Method and system for realizing service data request by using AI agent under private deployment service system

Abstract

The invention relates to a method for realizing a service data request by using an AI agent under a private deployment service system. The method comprises the following steps: generating and storing an authentication token and a privacy level according to an identity certificate; receiving a service access instruction of the AI agent, wherein the service access instruction is formed by the AI agent by carrying an authentication token and index information of a service sub-module selected for the natural language access request by the natural language access request; verifying an authentication token of the service access instruction, if the authentication token passes the verification, calling a service sub-module according to the index information, and reading data limited in the access authority by the authentication token from a corresponding area of the service knowledge base; according to the privacy level, processing the data in a corresponding level, and sending the processed data to the AI intelligent agent, so that the AI intelligent agent analyzes the data to obtain answer feedback data responding to the natural language access request. According to the invention, the privacy leakage risk caused by the deployment of the AI agent in the private deployment service system is avoided, and the security of the private deployment service system is improved.

Description

Technical Field

[0001] The present application relates to the field of AI agent technology, and in particular to a method and system for implementing business data requests using an AI agent in a privately deployed business system. Background Art

[0002] With the rise and popularization of AI agents, AI agents are increasingly being deployed by enterprises locally or in private clouds, and are connected to the enterprise's privately deployed business systems. By receiving natural language requests input by enterprise users and calling the API interface of the enterprise's privately deployed business system, AI agents can directly access the enterprise's business knowledge base and analyze the relevant data in the business knowledge base to achieve professional questions and answers in the private field, thereby providing convenience for enterprise users.

[0003] However, an enterprise's business knowledge base is usually a knowledge base built from documents in the private domain, which stores a large amount of private data. AI agents will have direct access to this private data. Currently, when enterprise users use AI agents to call local privately deployed business systems to obtain data from the business knowledge base, there is no identity authentication. Therefore, enterprise users can directly call the private data of the business knowledge base through AI agents, which poses a huge risk of privacy leakage.

[0004] Therefore, how to avoid the privacy leakage risk caused by deploying AI agents in privately deployed business systems, thereby improving the security of privately deployed business systems, is a technical problem that technical personnel in this field urgently need to solve. Summary of the Invention

[0005] This application provides a method and system for using AI agents to implement business data requests in a privately deployed business system, so as to avoid the privacy leakage risks caused by deploying AI agents in the privately deployed business system, thereby improving the security of the privately deployed business system.

[0006] To solve the above technical problems, this application provides the following technical solutions:

[0007] A method for implementing business data requests using an AI agent in a privately deployed business system, applied to a backend service system, includes the following methods: receiving identity credentials sent by the front end of the business system, and generating and storing an authentication token and a privacy level based on the identity credentials; sending the stored authentication token to the AI agent in response to logging into the AI agent; receiving a business access instruction from the AI agent, where the business access instruction is formed by the AI agent using a natural language access request carrying the authentication token and index information of a business submodule selected for the natural language access request; verifying the authentication token of the business access instruction, and if the verification is successful, calling the corresponding business submodule based on the index information, and reading data restricted by the authentication token to the access permission from the corresponding area of the business knowledge base; performing corresponding level processing on the read data based on the stored privacy level, and sending the processed data to the AI agent so that the AI agent analyzes the data to obtain answer feedback data in response to the natural language access request.

[0008] The method of using an AI agent to implement business data requests in a privately deployed business system as described above is applied to a backend service system, wherein preferably, if the verification fails, a request failure message is returned.

[0009] The method for implementing business data requests using an AI agent in a privately deployed business system as described above is applied to a backend service system, wherein preferably, the AI agent forms a business access instruction, which specifically includes the following sub-steps: receiving and parsing the semantics of a natural language access request; judging whether its intention is related to the business based on the semantics of the natural language access request; if it is related to the business, judging whether the authentication token is valid; if it is valid, selecting the corresponding business sub-module based on the intention of the natural language access request; and forming a business access instruction based on the semantics of the natural language access request, the index information of the business sub-module, and the authentication token.

