TWI844473B - System and method for processing confidential data using online large language model - Google Patents

Abstract

A system and method for processing confidential data using an online large language model, the system includes a data privacy enhancement module, a computing module, a packaging module and a transmission module. The data privacy enhancement module generates a plurality of computing data and provides them to the computing module. The computing module includes an online large language model, and the computing data is input into the online large language model for analysis or reasoning to generate a program code or a guiding principle. The packaging module packages the program code or guiding principle into a data package, and the data package is in a form that can be run by a closed large language model in a closed intranet, and then the data package is transmitted to the closed intranet through the transmission module. While protecting confidential data, the present invention utilizes the powerful analysis and reasoning capabilities of the online large language model to improve the capabilities of the closed large language model and achieve high-quality data analysis and reasoning effects.

Description

System and method for processing confidential data using online large language model

The present invention relates to a data processing method, and more particularly to a system and method for processing confidential data using an online large language model.

With the development of big data and artificial intelligence technology, large language models such as GPT-4 are already capable of high-quality data analysis and reasoning. However, due to the need to protect confidential data, directly sending confidential data to online large models for analysis may leak confidential data. Therefore, how to protect confidential data while utilizing the powerful analysis and reasoning capabilities of online large language models is an important topic in current technological development.

In view of this, the present invention proposes a system and method for processing confidential data using an online large language model in order to address the above-mentioned deficiencies in the prior art and future needs. The specific architecture and implementation methods are described in detail below:

The main purpose of the present invention is to provide a system and method for processing confidential data using an online large language model, which uses the online large language model to perform data analysis or reasoning on the data, and generates program codes or guiding principles that can be used by a closed large language model, so that the closed large language model can analyze and reason on the confidential data according to the program codes or guiding principles, thereby improving the ability of the closed large language model.

Another purpose of the present invention is to provide a system and method for processing confidential data using an online big language model, which can remove sensitive information from confidential data and then provide it to the online big language model for analysis, or generate simulated data for the online big language model, thereby obtaining high-quality data analysis results while preventing the leakage of sensitive information in the confidential data.

To achieve the above-mentioned purpose, the present invention provides a system for processing confidential data using an online large language model, including: a data privacy enhancement module, generating a plurality of operation data; an operation module, connected to the data privacy enhancement module, the operation module including an online large language model, the operation module receiving the operation data and inputting the operation data into the online large language model for analysis or reasoning to generate a program code or a guiding principle; a packaging module, connected to the operation module, packaging the program code or guiding principle into a data package, the data package is in a form that can be run by a closed large language model in a closed intranet; and a transmission module, connected to the packaging module, transmitting the data package to the closed intranet.

According to an embodiment of the present invention, the data privacy enhancement module includes a desensitization data processing module for removing sensitive information from confidential data of multiple customers to serve as computing data.

According to an embodiment of the present invention, the data privacy enhancement module includes a simulation data generation module for generating simulation data of a plurality of simulated clients and using the simulation data as calculation data.

According to an embodiment of the present invention, the program code generated by the online large language model is used for the closed large language model to perform data analysis.

According to an embodiment of the present invention, the guiding principle generated by the online big language model is to design the title engineering for various documents.

According to an embodiment of the present invention, the packaging module packages the program code or guiding principles into a docker container form, a compressed package form, a database form, or a json or csv file format.

According to an embodiment of the present invention, the transmission module is configured to use USB, hard disk, temporarily enabled network permission or encrypted one-way network to transmit data packets.

A method for processing confidential data using an online large language model includes the following steps: generating a plurality of computational data; inputting the computational data into an online large language model for analysis or reasoning to generate a program code or a guiding principle; packaging the program code or guiding principle into a data packet in a form that can be run by a closed large language model in a closed intranet; and transmitting the data packet to the closed intranet.

The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

It should be understood that when used in this specification and the appended patent applications, the terms "include" and "comprising" indicate the presence of described features, wholes, steps, operations, elements and/or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and/or combinations thereof.

It should also be understood that the terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit the present invention. As used in this specification and the appended patent applications, the singular forms "a", "an" and "the" are intended to include plural forms unless the context clearly indicates otherwise.

It should be further understood that the term "and/or" used in this specification and the appended patent applications refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

The present invention provides a system and method for processing confidential data using an online large language model, the purpose of which is to use the powerful analysis and reasoning capabilities of the online large language model to improve the capabilities of the closed large language model. When there is a piece of data that needs to be processed, but the data contains confidential information that cannot be leaked, or the data itself is confidential data and contains sensitive information, this data can only be analyzed and reasoned using a closed large language model on the intranet. However, the analysis and reasoning capabilities of the closed large language model are not strong enough, and are only at the level of elementary school students, while the analysis capabilities of the online large language model are approximately equivalent to the level of a postdoctoral student. Therefore, the present invention uses the online large language model to first analyze and reason a set of computational data that does not contain confidential information, obtain the program code or guidelines (Standard Operation Procedure, SOP), and then provide the program code or guidelines to the closed large language model for use. In this way, the ability of the closed large language model can be improved by using the online large language model. The system architecture and step flow of the present invention are described in detail as follows.

