processed sensitive word model using a vector database and construct an index route; S2.

5. Based on the partition index, sensitive words of different categories are stored in segments to construct a vector database storage structure; S2.6 Unified retrieval structure.

5. The enterprise digital transformation decision optimization method based on multi-agent collaboration as described in claim 4, characterized in that: The retrieval structure in S2.6 includes an exact matching channel and a semantic matching channel.

6. The enterprise digital transformation decision optimization method based on multi-agent collaboration as described in claim 5, characterized in that: The method for defining the sensitivity level of each contract in S2 includes the following steps: S2.

10. Input the contract content into the company's internal sensitive word storage database, and use the precise matching channel and semantic matching channel to identify sensitive words and obtain the number of sensitive words contained in the contract content. ; S2.11, Define the threshold for the number of sensitive words. Define sensitive areas; S2.12 Compare the number of sensitive words contained in the contract content. Obtain the corresponding risk range and classify the sensitivity level of the current contract based on the risk range.

7. The enterprise digital transformation decision optimization method based on multi-agent collaboration as described in claim 6, characterized in that: The sensitive range in S2.11 includes low-risk sensitive ranges. Medium-risk sensitive areas and high-risk sensitive areas ,in The threshold multiple for the number of sensitive words is defined internally by the company. This represents the maximum number of sensitive words.

8. The enterprise digital transformation decision optimization method based on multi-agent collaboration as described in claim 1, characterized in that: The step-by-step encryption method in S3, based on the sensitivity level of the contract, includes the following steps: S3.

1. Implement step-by-step encryption for contracts with different sensitivity levels; S3.2 For contracts in low-risk sensitive areas, file encryption is adopted, and the key and key of the contract file are generated by AES symmetric encryption. S3.

3. Contracts falling within the medium-risk sensitive range shall be encrypted using both file encryption and sensitive word masking. S3.4 Contracts in high-risk and sensitive areas shall be encrypted using file encryption, mask encryption, and encryption of sensitive paragraphs.

9. The enterprise digital transformation decision optimization method based on multi-agent collaboration as described in claim 8, characterized in that: The method for encrypting sensitive paragraphs in S3.4 includes the following steps: S3.4.

1. Mask and encrypt sensitive words in each paragraph; S3.4.2 Obtain the mask rate in each paragraph ,in The number of masked sensitive words in the paragraph. This represents the total number of words in the current paragraph. S3.4.3, Set the mask rate threshold and the mask rate in each paragraph Perform a comparison; When the mask rate of the current paragraph Exceeding the mask rate threshold When this happens, the current paragraph is marked as a sensitive paragraph, and placeholders are used to replace all characters in the sensitive paragraph. When the mask rate of the current paragraph Not exceeding the mask rate threshold If this happens, the current paragraph is marked as a regular paragraph, and placeholders are used to replace sensitive words in the regular paragraph.

10. The enterprise digital transformation decision optimization method based on multi-agent collaboration according to claim 1, characterized in that: The method for granting access permissions to the corresponding independent file storage blockchain based on the administrator's security level in S5 includes the following steps: S5.

1. Access rules are enforced through smart contracts, and administrators submit access requests through a blockchain client; The access request includes the administrator's digital certificate, the target partition, the file number, and the operation type; S5.2 After receiving the access request, the blockchain client verifies the administrator's identity by verifying the administrator's identity through the certificate. S5.3 The node of the corresponding partition locates the target file by the file number in the access request and sends the download path and the decryption path of the file back to the current administrator; S5.4 The administrator downloads the required file through the download path and obtains the access permissions for the corresponding file based on the corresponding decryption path.

Images