[0010] The method for implementing business data requests using an AI agent in a privately deployed business system as described above is applied to a backend service system, wherein preferably, calculating the intent of a natural language access request includes the following sub-steps: performing text cleaning on the natural language access request; performing keyword extraction on the natural language access request after text cleaning; calculating the importance value of each keyword and using the importance value as the weight of the keyword; determining the candidate score of the natural language access request belonging to each intent based on each keyword and the probability that each keyword belongs to the corresponding intent; and using the intent with the maximum candidate score as the intent of the natural language access request.

[0011] The method for implementing business data requests using an AI agent in a privately deployed business system as described above is applied to a backend service system, wherein preferably, the processing of the read numerical data includes the following sub-steps: correcting the preset standard mapping interval width according to the privacy level; comparing the corrected standard mapping interval width with the width of the overall numerical data of the numerical data set; if the corrected standard mapping interval width is not less than the width of the overall numerical data of the numerical data set, the actual mapping interval width is the width of the overall numerical data of the numerical data set; if the corrected standard mapping interval width is less than the width of the overall numerical data of the numerical data set, the actual mapping interval width is the corrected standard mapping interval width; replacing the standard mapping interval width in the numerical data processing model with the actual mapping interval width to obtain a corrected numerical data processing model; and processing the numerical data through the corrected numerical data processing model.

[0012] The method for realizing business data request using AI agent in a privately deployed business system as described above is applied to a background service system, wherein preferably, the processing of the read text data includes the following sub-steps: determining the preliminary number of characters of the private part of the text data according to the text data; correcting the preliminary number of characters of the private part of the text data according to the privacy level; comparing the preliminary number of characters of the corrected private part with the number of characters of the text data; if the preliminary number of characters of the corrected private part is not less than the number of characters of the text data, the number of characters of the random number to be generated is half of the number of characters of the text data; if the preliminary number of characters of the corrected private part is less than the number of characters of the text data, the number of characters of the random number to be generated is the preliminary number of characters of the corrected private part; generating a random number of the corresponding number of characters based on the determined number of characters of the random number, and inputting the random number and the text data into a text data processing model to process the text data.

[0013] A method for implementing business data requests using an AI agent in a privately deployed business system, applied to the AI agent, includes the following methods: in response to logging into the AI agent, receiving an authentication token stored in a background service system; receiving and selecting a business submodule for a natural language access request, carrying the authentication token and index information of the business submodule with the natural language access request to form a business access instruction; sending the business access instruction to the background service system, so that the background service system calls the corresponding business submodule according to the index information, and reads data restricted by the authentication token within the access permission from the corresponding area of the business knowledge base; receiving and analyzing the data processed by the background service system to obtain answer feedback data in response to the natural language access request, wherein the background service system performs corresponding level of processing on the read data according to the stored privacy level to obtain processed data.

[0014] The method for implementing business data requests using an AI agent in a privately deployed business system as described above is applied to an AI agent, wherein preferably, a business access instruction is formed, specifically including the following sub-steps: receiving and parsing the semantics of a natural language access request; judging whether its intention is related to the business based on the semantics of the natural language access request; if it is related to the business, judging whether the authentication token is valid; if it is valid, selecting the corresponding business sub-module based on the intention of the natural language access request; forming a business access instruction based on the semantics of the natural language access request, the index information of the business sub-module and the authentication token.

[0015] The method for implementing business data requests using an AI agent in a privately deployed business system as described above is applied to an AI agent, wherein preferably, the intent of a natural language access request is calculated, including the following sub-steps: text cleaning of the natural language access request; keyword extraction of the natural language access request after text cleaning; calculating the importance value of each keyword and using its importance value as the weight of the keyword; determining the candidate score of the natural language access request belonging to each intent based on each keyword and the probability that each keyword belongs to the corresponding intent; and using the intent with the maximum candidate score as the intent of the natural language access request.

[0016] A system for implementing business data requests using an AI agent in a privately deployed business system, comprising: a privately deployed business system and an AI agent; wherein the privately deployed business system comprises: a business system front end, a back-end service system and a business knowledge base, the business system front end runs on an enterprise business system terminal, and the back-end service system and the business knowledge base run on an enterprise server; the AI agent runs on an enterprise user's terminal; the back-end service system executes any of the above methods applied to the back-end service system, and the AI agent executes any of the above methods applied to the AI agent.