Please refer to Figure 1, which is a block diagram of the system 10 for processing confidential data using an online large language model of the present invention. The system 10 for processing confidential data using an online large language model of the present invention includes a data privacy enhancement module 12, a computing module 14, a packaging module 16, and a transmission module 18. The data privacy enhancement module 12 is connected to the computing module 14, the computing module 14 is connected to the packaging module 16, the packaging module 16 is connected to the transmission module 18, and the transmission module 18 is connected to a closed intranet 20 via the Internet.

The data privacy enhancement module 12 is used to generate a plurality of computing data. Specifically, the data privacy enhancement module 12 includes a desensitization data processing module 122 and a simulation data generation module 124. The desensitization data processing module 122 is used to remove sensitive information from confidential data of a plurality of customers to use as computing data. The simulation data generation module 124 is used to generate simulation data of a plurality of simulated customers and use the simulation data as computing data. Therefore, the present invention has two ways to generate computing data, one is to use the original confidential data but remove the sensitive information in the confidential data; the other is to generate virtual data as computing data.

The computing module 14 includes an online large language model 142, which has powerful analysis and reasoning capabilities, such as ChatGPT's fourth generation GPT-4. The computing module 14 receives computing data and inputs the computing data into the online large language model 142 for analysis or reasoning to generate a program code or a guiding principle. The program code can be a pseudocode for a closed large language model to perform data analysis, and the guiding principle is the design of inscription engineering for various documents. In other words, the program code and the guiding principle are both processes similar to step outlines, which allow the closed large language model to follow the steps of the program code or guiding principle to analyze and reason on other data.

The packaging module 16 is used to package the program code or guidelines into a data package in a form that can be run by a closed language model in a closed intranet. For example, the packaging module 16 can package the program code or guidelines into a docker container form, a compressed package form, a database form, or a json or csv file format.

The delivery module 18 is used to deliver the data packets packaged by the packaging module 16 to the closed intranet 20 for use by the closed large language model 202 in the closed intranet 20. In the embodiment of the present invention, the delivery module 18 can use USB, hard disk, temporarily opened network permissions or encrypted one-way network to deliver data packets.

Please also refer to FIG. 2, which is a flow chart of the method of processing confidential data using an online large language model of the present invention. In step S10, the data privacy enhancement module 12 generates a plurality of operation data. Then in step S12, the operation data is input into the online large language model 142 of the operation module 14, and the operation data is analyzed or inferred to generate a program code or a guiding principle. Then, as described in step S14, the packaging module 16 packages the program code or guiding principle into a data package in a form that can be run by a closed large language model in a closed intranet. Finally, as described in step S16, the transmission module 18 transmits the data packet to the closed intranet 20, and the closed large language model 202 analyzes and infers the real confidential data according to the code or guiding principles. In this way, the closed large language model 202 can obtain high-quality data analysis and inference effects with the help of the capabilities of the online large language model 142.

The desensitization data processing module 122 in the data privacy enhancement module 12 uses keywords to remove sensitive information. For example, multiple sets of sensitive word libraries are set first, such as personal information such as ID number, phone number, or privacy such as bank account number and amount, which will be set as sensitive words and stored in the sensitive word library. The desensitization data processing module 122 will delete the sensitive information in the confidential data, or mask it with mosaics or other forms.

The simulation data generation module 124 in the data privacy enhancement module 12 automatically generates simulation data using artificial intelligence after the user inputs the requirements. The artificial intelligence can be GPT-4. For example, assuming that the input requirement is "please generate 10 rows of virtual data of air pollution, including five fields of device_id, lat, lon, pm2_5, and time", the simulation data generation module 124 can generate simulation data 126 as shown in Figure 3.

The confidential data and simulation data 126 with sensitive information removed are used as computational data for the online big language model 142 to generate code or guiding principles. The guiding principles are the design of the title engineering for various documents. For example, assuming that there is a proposition principle related to air pollution data analysis, input "Please provide a title template for air pollution data analysis so that the closed big language model can make more exciting data analysis propositions based on these templates", then the online big language model 142 can generate the guiding principles in the following table: 1 Air pollution trend analysis: Analyze air pollution data from the past five years to find out if there are any significant trends or patterns. For example, is pollution particularly severe during a certain season, time of day, or weather conditions? You can use time series analysis to answer this question. 2 Analysis of the relationship between air pollution and population health: Use public health data (e.g., incidence of respiratory diseases) to explore the relationship between air pollution and population health. This may require the use of regression analysis or other statistical methods to determine the relationship between variables. 3 Identification of air pollution sources: Try to identify potential sources that may be causing air pollution, such as factory emissions, vehicle emissions, or agricultural activities. This may require factor analysis or cluster analysis. 4 Air pollution prediction model: Use historical data to build a prediction model to predict future air pollution levels. This can use machine learning prediction models such as decision trees, random forests, or deep learning. 5 Evaluation of the effectiveness of air pollution policies: Evaluating how a particular policy or intervention affects air pollution levels. This may require the use of controlled experiments or regional analyses. Table I

Please refer to Figure 4 for the code. Assuming that the data in Figure 3 is input into the online big language model, and the online big language model is required to "use python to calculate the average value, and this phphone should be packaged into docker and placed in the closed intranet for the closed big language model to call and execute, and package it in a function as much as possible", the online big language model can generate the code 128 shown in Figure 4. The user can also make further requirements to the online big language model in order to obtain the most suitable code for the closed big language model. For example, if the online big language model is further required to "please do not put the simulation data in the code, because I want to dynamically read the .csv data file", the online big language model can rewrite the code and output the code 128' shown in Figure 5.

The guiding principles in Table 1 and the program code 128' in FIG. 5 can be packaged into a data package that can be transmitted to the closed large language model 202 and directly executed through the packaging module 16, such as the docker container format required by FIG. 4 or the .csv file format required by FIG. 5. Finally, it is transmitted to the closed large language model 202 in the closed intranet 20 through the delivery module 18, and the program code 128' or the guiding principles can be used for analysis and reasoning by directly executing the data package.

In summary, the present invention provides a system and method for processing confidential data using an online big language model, which uses the online big language model to perform data analysis or reasoning on the data, and generates a program code or guiding principle that can be used by the closed big language model, so that the closed big language model can analyze and reason on the confidential data according to the program code or guiding principle to improve the ability of the closed big language model. Therefore, although the ability of the closed big language model is relatively weak, by following the program code or guiding principle provided by the online big language model to analyze and reason data, high-quality inference results can still be obtained, and confidential data leakage can be avoided to ensure information security.

However, the above is only the preferred embodiment of the present invention and is not intended to limit the scope of the present invention. Therefore, all equivalent changes or modifications based on the features and spirit described in the scope of the present invention should be included in the scope of the patent application of the present invention.

10: A system for processing confidential data using online large language models

12: Data privacy enhancement module

122: Desensitizing data processing module

124:Simulation data generation module

126:Simulation data

128, 128’: Program code

14: Computational Module

142: Online Big Language Model

16: Packaging Module

18: Delivery module

20: Closed Intranet

202: Closed Large Language Model

Figure 1 is a block diagram of the system of the present invention for processing confidential data using an online large language model. Figure 2 is a flow chart of the method of the present invention for processing confidential data using an online large language model. Figure 3 is a schematic diagram of the simulation data generated by the simulation data generation module of the present invention. Figures 4 and 5 are schematic diagrams of the program code generated by the present invention using an online large language model.

10: A system for processing confidential data using an online large language model

12: Data privacy enhancement module

122: Desensitizing data processing module

124:Simulation data generation module

14: Computation module

142: Online Big Language Model

16: Packaging module

18: Delivery module

20: Close the intranet

202: Closed Large Language Model

Claims (10)

A system for processing confidential data using an online large language model includes: a data privacy enhancement module, which generates a plurality of operation data, wherein the data privacy enhancement module includes a simulation data generation module, which is used to generate simulation data of a plurality of simulated customers and use the simulation data as the operation data; an operation module, which is connected to the data privacy enhancement module, and the operation module includes an online large language model, and the operation module receives the and other computing data and inputting the computing data into the online big language model for analysis or reasoning to generate a program code or a guiding principle; a packaging module connected to the computing module to package the program code or the guiding principle into a data packet in a form that can be run by a closed big language model in a closed intranet; and a transmission module connected to the packaging module to transmit the data packet to the closed intranet. A system for processing confidential data using an online large language model as described in claim 1, wherein the data privacy enhancement module includes a desensitization data processing module for removing sensitive information from confidential data of multiple customers to serve as the computational data. A system for processing confidential data using an online large language model as described in claim 1, wherein the program code generated by the online large language model is used for the closed large language model to perform data analysis. A system for processing confidential data using an online large language model as described in claim 1, wherein the guiding principle generated by the online large language model is the design of title engineering for various documents. A system for processing confidential data using an online large language model as described in claim 1, wherein the packaging module packages the program code or the guiding principle into a docker container form, a compressed package form, a database form, or a json or csv file format. A system for processing confidential data using an online large language model as described in claim 1, wherein the transmission module is configured to use a USB, a hard drive, a temporarily enabled network permission, or an encrypted one-way network to transmit the data packet. A method for processing confidential data using an online large language model includes the following steps: generating simulation data of multiple simulated customers and configuring the simulation data into multiple computational data; inputting the computational data into an online large language model for analysis or reasoning to generate a program code or a guiding principle; packaging the program code or the guiding principle into a data packet in a form that can be run by a closed large language model in a closed intranet; and transmitting the data packet to the closed intranet. A method for processing confidential data using an online large language model as described in claim 7, wherein the program code generated by the online large language model is used for the closed large language model to perform data analysis. A method for processing confidential data using an online large language model as described in claim 7, wherein the guiding principle generated by the online large language model is the design of title engineering for various documents. A method for processing confidential data using an online large language model as described in claim 7, wherein the program code or the guiding principle is packaged in a docker container format, a compressed package format, a database format, or a json or csv file format.