[0017] Compared with the above background technology, the method and system provided in this application for using AI agents to implement business data requests in a privately deployed business system can avoid the privacy leakage risks caused by deploying AI agents in a privately deployed business system, thereby improving the security of the privately deployed business system. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0019] Figure 1This is a flowchart of a method for implementing a business data request using an AI agent in a privately deployed business system, provided in an embodiment of the present application;

[0020] Figure 2 This is a schematic diagram of a system for implementing business data requests using an AI agent in a privately deployed business system, as provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] The following describes embodiments of the present invention in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended only to explain the present invention and are not to be construed as limiting the present invention.

[0022] AI agents are mostly designed as general models. They have the following defects in privately deployed business system scenarios: (1) Lack of identity recognition: AI agents cannot identify the identity of the user initiating the current request, especially in multi-user concurrent scenarios, and it is difficult to distinguish requests from different users; (2) Insufficient data isolation: The lack of an authentication mechanism may cause business data to be confused or accessed without authorization, posing a risk of privacy leakage; (3) Compliance challenges: In the sensitive data environment of privately deployed business systems, the requirements of data protection regulations (such as GDPR and CCPA) for user data isolation are not met; (3) Security risks: Access requests without identity authentication may be used maliciously, increasing the risk of system attacks.

[0023] like Figure 1 As shown, this application provides a method for implementing business data requests using an AI agent in a privately deployed business system, including the following steps:

[0024] Step S10: Log in to the front end of the business system and provide the enterprise user's identity credentials to the token verification module of the backend service system. The token verification module generates an authentication token and privacy level for the enterprise user based on the identity credentials of the enterprise user.

[0025] like Figure 2 As shown, a system for implementing business data requests using an AI agent in a privately deployed business system includes: a privately deployed business system 100, and the privately deployed business system 100 includes: a business system front end 110, a backend service system 120 and a business knowledge base 130; the business system front end 110 runs on the enterprise business system terminal, and the backend service system 120 and the business knowledge base 130 run on the enterprise server.

[0026] Among them, the background service system 120 includes: a token verification module 121, a business module 122, a data isolation module 123 and a data return module 124; in this application, according to the business type, the business module 122 is divided into different business sub-modules, such as: employee attendance sub-module 1221, department budget sub-module 1222, employee information sub-module 1223..., each business sub-module has a corresponding API interface, and each business sub-module stores corresponding data in the corresponding area of the business knowledge base.

[0027] Before using the AI agent 200 to make a natural language access request, all enterprise users need to log in to the backend service system 120 through the business system front end 110 on the enterprise business system terminal, and the enterprise user provides the enterprise user's identity credentials (for example, user name and password) to the token verification module 121 of the backend service system 120. After successful login, the token verification module 121 of the backend service system 120 verifies the identity credentials (for example, user name and password) submitted by the enterprise user. After the verification is passed, the token verification module 121 of the backend service system 120 generates a unique authentication token (Token) and privacy level based on the identity credentials, and saves the authentication token and privacy level of the enterprise user.

[0028] Step S20: Log in to the AI agent, obtain the enterprise user's authentication token, receive the enterprise user's natural language access request, select a business submodule for the natural language access request, and add the enterprise user's authentication token and the index information of the selected business submodule to the natural language access request to form a business access instruction, which is then sent to the token verification module of the backend service system.

[0029] like Figure 2 As shown, a system for implementing business data requests using an AI agent in a privately deployed business system includes: an AI agent 200, the AI agent 200 runs on the terminal of the enterprise user, and the AI agent 200 includes: a language parsing module 210, a request judgment module 220, a business selection module 230, an instruction sending module 240, a data receiving module 250 and a language response module 260.

[0030] When an enterprise user logs in to the AI agent 200 on the enterprise user's terminal, the token verification module 121 of the backend service system 120 will pass the generated authentication token (Token) to the agent 200. When the enterprise user enables the AI agent 200 to ask a natural language question, the language parsing module 210 receives the enterprise user's natural language access request. Since the enterprise user has logged in to the AI agent 200, the AI agent 200 can know the relevant information of the enterprise user, such as the authentication token (Token) generated by the token verification module 121 of the backend service system 120. Therefore, when the language parsing module 210 receives the enterprise user's natural language access request, it also receives the enterprise user's authentication token. The language parsing module 210 parses the semantics of the natural language access request through natural language so that it can be understood and processed by the machine.

[0031] The request judgment module 220 judges whether the intention of the natural language access request is relevant to the business based on the semantics of the natural language access request. First, the natural language access request is cleaned to remove noise; then the natural language access request A after the text cleaning is subjected to keyword extraction to obtain keywords {A1, A2, ..., A i ,…,A I}, where A1 is the first keyword of the natural language access request A, A2 is the second keyword of the natural language access request A, and A i For the i-th keyword of natural language access request A, A I is the I-th keyword of the natural language access request A; then the importance value of each keyword of the natural language access request A is calculated, and its importance value is used as the weight of the keyword; then each keyword of the natural language access request A and each keyword belongs to the j-th intent C j The probability of natural language access request A is determined to determine the candidate scores of each intent. The intent with the maximum candidate score is used as the intent of natural language access request A. Whether the intent of natural language access request A is related to the business is determined. For example, if the intent of natural language access request A is to obtain employee attendance records, then natural language access request A is determined to be related to the business.

[0032] Specifically, according to the formula Calculate the i-th keyword A i The importance value Q(A i , A); among them, TF(A i , A) is keyword A i The frequency of occurrence in the natural language access request A; N is the total number of documents used to train the AI agent 200; The total documents containing keyword A for training AI agent 200 iThe number of documents; μ is a constant 1; To round up. According to the formula Calculate the natural language access request A intent C j The candidate score S(A, C j ); where δ(A i , C j ) is keyword A i Belongs to the jth intention C j The probability that keyword A i For intention C j , then δ(A i , C j )=1, otherwise 0; I is the number of keywords in the natural language access request A.

[0033] If it is not related to the business, the language response module 260 gives a response based on public knowledge capabilities to respond to the natural language access request; if it is related to the business, the request judgment module 220 initially determines whether the enterprise user's authentication token is valid; if it is invalid, the language response module 260 gives a response based on public knowledge capabilities to respond to the enterprise user's access request; if it is valid, the business selection module 230 selects the corresponding business sub-module according to the intention of the natural language access request, for example: department budget sub-module 1222, thereby forming a business access instruction.

[0034] Specifically, the AI agent 200 is pre-configured with index information corresponding to different business sub-modules in the privately deployed business system 100. This allows for searching and matching within the index information based on the intent of the natural language access request, obtaining index information that matches the intent of the natural language access request, and then forming a business access instruction based on the semantics of the natural language access request, the obtained index information, and the enterprise user's authentication token. The authentication token is placed in the instruction header authentication field of the business access instruction, while the semantics of the natural language access and the index information are placed in the instruction body field of the business access instruction. Based on the obtained index information, the corresponding business sub-module's API interface is invoked, which is used to perform operational access to the selected business sub-module.

[0035] After forming the service access instruction, the instruction sending module 240 sends the service access instruction to the background service system 120 of the private deployment service system 100 .

[0036] Step S30: The token verification module verifies the validity and authenticity of the authentication token in the service access instruction. If the verification fails, the data return module of the backend service system returns a request failure message.

[0037] After the token verification module 121 of the background service system 120 receives the business access instruction from the AI agent 200, it verifies the authentication token in the business access instruction again through the authentication token of the enterprise user saved in the token verification module 121. If the verification fails, the data return module 124 of the background service system 120 returns a request failure message and ends the natural language access request.

[0038] Step S40: If the verification is successful, the token verification module calls the corresponding business submodule of the backend service system according to the instruction of the business access instruction, and sends the authentication token and privacy level of the enterprise user to the business submodule;

[0039] If the verification is successful, the token verification module 121 of the backend service system 120 calls the corresponding API interface using the index information in the business access instruction, thereby accessing the corresponding business sub-module communication and reading the data in the corresponding area of the business knowledge base 130 through the business sub-module. In addition, since the token verification module 121 of the backend service system 120 generates the enterprise user's privacy level based on the enterprise user's identity credentials when the enterprise user logs into the backend service system 120 through the business system front end 110, when processing the enterprise user's natural language access request and establishing communication between the token verification module 121 and the corresponding business sub-module, the token verification module 121 also sends the enterprise user's privacy level to the corresponding business sub-module.

[0040] Step S50: The business submodule reads the data restricted by the enterprise user's authentication token within the access rights from the corresponding area of the business knowledge base to the data isolation module of the backend service system, and sends the privacy level of the enterprise user to the data isolation module;

[0041] The business knowledge base 130 is divided into different areas, for example: an area for storing employee attendance, an area for storing department budgets, an area for storing employee information... The employee attendance submodule 1221 stores employee attendance data in the area for storing employee attendance, the department budget submodule 1222 stores department budget data in the area for storing department budgets, and the employee information submodule 1223 stores employee information in the area for storing employee information.

[0042] Different authentication tokens have different access rights. After verification, the called business sub-module will go to the area corresponding to the business sub-module in the business knowledge base 130, read the data within the access rights according to the access rights restricted by the authentication token of the enterprise user, and read the data to the data isolation module 123 of the background service system 120.

[0043] Since each business sub-module stores data in the corresponding area of the business knowledge base 130, the access scope of each business sub-module is limited from the storage area, thereby preliminarily improving the security of the data and preventing information leakage; and, access permissions are set through the enterprise user's authentication token, and when the business sub-module reads data, the reading scope is further limited, thereby further improving the overall security of the data and preventing information leakage.

[0044] In addition, the called business sub-module will also send the privacy level of the enterprise user to the data isolation module 123 of the background service system 120, so that the data isolation module 123 can perform different levels of processing on the read data according to the privacy level of the enterprise user, thereby further improving the security of the data and preventing information leakage.

[0045] Step S60: The data isolation module processes the read data according to the privacy level, and the data return module of the backend service system sends the processed data to the AI agent.

[0046] Since the read data is generally numerical data or text data, and the data isolation module 123 processes numerical data and text data differently, after the data is read to the data isolation module 123 of the background service system 120, the data isolation module 123 will divide the data into a numerical data set X = {x1, x2, ..., x m ,…,x M} or text dataset Y={y1, y2, ..., y t ,…,y T}. Where x1 is the first numerical data in the numerical data set, x2 is the second numerical data in the numerical data set, and x m is the mth numerical data in the numerical data set, x M is the Mth numerical data in the numerical data set; y1 is the first text data in the text data set, y2 is the second text data in the text data set, and y t is the tth text data in the text dataset, y T is the Tth text data in the text dataset.

[0047] When processing numerical data, the data isolation module 123 modifies the preset standard mapping interval width w according to the privacy level D of the enterprise user. bz , expressed as: Dw bz ; Compare the corrected standard mapping interval width Dw bz The width x of the overall numerical data set X M -x1; if the modified standard mapping interval width Dw bzThe width x of the entire numerical data set X is not less than M -x1, the actual mapping interval width is w sj The width x of the entire numerical data set X M -x1, that is: w sj =x M -x1; if the modified standard mapping interval width Dw bz If the width of the numerical data set X is smaller than the width of the entire numerical data, the actual mapping interval width w sj is the modified standard mapping interval width Dw bz , that is: w sj =Dw bz .

[0048] The numerical data processing model is pre-stored in the data isolation module 123 The data isolation module 123 sets the actual mapping interval width w sj Replace the standard mapping interval width w in the numerical data processing model bz Get the revised numerical data processing model Among them, x mv is the mth numerical data x m interval values mapped to intervals, To round down.

[0049] The data isolation module 123 uses the modified numerical data processing model The numerical data in the numerical data set X is processed to obtain processed numerical data.

[0050] When processing text data, the data isolation module 123 first t Determine the initial number of characters B of its private part tcb The data isolation module 123 then modifies the text data y by the privacy level D of the enterprise user t The initial number of characters of the privacy part B tcb , expressed as: DB tcb ; Compare the corrected preliminary character count DB of the private part tcb With text data y t The number of characters r t , if the initial number of characters of the corrected privacy part DB tcb Not less than text data y t The number of characters r t , then the number of characters of the random number B to be generated is B r is the text data y t The number of characters r t Half of B r =0.5r t; If the initial number of characters DB of the corrected privacy part tcb is less than the number of characters r of the text data y t , then the number of characters B of the random number B to be generated t is the initial number of characters DB of the corrected privacy part r , that is: B tcb = DB r . tcb .

[0051] The data isolation module 123 generates a random number B with the number of characters B r , and inputs the generated random number B and the text data y t into the text data processing model y pre-stored in the data isolation module 123 tu = (1 - k)left(y t , r1)+kB+(1 - k)right(y t , r2); where, left(y t , r1) selects r1 characters starting from the leftmost character among the characters of y t ; right(y t , r2) selects r2 characters starting from the rightmost character among the characters of y t ; r1 + r2 + B r = r t , and the (r1 + 1)-th character to the (r1 + 1 + B)-th character of the text data y t are privacy characters, which are replaced with the random number B; k is an adjustable factor, 0 < k < 1. The data isolation module 123 processes the text data in the text data set Y through the text data processing model y r = (1 - k)left(y tu , r1)+kB+(1 - k)right(y t , r2) to obtain the processed text data. t .

[0052] After the data isolation module 123 processes the read numerical data or text data, the data return module 124 of the background service system 120 sends the processed numerical data or text data to the data receiving module 250 of the AI intelligent agent 200.

[0053] Step S70, the AI intelligent agent receives the processed data and analyzes these data to obtain the answer feedback data of the natural language access request;

[0054] After the data receiving module 250 of the AI agent 200 receives the processed numerical data or text data, the language response module 260 of the AI agent 200 analyzes these numerical data or text data through a large language model to obtain answer feedback data corresponding to the natural language access request in response to the natural language access request.

[0055] In this application, the AI agent 200 is embedded in the private deployment business system 100, and the AI agent and the private deployment business system are connected through an API interface to achieve integration. By verifying the authentication token and data isolation mechanism, unauthorized access is prevented, the risk of privacy leakage is reduced, and the regulatory requirements for multi-user data protection in private deployment scenarios are met; the AI agent uses different API interfaces to access different areas of the business knowledge base of the private deployment business system, and limits the scope of access to data through authentication tokens, reducing the possibility of exposure and improving data security; in addition, the AI agent provides personalized responses based on user identity, improving business processing efficiency.

[0056] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

[0057] In addition, it should be understood that although this specification is described in terms of implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

Claims

1. A method for implementing business data requests using an AI agent in a privately deployed business system, characterized in that: Applied to the background service system, including the following methods: Receive the identity credentials sent by the business system front end, and generate and store the authentication token and privacy level based on the identity credentials; In response to logging into the AI agent, sending the stored authentication token to the AI agent; Receive a service access instruction from the AI agent, where the service access instruction is formed by the AI agent using a natural language access request carrying an authentication token and index information of a service submodule selected for the natural language access request; Verify the authentication token of the business access instruction. If the verification is successful, call the corresponding business sub-module according to the index information and read the data within the access permission restricted by the authentication token from the corresponding area of the business knowledge base; The read data is processed at a corresponding level according to the stored privacy level, and the processed data is sent to the AI agent so that the AI agent analyzes the data to obtain answer feedback data in response to the natural language access request.

2. The method for implementing business data requests using an AI agent in a privately deployed business system according to claim 1, characterized in that: Applied to the backend service system. If the verification fails, a request failure message will be returned.

3. The method for implementing business data requests using an AI agent in a privately deployed business system according to claim 1, characterized in that: Applied to the backend service system, the AI agent generates business access instructions, which specifically include the following sub-steps: Receive and parse the semantics of natural language access requests; Determine whether the intent of a natural language access request is business-related based on its semantics; If it is relevant to the business, determine whether the authentication token is valid; If valid, the corresponding business submodule is selected based on the intent of the natural language access request; A business access instruction is formed based on the semantics of the natural language access request, the index information of the business submodule and the authentication token.

4. The method for implementing business data requests using an AI agent in a privately deployed business system according to claim 3, characterized in that: Applied to the backend service system, calculating the intent of natural language access requests includes the following sub-steps: Perform text cleaning on natural language access requests; Extract keywords from natural language access requests after text cleaning; Calculate the importance value of each keyword and use it as the weight of the keyword; Determining a candidate score for each intent of the natural language access request based on each keyword and the probability that each keyword belongs to the corresponding intent; The intent with the largest candidate score is used as the intent of the natural language access request.

5. The method for implementing business data requests using an AI agent in a privately deployed business system according to any one of claims 1 to 4, characterized in that: Applied to the background service system, the processing of the read numerical data includes the following sub-steps: Modify the preset standard mapping interval width by privacy level; Compare the width of the corrected standard mapping interval with the width of the overall numerical data of the numerical data set; If the modified standard mapping interval width is not less than the width of the entire numerical data of the numerical data set, the actual mapping interval width is the width of the entire numerical data of the numerical data set; If the modified standard mapping interval width is smaller than the width of the entire numerical data of the numerical data set, the actual mapping interval width is the modified standard mapping interval width; The actual mapping interval width is substituted for the standard mapping interval width in the numerical data processing model to obtain a revised numerical data processing model; The numerical data is processed using the modified numerical data processing model.

6. The method for implementing business data requests using an AI agent in a privately deployed business system according to any one of claims 1 to 4, characterized in that: Applied to the background service system, the processing of the read text data includes the following sub-steps: Determine a preliminary number of characters of the private portion of the text data; Correcting the initial number of characters of the private portion of the text data by the privacy level; comparing the preliminary character count of the corrected private portion with the character count of the text data; If the preliminary number of characters of the revised private portion is not less than the number of characters of the text data, the number of characters of the random number to be generated is half the number of characters of the text data; If the number of characters in the revised initial private portion is less than the number of characters in the text data, the number of characters in the random number to be generated is the number of characters in the revised initial private portion; A random number with a corresponding number of characters is generated according to the determined number of characters of the random number, and the random number and text data are input into a text data processing model to process the text data.

7. A method for implementing business data requests using an AI agent in a privately deployed business system, applied to an AI agent, characterized in that: The following methods are included: In response to logging into the AI agent, receiving an authentication token stored in the backend service system; Receive and select a business submodule for a natural language access request, carry the authentication token and the index information of the business submodule to the natural language access request, and form a business access instruction; Send the business access instruction to the backend service system, so that the backend service system calls the corresponding business submodule according to the index information and reads the data within the access permission restricted by the authentication token from the corresponding area of the business knowledge base; Receive and analyze the data processed by the background service system to obtain answer feedback data in response to the natural language access request, wherein the background service system processes the read data at a corresponding level based on the stored privacy level to obtain the processed data.

8. The method for implementing business data requests using an AI agent in a privately deployed business system according to claim 7, characterized in that: Applied to the AI agent to form a business access instruction, which specifically includes the following sub-steps: Receive and parse the semantics of natural language access requests; Determine whether the intent of a natural language access request is business-related based on its semantics; If it is relevant to the business, determine whether the authentication token is valid; If valid, the corresponding business submodule is selected based on the intent of the natural language access request; A business access instruction is formed based on the semantics of the natural language access request, the index information of the business submodule and the authentication token.

9. The method for implementing business data requests using an AI agent in a privately deployed business system according to claim 8, characterized in that: Applied to AI agents, calculating the intent of natural language access requests includes the following sub-steps: Perform text cleaning on natural language access requests; Extract keywords from natural language access requests after text cleaning; Calculate the importance value of each keyword and use it as the weight of the keyword; Determining a candidate score for each intent of the natural language access request based on each keyword and the probability that each keyword belongs to the corresponding intent; The intent with the largest candidate score is used as the intent of the natural language access request.

10. A system for implementing business data requests using an AI agent in a privately deployed business system, characterized in that: include: Privately deploy business systems and AI agents; Among them, privately deployed business systems include: business system front-end, back-end service system and business knowledge base. The business system front-end runs on the enterprise business system terminal, and the back-end service system and business knowledge base run on the enterprise server; The AI agent runs on the enterprise user's terminal; The backend service system executes any one of claims 1 to 6 above, and the AI agent executes any one of claims 7 to 9 above.

Images